{ "papers": { "40522761": { "pmid": 40522761, "pmcid": "PMC12296512", "title": "", "journal": "", "date": "", "sections": [ "Introduction", "Search Strategy", "Aging and Neurodegenerative Diseases", "Genetics and Neurodegenerative Diseases", "Progressive Neuron Dysfunction/Death and Neurodegenerative Diseases", "Apoptosis and Neurodegenerative Diseases", "Protein Misfolding, Aggregation, Conformation, and Neurodegenerative Diseases", "Propagation of Protein Aggregation and Neurodegenerative Diseases", "Neuroinflammation and Neurodegenerative Diseases", "Mitochondrial Dysfunctions and Neurodegenerative Diseases", "Axon Transportation Defect and Neurodegenerative Diseases", "Gut Microbiota/Dysbiosis and Neurodegenerative Diseases", "Clinical Translation Based on Pathogenic Mechanisms", "Limitations", "Discussion and Conclusions" ], "chunk_count": 83 }, "41295902": { "pmid": 41295902, "pmcid": "PMC12847458", "title": "", "journal": "", "date": "", "sections": [ "Introduction", "Ongoing clinical trials", "SKY-0515 (", "SPK-10001 (", "Breaking news" ], "chunk_count": 27 }, "41827217": { "pmid": 41827217, "pmcid": "PMC12986383", "title": "", "journal": "", "date": "", "sections": [ "1. Introduction", "2. Typical Ex Vivo Cell Therapies, Excluding CAR-T", "3. Gene Therapies Without the Need for Ex Vivo Cell Modification", "4. CAR-T Cell Immunotherapies (Including CAR-NK and CAR-NKT)\u2014In Vivo and Ex Vivo Approaches, Genetic Modifications, and Cell Selection", "5. In Vivo CAR-T (CAR-X) Therapies", "6. Comparative Evaluation: Ex Vivo and In Vivo Gene Modification Strategies", "6.1. Vector Safety: Ex Vivo vs. In Vivo Approaches", "6.2. Durability of Expression: Ex Vivo vs. In Vivo", "6.3. Tissue and Cellular Tropism", "6.4. Tunability of Expression: Ex Vivo vs. In Vivo", "7. Summary" ], "chunk_count": 53 }, "41783572": { "pmid": 41783572, "pmcid": "PMC12955603", "title": "", "journal": "", "date": "", "sections": [ "Introduction", "Systemic Copper Metabolism", "Cellular Copper Metabolism", "Uptake", "Utilization", "Efflux", "Copper Homeostasis Dysregulation", "Copper Homeostasis and Oxidative Stress", "Copper Homeostasis and Inflammatory Response", "Copper Homeostasis and Mitochondrial Dysfunction", "Copper Homeostasis and Synaptic Function", "Mechanism of Cuproptosis", "Mitochondrial Lipoylated Protein Aggregation", "Loss of Iron-Sulfur Cluster Proteins", "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "Alzheimer\u2019s Disease", "Parkinson\u2019s Disease", "Huntington\u2019s Disease", "Amyotrophic Lateral Sclerosis", "Multiple Sclerosis", "Wilson Disease", "Menkes Disease", "Stroke", "Drugs Related to Neurological Diseases", "Copper Chelators", "D-Penicillamine", "Triethylenetetramine", "Tetrathiomolybdate", "Clioquinol", "Diacetyl-Bis(4-Methylthiosemicarbazonato)copper(II)", "Bis-Choline Tetrathiomolybdate", "Tetradentate Monoquinolines", "Copper Ion Carriers", "Elesclomol", "5,7-Dichloro-2-[(Dimethylamino)methyl]-8-Hydroxyquinoline", "Copper Nanoparticles", "Natural Compounds", "Conclusion and Future Perspectives" ], "chunk_count": 143 }, "41770933": { "pmid": 41770933, "pmcid": "PMC12974472", "title": "", "journal": "", "date": "", "sections": [ "Results", "LIG1 K845N Shows Reduced Ligase Activity In Vitro That Is Sensitive to the Ligation Context, Particularly Promutagenic 3\u2019 Mismatches.", "K845N Selectively Aborts Ligation of an 8-oxoG:A Mismatch to Enhance the Fidelity of Ligation.", "Cells Harboring the LIG1 K845N Variant Exhibit Increased Viability and Decreased Genomic DNA Mutation under Oxidative Stress.", "The Mouse K845N Orthologous Variant Suppresses Somatic CAG Expansion in ", "Discussion", "Materials and Methods", "Protein Expression and Purification.", "Preparation of DNA Substrates.", "Active Site Titration.", "Gel-Based Ligation Assays for Full-Length LIG1.", "Steady-State Kinetics with \u2206232 LIG1.", "HD Patient and Control LCLs.", "Cell Viability Assay and Duplex Sequencing.", "Generation of ", "Mouse Crosses with ", "Assaying ", "Statistical Analyses." ], "chunk_count": 68 }, "41639620": { "pmid": 41639620, "pmcid": "PMC12964631", "title": "", "journal": "", "date": "", "sections": [ "Introduction", "Background", "Current landscape of computational drug repurposing", "Clinical approaches to drug repurposing", "Bio-semantics-based computational approaches to drug repurposing", "Computational tools for constructing biological knowledge graphs", "Semi-automated pipeline for drug repurposing in biological networks", "Algorithm for semantic generation of negative samples", "Automating drug repurposing pipelines", "Node embeddings via Node2Vec", "Graph-based ML prediction of network associations", "Parallelisation in computing as a processing solution for big data", "Huntington\u2019s disease as a model rare disease for drug repurposing", "Results", "Pipeline upgrades", "Database APIs for biological, gene-drug associations and drug-drug similarity", "Generation of semantically-valid negative samples based on centrality measures", "Parallelisation in gene-drug embeddings", "ML training and prediction in enriched KG", "Prediction of repurposed drugs in binary models with non-binary datasets", "Prediction iterations as a solution for seed variability in ML methods", "Methods", "Pipeline upgrades", "Database APIs for biological, gene-drug associations and drug-drug similarity", "Generation of semantically-valid negative samples based on centrality measures", "Parallelisation in gene-drug embeddings", "ML training and prediction in enriched KG", "Prediction of repurposed drugs in binary models with non-binary datasets", "Prediction iterations as a solution for seed variability in ML methods", "Discussion", "Achievements", "Comparison to current state-of-the-art drug repurposing methods", "Limitations", "Contributions to drug repurposing", "Next steps", "Conclusion" ], "chunk_count": 230 }, "41492395": { "pmid": 41492395, "pmcid": "PMC12765053", "title": "", "journal": "", "date": "", "sections": [ "Introduction", "Methods", "Participants", "Cognitive assessment", "Neuroimaging acquisition and preprocessing", "Biosample collection and processing", "Statistical analysis", "Results", "Sociodemographic and clinical characteristics", "Neuroanatomical correlates of PD-CRS", "Associations between regional gray matter volumes and PD-CRS subdomains", "Association between plasma neurofilament levels and global cognitive function (PD-CRS)", "Association between plasma neurofilament levels and regional gray matter volume", "Discussion", "CRediT authorship contribution statement", "Ethical compliance statement", "Declaration of competing interest" ], "chunk_count": 54 }, "41828602": { "pmid": 41828602, "pmcid": "PMC12985888", "title": "", "journal": "", "date": "", "sections": [ "1. Introduction", "2. Results", "2.1. Validation of Striatal Atrophy", "2.2. Divergent Proteomic Signatures", "2.3. Independent Contributions to Atrophy", "2.4. Mediation Analysis", "3. Discussion", "3.1. Limitations", "3.2. Future Directions", "4. Materials and Methods", "4.1. Participants", "4.2. Neuroimaging", "4.3. CSF Collection and Handling", "4.4. Proteomics", "4.5. Statistical Analysis and Biomarker Screening", "4.6. Generative Artificial Intelligence", "5. Conclusions" ], "chunk_count": 42 }, "41674784": { "pmid": 41674784, "pmcid": "PMC12887895", "title": "", "journal": "", "date": "", "sections": [ "Introduction", "CRISPR\u2013Cas system", "An adaptive immune system of bacteria and archaea", "CRISPR-Cas systems and classes", "Genome editing tools: CRISPR-Cas9 and beyond", "Application of CRISPR\u2013Cas in neurodegenerative disorders", "CRISPR\u2013Cas in Alzheimer's disease", "CRISPR/Cas in Parkinson's disease", "CRISPR-Cas in Huntington's disease", "CRISPR-Cas in amyotrophic lateral sclerosis", "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "Delivery mechanisms", "Challenge", "Potential strategy", "Off-target effects", "Challenge", "Potential strategy", "Immune response", "Challenge", "Potential strategy", "Ethical and regulatory challenges", "Challenge", "Potential strategy", "CRISPR-based therapeutics for neurological disorders in clinical trials", "Future directions and emerging trends", "Treatment of genetic disorders", "Future applications", "Current findings to potential clinical translation", "Challenges", "Conclusion" ], "chunk_count": 215 }, "41708002": { "pmid": 41708002, "pmcid": "PMC12993198", "title": "", "journal": "", "date": "", "sections": [ "Results", "A FRET-based reporter of polyQ protein aggregation in the nucleus", "A CRISPRi screen identifies DNAJC7 as a suppressor of polyQ aggregation", "DNAJC7 suppresses mutant HTT exon 1 aggregation and colocalizes with inclusions in cells", "An FRET-based reporter for polyG aggregation reveals detergent-resistant, p62-positive nuclear inclusions", "PolyQ and polyG proteins aggregate independently in a homotypic repeat\u2013dependent manner", "Chaperone screening of polyG aggregation shows few modifiers, but polyG aggregation is suppressed by DNAJC7 overexpression", "Discussion", "Experimental procedures", "Animals", "Plasmid construction and lentivirus packaging", "Cell culture and cell line generation", "Transfections, fluorescence imaging, and immunofluorescence", "Primary CRISPRi screen and analysis", "Secondary assays based on flow cytometry", "Transfecting cell or tissue homogenates for seeding aggregation", "Immunoprecipitation and Western blot", "Graphing and statistical analysis", "Data availability", "Conflict of interest" ], "chunk_count": 80 }, "41849610": { "pmid": 41849610, "pmcid": "PMC12998519", "title": "", "journal": "", "date": "", "sections": [ "INTRODUCTION", "RESULTS", "CRISPR-Cas9 design and validation of the cleavage of human ", "SaCas9-HTTg1\u2013mediated human mHTT disruption in BAC226Q mice", "Robust and long-lasting rescue of motor deficits, weight loss, and life span by CRISPR-Cas9\u2013mediated m", "Rescue effects of gene editing at different pathogenic time points", "Highly effective and sequential deletion of mHTT and Cas9 by self-inactivating CRISPR-Cas9 rescued deficits in motor function and life span", "On and off-target effects of conventional and self-inactivating CRISPR-Cas9", "DISCUSSION", "MATERIALS AND METHODS", "Experimental design", "Molecular cloning", "Culture and transfection of HEK293T cell", "Culture and nucleofection of human iPSC", "Mice", "Western blot", "Immunofluorescent staining", "Immunohistochemistry", "AAV vector production", "Stereotaxic injection procedure", "Mouse behavioral analysis", "\nRotarod test\n", "\nOpen field\n", "\nCatwalk\n", "\nBody weight\n", "\nSurvival\n", "Primer extension\u2013mediated sequencing", "Sanger sequencing", "RNA-seq and data analysis", "Statistical analysis" ], "chunk_count": 97 }, "41618467": { "pmid": 41618467, "pmcid": "PMC12895692", "title": "", "journal": "", "date": "", "sections": [ "Introduction", "Circular RNAs (circRNAs): biogenesis and multimodal functions", "miRNA sponging", "Sponging and protein scaffolding", "Transcriptional regulation and splicing", "Cap-independent translation", "Enhancer RNAs (eRNAs): biogenesis and transcriptional roles", "Facilitation of enhancer-promoter looping", "Recruitment of transcriptional co-activators", "Modulation of chromatin accessibility", "Release of paused RNA polymerase II", "Involvement of circRNAs and eRNAs in the brain", "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "Alzheimer's disease (AD)", "CircRNAs in AD", "eRNAs in AD", "Clinical implications of CircRNAs and eRNAS in AD", "Parkinson\u2019s disease (PD)", "CircRNAs in PD", "eRNAs in PD", "Clinical implications circRNAs and eRNAs in PD", "Huntington\u2019s disease (HD)", "circRNAs in HD", "eRNAs in HD", "Clinical implications of circRNAs and eRNAs in HD", "Multiple sclerosis (MS)", "CircRNAs in MS", "eRNAs in MS", "Clinical implications of circRNAs and eRNAs in MS", "Converging subjects and integrated regulatory networks", "Cross-disease patterns of circRNA/eRNA dysregulation", "Synaptic plasticity and neuronal homeostasis", "circRNAs:", "eRNAs:", "Convergence:", "Neuroinflammatory cascades", "circRNAs:", "eRNAs:", "Cross-disease paradigm:", "Protein quality control systems", "Autophagy-lysosomal pathways:", "Ubiquitin-proteasome system:", "eRNA contributions:", "Emerging evidence of crosstalk:", "Transcriptional-posttranscriptional interface in neurodegeneration", "Beyond miRNA sponging: expanding the functional paradigm", "Protein splicing and scaffolding", "Translation regulation", "RBP coordination and phase separation", "eRNA functional diversity", "Challenges and future directions", "Technical and mechanistic challenges", "Detection and quantification:", "Cell-type specificity:", "Cause vs. consequence:", "Biomarker development", "Therapeutic potential", "Delivery challenges:", "Precision targeting:", "Emerging opportunities", "Multi-omics integration:", "CircRNA/protein interactions:", "Humanized models:", "Conclusion and perspective", "Electronic supplementary material" ], "chunk_count": 198 }, "41689141": { "pmid": 41689141, "pmcid": "PMC12903234", "title": "", "journal": "", "date": "", "sections": [ "Introduction", "mtDNA structure, maintenance, and vulnerability", "Mechanisms of mtDNA release", "Mitochondrial permeability transition pore (mPTP) opening and membrane rupture", "Dysfunctional mitophagy and impaired mitochondrial quality control", "Apoptosis and BAX/BAK-mediated outer membrane permeabilization", "Necrosis, necroptosis, and pyroptosis, inflammatory cell death modalities", "Extracellular vesicle (EV)-mediated secretion of mtDNA", "mtDNA: sensing and immune activation", "cGAS\u2013STING pathway\u2014cytosolic DNA sensing", "NLRP3 inflammasome activation\u2014cytosolic danger sensing", "TLR9\u2014endosomal sensing of CpG DNA", "Other potential mtDNA sensing pathways", "Neuroinflammation and mtDNA", "AD", "PD", "HD", "Therapeutic strategies targeting mtDNA-driven neuroinflammation", "Reducing mtDNA release", "Blocking downstream sensing pathways", "Promoting mtDNA repair", "Modulating the metabolic\u2013immune interface", "Challenges and future perspectives", "Conclusion" ], "chunk_count": 87 }, "41699331": { "pmid": 41699331, "pmcid": "PMC12909434", "title": "", "journal": "", "date": "", "sections": [ "Introduction", "Involvement of non-canonical cell death mechanisms in neurodegenerative disease", "Untitled", "The core mechanisms of necroptosis and its implications in neurodegeneration", "The core mechanisms of pyroptosis and its implications in neurodegeneration", "The core mechanisms of parthanatos and its implications in neurodegeneration", "Disease focused summary of non-canonical cell death pathways in major neurodegenerative disorders", "Alzheimer\u2019s disease", "Parkinson\u2019s disease", "Huntington\u2019s disease", "Amyotrophic lateral sclerosis", "Comprehensive biomarkers of non-canonical cell death pathways in neurodegeneration", "Ferroptosis biomarkers", "Necroptosis biomarkers", "Pyroptosis biomarkers", "Parthanatos biomarkers", "Untitled", "Emerging therapeutic strategies and pharmacological agents for non-canonical cell death mechanisms", "Integrative framework and future perspectives", "Conclusion" ], "chunk_count": 74 }, "41693955": { "pmid": 41693955, "pmcid": "PMC12893954", "title": "", "journal": "", "date": "", "sections": [ "Introduction", "Nutrition and neuronal metabolism", "Effects of nutrients on neuronal metabolism", "Nutrient metabolism and neuronal function", "Abnormal nutrient metabolism and neuronal injury", "Nutrition and neurodegenerative diseases", "AD", "PD", "HD", "Others", "Cell metabolic reprogramming and neurodegenerative diseases", "Concept and mechanism of metabolic reprogramming", "Regulation of metabolic reprogramming", "Nutritional interventions and therapeutic strategies", "Emerging interventions integrate dietary modification, probiotics, and pharmacologic agents", "Clinical trials and future directions", "Discussion", "Conclusion" ], "chunk_count": 75 } }, "chunks": [ { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Introduction", "text": "Neurodegenerative diseases (NDs), such as Alzheimer\u2019s disease (AD), Parkinson\u2019s disease (PD), amyotrophic lateral sclerosis (ALS), Huntington\u2019s disease (HD), frontotemporal lobar dementia (FTD), and Lewy body disease (LBD) are debilitating disorders that are gradually increasing in prevalence in our aging society. In recent decades, advancements in research methodologies have yielded novel insights into both the phenotypic manifestations and molecular underpinnings of these common NDs. Studies have identified several pathogeneses that are shared among these diseases (Ross and Poirier, 2004; Frost and Diamond, 2010; Amor et al., 2014; Soto and Pritzkow, 2018; Vaquer-Alicea and Diamond, 2019; Wells et al., 2019). For example, many of these NDs exhibit overlapping gene variants and share pathogenic proteins. Specific toxic proteins associated with the diseases include extracellular amyloid-beta (A\u03b2) and phosphorylated tau (pTau), which are linked to AD (Rajendran and Krishnan, 2024; Pradeepkiran et al., 2024); \u03b1-synuclein (\u03b1-syn, SNCA) in PD (Mehra et al., 2019); Cu/Zn superoxide dismutase (SOD1) (Kaur et al., 2016), TAR DNA-binding protein 43 (TDP-43) (Baralle et al., 2013), fused in sarcoma (FUS) (Mackenzie et al., 2010), and chromosome 9 open-reading frame 72 (C9ORF72) (Geng and Cai, 2024) in ALS; pTau, C9ORF7 (Balendra and Isaacs, 2018), and progranulin (Petkau and Leavitt, 2014) in FTD; huntingtin (HTT) (Zheng and Diamond, 2012) in HD; and \u03b1-syn (Noguchi-Shinohara and Ono, 2023) in LBD. These proteins serve as classic pathological hallmarks of their respective diseases. The underlying genetic mutations associated with the pathogenic proteins in these common NDs either dictate or contribute to the progression of these disorders. Therefore, a clear understanding of the relationship between genetic mutations and NDs is crucial for uncovering the disease-specific mechanisms as well as potential common pathways. Although each ND has its own unique pathogenesis, current studies have demonstrated that there are also shared pathogenic mechanisms across different NDs (Ross and Poirier, 2004; Frost and Diamond, 2010; Amor et al., 2014; Soto and Pritzkow, 2018; Vaquer-Alicea and Diamond, 2019; Wells et al., 2019). Based on these findings, a better understanding of the overlapping and interconnected pathogeneses among common NDs may reveal novel mechanisms or therapeutic targets for specific NDs or for NDs as a whole.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Introduction", "text": "Current investigations into the shared or interactive pathogeneses among different NDs are primarily focused on disease-candidate genes, molecular pathways, and gene networks, aiming to provide support for the existence of common molecular mechanisms (Arneson et al., 2018; Soto and Pritzkow, 2018; Liao et al., 2022; Vasconcelos et al., 2023). Associated studies have found valuable evidence for shared molecular dysfunctions and disorders with the pathogeneses of multiple NDs. However, these findings require further validation through extensive studies on the common pathogeneses of these diseases.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Introduction", "text": "For this review paper, we systematically and multi-dimensionally retrieved and analyzed research information on the pathogenesis of six NDs at the genetic, transcriptomic, molecular pathway and gene network levels. We focused on molecular pathogenesis with the aim of providing more objective and comprehensive insights into these common NDs, summarizing the shared molecular mechanisms identified among them to draw conclusive insights. Our review of the similarities among the potential pathogeneses of these six NDs is organized based on current investigations, and the shared factors associated with genetics, gene mutations, pathways, and networks are discussed. Our objective is to provide an overview of the existing research on pathogenesis, linking recent findings related to the six NDs to identify promising common mechanisms. The biological processes underlying neurodegeneration in various NDs ultimately lead to neural cell death, including that of neurons, which contributes to the development of these diseases. We believe that damaged neural cells can be prevented from dying or recovered through interventions that halt or reverse these related pathogenic processes, particularly those that target multiple pathways simultaneously. Understanding the exact pathogeneses of these diseases is crucial for developing effective treatments. To this end, we sought to identify potential common pathogeneses among the NDs by reviewing recent novel findings, offering clues for further research into more precise and valuable pathogenic mechanisms and optimal therapeutic strategies.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Search Strategy", "text": "In this review, we conducted a systematic literature retrieval on the pathogenesis of six common NDs using several databases, including ISI Processing, PubMed, ProQuest, OpenGray, Scopus, Web of Science, Cochrane Library, EmBase, Google Scholar, WanFang, and relevant medical guidelines. We used search terms such as \u201cneurodegenerative diseases,\u201d \u201cAlzheimer\u2019s disease,\u201d \u201cParkinson\u2019s disease,\u201d \u201camyotrophic lateral sclerosis,\u201d \u201cHuntington\u2019s disease,\u201d \u201cfrontotemporal lobar dementia,\u201d and \u201cLewy body disease,\u201d combined with \u201cpathogenesis.\u201d\n\nAdditionally, we refined key search terms to extensively explore the literature regarding the potential common pathogenesis of the six NDs. The retrieved literature primarily included review articles, major textbook chapters, abstracts, and reports from relevant meetings. We focused on literature published over nearly the past 30 years, up until the end of October 2024, while not excluding frequently referenced older studies with high citation counts. The final reference list was compiled based on both the originality and relevance of the literature to our review topic, as well as the eligibility of the studies, their quality, and their compatibility with the present review through evaluating and analyzing titles, abstracts, and full articles (Figure 1).\n\nTimeline showing the potential common pathogenesis of AD, PD, ALS, HD, FTD and LBD as documented in the literature.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Search Strategy", "text": "In this review, we analyzed literature from the past 30 years, spanning from 1997 to 2024. Although the major pathogenesis of NDs remains incompletely understood, researchers have identified numerous important mechanisms over the years, providing crucial clues and evidence for the clinical treatment of NDs. Key findings include the roles of excitotoxicity and mitochondrial dysfunction, which have contributed to the development of Edaravone and Riluzole for ALS treatment. Additionally, the discovery of SOD1 genetic mutations has opened new avenues for gene-modified therapeutics for NDs. Other important factors such as neuroinflammation, the toxic production and aggregation of specific pathogenic proteins, the failure of neural stem cells, and the disruption of gut microbiota homeostasis have also been recognized as contributing to the pathogenesis of NDs, creating new opportunities for novel treatment strategies. AD: Alzheimer\u2019s disease; ALS: amyotrophic lateral sclerosis; FTD: frontotemporal lobar dementia; HD: Huntington\u2019s disease; LBD: Lewy body disease; NDs: neurodegenerative diseases; PD: Parkinson\u2019s disease; SOD1: Cu/Zn superoxide dismutase.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Aging and Neurodegenerative Diseases", "text": "Aging is an inevitable biological process in the human life cycle that is characterized by the progressive physiological deterioration of various organs and an increased risk of multiple diseases, including NDs (Hou et al., 2019). Among the various conditions encountered during aging, NDs are the most prevalent in the older adult population (Hou et al., 2019). The mechanisms underlying neurodegeneration in these disorders are highly complex and not yet fully understood. Among the numerous risk factors associated with neurodegeneration in the pathogenesis of NDs, aging is considered one of the most significant (Hou et al., 2019).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Aging and Neurodegenerative Diseases", "text": "Aging is a crucial risk factor associated with several common NDs, such as AD (Sengoku, 2020), PD (Reeve et al., 2014), ALS (Niccoli et al., 2017), HD (Iskusnykh et al., 2024), FTD (Niccoli et al., 2017), and LBD (Walker et al., 2019). Nine biological hallmarks of aging are related to the development of NDs: genomic instability (Peze-Heidsieck et al., 2021), telomere attrition (Gao et al., 2022), epigenetic alterations (Ghosh and Saadat, 2023; la Torre et al., 2023), loss of proteostasis (Labbadia and Morimoto, 2015), mitochondrial dysfunction (Johnson et al., 2021; Klemmensen et al., 2024), cellular senescence (Lee et al., 2023; Wang et al., 2024), deregulated nutrient sensing (Guo et al., 2022), stem cell exhaustion (Jagaraj et al., 2024), and altered intercellular communication (Hill, 2019; Huo et al., 2021). These critical biological mechanisms of aging play significant roles in ND pathogenesis (L\u00f3pez-Ot\u00edn et al., 2013; Hou et al., 2019).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Aging and Neurodegenerative Diseases", "text": "The nine biological hallmarks of aging are typically categorized into primary, antagonistic, and integrative aging biomarkers. The primary biomarkers consist of genomic instability, epigenetic alterations, telomere attrition, and dysproteostasis. The antagonistic aging biomarkers encompass compensatory responses to the primary aging-related lesions and include mitochondrial dysfunction, cellular senescence, and downregulation of nutrient sensing. The integrative aging biomarkers result from the cumulative effects of the primary and antagonistic biomarkers and include stem cell exhaustion and altered intercellular communication (L\u00f3pez-Ot\u00edn et al., 2013).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Aging and Neurodegenerative Diseases", "text": "Among the aging biomarkers, cellular senescence, genomic instability, and dysproteostasis are closely associated with protein aggregation in various NDs, including AD and PD (Wilson et al., 2023). Neurodegeneration in these NDs is particularly linked to age-related changes, suggesting there are connections between NDs and aging-related alterations in the central nervous system (CNS) microenvironment, genomic instability, epigenetic modifications, and dysproteostasis. While aging is recognized to be one of the crucial risk factors for NDs, the precise mechanisms by which it contributes to neurodegeneration and whether age-related changes in the brain serve as early indicators of neurodegenerative processes remain unclear and require further investigation. Several recent molecular studies have reported that abnormal aggregations of \u03b1-syn (Li et al., 2023a, b), phosphorylated tau (Guarnieri et al., 2024), and A\u03b2 (Liu et al., 2024a, b) are related to aging (Sengupta and Kayed, 2022). However, there is currently no direct evidence that these pathogenic proteins of NDs actively participate in neurodegeneration during human aging.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Aging and Neurodegenerative Diseases", "text": "Nonetheless, some mechanisms linking aging and the pathogenesis of NDs have been made relatively clear. For example, CNS cells, such as neurons and glial cells, may be susceptible to age-dependent alterations such as DNA methylation. DNA methylation is an important epigenetic modification that occurs during aging (Unnikrishnan et al., 2019; Soto-Palma et al., 2022) and is involved in DNA methyltransferase-mediated DNA damage that can induce NDs (Mattson, 2003). Additionally, mitochondrial dysfunction, which is associated with aging, promotes neurodegeneration by generating reactive oxygen species (ROS) and enhancing neuroinflammation in NDs (Lin and Beal, 2006).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Aging and Neurodegenerative Diseases", "text": "Aging is a significant risk factor for AD. Age-related disorders include oxidative stress, DNA damage, and mitochondrial dysfunction. Moreover, a progressive accumulation of senescent cells leads to chronic inflammatory responses. This inflammation is accompanied by the accumulation of toxic proteins, resulting in tau aggregation and A\u03b2 deposition, which damage neuronal synapses and contribute to cognitive impairment in AD. These processes involve various forms of cellular damage, including DNA, protein, and lipid damage, as well as changes in DNA methylation, histone modification, and the dysregulation of non-coding RNAs. Additionally, mitochondrial dysfunction, deregulated nutrient sensing, and the senescence of neurons, astrocytes, microglia, oligodendrocytes, immune cells, and cerebrovascular endothelial cells all play roles in AD pathology. Other contributing factors include stem cell exhaustion, altered intercellular communication, chronic neuroinflammation, and dysbiosis (Liu et al., 2024a, b).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Aging and Neurodegenerative Diseases", "text": "A major pathological feature of PD is the formation of Lewy bodies, composed of the toxic protein \u03b1-syn, which result from mutations in the \u03b1-syn gene in dopaminergic neurons. This phenomenon is linked to various pathogenic mechanisms associated with aging and multiple NDs. These mechanisms include inflammatory pathway activation by \u03b1-syn in glial cells through interactions with immune cells. Natural aging impairs the cellular processes responsible for clearing toxic substances, leading to chronic neuroinflammatory responses and the accumulation of internal \u03b1-syn, which ultimately contributes to the death of dopaminergic neurons. The interplay between aging, \u03b1-syn pathogenicity, and \u03b1-syn interactions with specific types of neural cells in the CNS are indicators of the \u03b1-syn-driven pathogenesis of PD. This understanding may pave the way for novel therapeutic approaches for treating PD (Zhang et al., 2023).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Aging and Neurodegenerative Diseases", "text": "In patients with HD, aging primarily affects the morphology of the cerebellum, leading to alterations in cerebellar circuits predominantly involved in motor control. This results in a deficiency of synapses and dendrites, contributing to the pathogenesis of HD. Aging is additionally associated with the apoptosis of cerebellar cells that is driven by iron-dependent ferroptosis, oxidative stress, DNA damage, and alterations in DNA modifications. Other contributing factors include progressive mitochondrial dysfunction, the activation of microglia and oligodendrocytes, and neuroinflammation related to aging (Iskusnykh et al., 2024).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Genetics and Neurodegenerative Diseases", "text": "Genetic factors are recognized to be major contributors to the pathogenesis of neurodegeneration in NDs, either in the context of monogenic hereditary disorders or complex, multifactorial sporadic diseases (Pihlstr\u00f8m et al., 2017). Over the past 20 years, the identification of genes and risk loci associated with NDs has significantly advanced our understanding of the pathogenic mechanisms and pathways involved in these disorders. Through large-scale research efforts, new study designs and advancements in methodologies have been made that have rapidly expanded our knowledge of the genomic and genetic architectures underlying the pathogenesis of NDs (Pihlstr\u00f8m et al., 2017).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Genetics and Neurodegenerative Diseases", "text": "All NDs are associated with significant genetic phenotypes. Based on estimates, the genetic heritability of AD, PD, and ALS is approximately 58% (Gatz et al., 2006), 40% (Hamza and Payami, 2010), and 60% (Al-Chalabi et al., 2010), respectively. Rare genetic mutations associated with NDs display Mendelian hereditary forms. For example, the amyloid precursor protein (APP) (Orobets and Karamyshev, 2023) and presenilin genes (PSEN1 and PSEN2) (Lanoisel\u00e9e et al., 2017; Neuner et al., 2020; Andrade-Guerrero et al., 2023) are linked to AD, while mutations in SNCA (Stefanis, 2012), Parkin 2 (von Coelln et al., 2004), PTEN-induced putative kinase 1 (PINK1) (Cherian and Divya, 2020), microtubule-associated protein tau (MAPT) (Bekris et al., 2010), and leucine-rich repeat kinase 2 (LRRK2) (Xiong and Yu, 2024) are associated with PD. In ALS, genes such as cytosolic SOD1 (Kaur et al., 2016), alsin 2 (Finsterer and Burgunder, 2014), senataxin (Tsui et al., 2023), and vesicle-associated membrane protein B (VAPB) (Zhang et al., 2015) have been implicated as causative factors. However, the majority of NDs exhibit non-Mendelian hereditary forms and involve complex pathogenesis with numerous genetic variants. These complex genetic forms of NDs were difficult to detect and identify prior to the advent of genome-wide association studies (GWAS). However, with the increasing application of GWAS, several risk genes and loci associated with the pathogenesis of NDs have been identified. For instance, the apolipoprotein E locus has been shown to have a strong relationship with the onset of clinical AD and PD, although the specific alleles or loci may differ between these diseases (Li et al., 2017; Neuner et al., 2020). Additionally, HLA-DRB5 and MAPT have emerged as common potential genes for both AD and PD based on GWAS results (Zhou and Wang, 2017; Little et al., 2017; Kim, 2018; Pan et al., 2021), and the C9ORF72 repeat expansion has been identified as a shared genetic factor in both ALS and frontotemporal dementia (FTD) (Jiang et al., 2022). At the pathway level, vesicle-mediated transport is a common process involved in multiple NDs, including AD, PD, and ALS. Furthermore, nine pathways, including synaptic signaling, neuronal projection development, and proteolysis, represent common mechanisms in the pathogenesis of AD and PD (Ramanan and Saykin, 2013; Shang et al., 2015; Ahmad et al., 2017).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Genetics and Neurodegenerative Diseases", "text": "By applying current analytical strategies, numerous pathways have been identified that are common among prevalent NDs. These shared pathways involve various internal cellular processes, including apoptosis, autophagy, mitochondrial functions, oxidative damage, and proteasome activity. Additionally, they encompass partial tissue environments related to cellular adhesion, endocytosis, neurotransmission, and prionoid transmission factors. Systemic environments involving inflammation and immune responses, and lipid metabolism, as well as endocrine and vascular factors, have also been implicated as disease factors. Furthermore, biological processes related to development and aging\u2014such as epigenetics, neurotrophic factors, and telomere dynamics\u2014have been highlighted as being involved in NDs (Ahmad et al., 2017; Gan et al., 2018; Sanghai and Tranmer, 2023). From the overrepresented pathways identified via updated GWAS, it has been demonstrated that internal cellular processes\u2014including various cellular mechanisms, transportation, macromolecule complex binding, cellular activation, vesicle-mediated transport, proteolysis, and kinase binding\u2014are significant factors in common NDs. Additionally, pathways related to neural systems, such as synaptic signaling and neuronal projections, are also involved (Cooper-Knock et al., 2012). In a study examining gene interactions in AD and PD, a gene model was constructed using a transcription factor binding network curated through MetaCore software. This study revealed that several GWAS candidate genes for AD and PD are closely related to coherent networks coordinated by transcription factors SP1 and AP-1. Moreover, overlapping genes associated with ALS, PD, and AD were identified, particularly those related to neuroinflammatory gene signatures. These signatures were found to correlate with RNA splicing and protein turnover in ALS and PD. Mitochondrial dysfunction genes were also identified as being common among PD and AD cases (Cooper-Knock et al., 2012).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Genetics and Neurodegenerative Diseases", "text": "The first mutation in the SNCA gene was discovered in a subset of PD patients in 1997, highlighting the strong genetic link to PD pathogenesis. Currently, seven pathogenic mutations and/or potential risk genes associated with PD have been identified in patients. The established PD genes among these include SNCA, LRRK2, VPS35, RAB32, PRKN, PINK1, and PARK7/DJ-1. The pathogenic mutations contribute directly or indirectly to the aggregation and production of toxic SNCA protein or are closely associated with these processes (Lim and Klein, 2024).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Genetics and Neurodegenerative Diseases", "text": "Although most ALS cases are sporadic, a smaller proportion are familial (fALS). The hexanucleotide repeat expansion (GGGGCC) located in the first intron of the C9ORF72 gene is a common genetic cause of both fALS and sporadic ALS, underscoring the disease\u2019s strong genetic component. ALS primarily involves the degeneration and death of motor neurons; however, glial cells also play significant protective roles in the nervous system and contribute to the pathophysiological processes of ALS through non-cell-autonomous mechanisms. Microglia regulate phagocytosis, immune responses, neuroinflammation, and immune surveillance and activation. In addition, oligodendrocytes myelinate neuronal axons in the CNS to facilitate synaptic transmission and provide metabolic support to neurons. Pathogenic mutations in various ALS-related genes are known to lead to the activation of microglia and oligodendrocytes, triggering the pathogenesis of ALS through these mechanisms (Jagaraj et al., 2024).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Genetics and Neurodegenerative Diseases", "text": "FTD shares mutations of several common genes with ALS, including C9ORF72, TDP-43, sequestosome 1, FUS, VCP, CHCHD10, and tank-binding kinase 1. These genes are involved in a number of the mechanisms that contribute to the pathogenesis of FTD, such as protein misfolding, oxidative stress, axonal transport impairment, neuroinflammation, and immune responses. These processes are closely associated with neuroinflammatory and immune cells, including astrocytes, oligodendrocytes, microglia, and lymphocytes. Overall, cellular microenvironment processes closely linked to genetic disease factors\u2014particularly neuroinflammation and immune responses\u2014are suggested to be pivotal factors in the pathogenesis of FTD (De Marchi et al., 2023; Table 1).\n\nGenes and signaling pathways associated with the risk of neurodegenerative diseases\n\nAD: Alzheimer\u2019s disease; ALS: amyotrophic lateral sclerosis; APOE: apolipoprotein E; APP: amyloid precursor protein; A\u03b2: amyolid-beta; C9orf72: chromosome 9 open-reading frame 72; FTD: frontotemporal lobar dementia; GBA: glucocerebrosidase; HD: Huntington\u2019s disease; HTT: huntingtin; LBD: Lewy body disease; LRRK2: leucine-rich repeat kinase 2 or dardarin; MAPT: microtubule-associated protein tau; PD: Parkinson\u2019s disease; SNCA: \u03b1-synuclein; SOD1: Cu/Zn superoxide dismutase.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Progressive Neuron Dysfunction/Death and Neurodegenerative Diseases", "text": "Abnormal neuronal death induced by excessive programmed cell death (PCD) in specific regions of the CNS is a hallmark of the pathology associated with NDs such as AD, PD, ALS, and HD (Moujalled et al., 2021). The pathogenesis of NDs is multifactorial and complex, involving various defects in different cellular processes, including oxidative stress (Niedzielska et al., 2016), excitotoxicity (Doble, 1999), mitochondrial dysfunction (Klemmensen et al., 2024), protein misfolding (Sami et al., 2017), and inflammation (Stephenson et al., 2018). There is a considerable amount of evidence supporting the role of PCD in the pathogenesis of various NDs (Moujalled et al., 2021). However, a critical question remains: are the defects in PCD signaling and neuronal death key contributors to the insults that lead to NDs, or are they merely secondary responses? Additionally, how different PCD pathways and related processes interact to cause the demise of neuronal cells and other cell types in the context of NDs remains unclear.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Progressive Neuron Dysfunction/Death and Neurodegenerative Diseases", "text": "The mechanisms behind neuronal death in the CNS have long been challenging to decipher. Under normal physiological conditions, neuronal death in the adult brain is strictly regulated, even among the older adult population. However, excessive neuronal death in specific regions of the CNS represents a fundamental pathological feature of various NDs (Chi et al., 2018; Moujalled et al., 2021). In a healthy mature CNS, extensive neuronal death is rare; however, neuronal death tends to increase with age in certain brain regions. Although studies have detected differences in neuronal numbers between younger and older populations, the overall loss of neurons is still relatively limited. In contrast, significantly more neuronal death occurs in specific regions of the brain in individuals with various NDs than in those of age-matched controls, and this loss is correlated with the onset and progression of these diseases (Chi et al., 2018; Moujalled et al., 2021). Generally, mature neurons in the CNS exhibit a strong resistance to cell death compared to immature neurons. They maintain cellular homeostasis and longevity by managing various stresses. However, when faced with severe pathological environments, such as those present in NDs, this resistance can be overwhelmed, leading to neuronal death (Chi et al., 2018; Moujalled et al., 2021).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Progressive Neuron Dysfunction/Death and Neurodegenerative Diseases", "text": "Current studies on the pathogenesis of NDs are focused on identifying key pathological changes involved in the formation and aggregation of specific pathogenic proteins (Ross and Poirier, 2004; Soto and Pritzkow, 2018; Wilson et al., 2023). These pathogenic proteins create pathological stress and are cytotoxic to neurons, leading to excessive excitotoxicity and the generation of ROS. This process results in synaptic dysfunction, the impairment of protein degradation systems, endoplasmic reticulum stress, DNA damage, mitochondrial dysfunction, inflammation, and reentry into the cell cycle (Chi et al., 2018). Neuronal death in NDs poses significant challenges to understanding the neurodegenerative pathogenesis. While many mechanisms by which these pathogenic proteins initiate neuronal death in NDs have been partially elucidated, further research is needed.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Progressive Neuron Dysfunction/Death and Neurodegenerative Diseases", "text": "Selective neuron death in specific regions of the CNS is a prominent pathological feature of various NDs. Thus identifying and measuring neuron numbers in these distinct regions can significantly aid our understanding of the occurrence and pathogenesis of NDs (Chi et al., 2018; Moujalled et al., 2021). Another feature prevalent across various NDs is the abnormal aggregation of proteins. Although the composition and location of these aberrant toxic protein aggregates differ among the various NDs, it is clear that their generation and accumulation represent significant changes and shared mechanisms of pathogenesis in diseases such as AD, PD, ALS, HD, FTD, and LBD (Taylor et al., 2002; Ross and Poirier, 2004; Chung et al., 2018; Ruz et al., 2020). The aggregation of toxic proteins in neural cells is closely associated with critical pathological developments, such as neuron death, in various NDs. However, we should exercise caution when evaluating the role of specific aberrant proteins in the pathogenesis of NDs, as it remains unclear whether the aggregation of certain proteins is harmful or has protective effects against neuronal death.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Progressive Neuron Dysfunction/Death and Neurodegenerative Diseases", "text": "The pathways of neuron death in the pathogenesis of NDs may involve apoptosis or necroptosis. It has been suggested that initiators or executors of these processes are activated in NDs (Hajimohamadreza and Treherne, 1997; Nishimoto et al., 1997; Desjardins and Ledoux, 1998; Friedlander, 2003). Additionally, some authors have proposed that other cell death pathways, including necroptosis, may also contribute to the pathogenesis of NDs (Shao et al., 2017; Zhang et al., 2017; Chi et al., 2018; Dion\u00edsio et al., 2020).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Apoptosis and Neurodegenerative Diseases", "text": "Apoptosis primarily involves two pathways: the intrinsic (mitochondrial) pathway and the extrinsic (death receptor) pathway. The intrinsic pathway is regulated by pro-apoptotic and anti-apoptotic members of the BCL-2 protein family, including BCL-2, BCL-XL, MCL-1, BCL-W, and A1/BFL1. Critical apoptotic initiators known as BH3-only proteins, such as BIM, PUMA, BID, BMF, BAD, HRK, BIK, and NOXA, bind with high affinity to anti-apoptotic BCL-2 proteins. These interactions release or directly activate BAX, leading to the formation of oligomers that permeabilize the outer mitochondrial membrane. Consequently, this process promotes the release of cytochrome c and activates caspase-activating factors, which in turn trigger a caspase cascade that cleaves hundreds of proteins, ultimately inducing cell death. The extrinsic pathway is initiated by the binding of ligands to members of the tumor necrosis factor (TNF) receptor superfamily that contain intracellular death domains. This binding activates the formation of signaling complexes that lead to the activation of caspase-8 and downstream effector caspases, such as caspase-3 and -7. The extrinsic pathway can also interact with the intrinsic apoptosis pathway by activating the pro-apoptotic BH3-only protein BID, thus modulating cellular apoptosis through caspase-8-mediated proteolytic processes (Offen et al., 2000; Harada and Grant, 2003).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Apoptosis and Neurodegenerative Diseases", "text": "With regards to the neuron death pathways involved in ND pathogenesis, apoptosis has long been suggested to be a primary mechanisms of neuron death in NDs. Research has shown that nearly all Lewy-body-positive neurons exhibit positivity for the pro-apoptotic factor BAX, indicating a significant aggregation of pro-apoptotic proteins occurs in the pathogenesis of NDs (Camins et al., 2008). Additionally, the identification of mediators associated with apoptosis and necroptosis, such as tumor suppressor p53, mitogen-activated protein kinases, and necrostatin-1, suggests that both apoptotic and necroptotic processes contribute to neuron death pathways in NDs (Deigner et al., 2000; Yuan and Yankner, 2000; Jellinger, 2001; Li, 2021). Therefore, apoptosis and necroptosis represent critical common mechanisms in the pathogenesis of NDs.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Protein Misfolding, Aggregation, Conformation, and Neurodegenerative Diseases", "text": "The three-dimensional architectures of proteins are determined by their distinct amino acid sequences, and the construction of functional frameworks in cells and tissues is essential for cellular metabolism and biological signal communication. While the majority of protein folding is specific, some proteins exhibit a degree of structural flexibility that allows them to adapt their shape to perform certain physiological functions. As with other biological processes, the structures of proteins dictate their functions. Consequently, protein folding is a critical biological process for maintaining efficient cellular quality control systems, which ensures normal protein production and elimination are maintained (Dobson, 2003).\n\nWhen certain pathogenic factors alter the cellular environment, they can lead to protein misfolding, resulting in the production of abnormal protein structures. These aberrant proteins disrupt the normal processes involved in scavenging and eliminating abnormal, pathogenic, or toxic proteins, causing them to aggregate and progressively deposit in intracellular and/or extracellular spaces. This accumulation contributes to the development of specific diseases, such as NDs (Soto and Pritzkow, 2018). Recent insights into the mechanisms of protein misfolding and aggregation indicate that the characteristic aggregation patterns and distinctive deposition sites of specific proteins are closely associated with the pathogenesis of NDs (Ross and Poirier, 2004).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Protein Misfolding, Aggregation, Conformation, and Neurodegenerative Diseases", "text": "Another common pathogenic process in NDs is the progressive accumulation of protein aggregations, typically referred to as amyloid proteins, in specific intercellular or extracellular compartments (Soto, 2003). Although the protein aggregation processes involved in different NDs are distinct, the intermediates, end-products, and major features of protein misfolding are remarkably similar (Soto, 2003). The seeding-nucleation mechanisms of protein misfolding and aggregation have been increasingly characterized, described in detail, and modeled kinetically (Meisl et al., 2017; Zampar et al., 2024). While there are commonalities in the pathogeneses of NDs, there are also potential differences, including clinical symptoms, prevalence, risk factors, affected regions of the CNS, injured cell types, and implicated genes. Moreover, each ND is often associated with the misfolding and aggregation of a distinct protein that forms deposits in certain cellular locations, such as the cytoplasm, nuclei, cytoplasmic membranes, or extracellular space (Ross and Poirier, 2004; Sami et al., 2017; Soto and Pritzkow, 2018). Despite the similarities in the gross structural signatures of aggregation and cross-beta conformations, alterations in the detailed structures reveal significant differences among the proteins involved in specific NDs. Variations in the sequences, cellular locations, and biophysical characteristics of the aggregated proteins likely result in different mechanisms of cellular toxicity (Roy et al., 2022).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Protein Misfolding, Aggregation, Conformation, and Neurodegenerative Diseases", "text": "The presence of protein aggregations is a hallmark of multiple NDs, including AD, PD, ALS, HD, FTD, and LBD. Traditionally, each disease is associated with specific protein aggregations. For example, AD is characterized by the aggregation of A\u03b2 and tau proteins, while \u03b1-synucleinopathies such as PD and LBD involve the aggregation of \u03b1-syn. In ALS, protein aggregates include SOD1, TDP-43, FUS, and C9ORF72 dipeptide repeat proteins, while HD features the aggregation of HTT protein. FTD is associated with the aggregation of tau, C9ORF72, and Pick bodies, and LBD is characterized by the aggregation of insoluble Lewy bodies. However, it has been observed that the aggregation of a single protein can occur across multiple NDs, and one ND may involve the aggregation of several different proteins. For instance, protein aggregation has been identified in AD, HD, and various forms of FTD. Similarly, A\u03b2 aggregation has also been found in PD and LBD. Tau aggregation, while primarily associated with tauopathies, has also been detected in prion diseases, \u03b1-synucleinopathies, and cognitively healthy aging individuals. TDP-43 aggregation is typically linked to FTD and ALS, but it has also been observed in AD, progressive supranuclear palsy, corticobasal degeneration, multiple system atrophy, LBD, and other NDs. These findings highlight the complexities of protein aggregations in NDs, suggesting that common interactions and pathogeneses may exist across the different diseases. Further research into these intricate shared mechanisms may provide a better understanding of the distinctive and common pathogeneses of these fatal conditions (Ross and Poirier, 2004; Calabrese et al., 2022).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Protein Misfolding, Aggregation, Conformation, and Neurodegenerative Diseases", "text": "One of the most important findings in the pathogenesis of NDs is the misfolding, aggregation, and accumulation of proteins, which lead to cellular dysfunction, loss of synaptic connections, and lesions in the CNS. Although distinct proteins are involved in different NDs, the processes of protein misfolding and aggregation show remarkable similarities across these diseases. NDs are increasingly understood to share common cellular and molecular mechanisms, including protein aggregation and the formation of specific inclusion bodies. Protein aggregates typically consist of fibrils formed by proteins with cross-beta conformations, including amyloid proteins. There is a partial, but not complete, overlap among the types of abnormal proteins deposited in affected cells and the degenerative cells themselves. These inclusions and visible protein aggregates may represent a common stage in the molecular cascades underlying the pathogenesis of NDs. Furthermore, the earlier steps in these cascades may represent more direct mechanisms of disease pathogenesis than the formation of inclusions (Ross and Poirier, 2004).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Protein Misfolding, Aggregation, Conformation, and Neurodegenerative Diseases", "text": "The shared characteristics of NDs suggest there is a common underlying pathogenesis. The most common feature of these diseases is the deposition of amyloid proteins and the degeneration of the CNS network. Amyloids are ordered assemblies of proteins that may act as self-replicating templates that reproduce by adding monomers. Various amyloid proteins, such as A\u03b2 and tau, exhibit properties similar to those of infectious prion-like proteins, including discrete self-replicating amyloid structures and the ability to propagate aggregates across cells, leading to transmissible neuropathologies. This indicates that the prion-like phenomenon of amyloid transcellular propagation may be a common mechanism in the pathogenesis of various NDs, including AD and PD. The fundamental and distinct patterns of neuropathology, such as cell entry, amplification, transcellular movement, and the spread of amyloid proteins throughout the nervous system, represent potential common pathways in the development of NDs (Vaquer-Alicea and Diamond, 2019).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Protein Misfolding, Aggregation, Conformation, and Neurodegenerative Diseases", "text": "The most common NDs, AD, PD, ALS, and FTD, share several features of protein aggregation, despite exhibiting various distinct clinical symptoms. For instance, AD is characterized by the presence of senile plaques composed of aggregated A\u03b2 peptides and neurofibrillary tangles formed by intraneuronal tau aggregation. Similar protein accumulations are observed in PD (Bourdenx et al., 2017), indicating a common pathogenesis. TDP-43 aggregation is associated with ALS, while tau protein aggregation (MAPT) is observed in both FTD and AD, implying there are shared pathogenic mechanisms among these diseases (Montalbano et al., 2020).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Protein Misfolding, Aggregation, Conformation, and Neurodegenerative Diseases", "text": "Protein aggregations are often composed of a single protein or a mixture of multiple heterogeneous proteins of varying sizes and structures (Hayden and Teplow, 2013). The latest studies have demonstrated that several common proteins, including A\u03b2 and tau, often coexist in aggregates (Sengupta and Kayed, 2022; Li et al., 2023a, b; Guarnieri et al., 2024; Liu et al., 2024a, b). These findings suggest that there may be a significant common pathogenesis underlying AD and PD, despite the considerable heterogeneity in their clinical manifestations (Melki, 2015; Narasimhan et al., 2017). Research has shown that more than seven different diseases are associated with abnormal tau aggregation, including AD, FTD, progressive supranuclear palsy, corticobasal degeneration, argyrophilic grain disease, and chronic traumatic encephalopathy (Williams, 2006). While three diseases are linked to tau deposition, PD, multiple system atrophy, and LBD (Goedert et al., 2017). These findings further support the notion that multiple neurodegenerative diseases, including AD, PD, FTD, and LBD, share a common pathogenesis.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Protein Misfolding, Aggregation, Conformation, and Neurodegenerative Diseases", "text": "The coexistence of multiple distinct protein aggregations among various NDs is supported by substantial amounts of evidence (Brown et al., 1998; Giasson et al., 2003; Clinton et al., 2010; Morales et al., 2013; Spires-Jones et al., 2017). A typical pathogenic sign in AD is the simultaneous presence of intracellular tau neurofibrillary tangles and extracellular A\u03b2 plaques (Morales et al., 2013). Although neurofibrillary tangles and plaques may form independently, recent studies have provided evidence that A\u03b2 promotes tau abnormalities through direct protein-protein interactions (Guo et al., 2006; Soto and Pritzkow, 2018). Neuropathological studies have also shown that nearly half of AD patients exhibit TDP-43 deposition (Uchikado et al., 2006) and/or aggregation (Josephs et al., 2016). In PD and associated synucleinopathies, the prevalence of mixed pathologies is even higher: approximately 80% of patients show detectable A\u03b2 deposition, around 50% exhibit tau aggregation, and about 30% display TDP-43 deposition (Spires-Jones et al., 2017).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Protein Misfolding, Aggregation, Conformation, and Neurodegenerative Diseases", "text": "The extensive pathological overlap in protein aggregations across distinct CNS regions in different NDs may be closely linked to the complex and multifaceted phenotypes of these disorders. Evidence from pathological and clinical analyses suggests that NDs can be triggered by a specific type of protein aggregation that acts as a driving force for the initial clinical manifestations. This initial aggregation leads to the accumulation of other protein aggregates, inducing secondary changes that alter or expand CNS impairments and clinical symptoms (Spires-Jones et al., 2017). Supporting this hypothesis is the fact that the pathogenesis of PD typically presents with classic motor symptoms in the early stages. Over time, the majority of PD patients develop a series of non-motor symptoms and more complex motor symptoms (Poewe et al., 2017; Outeiro et al., 2019; Chahine et al., 2023). The emergence of these symptoms is speculated to be induced by the initial protein aggregation. Additionally, protein deposits in certain CNS regions can contribute to non-motor symptoms on their own, as some of these symptoms arise from Lewy bodies that contain pathogenic protein aggregates (Poewe et al., 2017; Outeiro et al., 2019; Tsoi et al., 2023).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Protein Misfolding, Aggregation, Conformation, and Neurodegenerative Diseases", "text": "Although the occurrence of cross-seeding via direct interactions among several common pathogenic proteins, such as tau, A\u03b2 (Murakami and Ono, 2022), TDP-43, and SOD1 (Koch et al., 2016; Pokrishevsky et al., 2024), has been confirmed by in vitro experiments, it is important to note that there are alternative explanations for the cross-seeding observed in different NDs (Montalbano et al., 2020). Other factors contributing to cross-seeding include enhanced cell vulnerability, impaired clearance mechanisms, neuroinflammation, and the activation of indirect signaling transduction pathways, all of which can lead to increased protein misfolding (Spires-Jones et al., 2017). Furthermore, properties closely associated with seeding or cross-seeding, such as the stereotypical pathological progressions observed in certain NDs, should also be considered, as they may help explain the existence of selective neuronal vulnerability (Lee et al., 2024a, b).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Protein Misfolding, Aggregation, Conformation, and Neurodegenerative Diseases", "text": "Protein aggregations in NDs can also be cross-seeded by aggregates from systemic disorders related to peripheral tissues. One of the strongest pieces of evidence supporting this mechanism is the interaction between AD and type 2 diabetes (T2D) (Hamz\u00e9 et al., 2022; Michailidis et al., 2022). T2D is associated with the accumulation of islet amyloid polypeptide in the pancreas, and interestingly, T2D patients are at an increased risk of developing AD (Kim and Feldman, 2015), with approximately 80% of AD cases showing a history of T2D or abnormal glucose metabolism (Ryu et al., 2019). Transgenic animal models expressing both human A\u03b2 and IAPP have been shown to have exacerbated AD-like pathologies (Moreno-Gonzalez et al., 2017). Both islet amyloid polypeptide and amyloid plaques co-localized in parenchymal deposits in the brains of these animals, suggesting A\u03b2 and IAPP may interact directly to worsen diseases such as T2D and AD. Furthermore, inoculating IAPP aggregates from the pancreas into the brains of AD transgenic mice resulted in more severe AD pathologies and significantly greater memory impairment compared to control mice (Moreno-Gonzalez et al., 2017). Cross-seeding mechanisms have been confirmed in in vitro experiments, which demonstrated that the presence of IAPP seeds accelerated A\u03b2 aggregations, with both types of peptides forming partially similar fibrils (Moreno-Gonzalez et al., 2017).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Protein Misfolding, Aggregation, Conformation, and Neurodegenerative Diseases", "text": "In recent years, emerging evidence suggests that pathological protein aggregations leading to NDs may be initiated by seeds from functional amyloids (Ross and Poirier, 2004). Multiple proteins have been found to naturally aggregate as non-pathogenic amyloid structures that contribute to protein function modulation or even the acquisition of novel biological activities. These functional amyloids are present in organisms ranging from bacteria to humans (Fowler et al., 2007), indicating that the formation of these structures is not necessarily linked to pathological processes. The possibility that the protein misfolding and aggregations that cause NDs might be triggered by cross-seeding with functional amyloids has not been extensively researched (Chiti and Dobson, 2006). However, one recent study reported that bacterial amyloids might exert effects on such aggregations (Soto and Pritzkow, 2018).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Protein Misfolding, Aggregation, Conformation, and Neurodegenerative Diseases", "text": "Calmodulin-binding proteins and calcium dysregulation are early and persistent factors in the pathogenesis of multiple NDs. Calmodulin serves as the initial and primary sensor of cytosolic calcium levels and has garnered interest as a major regulator of neurodegeneration due to its crucial role in calcium dysregulation and its effects on calcium signaling. Calmodulin plays a critical role in NDs by binding to proteins and regulating protein abundance, with many of its functions participating across several NDs. Notably, calmodulin has a common function in the production of protein aggregates in AD, PD, ALS, FTD, HD, and LBD. Calmodulin not only binds to A\u03b2, pTau, and mutant HTT but also converts them into toxic protein aggregates by regulating transglutaminase 2. Additionally, multiple calmodulin-binding proteins have been suggested as promising primary drug targets, and using combinations of these proteins could offer insights into the pathogenesis and progression of several NDs (O\u2019Day, 2023), including AD (O\u2019Day et al., 2015), PD, ALS, FTD, HD, and LBD (Bourdenx et al., 2017; Azam et al., 2021; O\u2019Day and Huber, 2022).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Propagation of Protein Aggregation and Neurodegenerative Diseases", "text": "The present findings in the field of protein propagation have revealed that aggregating proteins can self-propagate through seeding\u2014the spread of pathological abnormalities between cells and tissues in a manner similar to the behavior of infectious prions in prionoid diseases. These findings have important implications for our understanding of the mechanisms that initiate and drive NDs (Vaquer-Alicea and Diamond, 2019).\n\nA growing body of evidence demonstrates that prionoid protein propagation of misfolded proteins is one of the core mechanisms underlying NDs. The discovery of several common pathogenic elements, pathological effects, and protein homeostasis involved in the pathogenesis of AD, PD, ALS, and HD has extended the pathogenic processes from neurons to glial cells. These add to the complexity of disease progression and can promote the initiation of protective processes in neural cells that reduce the propagation of pathological protein aggregation (Wells et al., 2019).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Propagation of Protein Aggregation and Neurodegenerative Diseases", "text": "NDs are a group of debilitating disorders primarily characterized by the gradual loss of function and cell death in the CNS and peripheral nervous system. The majority of neurons cannot regenerate to replace dead neural cells, leading to the accumulation of lesions and permanent damage. Although these NDs exhibit various etiologies and pathologies, each type involves the degeneration of distinct neurons and the gradual loss of different nervous functions, with the specific neurons and functional losses varying among the different NDs. Prionoid propagation represents a common and classic form of protein aggregation in the pathogenesis of NDs and is driven by template-directed protein misfolding. Prionoids, similar to prion proteins, utilize a seeding mechanism to induce template-based conformational changes that transform native proteins into their prionoid form. Prionoid transmission then occurs between host neurons. The seeding properties of prionoids resemble those of aggregated proteins, with misfolded and aggregated proteins spreading to neighboring neural cells in a manner akin to that of infectious prion particles (Aguzzi and Rajendran, 2009; Aguzzi and Lakkaraju, 2016; Wells et al., 2019).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Propagation of Protein Aggregation and Neurodegenerative Diseases", "text": "Multiple common NDs, such as AD, PD, HD, and ALS, are now considered to be based on prionoid pathology. Despite extensive research, our understanding of the details of prionoid propagation processes in the context of NDs remains limited. In addition, there are few effective treatments or cures for prionoid diseases. Prionoid proteins can misfold into multiple pathological conformations, which can then propagate to normal proteins, and this template function is facilitated by the exposure of hydrophobic amino acid side chains that are normally buried within the protein\u2019s interior (Prusiner, 1998). Prionoid proteins typically possess two common structures, \u03b2 sheets and polypeptides, making them prone to forming aggregates, including fibrils (Aguzzi and Lakkaraju, 2016). These properties of prionoid proteins promote the intracellular aggregation of misfolded proteins, facilitating the formation of stable inclusion bodies by recruiting normal proteins. Extracellular protein fibrils can also enter and aggregate in other cells, enabling the intercellular transmissibility of prionoid proteins (Aguzzi and Lakkaraju, 2016).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Propagation of Protein Aggregation and Neurodegenerative Diseases", "text": "Various NDs demonstrate different mechanisms of pathogenesis, yet they all ultimately result in distinctive neuronal death within the CNS, leading to symptoms associated with the loss or gain of neuronal functions. The major gain-of-function mechanisms include prionoid protein misfolding, which is often accompanied by some functional losses (Allison et al., 2017). Recently, the mechanisms underlying the gain and loss of neuronal functions have been identified as participating in the pathogenesis of AD, HD, and ALS (Leighton and Allison, 2016). Generally, gain-of-function mechanisms include the activation of autophagy, protein aggregation, axon dysfunction, and cellular stress, while loss-of-function mechanisms involve protein sequestration, synaptic dysfunction, and DNA impairment. Symptoms of NDs may arise from the loss and/or gain of functions, contributing to conditions such as denervation, mitochondrial dysfunction, excitotoxicity, and oxidative stress (Wells et al., 2019).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Propagation of Protein Aggregation and Neurodegenerative Diseases", "text": "When a pathological A\u03b2 protein homogenate derived from the brains of AD patients was injected into APP model transgenic mice, the animals exhibited time- and concentration-dependent development of plaques as well as neuronal pathology (Meyer-Luehmann et al., 2006). The A\u03b2 protein is known to be capable of propagating directly between the neurites of interfacing neurons (Nath et al., 2012). Similarly, extracellularly aggregated tau protein, but not tau protein monomers, propagates from the extracellular space to the intracellular environment during the pathogenesis of AD (Frost and Diamond, 2009). Microglia package the misfolded and aggregated tau proteins into nanoscale vesicles, which are secreted as exosomes by all mammalian cells. While this process can help control the accumulation of pathogenic proteins within cells, it can also propagate aberrant proteins between adjacent brain regions, thereby accelerating the spread of pathological proteins in the pathogenesis of NDs (Asai et al., 2015).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Propagation of Protein Aggregation and Neurodegenerative Diseases", "text": "Similarly, soluble endogenous proteins are recruited and converted into fibrils, which propagate among cells via the intracellular spaces and form Lewy body-like aggregates in the cytoplasm during the pathogenesis of PD (Luk et al., 2009). Evidence suggests that these proteins are secreted into the extracellular space via atypical exocytotic processes under normal physiological conditions. However, the amounts secreted significantly increase due to both proteasomal and mitochondrial dysfunctions in PD pathogenesis (Rocha et al., 2018), and this extracellular internalization may lead to seeded endogenous aggregation as the disease progresses (Rodriguez et al., 2018; Hoffmann et al., 2019). While the precise mechanisms of the cellular protein internalization involved in PD pathogenesis remain a topic of debate, potential mechanisms include endocytosis, micropinocytosis, and cell surface-protein-mediated uptake (Rodriguez et al., 2018).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Propagation of Protein Aggregation and Neurodegenerative Diseases", "text": "Large internalizing aggregates of HTT polyQ in the neurons of HD patients bind to cellular plasma membranes to form HTT inclusion bodies. The aggregations of HTT protein, along with their self-sustaining seeding and fragmentation abilities, are inherited as cells differentiate (Bonavita et al., 2023). Experimental evidence shows that mutant HTT (mHTT) protein may propagate across synapses in transgenic mice and Drosophila melanogaster models (Pecho-Vrieseling et al., 2014; Babcock and Ganetzky, 2015). The propagated HTT protein can either recruit endogenous HTT monomers and oligomers to propagate seed aggregations (Sancho et al., 2011) or translocate to the nuclei, resulting in cell death. Interestingly, shorter peptides with fewer polyQ repeats are no less toxic than longer peptides, suggesting that the pathological roles of polyQ repeats may be more related to their aggregation efficiencies than their toxicities (Wellington et al., 2002).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Propagation of Protein Aggregation and Neurodegenerative Diseases", "text": "In ALS, mutant SOD1 (mSOD1)-positive aggregates form granules with a fibril morphology rich in \u03b2-sheet structures, facilitating the template-based misfolding of wild-type SOD1 protein in fALS patients and fALS transgenic mouse models (Kato et al., 2000). In in vitro experiments, mSOD1 aggregation was shown to accelerate or seed the aggregation of soluble SOD1 protein (Brotherton et al., 2013). The intercellular propagation of mSOD1 requires the aggregated proteins to be released into extracellular spaces, followed by their internalization into other cells through micropinocytosis, which allows the internalized protein aggregates to seed endogenous SOD1 aggregation (Cicardi et al., 2021). The ability of self-propagation via fibril fragments suggests that fibril breakage may be a critical component of the self-propagation process (Lee and Kim, 2015; Wells et al., 2019).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Neuroinflammation and Neurodegenerative Diseases", "text": "NDs are chronic, progressive conditions primarily affecting the CNS and are characterized by the occurrence of progressive damage to neuronal structures and functions. Current evidence indicates that neuroinflammation plays a significant role in the pathophysiological processes underlying NDs and is closely associated with major conditions, including AD (Calsolaro and Edison, 2016), PD (Wang et al., 2022), ALS (McCauley and Baloh, 2019), HD (Soto, 2003), FTD (Bright et al., 2019), and LBD (Surendranathan et al., 2015).\n\nThe roles of cellular and molecular pathways, along with the activation of inflammatory mediators within neuroinflammatory systems, are critical biological processes involved in the pathogenesis of AD (Ahluwalia and Vellas, 2003; Calsolaro and Edison, 2016). AD is characterized by the significant accumulation of A\u03b2 protein; however, we still lack a clear understanding of its causative pathogenesis. The disappointing results from several anti-amyloid treatment strategies have raised concerns that the amyloid cascade hypothesis alone cannot fully explain the neuronal damage associated with AD pathogenesis.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Neuroinflammation and Neurodegenerative Diseases", "text": "Neuroinflammation plays significant roles in the pathogenesis of NDs, although there is ongoing debate about whether its effects are protective or harmful (Calsolaro and Edison, 2016). The deposition of A\u03b2 protein in the context of AD initiates a cascade of neuroinflammatory responses, including the activation of microglial cells and astrocytes, which leads to the expression of various pro-inflammatory and anti-inflammatory cytokines. These neuroinflammatory reactions are crucial pathophysiological processes in AD. Pro-inflammatory cytokine signaling exerts multiple effects, both neurodegenerative and neuroprotective, during the disease\u2019s progression. The induction of pro-inflammatory signals results in the release of immunological mediators that affect neuronal function and lead to cell death, and a reduction in anti-inflammatory mechanisms can further activate neuroinflammation. Various pathways, including nuclear factor kappa B (NF-\u03baB), p38 MAPK, Akt/mTOR, caspase, nitric oxide, and COX, are involved in stimulating neural immune cells, such as astrocytes and microglia, to release inflammatory cytokines, including tumor necrosis factor, interleukins, and chemokines, all of which participate in the pathogenesis of AD. Research has found that PPAR-\u03b3 agonists enhance the phagocytosis of A\u03b2 protein (Mandrekar-Colucci et al., 2012) and decrease the levels of inflammatory cytokine IL-1\u03b2 (Thakur et al., 2023). Moreover, gut microbiota, and its connection to neuroinflammation, (Thakur et al., 2023) also plays a role in the pathogenesis of AD.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Neuroinflammation and Neurodegenerative Diseases", "text": "PD is primarily caused by the aberrant accumulation of pathogenic proteins in dopaminergic neurons of the substantia nigra in the midbrain, and the consequent neuroinflammation exerts key effects in the neurodegenerative pathogenesis of PD. Neuroinflammatory processes that contribute to the neurodegeneration involve the activation of microglial cells, the upregulation of pro-inflammatory factors, and disturbances in gut microbiota, all of which facilitate the release of inflammatory cytokines such as IL-1\u03b2, IL-2, IL-6, IL-10, and TNF-\u03b1. Neuroinflammation has been closely linked to the pathogenesis of PD, as evidenced by increased levels of pro-inflammatory cytokines in the blood and/or CSF (Cohen et al., 2024). Activated microglial cells secrete several pro-inflammatory cytokines, including TNF-\u03b1, IL-1\u03b2, and IL-6 (Dzamko, 2023), as well as major histocompatibility complex class II (Kulski et al., 2024), all of which are associated with neuronal damage in the context of PD.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Neuroinflammation and Neurodegenerative Diseases", "text": "The abnormal aggregation of insoluble proteins exerts significant effects in the pathogenesis of PD. Protein misfolding contributes to the dysregulation of microglial Toll-like receptor (TLR)2 or TLR4-mediated signaling pathways, which are associated with pathogen-associated molecular patterns and damage-associated molecular patterns. This dysregulation activates myeloid differentiation primary response 88 and NF\u03baB, ultimately triggering the production of TNF-\u03b1 and IL-1\u03b2 (Li et al., 2021). Neuroinflammation is activated through immunological reactions involved in the degeneration of dopaminergic neurons in PD. Additionally, gut microbiota, metabolites, and cytokines may play roles in the interplay of processes underlying the pathogenesis of PD (Salim et al., 2023).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Neuroinflammation and Neurodegenerative Diseases", "text": "Neuroinflammation is a pathophysiological reaction involving resident glial cells, such as astrocytes and microglia, as well as circulating immune cells, e.g., monocytes, neutrophils, and lymphocytes, in response to infections, injury, or the degeneration of the CNS. Neuroinflammation is primarily characterized by innate immune responses rather than adaptive immune responses in the pathogenesis of ALS, similar to that of other NDs (Prinz and Priller, 2017). In addition to astrocytes and microglial cells, T cells and non-resident innate immune cells, such as dendritic cells, macrophages, and mast cells, are also activated during the pathogenesis of ALS. However, the roles of these immune cells remain poorly understood in the context of this condition (McCauley and Baloh, 2019).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Neuroinflammation and Neurodegenerative Diseases", "text": "HD is a genetic neurodegenerative disorder caused by an expansion of CAG repeats in the HTT gene. The clinical manifestations of HD include movement, cognitive, and behavioral dysfunctions. Inflammation occurs in HD patients prior to symptom onset, with reactive microglial cells, astrocytes, and inflammatory factors frequently observed in the pathogenesis, indicating changes in microglial and astrocyte function are involved in HD onset (Soto, 2003). The activation of the innate immune system contributes to rapid general responses to various pathogens and stressors, while the adaptive immune system generates highly specific responses primarily modulated by T or B lymphocytes. There is a considerable amount of evidence supporting the activation of innate immune cells within the CNS, while fewer studies have reported on the roles of the adaptive immune system in HD. Inflammation may be stimulated by glial responses to neurodegeneration or by the immune cell activation that results from mHTT expression; in addition, these two pathways may exert synergistic effects in HD (Soto, 2003). In general, neither the mechanisms nor the contributions of inflammation to the pathogenesis of HD have been fully elucidated, and the roles of neuroinflammation in HD are less well understood than those in other HDs.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Neuroinflammation and Neurodegenerative Diseases", "text": "FTD encompasses a spectrum of NDs characterized by various pathological features with genetic variability and complex pathogenic mechanisms. As the study of pathogenesis of FTD progresses, evidence is mounting that suggests neuroinflammation plays a role in its development. This includes cortical inflammation, microglial activation, astrogliosis, and the abnormal expression of inflammation-related proteins. Moreover, significant evidence from genetic studies indicates that multiple FTD-related genes are related to neuroinflammation within the pathogenesis of FTD (Bright et al., 2019; De Marchi et al., 2023).\n\nNeuroinflammation is increasingly being recognized as a key factor in the pathogenesis of NDs, but it remains unclear whether it plays a protective or damaging role in LBD. Research on LBD has provided substantial evidence for the involvement of inflammatory pathologies in its neurodegeneration (Noori et al., 2021; Becchi et al., 2023), with much of the evidence being derived from studies on LBD pathology and biomarkers (Griffin et al., 2006; Mrak and Griffin, 2007).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Mitochondrial Dysfunctions and Neurodegenerative Diseases", "text": "Mitochondrial dysfunction is proposed to play a significant role in various NDs, including AD (Hroudov\u00e1 et al., 2014; Wang et al., 2020; Kalani et al., 2023), PD (Gautier et al., 2014), ALS (Cozzolino and Carr\u00ec, 2012), HD (Carmo et al., 2018), FTD (Anoar et al., 2021), and LBD (Spano et al., 2015). Increasing evidence suggests that defects in mitochondrial metabolism, particularly within electron transport chains, exert key effects in the pathogenesis of NDs. Structural and functional damage to mitochondria hampers their ability to generate sufficient ATP and, conversely, leads to the production of pro-apoptotic factors and ROS, which are involved in the initial stages of multiple mitochondrial disorders, including NDs. Mitochondrial dysfunction can be triggered by mutations in mitochondrial and nuclear DNA sequences that code for mitochondrial components or by environmental factors. Common pathogenic mechanisms associated with mitochondrial damage include ROS generation, impairments in mitochondrial dynamics, the dysfunction of electron transport chain complexes, and interactions between pathogenic proteins and mitochondria (Hroudov\u00e1 et al., 2014). Normal mitochondrial function is critical for maintaining neuronal structure and function due to the high energy consumption of neurons. In the context of AD, mitochondrial dysfunction disrupts mitochondrial autophagy, which is responsible for eliminating dysfunctional mitochondria. This impairment directly or indirectly induces high levels of iron in neural cells and subsequently interferes with mitochondrial autophagy and mtDNA release. This cascade of events leads to the production of pro-inflammatory factors and activates cyclic guanosine monophosphate\u2013adenosine monophosphate synthase and stimulators of interferon genes pathways, promoting and exacerbating the development of NDs, including AD (Kalani et al., 2023).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Mitochondrial Dysfunctions and Neurodegenerative Diseases", "text": "NDs comprise a group of disorders characterized by progressive neuronal dysfunction and degeneration. NDs are highly heterogeneous, affecting various neuronal subpopulations and involving multiple origins and pathogenic mechanisms. Despite this diversity, evidence for some common pathological features among these diseases has emerged. Increasingly, observations are being made that suggest mitochondrial dysfunction plays a pivotal role in neurodegeneration across various NDs, including AD, PD, ALS, HD, FTD, and LBD. Impairments in mitochondrial function can occur at any stage of ND pathogenesis. Within the spectrum of NDs, current evidence indicates that mitochondrial dysfunction exerts similar effects during neuronal demise. PD has shown a particularly strong association with mitochondrial damage events, such as reduced complex I activity and oxidative impairment, in postmortem studies (Camilleri and Vassallo, 2014; Gautier et al., 2014). Additionally, the pathogenesis of PD is closely linked to mutations in more than five distinct genes, including parkin, DJ-1, PINK1, and LRRK2. Understanding the pathogenesis of neurodegeneration in selectively vulnerable dopaminergic neurons in PD could provide valuable insights into potential treatment options. To investigate these mechanisms, numerous in vitro and in vivo models of PD have been developed to mimic the disease\u2019s insults. These model studies have yielded valuable data regarding the damaging mechanisms affecting dopaminergic neurons, including mitochondrial dysfunction (Gautier et al., 2014). While PD is recognized as an ND related to aging, the exact relationship between aging and PD remains unclear. Mitochondrial dysfunction is considered a major pathogenic factor and a central part of the etiology and is closely associated with the selective death of dopaminergic neurons in PD. Specifically, mitochondrial dysfunction contributes to various aspects of PD pathogenesis, including ROS production, calcium overload, the activation of neuroinflammatory pathways, apoptosis, mitophagy, mitochondrial biogenesis, and mitochondrial dynamics (Moradi Vastegani et al., 2023).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Mitochondrial Dysfunctions and Neurodegenerative Diseases", "text": "Abundant evidence implicates mitochondrial dysfunction as a major element in the pathogenesis of ALS. Recent studies suggest that various interrelated aspects of mitochondrial dysfunction play significant roles in ALS pathogenesis, challenging previous perceptions (Cunha-Oliveira et al., 2024; Landry et al., 2024). Mitochondrial dysfunction alters numerous physiological functions and properties of mitochondria, and these alterations may contribute to the internal susceptibilities of long-lived and postmitotic motor neurons to energy deficits, calcium mishandling, and oxidative stress, ultimately facilitating motor neuron death in the context of ALS (Cozzolino and Carr\u00ec, 2012).\n\nMitochondrial dysfunction, which has been identified as an early factor contributing to the selective neurodegeneration of the striatum and cerebral cortex in HD, is driven by a polyglutamine expansion in the pathogenic HD gene. Mitochondria synthesize energy and several metabolites, and their maximal bioenergetic performance relies on their transportation along microtubules. They also regulate intracellular calcium homeostasis through interactions with the endoplasmic reticulum and produce free radicals, which participate in cell death processes. The majority of these mitochondrial activities have been shown to play roles in the pathogenesis of HD, potentially contributing to neuronal dysfunction in this condition (Carmo et al., 2018).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Mitochondrial Dysfunctions and Neurodegenerative Diseases", "text": "FTD and ALS are NDs characterized by a combined decline in movement and cognitive function. Although these two conditions exhibit distinct clinical manifestations, they are considered to be similar subtypes within the neurodegenerative spectrum, as they show significant overlap in their pathogeneses. Mitochondrial dysfunction is one these overlapping mechanisms and has been linked to both FTD and ALS. Several studies have identified mitochondrial abnormalities in various animal models, such as the Drosophila transgenic model, and partially elucidated the molecular mechanisms underlying the toxicity of gene mutations associated with FTD and ALS pathogenesis (Anoar et al., 2021; Nemtsova et al., 2023).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Mitochondrial Dysfunctions and Neurodegenerative Diseases", "text": "The main hallmark of LBD pathology is the formation of Lewy bodies, which are composed of aberrant \u03b1-syn in specific regions of the CNS. The CNS is particularly vulnerable to oxidative damage, such as that caused by ROS, which can easily lead to NDs. Comparative studies on the potential pathogenesis of LBD have been conducted using various models of mitochondrial dysfunction (Gu et al., 1998; Jin et al., 2024). The results from these studies confirmed the hypothesis that \u03b1-syn damages mitochondrial functions and increases sensitivity to cell death through factors such as cytochrome C (Hashimoto et al., 1999). Mitochondria play crucial roles in neurodegeneration and the production of Lewy bodies. The abnormal accumulation of \u03b1-syn can lead to the release of cytochrome C, which subsequently activates apoptosis cascades (Spano et al., 2015). LBD and AD are the most common forms of dementia in older adult populations. Previous studies have shown that cognitive alterations in NDs are related to synaptic loss, particularly via processes involving synaptic proteins such as \u03b1-syn. Furthermore, recent research has demonstrated that the aberrant aggregation and accumulation of the synaptic protein \u03b1-syn are critical for plaque formation in the pathogenesis of AD (Del Giudice et al., 2024). The presence of Lewy bodies in the pathogenesis of both LBD (Leak et al., 2024) and PD (Bayati and McPherson, 2024) further reinforces the hypothesis that \u03b1-syn plays a crucial role in the pathogenesis of NDs. Although the mechanisms underlying the functional alterations and \u03b1-syn aggregation that lead to neurodegeneration in diseases such as AD, PD, and LBD are not yet fully understood, recent evidence suggests \u03b1-syn has roles in synaptic lesions and neurotoxicity, potentially through the generation of amyloid-like fibrils and the dysfunction of mitochondria (Hashimoto and Masliah, 1999).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Axon Transportation Defect and Neurodegenerative Diseases", "text": "Axonal transport refers to the long-range, bidirectional transport of cellular components within axons, which allows cargo to be moved from the cell body to the terminals (anterograde axonal transport) and from the synapse back to the cell body (retrograde axonal transport). The directionality of cargo movement is controlled by the molecular motor proteins kinesin and dynein, which operate based on their polarity. Kinesins are essential for synaptic function and facilitate the anterograde transport of cargo, while dynein governs retrograde axonal transport. These molecular motor proteins are responsible for carrying various types of cargo, including synaptic vesicles, organelles, and even RNA (Brady ST, Morfini, 2017).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Axon Transportation Defect and Neurodegenerative Diseases", "text": "Neurons communicate within neural systems by transmitting information along their axons to synapses. The normal function of axonal transport plays a crucial role in the transportation of proteins, organelles, vesicles, neurotrophins, and other elements between neural cells and between neural cells and body cells. Molecular motors, i.e., kinesin and dynein, are responsible for transporting these elements along the neuronal axon. Proper axonal transport is essential for maintaining neuronal viability and function (Chevalier-Larsen and Holzbaur, 2006; Peng et al., 2020), and defects in axonal transport are considered significant contributors to the development and progression of NDs (Chevalier-Larsen and Holzbaur, 2006; De Vos and Hafezparast, 2017; Berth and Lloyd, 2023; Vasconcelos et al., 2023).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Axon Transportation Defect and Neurodegenerative Diseases", "text": "The core pathogenesis of NDs involves the misfolding, fibrillization, and aggregation of pathogenic proteins, which produce the hallmark lesions associated with specific NDs. Impaired axonal transport is believed to play a mechanistic role in several common NDs, such as AD (Wang et al., 2015; Vasconcelos et al., 2023), PD (Berth and Lloyd, 2023), ALS (Chevalier-Larsen and Holzbaur, 2006; De Vos and Hafezparast, 2017), HD (Roze et al., 2010), FTD (Adalbert et al., 2018; Combs et al., 2021), and LBD (Yamamoto, 2021). However, this hypothesis has not been thoroughly validated through extensive research. Mutations in genes encoding the motor proteins responsible for axonal transport have provided direct evidence that defective transport is a contributor to the pathogenesis of NDs, and this has been further supported by findings from neurodegeneration animal models (Roy et al., 2005). Increasing evidence suggests that A\u03b2 peptides, the characteristic pathogenic proteins of AD, disrupt axonal transport, contributing to the development of the disease. There appear to be both direct and indirect relationships between defective axonal transport and the pathogenesis of AD (Vasconcelos et al., 2023).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Axon Transportation Defect and Neurodegenerative Diseases", "text": "A substantial body of evidence indicates that defects in axonal transport play critical roles in the pathogenesis of AD. Axonal transport is crucial for maintaining neuronal function and requires four basic components, cytoskeletons, motor proteins, ATP, and adaptor proteins, which work together to ensure the specific transport of materials and the active regulation of movement. Pathogenic factors associated with AD disrupt these four components of axonal transport through various signaling pathways, leading to axonal transport deficits. These factors interact in overlapping, synergistic, and cascading pathways during pathogenesis (Wang et al., 2015). Axonal transport defects have been found to initiate neurodegeneration or axonopathy, resulting in neuronal death in several common NDs. Although the exact mechanisms underlying axonal degeneration remain unclear, multiple elements involved in axonal transport appear to be potential contributors to neurodegeneration in the pathogenesis of NDs such as AD. The destabilization and fragmentation of microtubules, loss of molecular motor function, deregulation of motor activity, and protein aggregation\u2014exacerbated by deficits in cellular scavenging pathways such as the autophagy-lysosome pathway and the ubiquitin-proteasome pathway\u2014can induce axonal degeneration and disrupt axonal transport across many NDs. The maintenance of microtubules and regulation of motor proteins are essential for normal axonal transport and the maturation of autophagosomes during axonal growth. The disruption of long-distance transport further contributes to axonal degeneration in the pathogenesis of AD (White et al., 2016).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Axon Transportation Defect and Neurodegenerative Diseases", "text": "PD is an ND characterized by prominent damage to dopaminergic neurons in the substantia nigra pars compacta. Misfolded presynaptic proteins aggregate during axonal transport, forming Lewy bodies within the cytoplasm of dopaminergic neurons, which contributes to neuronal damage in the pathogenesis of PD. Postmortem analyses and animal model studies have determined that axonal dysfunction, a loss of presynaptic terminals, and a loss of motor proteins contribute to deficits in axonal transport. As a result, misfolded proteins cannot be transported out of neuronal bodies, leading to their progressive aggregation in the cytoplasm and the formation of the Lewy body inclusions involved in the pathogenesis of PD (Volpicelli-Daley, 2017).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Axon Transportation Defect and Neurodegenerative Diseases", "text": "ALS, FTD, and LBD also share similar pathogeneses, characterized by the progressive aggregation of misfolded pathogenic proteins such as SOD1, FUS, C9ORF72, and tau. These aggregations represent hallmark pathological alterations and key pathogenic elements in the development of these NDs. Therefore, it is hypothesized that deficits in axonal transport hinder the removal of these aberrant misfolded proteins from the neuronal body and cytoplasm, accelerating their aggregation and toxicity to neural cells, including neurons. This damage to specific neuronal populations may contribute to the onset of these NDs (Chevalier-Larsen and Holzbaur, 2006; Berth and Lloyd, 2023). While our understanding of the precise pathogeneses of these diseases is limited, a deficit in axonal transport appears to be a significant pathophysiological feature in the development of several common NDs and thus warrants further exploration.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Gut Microbiota/Dysbiosis and Neurodegenerative Diseases", "text": "The human gut microbiota, which includes a diverse array of microorganisms, engages in bidirectional communication with the CNS through the brain\u2013gut axis. This interaction exerts significant effects on physiological processes and the pathogenesis of multiple diseases, including neurodegenerative disorders. Increasingly, evidence is highlighting the critical role of gut microbes in both normal brain function and pathological conditions, particularly in NDs such as AD, PD, and ALS (Quigley, 2017; Chen et al., 2021; Loh et al., 2024).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Gut Microbiota/Dysbiosis and Neurodegenerative Diseases", "text": "The aberrant aggregation of toxic proteins is a major cause of NDs such as AD (Jiang et al., 2017), PD (Hirayama and Ohno, 2021; Pant et al., 2022), ALS (McCombe et al., 2019), HD (Khoshnan, 2024), FTD (Hashim and Makpol, 2022), and LBD (Mateo et al., 2023). As research on the relationship between the pathogenesis of NDs and gut microbiota deepens, it has been suggested that a loss of gut microbiota homeostasis plays a significant role in the pathogenesis of these diseases and is thus drawing increasing attention from researchers. Studies have reported that disturbances in gut microbiota are closely associated with the accumulation of toxic proteins in NDs (Hashim and Makpol, 2022; Zhang et al., 2023a,b; Ekwudo et al., 2024; Lee et al., 2024a, b; Mateo et al., 2024); some of these pathogenic proteins may be directly produced by gut microbiota and transported to the CNS via vagus nerve fibers. Additionally, toxic proteins generated by gut microbiota can enter the peripheral circulation through the intestinal barrier or cross the blood\u2013brain barrier, leading to neuroinflammation and the production of inflammatory factors and cytokines that further induce the generation of toxic proteins. These toxic proteins activate microglia, contributing to the development of NDs. Interestingly, certain gut microbiota and their metabolites have also been shown to exert neuroprotective effects. Therefore, disruptions to the dynamic homeostasis of gut microbiota may represent a common underlying mechanism in the pathogenesis of NDs (Quigley, 2017; Loh et al., 2024).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Gut Microbiota/Dysbiosis and Neurodegenerative Diseases", "text": "The gut microbiota consists of complex communities of microbial species, and alterations in these communities can affect not only various gut diseases but also numerous CNS disorders, including AD. A hallmark of AD pathogenesis is the progressive cognitive impairment driven by the accumulation of A\u03b2. Notably, the microbiota\u2013gut\u2013brain axis, a bidirectional communication system, involves nervous, immune, endocrine, and metabolic pathways. These biological pathways play crucial roles in cognitive impairment by increasing the permeability of the gut\u2013 and blood\u2013brain barriers, a process induced by microbiota dysbiosis in the pathogenesis of AD. Moreover, gut microbiota have been reported to produce significant amounts of amyloid proteins and lipopolysaccharides, which modulate AD-related signaling pathways and promote the production of pro-inflammatory cytokines. Additionally, a loss of gut microbiota homeostasis was shown to trigger certain inflammatory cascades linked to AD pathogenesis. These findings have helped partially elucidate the role of the gut microbiota in the development and pathogenesis of AD (Jiang et al., 2017).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Gut Microbiota/Dysbiosis and Neurodegenerative Diseases", "text": "The characteristic pathogenesis of PD involves the formation of Lewy body inclusions caused by the aberrant aggregation of misfolded proteins. Accumulating evidence suggests that these misfolded proteins originate from the dorsal vagus nuclei in the CNS and are transported via the vagus nerve to the locus coeruleus and substantia nigra, where they extensively spread to various brain areas, including basal neuronal tangles and the cerebral cortices. The gut microbiota contribute to the production of intestinal proteins, which can become misfolded in the intestines and are subsequently transported to distinct brain areas such as the substantia nigra and striatum, potentially extending to the cerebral cortices. This process can damage neural cells, including dopaminergic neurons, thereby contributing to the development of PD (Hirayama and Ohno, 2021).\n\nAbnormal changes associated with gut microbiota are thought to play a vital role in the pathogenesis of PD. The aggregation of pathogenic proteins may originate in the gut and migrate to the CNS via the vagus nerve. Alterations in gut microbiota can trigger gut inflammation, which in turn promotes neuroinflammation in the CNS through gut-brain communication, and these communication events are closely linked to the development of PD. Gut dysbiosis can directly or indirectly alter gene expression, affect multiple biological pathways, and modify the microbial metabolites involved in the pathogenesis of PD (Pant et al., 2022).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Gut Microbiota/Dysbiosis and Neurodegenerative Diseases", "text": "The gut microbiota and its metabolites play key roles in the pathogenesis of ALS. A previous study has provided some evidence regarding gut dysbiosis in this condition (McCombe et al., 2019). However, research on the relationship between gut dysbiosis and ALS pathogenesis is limited (Zheng et al., 2023; Lee et al., 2024a, b) and warrants further investigation.\n\nThe HTT protein is a pathogenic factor in the pathogenesis of HD, and it is suggested that mutant HTT protein may disrupt the homeostasis of gut microbiota in this context. Gut microbiota dysbiosis could directly or indirectly induce the production and aggregation of the mutant HTT protein, leading to potential toxicity in cells, tissues, and organs through the microbiota-gut-immune-CNS axis in HD pathogenesis (Khoshnan, 2024). Gut dysbiosis influences CNS functions via the brain\u2013gut microbiota axis, which represents a bi-directional link between the gastrointestinal tract and the CNS closely related to the pathogenesis of NDs (Hashim and Makpol, 2022; Mateo et al., 2023). Similar to ALS, evidence for the relationship between gut dysbiosis and the pathogenesis of FTD and LBD is currently very limited, and this area warrants further investigation (Figure 2).\n\nPotential common mechanisms in six neurodegenerative diseases including AD, PD, ALS, HD, FTD, and LBD.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Gut Microbiota/Dysbiosis and Neurodegenerative Diseases", "text": "Based on current evidence, the potential common mechanisms of NDs may include aging, genetics, progressive neuronal dysfunction and death, protein misfolding and aggregation, propagation of protein aggregates, neuroinflammation, mitochondrial dysfunction, axonal transport defects, and gut microbiota dysbiosis. These mechanisms may interact with one another, collectively contributing to the pathogenesis of NDs. AD: Alzheimer\u2019s disease; ALS: amyotrophic lateral sclerosis; FTD: frontotemporal lobar dementia; HD: Huntington\u2019s disease; LBD: Lewy body disease; NDs: neurodegenerative diseases; PD: Parkinson\u2019s disease; SOD1: Cu/Zn superoxide dismutase.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Clinical Translation Based on Pathogenic Mechanisms", "text": "This section of the review provides a brief overview of contemporary research focused on developing new therapeutic approaches for NDs, such as AD, PD, ALS, HD, FTD, and LBD, utilizing gene editing, immunoregulation, and neuroregeneration. Gene therapy is experiencing a resurgence due to its potential to target underlying etiological factors and provide long-term modifications. Gene editing therapeutics are particularly appealing for NDs, given the disappointing results of traditional pharmacologic treatments (Sun and Roy, 2021). With the continuous evolution of gene editing technologies, gene therapies now offer more precise transgene delivery for treating genetic diseases. Currently, the three main gene editing techniques used employ zinc finger nucleases, transcription activator-like effector nucleases, and the clustered regularly interspaced short palindromic repeat/CRISPR-associated proteins 9 system (CRISPR/Cas9), with the latter being the most advanced technology available at present. The pathogenic mechanisms so far characterized in AD, PD, and HD are primarily associated with the generation of A\u03b242, phosphorylated tau, \u03b1-syn, and mHTT aggregation, respectively. However, the most effective current treatment measures for these NDs only result in temporary symptom relief, and improvements to disease-modifying therapies against the production, accumulation, and toxicity of A\u03b242, phosphorylated tau, \u03b1-syn, and mHTT have yet to achieve satisfactory results. CRISPR/Cas9 gene editing is gaining significant attention from researchers due to its potential benefits in managing and treating AD, HD, and PD (Bhardwaj et al., 2022). While gene editing is typically used for gene overexpression and knockout, CRISPR/Cas9 can also be applied to alter single nucleotides, modify DNA domains, or modulate the expression of multiple genes. The CRISPR/Cas9 guiding domain can also excise pathogenic mutations or small DNA fragments during nonsense mRNA decay. Additionally, transcription repressors or enhancers attached to the catalytic dead Cas9 can target specific genomic sites using guiding RNAs, resulting in the inhibition or activation of genes. The gene-editing capabilities of CRISPR/Cas9 can expand the potential for genetic manipulation beyond traditional gene therapies, which are typically used to modify a single gene in monogenic hereditary diseases (Sun and Roy, 2021). Although several issues related to the CRISPR/Cas9 system remain unresolved, such as off-target effects, cytotoxicity from DNA lesions, chromosomal rearrangements, and the frequent activation of p53 (Hendriks et al., 2020; Bravo et al., 2022), this strategy holds promise for treating incurable NDs through the targeted editing of pathogenic genes (Table 1) and the subsequent elimination of pathogenic proteins (Table 2).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Clinical Translation Based on Pathogenic Mechanisms", "text": "Different proteins and their associated neurodegenerative diseases\n\nAD: Alzheimer\u2019s disease; ALS: amyotrophic lateral sclerosis; ApoE: apolipoprotein E; APP: amyloid precursor protein; A\u03b2: amyloid-beta; C9ORF72: chromosome 9 open-reading frame 72 TBK-1: tank binding kinase 1; CHMP2B: charged multivesicular body protein 2B; FTD: frontotemporal lobar dementia; FUS: fused in sarcoma; HD: Huntington\u2019s disease; HSPs: heat-shock proteins; IFITM3: Interferon-induced transmembrane protein 3; LBD: Lewy body disease; LRP1: low-density lipoprotein receptor-related protein 1; NEK1: NIMA-related kinase 1; parkin E3: Parkin E3 ligase; PD: Parkinson\u2019s disease; PINK1: PTEN-induced putative kinase I; PSEN1: presenilin 1; PSEN2: presenilin 2; SOD1: superoxide dismutase 1; TDP-43: 43-kDa TAR DNA-binding protein; \u03b1-Syn: \u03b1-synuclein.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Clinical Translation Based on Pathogenic Mechanisms", "text": "Emerging research suggests that the activation of the immune system is one of the most important pathogenic mechanisms in NDs. Research has demonstrated that conditions such as AD (Mary et al., 2024), PD (Morris et al., 2024), ALS, HD, FTD, and LBD involve immune system activation. The release of inflammatory factors leads to the characteristic pathological alterations of NDs, including abnormal proteostasis, the aggregation of pathogenic proteins, cytoskeletal abnormalities, alterations in energy homeostasis, RNA and DNA deficiencies, and synaptic and network dishomeostasis, ultimately resulting in neuronal cell death (Castro-Gomez and Heneka, 2024). Currently, AD lacks effective treatments and drug therapies, and clinical approaches only provide temporary relief from symptoms. An emerging effective treatment for AD is the transplantation of mesenchymal stem cells (MSCs), which has immunoregulation potential, in addition to apoptotic or degenerative neuron replacement. MSC transplantation can be used to modulate the activation of microglia and astrocytes, helping to prevent neuroinflammation through various transcription factor signaling pathways. The immunoregulation associated with MSC treatment may play a major role in its therapeutic effects in AD (Zhang et al., 2020).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Clinical Translation Based on Pathogenic Mechanisms", "text": "Both oxidative stress and neuroinflammation are critical factors in the pathogenesis of NDs. Therapeutics aimed at reducing neuronal damage caused by oxidative free radicals and anti-inflammatory treatments, such as non-steroidal anti-inflammatory drugs, may help slow disease progression. However, these treatments can have serious side effects, including gastrointestinal, cardiovascular, and renal complications. Furthermore, their efficacy in treating NDs remains a subject of debate (Gong et al., 2018). Neurotrophic drugs designed to support and protect the survival and function of neurons are also being explored as potential therapies for NDs. However, effective delivery of neurotrophic factors to damaged brain regions remains a significant challenge. Additionally, the long-term efficacy and safety of these treatments need to be extensively evaluated (Pern\u00e8gre et al., 2019).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Clinical Translation Based on Pathogenic Mechanisms", "text": "Emerging evidence regarding gut microbiota and its role in the pathogenesis of AD suggests that regulating microbial dysbiosis is a promising target for treatment. Strategies to modify the microbiome to restore gut homeostasis, such as probiotics, prebiotics, synbiotics, postbiotics, and fecal microbiota transplantation, are being explored (Zhang et al., 2023). A\u03b2 has long been a primary focus in the development of passive immunotherapies for AD (Karran et al., 2011). When the Fab regions of anti-A\u03b2 antibodies bind to A\u03b2 in the brain, the Fc regions can interact with Fc gamma receptors on microglial surfaces, activating microglial phagocytosis (Wilcock et al., 2004). However, anti-A\u03b2 monoclonal antibodies can alter the status of microglia, shifting them from anti-inflammatory M2-like phenotypes to pro-inflammatory M1-like phenotypes, which may lead to neuroinflammation and decreased microglial ability to clear A\u03b2 (Wilcock et al., 2011). Given that Fc gamma receptors are categorized into both activating and inhibitory types, the affinities of the Fc domains of anti-A\u03b2 monoclonal antibodies for different Fc gamma receptor types should be considered in antibody design (Fuller et al., 2014). Moreover, when anti-A\u03b2 monoclonal antibodies are not present at optimal levels in the brain, A\u03b2 monoclonal antibody compounds can bind to C1q and be taken up by microglial cells through the C1q receptor (Webster et al., 2001). Despite numerous challenges and failures in developing effective anti-A\u03b2 immunotherapies, several monoclonal antibodies have shown promising efficacy in treating AD, particularly concerning antibody epitopes and targeted A\u03b2 formation. Notable monoclonal antibodies such as gantenerumab, aducanumab, lecanemab, and donanemab exhibit strong scavenging capabilities for amyloid plaques (Guo et al., 2024).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Clinical Translation Based on Pathogenic Mechanisms", "text": "Immunoregulatory therapeutics, particularly monoclonal antibodies, have emerged as promising strategies to target pathological proteins in NDs. In the case of AD, amyloid-directed antibodies may help reduce the aggregation of A\u03b2, thereby delaying cognitive decline (Guo et al., 2024). Research and development of these antibodies targeting pathogenic proteins in NDs have accelerated in recent years. Although clinical trials assessing the efficacy and safety of monoclonal antibodies remain inconclusive, the potential of these strategies warrants further investigation for the treatment of NDs (Cummings et al., 2019; Mahboob et al., 2024). Monoclonal antibodies currently undergoing clinical trials for PD include Cinpanemab, Prasinezumab, UCB7853, LUAF82422, PRX002, and TAK-341/MEDI1341 (Mahboob et al., 2024).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Clinical Translation Based on Pathogenic Mechanisms", "text": "MSCs present a lower risk of tumorigenicity and immunogenicity, and their tissue sources are readily accessible. These advantages make MSCs an optimal option for treating NDs through neuroregeneration, particularly when compared to other neural stem cells such as induced pluripotent stem cells (Regmi et al., 2022). MSCs are considered promising therapies for NDs due to their regenerative properties and their ability to secrete growth factors, exert anti-inflammatory effects, block apoptosis, and stimulate neurogenesis, synaptogenesis, and angiogenesis (Mattei and Delle Monache, 2024). Their anti-apoptotic and antioxidant properties may help prevent neuronal cell death. Furthermore, MSCs enhance neurogenesis by secreting growth factors such as glial cell line-derived neurotrophic factor, brain-derived neurotrophic factor, vascular endothelial growth factor, hepatocyte growth factor, nerve growth factor, and neurotrophin-3. These factors can migrate directly to damaged brain regions and promote the differentiation of neural cells, including neurons, thereby facilitating neuritic formation, neurorestoration, and the recovery of damaged functions (Trinh et al., 2024).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Limitations", "text": "While the article addresses various pathogenic mechanisms of NDs, it lacks an in-depth analysis of their specific roles across different diseases. The focus on summarizing the potential common pathogenesis of six NDs limits a more detailed exploration and comparison of these mechanisms. As a result, the discussion of each pathogenic mechanism does not delve deeply into its specific role in individual diseases. For instance, the article does not thoroughly analyze the different types and distributions of misfolded and aggregated proteins among various diseases, nor does it address their impact on neuronal functions. Additionally, the review does not expand on the specific role of each pathogenic mechanism in different diseases.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Discussion and Conclusions", "text": "Several common NDs pose a serious threat to human health in modern society due to the lack of effective therapeutics. However, with the rapid advancement of research technologies, diverse studies are helping to better understand the fundamental pathogenesis of these diseases and identify new targets for intervention.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Discussion and Conclusions", "text": "In this review, the authors examine the relationships and potential roles of aging, genetics, progressive neuronal dysfunction and death, protein misfolding, aggregation and conformation, propagation of protein aggregation, neuroinflammation, mitochondrial dysfunctions, axonal transport defects, and gut microbiota/dysbiosis in six common NDs, including AD, PD, ALS, HD, FTD, and LBD. The accumulating evidence suggests that the loss of homeostasis and/or dysfunction of these biological processes represents a common pathogenesis among the six NDs. These processes act as bridges linking the different pathologies, contributing to distinct neuronal degeneration in the CNS through similar biological mechanisms. The authors propose that different NDs may be induced by multiple overlapping biological processes, indicating that these processes are critical but not the sole contributors to the pathogenesis of these diseases. This perspective is supported by the observation that clinical treatments targeting a single biological process have not yielded ideal outcomes. Therefore, the authors suggest that the pathogenesis of these NDs may exist within a pathogenic network composed of multiple biological processes, with the currently identified pathophysiological mechanisms playing a relatively small and limited role within this larger network. Based on recent findings, the exploration of the pathogenesis of ND should focus on identifying more pathogenic targets and conducting in-depth analyses of the networks formed by these different targets. By detailing the pathogenic networks associated with these diseases and examining their commonalities and differences, researchers may uncover clues to better understand the complex and multifaceted nature of NDs.", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Discussion and Conclusions", "text": "Based on the current review findings from various researchers, we conclude that although our literature retrieval encompasses a range of biological processes, nearly all pathogenic processes associated with six NDs involve pathways leading to the death of neural cells. These pathogenic mechanisms suggest several avenues for promoting nerve regeneration by addressing the following aspects: antagonizing aging, regulating genes related to genetics, reversing progressive neuronal dysfunction, death, and apoptosis, preventing protein misfolding, aggregation, and conformational changes, halting the propagation of protein aggregation, eliminating neuroinflammation, recovering mitochondrial dysfunction, repairing axonal transport defects, and managing gut microbiota dysbiosis. A significant challenge in this field remains the lack of effective measures and medications to prevent and reverse the pathogenic processes associated with NDs. We believe that treatment for NDs could greatly improve, and potentially lead to cures, if effective strategies can be developed to target all aspects of their common pathogenesis. The novelty of this review lies in its examination of the shared potential pathogenesis across different NDs, aiming to uncover new insights into their underlying mechanisms by analyzing the current findings related to these six prevalent NDs. The significance of this review is that it first highlights the potential common pathogenesis of these six different NDs, which may reveal shared features in their mechanisms and provide valuable information and clues for further research into the novel pathogenesis of both multiple and individual NDs. Future recommendations from experts in this field emphasize the need for a deeper exploration and investigation of the synthesized pathogenic networks and the mechanisms through which these processes interact, ultimately leading to neuronal cell death in NDs. We believe that these six NDs are driven by multiple pathogenic pathways, molecular targets, and mechanisms, making it unlikely that they are solely caused by isolated biological processes. This review aims to share our insights derived from our research experiences and clinical practice, encouraging other researchers to further investigate the novel pathogenesis of NDs based on the direction we provide. We also wish to disseminate our opinions to the broader research community in this field. Additionally, we have conducted related studies using several experimental mouse models of NDs, aiming to observe behavioral and lifespan changes by regulating the potential common pathogenesis of these diseases. Our goal is to identify shared pathogenic mechanisms and explore the networks associated with the complex biological processes involved in NDs. So far, we have obtained some intriguing and valuable results, but further experiments are necessary to validate our findings. We are confident in our conclusions and hope that this review will draw significant attention from other researchers. Ultimately, we aim to contribute our ideas to the study of ND pathogenesis and provide crucial insights and evidence that could lead to potential cures for these diseases (Figure 3).", "title": "" }, { "pmid": 40522761, "pmcid": "PMC12296512", "section": "Discussion and Conclusions", "text": "Common biological processes in the pathogenesis of six neurodegenerative diseases including AD, PD, ALS, HD, FTD, and LBD.\n\nThese biological processes include aging, genetics, progressive neuronal dysfunction and death, protein misfolding, aggregation and conformational changes, propagation of protein aggregation, neuroinflammation, mitochondrial dysfunction, axonal transport defects, and gut microbiota dysbiosis. Each of these factors may contribute to neurodegeneration through multiple pathways in the pathogenesis of NDs. We propose that these processes interact synergistically to form a pathogenic network that drives neurodegeneration in NDs. AD: Alzheimer\u2019s disease; ALS: amyotrophic lateral sclerosis; FTD: frontotemporal lobar dementia; HD: Huntington\u2019s disease; LBD: Lewy body disease; PD: Parkinson\u2019s disease.", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "Introduction", "text": "The Clinical Trials Update is a regular feature devoted to highlighting ongoing and recently completed clinical trials in Huntington's disease (HD). Clinical trials previously reviewed in this section are listed in Table 1. For current and completed studies discussed here, endorsement status by the European Huntington's Disease Network (EHDN) is now also indicated.\n\nClinical trials previously reviewed by the Huntington's disease clinical trials update.\n\nIONIS-HTTRx, RG6042 and tominersen refer to the same molecule.\n\nVX15/2503 and pepinemab refer to the same molecule.\n\nAAVrh10.CAG.hCYP46A1, BV-101 and AB-1001 refer to the same molecule.", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "Introduction", "text": "In this edition, we expand on the launch of the phase II/III clinical trial of SKY-0515 (NCT06873334)\n13\n (FALCON-HD) from Skyhawk Therapeutics and the phase I/II clinical trial of SPK-10001 (NCT06826612)\n14\n from Spark Therapeutics. We also report positive topline data from uniQure's phase I/II clinical trial of AMT-130 (NCT04120493; NCT05243017)15,16 after 36 months of follow-up. Further updates include recent developments in Roche's tominersen programme within GENERATION HD2 (NCT05686551),\n17\n progress with votoplam (PTC518) in PIVOT-HD (NCT05358717; NCT06254482)18,19 by PTC Therapeutics and developments in the collaborative PTC Therapeutics/Novartis programme. We additionally discuss regulatory developments regarding pridopidine following the negative PROOF-HD (NCT04556656)\n20\n study. Finally, we provide an updated listing of all registered and ongoing clinical trials in Huntington's disease.\n\nWe tabulate all currently registered and ongoing clinical trials in Tables 2\u20134. For further details on the methodology used, please refer to the first edition of this series.\n1\n\nPharmacological clinical trials registered at the World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP) for people with Huntington's disease (HD) since the first edition of the \u201cHuntington's Disease Clinical Trials Corner\u201d.", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "Introduction", "text": "AAV: adeno-associated virus; AMP: adenosine monophosphate; AMPK: AMP-activated protein kinase; ASO: antisense oligonucleotide; HD: Huntington's disease; HD-ISS: Huntington's Disease Integrated Staging System; HTT: huntingtin; MAPK: mitogen-activated protein kinase; mHTT: mutant huntingtin; miRNA: microRNA; MRI: magnetic resonance imaging; mRNA: messenger ribonucleic acid; NAC: N-acetylcysteine; NAD+: nicotinamide adenine dinucleotide (oxidised form); NMDA: N-methyl-D-aspartate; N/S: not specified; PD: Parkinson's disease; PMS1: post-meiotic segregation 1; PPAR\u03b1: peroxisome proliferator-activated receptor alpha; RNA: ribonucleic acid; SCA1: spinocerebellar ataxia 1; SCA3: spinocerebellar ataxia 3; TD: tardive dyskinesia; TEAEs: treatment-emergent adverse events; TFC: Total Functional Capacity; UHDRS: Unified Huntington's Disease Rating Scale; VMAT2: vesicular monoamine transporter 2. Note: IONIS-HTTRx, ISIS 443139, RG6042 and tominersen refer to the same molecule. New trials added since the last Clinical Trials Update are indicated by *. Pharmacological trials terminated are indicated by \u2020.\n\nInvasive non-pharmacological clinical trials registered at the WHO ICTRP for people with HD since the first edition of the \u201cHuntington's Disease Clinical Trials Corner\u201d.", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "Introduction", "text": "AD: Alzheimer's disease; CBD: corticobasal degeneration; DBS: deep brain stimulation; ET: essential tremor; GP: globus pallidus; GPi: globus pallidus internus; HD: Huntington's disease; HT: Holmes tremor; ICTRP: International Clinical Trials Registry Platform; MNC: mononuclear cells; MR: magnetic resonance; MS: multiple sclerosis; N/S: not specified; RNA: ribonucleic acid; PD: Parkinson's disease; TD: tardive dyskinesia; WD: Wilson's disease; WHO: World Health Organization.\n\nNon-invasive non-pharmacological clinical trials registered at the WHO ICTRP for people with HD since the first edition of the \u201cHuntington's Disease Clinical Trials Corner\u201d.", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "Introduction", "text": "3D: three-dimensional; AD: Alzheimer's disease; ALS: amyotrophic lateral sclerosis; BOLD: blood oxygen level-dependent; DSCT: Digit Symbol Coding Test; EEG: electroencephalography; ET: essential tremor; FAB: Frontal Assessment Battery; fMRI: functional magnetic resonance imaging; HD: Huntington's disease; HT: Holmes tremor; ICTRP: International Clinical Trials Registry Platform; mHTT: mutant huntingtin; MRI: magnetic resonance imaging; MS: multiple sclerosis; N/S: not specified; PBA-s: Problem Behaviors Assessment - short; PD: Parkinson's disease; PET: positron emission tomography; SAGE: Self-Administered Gerocognitive Examination; SDMT: Symbol Digit Modalities Test; SV2A: synaptic vesicle glycoprotein 2A; SWR: Stroop Word Reading; TD: tardive dyskinesia; tDCS: transcranial direct current stimulation; TRE: time-restricted eating; VT: volume of distribution; WHO: World Health Organization. New trials since the last Clinical Trials Update are indicated by *.\n\nIf you would like to draw attention to specific trials, please feel free to email us at: mena.farag@ucl.ac.uk and e.wild@ucl.ac.uk.", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "Ongoing clinical trials", "text": "A list of all registered clinical trials is given in Tables 2\u20134.\n\nStudy Title: Study of SKY-0515 for Safety, Efficacy and Pharmacodynamics in Participants with Huntington's Disease (FALCON-HD).\n\nIntervention: Oral SKY-0515 (low, mid, or high dose) or matched placebo, administered once daily for 12 months. SKY-0515 is an mRNA-splicing modulator designed to reduce both wild-type HTT and mutant huntingtin (mHTT) protein by lowering HTT mRNA levels, and to suppress post-meiotic segregation 1 (PMS1) protein, a mediator of somatic CAG repeat expansion and known driver of HD progression.13,21,22", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "Ongoing clinical trials", "text": "Description: The FALCON-HD study, sponsored by Skyhawk Therapeutics, Inc., is a phase II/III randomised, double blind, placebo-controlled, dose ranging clinical trial to evaluate the pharmacodynamics, safety and efficacy of SKY-0515 in people with HD. This multi-centre clinical trial will enrol 120 adults (\u226525 years) with genetically confirmed HD (CAG repeat length \u226540) who meet defined functional and motor thresholds (specifically Unified Huntington's Disease Rating Scale (UHDRS) Total Functional Capacity (TFC)\u2009\u2265\u200910; Total Motor Score (TMS)\u2009\u2265\u20096). The clinical trial design comprises a screening period lasting up to four weeks, a 12-month double-blind treatment period and a four-week follow-up period thereafter. Participants will be randomised (1:1:1:1) to one of three once-daily dose levels of SKY-0515 or placebo. Safety and efficacy will be assessed through regular clinical trial visits, laboratory tests, brain MRI and standardised clinical assessments, including the UHDRS.\n\nPrimary outcome measures are change from baseline to 12 months in blood mHTT protein levels, MRI-derived brain volumes and UHDRS scores. Other outcome measures include changes in other HD-related biomarkers and long-term safety of SKY-0515.", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "Ongoing clinical trials", "text": "Comments: On 9th July 2024, Skyhawk Therapeutics announced positive results from its phase I trial of SKY-0515 (ACTRN12623001161617)\n21\n in healthy volunteers, initiated in late 2023.\n22\n SKY-0515 produced dose-dependent reductions in HTT mRNA, with maximum decreases of 67% after a single 16\u2005mg dose and 72% after 14 days of 9\u2005mg daily dosing.\n22\n Reductions were observed across all dose levels, and the drug was generally well-tolerated. On 17th September 2025, Skyhawk Therapeutics reported further positive findings from the first interim analysis of this phase 1 trial.\n23\n At day 84, treatment with SKY-0515 resulted in dose-dependent reductions of mHTT protein level in blood, including a 29% reduction at the 3\u2005mg dose and a 62% reduction at the 9\u2005mg dose. SKY-0515 was generally well tolerated at both dose levels.\n23\n\nOn 20th June 2025, Skyhawk Therapeutics announced that the first patient was dosed in the phase II/III FALCON-HD study (NCT06873334)\n13\n in Adelaide, Australia,\n24\n marking the initiation of the registrational trial programme across multiple sites in Australia and New Zealand.", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "Ongoing clinical trials", "text": "SKY-0515 is designed to reduce both HTT and PMS1 proteins, the latter implicated in somatic CAG repeat expansion, a mechanistic process increasingly recognised as a key driver of neuronal loss in HD.25,26 Genome-wide association and mismatch repair studies indicate that targeting PMS1 and other DNA repair genes could help prevent progressive neuronal pathogenesis in striatal and cortical regions.27,28 This important phase II/III clinical trial of SKY-0515 will assess whether HTT and PMS1 lowering can offer greater therapeutic benefit than HTT reduction alone in early-stage HD.\n\nStudy Title: A Randomized Study of SPK-10001 Gene Therapy in Participants with Huntington's Disease.\n\nIntervention: One-time bilateral intraparenchymal infusion of SPK-10001 (also known as RG6662) into the caudate and putamen, compared with placebo surgery.\n14\n SPK-10001 is an engineered adeno-associated virus (AAV) vector delivering an artificial microRNA designed to selectively target human HTT mRNA.\n29", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "Ongoing clinical trials", "text": "Description: The SPK-10001 clinical trial, sponsored by Spark Therapeutics, Inc., is a phase I/II multicentre, randomized, sequential, dose-escalation study evaluating the safety, tolerability, and preliminary efficacy of SPK-10001, an investigational gene therapy, in people with HD. Eligible participants receive a one-time bilateral intraparenchymal infusion of SPK-10001 into the caudate and putamen, or undergo a placebo surgical procedure. The clinical trial is structured in four sequential parts: an initial open-label cohort (Part A) in which all participants receive SPK-10001; a randomized, double-blind, placebo-surgery-controlled cohort (Part B); an open-label crossover (Part C) in which participants initially assigned to placebo surgery receive SPK-10001; and a long-term follow-up phase (Part D) for all participants following completion of active treatment. Up to 53 adults (aged 25\u201365 years) with genetically confirmed HD (CAG repeat length \u226540), evidence of striatal atrophy and preserved functional capacity (defined as UHDRS TFC \u226511) will be enrolled into the clinical trial.", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "Ongoing clinical trials", "text": "Primary outcome measures are safety and tolerability, measured by the incidence and severity of treatment-emergent adverse events (TEAEs) over a five-year follow-up period, and change from baseline in UHDRS TFC score at 24 months. Secondary outcome measures include changes from baseline in motor progression assessed by digital motor testing at 12, 18 and 24 months; changes in composite UHDRS (cUHDRS) scores at 12, 18 and 24 months; and changes in UHDRS TFC scores at 12 and 18 months.", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "Ongoing clinical trials", "text": "Comments: Data on SPK-10001 were presented at the 31st Annual Meeting of the Huntington Study Group (HSG), held in Cincinnati, Ohio from 7th to 9th November 2024, and at the Cure Huntington's Disease Initiative (CHDI) 20th Annual HD Therapeutics Conference, held in Palm Springs, California from 24th to 27th February 2025.29,30 Preclinical studies of SPK-10001, an engineered AAV vector delivering an artificial microRNA targeting human HTT mRNA, demonstrated a consistent, dose-dependent reduction of HTT mRNA and protein in HD mouse models following direct intrastriatal injection, compared with untreated controls. Reductions of up to 40% in HTT mRNA and up to 70% in HTT protein were well-tolerated and remained stable over a 12-month period. In contrast to uniQure's AMT-130 gene therapy, SPK-10001 targets an RNA sequence downstream of exon 1, promising valuable insights into the relative importance of the exon 1a splice variant as opposed to full length mHTT. In nonhuman primate studies, robust and consistent suppression of HTT protein throughout the caudate and putamen was achieved only through direct intraparenchymal administration, whereas intra cerebrospinal fluid (CSF) routes (intraventricular, intrathecal or intracisternal) failed to achieve satisfactory target engagement in these regions. Safety data in cynomolgus macaques indicated that the surgical procedure was generally well-tolerated with minimal perioperative immunosuppression. Transient elevations in CSF neurofilament light chain (NfL) protein were observed in all animals but returned to baseline within 12 months. Levels of HTT suppression were stable over the same period and highly consistent within dose groups.\n29\n On 12th June 2025, Spark Therapeutics and Roche issued a joint community update via a letter announcing that the first participant had been dosed with RG6662, also known as SPK-10001, and that the programme will be fully transitioned to Roche.\n31", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "Ongoing clinical trials", "text": "These findings and the update from Spark Therapeutics and Roche represent an important step in the development and progress of gene therapy for HD, particularly in light of the news below from the first gene therapy to be tested in HD patients. Early preclinical data support target engagement and tolerability of SPK-10001; however, safety, efficacy and clinical benefit in humans remain to be established. Long-term monitoring will be essential to determine the durability of any therapeutic effects, ongoing safety and potential functional impact as results from the ongoing phase I/II clinical trial become available.", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "SKY-0515 (", "text": "Study Title: Study of SKY-0515 for Safety, Efficacy and Pharmacodynamics in Participants with Huntington's Disease (FALCON-HD).\n\nIntervention: Oral SKY-0515 (low, mid, or high dose) or matched placebo, administered once daily for 12 months. SKY-0515 is an mRNA-splicing modulator designed to reduce both wild-type HTT and mutant huntingtin (mHTT) protein by lowering HTT mRNA levels, and to suppress post-meiotic segregation 1 (PMS1) protein, a mediator of somatic CAG repeat expansion and known driver of HD progression.13,21,22", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "SKY-0515 (", "text": "Description: The FALCON-HD study, sponsored by Skyhawk Therapeutics, Inc., is a phase II/III randomised, double blind, placebo-controlled, dose ranging clinical trial to evaluate the pharmacodynamics, safety and efficacy of SKY-0515 in people with HD. This multi-centre clinical trial will enrol 120 adults (\u226525 years) with genetically confirmed HD (CAG repeat length \u226540) who meet defined functional and motor thresholds (specifically Unified Huntington's Disease Rating Scale (UHDRS) Total Functional Capacity (TFC)\u2009\u2265\u200910; Total Motor Score (TMS)\u2009\u2265\u20096). The clinical trial design comprises a screening period lasting up to four weeks, a 12-month double-blind treatment period and a four-week follow-up period thereafter. Participants will be randomised (1:1:1:1) to one of three once-daily dose levels of SKY-0515 or placebo. Safety and efficacy will be assessed through regular clinical trial visits, laboratory tests, brain MRI and standardised clinical assessments, including the UHDRS.\n\nPrimary outcome measures are change from baseline to 12 months in blood mHTT protein levels, MRI-derived brain volumes and UHDRS scores. Other outcome measures include changes in other HD-related biomarkers and long-term safety of SKY-0515.", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "SKY-0515 (", "text": "Comments: On 9th July 2024, Skyhawk Therapeutics announced positive results from its phase I trial of SKY-0515 (ACTRN12623001161617)\n21\n in healthy volunteers, initiated in late 2023.\n22\n SKY-0515 produced dose-dependent reductions in HTT mRNA, with maximum decreases of 67% after a single 16\u2005mg dose and 72% after 14 days of 9\u2005mg daily dosing.\n22\n Reductions were observed across all dose levels, and the drug was generally well-tolerated. On 17th September 2025, Skyhawk Therapeutics reported further positive findings from the first interim analysis of this phase 1 trial.\n23\n At day 84, treatment with SKY-0515 resulted in dose-dependent reductions of mHTT protein level in blood, including a 29% reduction at the 3\u2005mg dose and a 62% reduction at the 9\u2005mg dose. SKY-0515 was generally well tolerated at both dose levels.\n23\n\nOn 20th June 2025, Skyhawk Therapeutics announced that the first patient was dosed in the phase II/III FALCON-HD study (NCT06873334)\n13\n in Adelaide, Australia,\n24\n marking the initiation of the registrational trial programme across multiple sites in Australia and New Zealand.", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "SKY-0515 (", "text": "SKY-0515 is designed to reduce both HTT and PMS1 proteins, the latter implicated in somatic CAG repeat expansion, a mechanistic process increasingly recognised as a key driver of neuronal loss in HD.25,26 Genome-wide association and mismatch repair studies indicate that targeting PMS1 and other DNA repair genes could help prevent progressive neuronal pathogenesis in striatal and cortical regions.27,28 This important phase II/III clinical trial of SKY-0515 will assess whether HTT and PMS1 lowering can offer greater therapeutic benefit than HTT reduction alone in early-stage HD.", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "SPK-10001 (", "text": "Study Title: A Randomized Study of SPK-10001 Gene Therapy in Participants with Huntington's Disease.\n\nIntervention: One-time bilateral intraparenchymal infusion of SPK-10001 (also known as RG6662) into the caudate and putamen, compared with placebo surgery.\n14\n SPK-10001 is an engineered adeno-associated virus (AAV) vector delivering an artificial microRNA designed to selectively target human HTT mRNA.\n29", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "SPK-10001 (", "text": "Description: The SPK-10001 clinical trial, sponsored by Spark Therapeutics, Inc., is a phase I/II multicentre, randomized, sequential, dose-escalation study evaluating the safety, tolerability, and preliminary efficacy of SPK-10001, an investigational gene therapy, in people with HD. Eligible participants receive a one-time bilateral intraparenchymal infusion of SPK-10001 into the caudate and putamen, or undergo a placebo surgical procedure. The clinical trial is structured in four sequential parts: an initial open-label cohort (Part A) in which all participants receive SPK-10001; a randomized, double-blind, placebo-surgery-controlled cohort (Part B); an open-label crossover (Part C) in which participants initially assigned to placebo surgery receive SPK-10001; and a long-term follow-up phase (Part D) for all participants following completion of active treatment. Up to 53 adults (aged 25\u201365 years) with genetically confirmed HD (CAG repeat length \u226540), evidence of striatal atrophy and preserved functional capacity (defined as UHDRS TFC \u226511) will be enrolled into the clinical trial.", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "SPK-10001 (", "text": "Primary outcome measures are safety and tolerability, measured by the incidence and severity of treatment-emergent adverse events (TEAEs) over a five-year follow-up period, and change from baseline in UHDRS TFC score at 24 months. Secondary outcome measures include changes from baseline in motor progression assessed by digital motor testing at 12, 18 and 24 months; changes in composite UHDRS (cUHDRS) scores at 12, 18 and 24 months; and changes in UHDRS TFC scores at 12 and 18 months.", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "SPK-10001 (", "text": "Comments: Data on SPK-10001 were presented at the 31st Annual Meeting of the Huntington Study Group (HSG), held in Cincinnati, Ohio from 7th to 9th November 2024, and at the Cure Huntington's Disease Initiative (CHDI) 20th Annual HD Therapeutics Conference, held in Palm Springs, California from 24th to 27th February 2025.29,30 Preclinical studies of SPK-10001, an engineered AAV vector delivering an artificial microRNA targeting human HTT mRNA, demonstrated a consistent, dose-dependent reduction of HTT mRNA and protein in HD mouse models following direct intrastriatal injection, compared with untreated controls. Reductions of up to 40% in HTT mRNA and up to 70% in HTT protein were well-tolerated and remained stable over a 12-month period. In contrast to uniQure's AMT-130 gene therapy, SPK-10001 targets an RNA sequence downstream of exon 1, promising valuable insights into the relative importance of the exon 1a splice variant as opposed to full length mHTT. In nonhuman primate studies, robust and consistent suppression of HTT protein throughout the caudate and putamen was achieved only through direct intraparenchymal administration, whereas intra cerebrospinal fluid (CSF) routes (intraventricular, intrathecal or intracisternal) failed to achieve satisfactory target engagement in these regions. Safety data in cynomolgus macaques indicated that the surgical procedure was generally well-tolerated with minimal perioperative immunosuppression. Transient elevations in CSF neurofilament light chain (NfL) protein were observed in all animals but returned to baseline within 12 months. Levels of HTT suppression were stable over the same period and highly consistent within dose groups.\n29\n On 12th June 2025, Spark Therapeutics and Roche issued a joint community update via a letter announcing that the first participant had been dosed with RG6662, also known as SPK-10001, and that the programme will be fully transitioned to Roche.\n31", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "SPK-10001 (", "text": "These findings and the update from Spark Therapeutics and Roche represent an important step in the development and progress of gene therapy for HD, particularly in light of the news below from the first gene therapy to be tested in HD patients. Early preclinical data support target engagement and tolerability of SPK-10001; however, safety, efficacy and clinical benefit in humans remain to be established. Long-term monitoring will be essential to determine the durability of any therapeutic effects, ongoing safety and potential functional impact as results from the ongoing phase I/II clinical trial become available.", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "Breaking news", "text": "In this section we provide brief updates about ongoing or recently terminated clinical trials.", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "Breaking news", "text": "The AMT-130 (NCT04120493; NCT05243017)15,16 study, sponsored by uniQure Biopharma B.V., is investigating the effects of AMT-130, a one-time intracranially administered gene therapy that uses an AAV vector (AAV5-miHTT) to deliver a microRNA targeting the HTT gene. The treatment is intended to provide long-lasting supression of HTT expression. On 24th September 2025, uniQure released positive topline results from the phase I/II clinical trial of AMT-130.\n32\n The prespecified analysis included 29 treated participants (17 high-dose, 12 low-dose), with 12 patients per cohort reaching 36 months of follow-up. Outcomes were compared with propensity score-matched external controls from the Enroll-HD natural history dataset (n\u2009=\u2009940 for high dose; n\u2009=\u2009626 for low dose). At 36 months, high-dose AMT-130 met the primary endpoint, demonstrating a statistically significant 75% slowing of disease progression versus controls, as measured by the cUHDRS (P\u2009=\u20090.003; mean change from baseline \u22120.38 vs \u22121.52). A key secondary endpoint, UHDRS TFC, showed a 60% slowing of decline (P\u2009=\u20090.033; \u22120.36 vs \u22120.88). Additional secondary measures indicated favourable trends, including Symbol Digit Modalities Test (SDMT) (88% slowing; P\u2009=\u20090.057; \u22120.44 vs. \u22123.73), Stroop Word Reading (SWR) (113% slowing; P\u2009=\u20090.002; 0.88 vs. \u22126.98) and UHDRS TMS (59% slowing; P\u2009=\u20090.174; 2.01 vs. 4.88). Fluid biomarker analysis also showed a mean reduction of 8.2% in CSF NfL from baseline at 36 months, which provides supportive biomarker evidence of target engagement and disease modification. AMT-130 was overall well-tolerated, with an acceptable safety profile observed in both the high-dose and low-dose cohorts.\n32\n The sponsor reported that the consistently sustained improvements across functional, motor and cognitive outcomes at 36 months in the high-dose cohort, compared with the more variable findings at the lower dose, suggest a dose-dependent effect of AMT-130. Overall, the press release indicates that AMT-130 may have disease-modifying potential in HD and has the potential to meaningfully slow disease progression. The sponsor has also provisionally announced plans to submit a Biologics Licence Application to the US Food and Drug Administration in early 2026, with a potential US launch later that year pending regulatory approval.\n32\n Peer-reviewed publication is expected and will permit close scrutiny of all the data, but the topline results appear to be the most encouraging seen so far in a clinical trial of a therapy intended to deflect the course of HD.", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "Breaking news", "text": "The GENERATION HD2 (NCT05686551)\n17\n study, sponsored by Roche, released an update on 17th April 2025 regarding the tominersen programme.\n33\n Tominersen is an intrathecally administered antisense oligonucleotide (ASO) therapy designed to lower HTT expression. GENERATION HD2 is a phase II, randomised, double-blind, placebo-controlled, dose-finding clinical trial comparing 60\u2005mg tominersen, 100\u2005mg tominersen and placebo. In the most recent update provided by Roche, the Independent Data Monitoring Committee (IDMC) reported that tominersen continues to demonstrate an acceptable safety profile and recommended continuation of the trial, while advising discontinuation of the 60\u2005mg arm while continuing with the 100\u2005mg dose, which was judged more likely to provide clinical benefit.\n33\n This update is encouraging following the IDMC decision in March 2021 to halt the phase III GENERATION HD1 (NCT03761849)\n34\n study, due to an unfavourable benefit-risk profile.\n35\n In GENERATION HD2 (NCT05686551),\n17\n participants who had been receiving the 60\u2005mg dose will transition to the 100\u2005mg dose in a blinded manner, while those in the placebo group will remain in their original assignment. Although this update does not establish efficacy, it represents a data-driven refinement of the study design to prioritise the dose most likely to demonstrate therapeutic benefit. The clinical trial remains ongoing, and definitive conclusions regarding clinical benefit will only be possible once the data are unblinded and fully analysed. This is anticipated to be late 2026.", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "Breaking news", "text": "The PIVOT-HD (NCT05358717; NCT06254482)18,19 study, sponsored by PTC Therapeutics, released an update on 5th May 2025 regarding PTC518 (also known as votoplam), an orally administered huntingtin mRNA splice modulator.\n36\n The study met its primary endpoint, showing dose-dependent reductions in blood HTT protein levels at week 12, along with a favourable safety and tolerability profile across all dose levels and disease stages. Results from the full 12-month cohort show that HD-ISS stage 2 participants showed dose-dependent reductions in blood HTT protein (up to 39% with the 10\u2005mg dose) and emerging trends suggesting potential clinical benefit, including favourable changes on the composite UHDRS and TMS subscale. For HD-ISS stage 3 participants, more modest trends were observed, with apparent benefit in the 5\u2005mg group but less consistent effects in the 10\u2005mg group, raising the possibility that treatment response may differ by disease stage. At 24 months, data from an early cohort (n\u2009=\u200921) demonstrated dose-dependent reductions in plasma NfL levels (\u22128.9% at 5\u2005mg and \u221214% at 10\u2005mg) and favourable trends on cUHDRS, TFC and SDMT subscales, compared with a propensity-matched natural history cohort from the Enroll-HD registry. Importantly, for all dose levels and disease stages, PTC518 showed a favourable safety and tolerability profile with no treatment-related serious adverse events or NfL spikes observed. These findings support ongoing investigation of PTC518, which is currently being assessed in the open-label extension (NCT06254482).\n19\n While these results are promising and suggestive of early signs of clinical benefit, definitive evidence of efficacy, functional impact and long-term safety will require confirmation in the ongoing open-label extension and potential subsequent phase III study. On that note, Novartis released a community update on 15th July 2025 announcing a global licence and collaboration agreement with PTC Therapeutics for the development of votoplam (PTC518).\n37\n Under this agreement, Novartis will assume sponsorship of the long-term extension study and lead any future phase III development. Novartis has indicated that analysis of the phase II data is ongoing and that discussions with regulatory authorities, clinical experts and patient advocacy groups are underway to inform the design of the next phase of development.", "title": "" }, { "pmid": 41295902, "pmcid": "PMC12847458", "section": "Breaking news", "text": "On 25th July 2025, Prilenia and Ferrer announced that the European Medicines Agency's Committee for Medicinal Products for Human Use had recommended refusal of marketing authorisation for pridopidine in HD.\n38\n This decision follows the outcome of the phase III PROOF-HD (NCT04556656)\n20\n study, which enrolled 499 participants and evaluated a single daily dose of 45\u2005mg pridopidine versus placebo. The primary endpoint, change in UHDRS TFC score from baseline to week 65, was not met, indicating no significant slowing of disease progression over the 15-month treatment period.\n38", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "1. Introduction", "text": "Although gene and cell therapies long fell short of expectations, significant progress has been made in recent years. Kohn et al., in their article, present both successful and less effective examples of therapies and aim to reflect on why such a positive shift has become apparent only now [1]. To illustrate the drivers of this change, we begin the review with disease areas in which direct cell administration remains central to therapeutic efficacy. Therefore, the first part discusses examples of approaches in which cell administration is currently considered crucial\u2014for instance, in the treatment of type 1 diabetes or Parkinson\u2019s disease. In these cases, patient-specific cell therapies are being developed. These cells can undergo ex vivo genetic modifications to achieve selected traits [2,3].\n\nThe results of work on the development and improvement of transgene delivery systems increasingly enable solutions in which cell collection and ex vivo procedures are no longer required [4]. Therefore, the second part of the article discusses in vivo approaches that serve as an alternative to generating therapeutic cells outside the patient\u2019s body, including in the context of genetic disorders.", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "1. Introduction", "text": "The third part discusses cell therapies that use modified cells engineered to express synthetic genes, such as CAR (e.g., CAR-T cells). Synthetic biologists are working to design an increasingly broad repertoire of such constructs [5]. In this context, both in vivo and ex vivo approaches are explored. Gradually, a paradigm shift is taking place\u2014a move away from strictly ex vivo therapies in favor of in vivo approaches [6]. On the other hand, ex vivo strategies still allow for more extensive and controllable cellular modifications since comprehensive long-term safety evaluations of effects, including mutagenesis, immune-mediated reactions, and carcinogenesis of in vivo approaches, may only become evident after extended periods of observation.\n\nTwo main strategies can also be observed: one focuses on less extensively modified immune cells with CAR transgene only (e.g., CAR-NKT cells), while the other explores progressively broader genetic alterations of therapeutic cells. Certainly, the presented examples do not exhaust all recent achievements in the field; however, in the authors\u2019 opinion, it represents an important and illustrative selection.", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "1. Introduction", "text": "The current breakthrough in gene and cell therapies, particularly evident between 2023 and 2026, results from the convergence of several key technological advances that have overcome long-standing limitations in delivery, precision, scalability, and safety. By presenting a broad range of therapies for different patient populations across a wide spectrum of diseases, it is possible to explain this phenomenon. The acceleration of lipid nanoparticle (LNP) technologies\u2014initiated by the success of mRNA COVID-19 vaccines\u2014has enabled efficient in vivo delivery of transgenes and gene-editing tools to organs such as the liver and to cells of the immune system. These advances have enabled a transition from complex ex vivo procedures to simpler, more cost-effective, and more accessible in vivo applications, significantly reducing both the cost and logistical burden of therapy [7]. At the same time, improvements in CRISPR-derived technologies\u2014such as base editing (ABE) and prime editing\u2014have enabled safer and more precise genetic modifications without inducing double-strand DNA breaks. This has opened the door to therapies for rare diseases, including suppressor tRNA editing for nonsense mutations [8] and correction of mutations in genes such as NCF1 in chronic granulomatous disease [9]. Hypoimmunogenic engineering\u2014based on the disruption of B2M and CIITA genes combined with the overexpression of CD47\u2014has overcome immune rejection, enabling the development of universal, off-the-shelf allogeneic therapies. Examples include hypoimmunogenic CAR-NK cells for systemic sclerosis [10] and pancreatic islet transplantation in type 1 diabetes [11]. Finally, tools from synthetic biology, including tissue-specific promoters (e.g., Myo15 in otoferlin-related deafness therapy, where the transgene is delivered via dual AAV without CRISPR editing [12], and GfaABC1D in in vivo CAR expression in astrocytes [13]), as well as multigene modifications, have extended these approaches to complex tissues such as the central nervous system and the inner ear. The synergy of these elements\u2014previously absent due to limitations in delivery platforms, editing fidelity, and immunogenicity\u2014has driven a paradigm shift from autologous ex vivo therapies toward more accessible in vivo and universal strategies, thereby opening the path to broader clinical application [1,6,7,14]. Importantly, in 2025, the FDA approved a record number of gene therapies (more than ten), including label expansions for existing products such as Zolgensma, underscoring the maturity of the field and growing regulatory confidence [7]. Nevertheless, challenges such as off-target effects, durability of expression, manufacturing costs, and the risk of immunogenicity associated with novel editors continue to require further innovation, including the use of artificial intelligence to optimize vectors and editing strategies [14].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "2. Typical Ex Vivo Cell Therapies, Excluding CAR-T", "text": "The first diseases in which gene and cell therapies have been successfully applied in recent years are sickle cell disease and \u03b2-thalassemia. The erythroid-specific enhancer region of BCL11A in hematopoietic stem cells was modified using the CRISPR/Cas9. After transplantation, the defective erythrocytes were rapidly replaced by corrected erythrocytes with improved hemoglobin composition. The circumstances were relatively favorable because hematopoietic stem cells could be modified ex vivo [15]. Similar ex vivo strategies can be applied to various genetic disorders affecting blood or bone marrow cells. In the case of BEAM-101 therapy for sickle cell disease, instead of CRISPR/Cas9, a precise base editing technique (specifically A-to-G base editing) is used [16]. A related example is ex vivo gene therapy using a lentiviral vector to introduce a modified \u03b2-globin gene (\u03b2T87Q-globin) into the patient\u2019s hematopoietic stem cells. The therapy was approved by the FDA in December 2023 [17]. Another example of an ex vivo developed therapy is the treatment of patients with Wiskott\u2013Aldrich syndrome (WAS), a rare, X-linked genetic disorder, characterized by a classic triad of immunodeficiency caused by a mutation in the WAS gene. Using a lentiviral vector carrying WAS cDNA, corrected protein production was achieved in various cell types derived from hematopoietic stem/progenitor cells (CD34+) [18]. In some cases, therapeutic success depends on the classical administration of cells rather than gene correction. In one example, cardiomyocytes derived from induced pluripotent stem cells (iPSCs) were used (Figure 1A). This was not an autologous transplant and represents one of the few examples in which allogeneic therapy was chosen. The treatment was used as a temporary solution to \u201cpatch up\u201d the heart and stabilize cardiac function, allowing the patient to survive until a heart transplantation can be performed. Preclinical studies in macaques were successful and led to approval for the temporary use of cell therapy in a woman awaiting heart transplantation (bridge-to-transplant). Post-transplantation analysis of the \u201cpatched\u201d heart confirmed a positive effect of the cell therapy. However, immune-cell infiltration was also detected, indicating that this approach is likely to remain a temporary solution rather than a long-term effective therapy using allogeneic cardiac cells [19].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "2. Typical Ex Vivo Cell Therapies, Excluding CAR-T", "text": "Another situation in which cell therapy appears feasible without introducing genetic modification into these cells involves defects in microglia. According to Wu et al., such conditions can be referred to as microgliopathies [20]. Surprisingly positive therapeutic effects have been observed in all cases where healthy microglial cells acted beneficially within the central nervous system (CNS). In this example, the therapy concerned the brain disorder called Adult-Onset Leukoencephalopathy with Axonal Spheroids and Pigmented Glia associated with mutations in the CSF1R gene. In the CNS, the defective CSF1R protein is abundantly expressed (with high membrane density) in microglial cells. Mutations in both alleles of this gene are perinatally lethal, while a mutation in a single allele leads to a rare, essentially fatal disorder\u2014a \u201cmicrogliopathy\u201d. Therefore, a large proportion of the patient\u2019s defective microglial cells were replaced. To achieve this, it was necessary to ablate the patient\u2019s bone marrow and perform a transplant from a healthy donor. Over time, the affected cells in the central nervous system (CNS) underwent cell death and were replaced by corrected donor-derived microglial cells. In the past, attempts had been made to treat these patients using iPSCs in which the pathogenic mutation had been corrected with CRISPR technology. Now, according to Chadarevian et al., autologous transplantation of CSF1R-corrected microglia is not the only option; allogeneic CSF1R-wildtype iPSC-derived microglia can also reduce the diverse neuropathologies found in ALSP, without requiring the use of corrective CRISPR editing [21]. Mader et al. argue that the rescuing transplant can be allogeneic in part because the central nervous system (CNS) exhibits a certain degree of immunological tolerance. Nevertheless, immunosuppression may still be necessary [22].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "2. Typical Ex Vivo Cell Therapies, Excluding CAR-T", "text": "Implementation of bone marrow transplantation-based therapy also had a significant impact on positive long-term effects in patients with childhood-onset cerebral X-linked adrenoleukodystrophy. Before this therapeutic approach became available, patients typically survived only three to five years after the onset of clear symptoms [23]. If autologous bone marrow transplantation can be life-saving in certain CNS disorders, this represents a major advancement [24].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "2. Typical Ex Vivo Cell Therapies, Excluding CAR-T", "text": "An analogous approach has been applied in the treatment of type 1 diabetes, whereby the administered cells were not genetically modified (Figure 1D). Patients received islet-like cell surrogates derived from embryonic stem cells. In 10 out of 12 patients, the therapy yielded substantial clinical improvement. However, it remains unclear how durable the therapeutic effect will be, as this strategy does not address the underlying cause of the disease, which is the destruction of pancreatic islets by the patient\u2019s own autoreactive cytotoxic lymphocytes. After one year of observation, however, the patients\u2019 conditions remained stable, although they required maintenance of immunosuppressive therapy. Despite significant advances, differentiating stem cells into pancreatic cells is one of the most challenging processes [25]. An alternative approach involves harvesting cells from a deceased donor and subsequently making them \u201cuniversal.\u201d Donislecel, an allogeneic but non-modified pancreatic islet cell therapy derived from deceased donors, was designed to replace destroyed insulin-producing \u03b2 cells in adult patients with type 1 diabetes. In June 2023, it was approved by the FDA as the first cell-based therapy for type 1 diabetes in adults who, despite intensive insulin therapy, experience severe hypoglycemic episodes and have difficulty achieving target HbA1c levels. The mechanism of action involves the infusion of pancreatic islet cells into the hepatic portal vein, allowing them to engraft in the liver and secrete insulin in a physiologic manner. In some patients, this approach may reduce or even eliminate the need for continuous exogenous insulin administration [26]. In another project, complex genetic modifications were introduced to prevent the destruction of these cells by the patient\u2019s immune system. Interestingly, the cells were implanted into muscle tissue because pancreatic surgeries required for direct cell implantation are technically difficult. Interestingly, immunosuppression was not required, although this was an early-stage study involving a single patient. Using CRISPR technology, two HLA genes were knocked out in the donor cells to prevent T-cell\u2013mediated recognition: B2M (responsible for MHC I presentation) and CIITA (responsible for MHC II). However, cells lacking B2M and CIITA were still destroyed by natural killer (NK) cells and macrophages. To overcome this, an additional CD47 transgene was introduced into the donor cells using a lentiviral vector. Only cells combining knockout of B2M and CIITA with CD47 overexpression were protected from destruction by any components of the recipient\u2019s immune system. Following transplantation, diabetes-related parameters improved and stabilized, although insulin administration was not discontinued, but its dosage was reduced [11]. Pancreatic surgeries required for direct cell implantation are technically challenging, which reduces the therapeutic potential (for instance, due to a lack of response to paracrine intestinal hormones), but does not eliminate it. Importantly, all genetic modifications were performed ex vivo, and in the reported case, genetic alterations by lentiviral vector in oncological CAR-T therapies are classified as a precedent for safety [27,28]. Another team conducted similar preclinical research and successfully differentiated iPSCs into cells capable of producing and releasing insulin properly [29]. Similar genetic modifications were also performed in pancreatic cells [30]. This model will also be discussed in the context of CAR-T cells. Alternatively, CD4+ CD25highCD127 Treg-based therapy offers a promising immunotherapeutic strategy for type 1 diabetes patients. The Tregs+antiCD20 patient group was superior to the Standard of Care control group in terms of C-peptide secretion in the mixed-meal tolerance test, the time of insulin independence, and remission [31].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "2. Typical Ex Vivo Cell Therapies, Excluding CAR-T", "text": "Another example of ex vivo therapy application is chronic granulomatous disease (CGD) (Figure 1B and Figure 2). The most common cause of CGD is a two-nucleotide deletion (delGT) in the Neutrophil Cytosolic Factor 1 (NCF1) gene encoding the p47phox protein. Worth noticing is the complex genetic context of NCF1, which is flanked by highly homologous pseudogenes (NCF1B and NCF1C) naturally carrying the delGT variant (Figure 2) [32]. The absence of p47phox impairs host defense against pathogens, primarily bacteria [33]. CDG occurs with a frequency of approximately 1 in 150,000 births. The therapeutic approach involved autologous hematopoietic stem cells (HSCs) edited ex vivo using prime editing to restore the correct GTGT reading frame. This represents the first therapeutic use of prime editing, targeting the pathogenic delGT mutation in the NCF1 gene. A single dose of intravenously administered modified HSCs restored NADPH oxidase activity in 66% of neutrophils (Figure 2) [9].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "2. Typical Ex Vivo Cell Therapies, Excluding CAR-T", "text": "Other solutions have been proposed for recessive dystrophic epidermolysis bullosa (RDEB), a particularly severe and difficult-to-treat form of epidermolysis bullosa (Figure 1C). The skin of affected patients is sometimes compared to the delicate wings of a butterfly. Patient-derived cells obtained by biopsy were transduced with elements from a replication-deficient retroviral vector carrying the cDNA of the gene encoding type VII collagen (COL7A1), which is mutated in affected individuals. Next, the patients\u2019 wounds were covered with skin grafts composed of their own modified cells [34]. Among 43 wounds treated with a single application, 81% showed at least a 50% improvement in healing compared to approximately 15% of 43 matched control wounds treated with standard therapy (assessed after six months). In seven patients with a total of 38 wounds, a single surgical application of the therapeutic product resulted in long-term improvement at the treated sites, lasting on average nearly seven years.", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "3. Gene Therapies Without the Need for Ex Vivo Cell Modification", "text": "Recently, therapeutic approaches have emerged for more complex cases in which treatment does not necessarily require ex vivo cell modification. Notably, early clinical studies have already shown efficient transduction with sustained CAR expression and initial signs of antitumor activity due to T-cell\u2013tropic lentiviral vectors and LNP delivering CAR-encoding mRNA in vivo. This aims to facilitate manufacturing, reduce costs, and simplify complex logistics [35]. For example, progress has been made in developing therapies for certain genetic disorders of the CNS. One such condition is alternating hemiplegia of childhood (AHC), which affects approximately one child per million [36,37]. The disease is caused by mutations in the ATP1A3 gene, which encodes the \u03b13 subunit of the Na+/K+-ATPase transmembrane ion pump, causing highly complex neurological symptoms [38]. In a mouse model, postnatal AAV9-mediated correction of pathogenic ATP1A3 variants in the CNS was successful in approximately 50% of cortical neurons. AAV viral vectors are episomal; thus, they do not replicate with cell division, but still create a clinically important phenotype and extend lifespan [37]. A particularly striking case involved a boy with a rare metabolic disorder associated with liver dysfunction\u2014severe carbamoyl phosphate synthetase 1 (CPS1) deficiency [39] (Figure 3D). Previously, liver transplantation was a standard therapy for children with carbamoyl phosphate synthetase 1 (CPS1) deficiency or related urea cycle disorders [40]. Without transplantation, death was common; even when a donor was found, severe brain damage often occurred beforehand [1,41]. Genomic sequencing showed compound heterozygous CPS1 deficiency (one pathogenic allele inherited from each parent\u2014Q335X inherited from the father and E714X inherited from the mother); thus, the research team selected a CRISPR-based-editing paternal variant approach rather than a double-strand-break nuclease strategy [42]. To optimize the therapy for safety and efficacy, researchers created both a cellular model mimicking the patient\u2019s altered human hepatocytes and an animal model replicating the boy\u2019s genetic mutations. During this period\u2014spanning several months\u2014clinicians were fighting to keep the child alive. These models were used to test different gRNAs and select optimal PAM sequences, refining the therapeutic design to ensure efficient correction at the mutation sites while minimizing off-target effects [43].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "3. Gene Therapies Without the Need for Ex Vivo Cell Modification", "text": "Instead of correcting cells ex vivo, researchers employed LNPs to encapsulate the mRNA therapeutic molecules for intravenous administration. The LNPs were specifically engineered and delivered in a way that ensured predominant accumulation in the liver. In animal studies, the gene correction rate reached as high as 40% of hepatocytes, demonstrating significant therapeutic potential.\n\nIn liver diseases, the situation is somewhat more favorable than in CNS disorders, because after intravenous administration, LNPs predominantly accumulate in the liver [44]. From this perspective, the application of this therapy was, therefore, easier than in the previously described case of alternating hemiplegia.\n\nImportantly, this patient will not be the only beneficiary. The therapeutic strategy developed for him may be applied to other children with genetic liver diseases, and potentially to disorders affecting other organs as well. A key aspect of this breakthrough is that it represents an in vivo therapy enabled by lipid nanoparticle delivery, rather than an ex vivo approach involving modification of stem cells outside the patient\u2019s body followed by reinfusion. Reflecting the growing promise of such personalized in vivo gene editing approaches, the FDA has recently approved the introduction of therapies based on this approach for several hepatocyte-related genetic diseases [45,46].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "3. Gene Therapies Without the Need for Ex Vivo Cell Modification", "text": "Moreover, the use of genome editing in vivo is being considered in patients with hypercholesterolaemia, not only in those with the familial form of the disease. Researchers have disrupted the ANGPTL3 gene in hepatocytes using CRISPR/Cas9 and mRNA lipid nanoparticles in order to lower levels of \u201cbad cholesterol\u201d (LDL). Individuals who naturally have reduced activity of this gene are less likely to develop LDL-related disease. Results from the first phase of clinical trials conducted in a group of several dozen patients have been published [47]. The LNPs were additionally tuned to bind the LDL receptor and scavenger receptors on hepatocytes (in the liver) and to be internalized in a manner similar to LDL. So far, this approach has been used only in seriously ill patients. However, not all of them had specific hereditary diseases or syndromes. This suggests that the boundary between treatment, prevention, and enhancement may gradually begin to blur. At present, the liver has become an especially intensively studied organ because of its high susceptibility to nanoparticle internalization. Aside from the elimination of HLA genes to produce universal CAR-T or CAR-NK cells, this is the only case in which a normal (non-mutated) gene has been deliberately inactivated [47].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "3. Gene Therapies Without the Need for Ex Vivo Cell Modification", "text": "Another significant advance in in vivo therapies involved the treatment of progeria (Figure 3B). This represented a different application of CRISPR-based technology\u2014this time used not to repair or remove a gene, but to target and destroy the mRNA (Cas13 instead of Cas9) encoding the protein responsible for the disease. The disorder, known as Hutchinson\u2013Gilford progeria syndrome (HGPS), is extremely severe and typically fatal within the first two decades of life. It affects approximately one child per one to four million births [48,49]. In the particular form of progeria for which this therapy shows promise, the disease is caused by a mutation in the lamin A gene (LMNA) [50]. Interestingly, in this case, a lentiviral vector was used in vivo. This was necessary because the vector needed to reach dividing cells. By contrast, in ex vivo CAR-T cell therapy\u2014which will be discussed in the following section\u2014the use of an integrating viral vector is more acceptable, as it is applied only to isolated cells outside the patient\u2019s body. Here, however, various cells in the organism were deliberately exposed to the vector, which raises safety concerns. This issue is also a topic of ongoing debate in the context of in vivo CAR-T therapies. The mutation targeted in this approach does not alter an amino acid and might appear \u201csilent.\u201d Nevertheless, it disrupts mRNA splicing, leading to the production of defective forms of lamin A. The therapy enables the selective degradation of abnormal transcripts. In this case, the use of Cas13 was required [48]. The studies were conducted in mouse models. Importantly, this therapeutic concept extends beyond progeria. It is estimated that at least 10% of genetic diseases and syndromes result from the presence of defective mRNA molecules\u2014often caused by aberrant splicing, as in this variant of progeria [38,50].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "3. Gene Therapies Without the Need for Ex Vivo Cell Modification", "text": "A particularly interesting combination of gene therapy in vivo and neurosurgery was implemented at University College London (UCL) in patients with Huntington\u2019s disease (Figure 3A). An AAV vector carrying DNA encoding a microRNA (miRNA) designed to degrade mRNA transcripts of both the normal and mutant forms of huntingtin was delivered directly into specific structures of the central nervous system (CNS) using a specialized catheter system under MRI guidance. It is estimated that the progression of the disease was significantly slowed\u2014a degenerative process that typically advances over about one year was delayed to approximately four years following treatment [51].\n\nAnother example is the gene therapy for spinal muscular atrophy approved by the FDA in November 2025. The therapy is based on an AAV9 vector that delivers a functional copy of the SMN1 gene to motor neurons and other target cells, restoring SMN protein production [52]. Promising results have also been reported from clinical trials of gene therapy for one form of congenital deafness caused by mutations in the OTOFERLIN gene. According to the authors, this gene therapy may offer advantages over cochlear implants for this particular type of deafness [12].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "3. Gene Therapies Without the Need for Ex Vivo Cell Modification", "text": "There are numerous genetic causes of hearing loss\u2014in the United States, approximately 1 in 1000 children is affected. Mutations in OTOFERLIN account for about 1\u20133% of congenital deafness cases (Figure 3C). The therapeutic approach employed an AAV viral vector in collaboration with surgical teams. In this case, two vectors were required to generate such a large mRNA fragment encoding otoferlin. The transgene fragments delivered by the two cooperating vectors were joined through the SD-AP and AP-SA sequences. These sequences facilitate transcription and promote recombination, resembling splicing [53]. Together, these examples illustrate that CRISPR-based methods are not always the superior solution. Each mutation requires a separate gRNA. In contrast to genome-editing approaches, the introduction of a therapeutic transgene allowed for the correction of multiple distinct mutations simultaneously. The disease is autosomal recessive. Moreover, the therapeutic transgene was expressed primarily in the hair cells of the organ of Corti, which enhanced both efficacy and safety. To achieve selective expression, the Myo15 promoter was used. Only one child in the trial did not show improvement. In some patients, cochlear implants were deactivated following successful gene therapy. Remarkably, hearing improvement was observed even in a 16-year-old patient. As in the case of carbamoyl phosphate synthetase 1 deficiency, researchers considered the possibility of an immune response against the newly produced therapeutic protein, which the patient\u2019s immune system might recognize as foreign. However, no clinically significant immune reactions have been observed so far [12].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "3. Gene Therapies Without the Need for Ex Vivo Cell Modification", "text": "In contrast to gene editing, gene therapies based on transgene delivery require considering the sizes of the genes. Duchenne muscular dystrophy (DMD) micro-dystrophin constructs (typically shortened to ~3.5\u20134.4 kb, compared with the full-length dystrophin of ~11\u201314 kb) represent the most well-known example of a disease in which the transgene had to be radically truncated to fit within the packaging capacity of AAV vectors (Table 1). Ongoing studies conducted within programs such as Elevidys (Sarepta) and RGX-202 (REGENXBIO) demonstrate that this strategy can be therapeutically effective.\n\nGene therapies targeting large genes inherently require compromises: micro-versions of the transgene lose some functional domains but retain sufficient activity to produce clinical benefit. In the case of DMD, the spectrum of mutations observed to date has not yet been particularly amenable to genome-editing approaches, making transgene replacement strategies the preferred option at present [54].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "3. Gene Therapies Without the Need for Ex Vivo Cell Modification", "text": "A very interesting approach has been developed by the team led by Suki Albers in the therapy of several rare diseases. The prerequisite is the presence of mutations that introduce a premature STOP codon. They modified the tRNA that normally recognizes the codon for an amino acid so that it recognizes the STOP codon instead. In theory, this approach has been explored for a long time [55]. However, earlier attempts introduced a transgene encoding such a tRNA using AAV or lentiviral vectors [56]. In this approach, the authors proposed modifying the tRNA gene itself using a specialized variant of prime editing.", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "4. CAR-T Cell Immunotherapies (Including CAR-NK and CAR-NKT)\u2014In Vivo and Ex Vivo Approaches, Genetic Modifications, and Cell Selection", "text": "In the field of CAR-T cell therapies, remarkable progress has been achieved in recent years, with currently approved CAR products being predominantly autologous ex vivo CAR-T therapies (Table 2). One of the major challenges now is to make such treatments available \u201coff-the-shelf.\u201d Several strategies are being pursued toward this goal [57].\n\nThe first strategy involves genetic modification of CAR-T cells in a manner similar to that described earlier for pancreatic \u03b2-cells\u2014introducing changes that allow these therapeutic cells to be used in any recipient.\n\nIn a notable study, Hypoimmune CD19 CAR-T cells were engineered to evade allorejection in patients with cancer and autoimmune diseases [58]. The approach was conceptually similar to that used by the same team in their research on pancreatic cells, which were derived not from deceased donors but from induced pluripotent stem cells (iPSCs). The HIP-engineered primary CD19 CAR-T cells required three major genetic modifications to ensure effective function and prevent graft rejection. Undoubtedly, insertion of the CAR transgene was essential in this case. As discussed later, other research groups have introduced additional modifications to further improve the performance and controllability of CAR-T cells.", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "4. CAR-T Cell Immunotherapies (Including CAR-NK and CAR-NKT)\u2014In Vivo and Ex Vivo Approaches, Genetic Modifications, and Cell Selection", "text": "Another approach to achieving universal applicability\u2014not mutually exclusive with the first\u2014involves the selection of alternative immune cell types, such as CAR-NKT cells. This concept has since been extended to other immune cell subtypes.\n\nFinally, there is the emerging possibility of applying CAR-T therapies in vivo. A research team successfully generated CAR-NK cells compatible with any recipient. These cells carried as many as nine genetic modifications in total: three genes were deleted, four natural transgenes were introduced, and two synthetic genes were added. The researchers derived their CAR-NK cells from iPSCs, which are considerably easier to modify genetically than CAR-NK cells isolated directly from donors\u2014and, for many research groups, easier still than modifying pancreatic cells as described earlier [10].\n\nThe CAR-NK cells, however, required a greater number of genetic modifications, as they were designed to enable controlled activation and targeting. By contrast, fewer modifications were needed in the pancreatic cell models described earlier, although pancreatic cells are far more difficult to derive from iPSCs than CAR-NK cells.", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "4. CAR-T Cell Immunotherapies (Including CAR-NK and CAR-NKT)\u2014In Vivo and Ex Vivo Approaches, Genetic Modifications, and Cell Selection", "text": "Interestingly, these CAR-NK cells were not used to treat an oncological patient, as is often the case with CAR-T or CAR-NK therapies, but rather a patient suffering from an autoimmune disease\u2014systemic sclerosis (scleroderma). This condition is characterized by fibrosis of the skin and, in some cases, fibrosis of internal organs [59]. Although systemic sclerosis today less frequently leads to death, the selected patient was a critical case, and the study also highlighted the broader potential of such genetic and epigenetic cell modifications. The researchers used a variant of CRISPR technology to perform precise gene edits. Specifically, they knocked out B2M, CIITA, and CD16 while introducing two HLA genes to make the cells universal. They further enhanced the cells with IL-2RF to boost therapeutic efficacy and added tEGFR as a safety switch, allowing physicians to mitigate potential complications if needed. Additionally, they inserted two CAR transgenes targeting BCMA and CD19 [10].\n\nMuch recent attention has focused on the concept of in vivo CAR-T generation within the patient\u2019s body. This is a highly promising strategy\u2014discussed later\u2014but the extent of genetic modification achievable in vivo remains far more limited than that possible ex vivo.", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "4. CAR-T Cell Immunotherapies (Including CAR-NK and CAR-NKT)\u2014In Vivo and Ex Vivo Approaches, Genetic Modifications, and Cell Selection", "text": "An intriguing alternative to cancer immunotherapy involves the use of CAR-NKT cells, which can be developed as universal therapeutic products. Preclinical studies have been conducted in ovarian cancer, a disease that remains a major medical challenge but may benefit substantially from this approach [60].\n\nNKT cells are relatively easy to derive for any recipient (compared with CD8 cytotoxic lymphocytes). They represent a natural \u201cchimera\u201d of NK and T lymphocytes, serving as a potential alternative and competitor to the universal CAR-NK approach developed in China. Chinese researchers achieved broad compatibility by introducing nine genetic modifications into CAR-NK cells.\n\nCAR-NKT cells are based on naturally occurring immune cells, which require fewer genetic alterations to achieve universality. Moreover, these engineered CAR-NKT cells were found to modulate the tumor microenvironment, making it more favorable for tumor cell destruction. These results were obtained in preclinical studies.\n\nA very interesting approach is presented in the work by Chiesa et al. [61].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "4. CAR-T Cell Immunotherapies (Including CAR-NK and CAR-NKT)\u2014In Vivo and Ex Vivo Approaches, Genetic Modifications, and Cell Selection", "text": "Three key genes were edited to address the main challenges associated with using CAR-T cells targeting CD7 in patients with T-ALL. Knockout of the CD7 gene prevents CAR-T cells from self-reactivity (fratricide). Knockout of CD52 renders the cells resistant to alemtuzumab, an anti-CD52 antibody used during lymphodepletion. Finally, knockout of TRAC (TCR\u03b1 chain) eliminates the risk of graft-versus-host disease (GVHD) [61].\n\nThe modified cells were subsequently transduced with a lentiviral vector encoding a CAR recognizing CD7. This represents a highly elegant solution for T-ALL therapy. However, the approach requires more advanced manufacturing procedures than leukemias derived from B-cell precursors, since the target antigen must be preserved on the T lymphocytes used for CAR-T production, as the leukemia originates from these progenitor cells (Figure 4).\n\nTo date, no NK- or NKT-based products have received regulatory approval, with ongoing research confined to the clinical trial stage [62].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "5. In Vivo CAR-T (CAR-X) Therapies", "text": "On top of that, CAR-T cells designed for the treatment of oncological and autoimmune diseases can now be prepared directly within the patient\u2019s body (in vivo), a development made possible largely through advances in mRNA vaccine and viral vector technologies [63]. In contrast to classical cell therapy, in vivo CAR-T represents a hybrid modality at the interface of cell and gene therapy, in which endogenous immune cells are genetically reprogrammed in situ to express chimeric antigen receptors [64,65].\n\nCompared with traditional ex vivo preparation, in vivo generation offers important advantages. Cells that are harvested from patients for in vitro modification often exhibit poor adaptability, particularly when obtained from individuals who are severely ill or have undergone intensive chemotherapy. The overall process of ex vivo CAR-T production is highly complex\u2014requiring specialized laboratories, costly reagents, strict regulatory oversight, and significant time investment [66]. Consequently, the cost of ex vivo-manufactured CAR-T therapies remains extremely high [67]. One of the clear benefits of the in vivo approach is the elimination of both leukapheresis and lymphodepletion, procedures typically required before ex vivo therapy [68].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "5. In Vivo CAR-T (CAR-X) Therapies", "text": "Mechanistically, in vivo CAR-T therapies rely on systemic or localized delivery of CAR-encoding genetic material using viral vectors (such as AAV or lentivirus) or non-viral carriers, including LNPs engineered for preferential uptake by lymphocytes [69]. Following delivery, resident T cells begin to express the CAR, undergo antigen-dependent activation, and expand within the patient, thereby generating a functional CAR-T population without external manipulation [70]. Preclinical studies have shown that in vivo\u2013generated CAR-T cells can mediate potent and antigen-specific cytotoxic responses, particularly in CD19-targeted models, supporting the feasibility of this approach [71].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "5. In Vivo CAR-T (CAR-X) Therapies", "text": "Nevertheless, while several in vivo gene therapies have already achieved FDA/EMA approval (Table 2), it remains at a preclinical/early clinical stage, highlighting the remaining delivery and safety difficulties. CAR-T and other CAR-modified immune cells (such as CAR-M, CAR-NK, or CAR-\u03b3\u03b4T) can be more extensively engineered in vitro than within the human body. Techniques such as CRISPR-based genome editing allow for the creation of \u201coff-the-shelf\u201d universal products, tailored for any recipient [72]. In addition, ex vivo manufacturing enables precise control over transgene copy number, integration site, and cell composition, features that remain difficult to regulate in in vivo approaches [73]. Safety mechanisms commonly used in ex vivo CAR-T products, such as suicide switches or stringent product quality control, are also more challenging to implement when CAR expression is induced directly within the patient [74].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "5. In Vivo CAR-T (CAR-X) Therapies", "text": "Creating universal CAR-T cells remains an enormous challenge. However, growing knowledge of how to modify or neutralize genes responsible for graft rejection and graft-versus-host responses is steadily advancing the field\u2014a crucial development for CAR-T therapies [75]. Such modifications are technically easier to perform in stem cells, yet differentiating genetically modified stem cells (iPSCs or HSCs) into mature lymphocytes remains a major obstacle [76]. Consequently, despite the higher susceptibility of stem cells to genetic manipulations, greater progress has been achieved in universalizing CAR-T cells derived directly from lymphocytes. Although genetic modifications in mature lymphocytes are less efficient than in stem cells, the practical difficulty of obtaining fully functional lymphocytes from stem cells often tips the balance in favor of the former.", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "5. In Vivo CAR-T (CAR-X) Therapies", "text": "This situation is somewhat analogous to the example discussed earlier in type 1 diabetes therapy: some research groups generate pancreatic cells from iPSCs or embryonic stem cells, while others focus on modifying cells harvested from deceased donors. The analogy lies in the question of whether it is easier to work with mature or stem cells. Stem cells divide indefinitely and are easier to genetically modify, but some target cell types are extremely difficult to derive from them. In such cases, modifying mature cells\u2014despite their limited proliferative capacity\u2014may be the more practical solution. Interestingly, CAR-NKT cells appear to have slightly greater proliferative potential than conventional T lymphocytes.\n\nThere are, of course, many additional arguments for and against each approach. One important factor supporting in vivo strategies is the rapid development of lipid nanoparticle (LNP) technology following COVID-19 vaccination advances. Technology is now evolving into targeted LNPs (tLNPs) capable of directing delivery even to lymphocytes, facilitating in vivo therapeutic applications. As a result, CAR-T generation within the patient\u2019s body is no longer a theoretical concept (Figure 5).", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "5. In Vivo CAR-T (CAR-X) Therapies", "text": "Redosing is another important consideration for CAR-T treatments. mRNA/LNP in vivo platforms may enable easier redosing, as LNPs can support repeated cycles of B-cell depletion with good tolerability, thereby enabling therapeutic cycles without repeatedly devastating the immune system. Allogeneic ex vivo platforms offer off-the-shelf advantages (e.g., rapid availability and standardized products); however, repeat dosing can be limited by alloimmunization and persistence issues. However, when the priority is clinician-directed control, allogeneic platforms incorporating switches and logic gates may provide a clear advantage in 2026 designs.\n\nIf, however, non-universal ex vivo CAR-T therapies are to compete with these in which CAR-T cells are generated in vivo, the latter may ultimately prove to be the winning strategy. The transgene encoding the CAR is now being integrated not only into classical immune cells. It has also been introduced, appropriately adapted or modified, into other cell types, such as astrocytes. These interventions were performed in vivo using a promoter that is active primarily in astrocytes (GfaABC1D). An AAV vector was used for delivery. This demonstrates that work on synthetic receptors is extending beyond the cell types traditionally associated with the immune system. CAR-X, in the subtitle, indicates that many different cell types are being targeted [13].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "5. In Vivo CAR-T (CAR-X) Therapies", "text": "To contextualize CAR-based platforms within the broader regulatory landscape of gene and cell therapies, Table 2 summarizes FDA/EMA-approved products.", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "6. Comparative Evaluation: Ex Vivo and In Vivo Gene Modification Strategies", "text": "Vector safety remains one of the major challenges in gene and cell therapies, with risks and mitigation strategies differing substantially between ex vivo and in vivo approaches. In ex vivo therapies (e.g., modified hematopoietic stem cells, CAR-T/CAR-NK cells, and iPSC-derived cells), lentiviral and retroviral vectors\u2014which integrate into the genome\u2014are predominantly used. The principal risk associated with these vectors is insertional mutagenesis leading to proto-oncogene activation. This risk was particularly evident in early generations of gene therapies, such as the development of leukemia in ADA-SCID trials.\n\nContemporary solutions have markedly improved safety, most notably through the use of self-inactivating (SIN) lentiviral vectors. These improvements have been achieved, for example, by incorporating chromatin insulators (e.g., cHS4) [77], employing weaker promoters (such as PGK instead of MSCV), and limiting vector copy number. In addition, CAR-T therapies commonly incorporate so-called safety switches (e.g., tEGFR, iCasp9, RQR8), which enable selective elimination of the modified cells in the event of toxicity or oncogenic transformation [10,58].\n\nA different situation applies when induced pluripotent stem cells (iPSCs), rather than lymphocytes or HSCs, are used. In this case, initial reprogramming typically relies on episomal vectors, after which the introduction of therapeutic transgenes can be achieved using a variety of alternative delivery systems.", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "6. Comparative Evaluation: Ex Vivo and In Vivo Gene Modification Strategies", "text": "In in vivo therapies, the risk of genomic integration must be minimized; therefore, episomal vectors such as adeno-associated viruses (AAVs) are commonly used. However, this approach introduces other challenges, including strong immune responses against the AAV capsid\u2014particularly the presence of neutralizing antibodies\u2014as well as potential immunogenicity of the newly expressed transgene product. Strategies to improve safety include AAV capsid detargeting, for example, through surface modifications that reduce immune recognition [78].\n\nLipid nanoparticle (LNP)\u2013delivered mRNA, which enables transient expression, offers an even lower risk of permanent genomic alteration and generally exhibits reduced immunogenicity. A remaining challenge for LNP-based approaches is precise control of tissue tropism; however, as demonstrated in the context of hypercholesterolemia, this is increasingly becoming feasible [47].\n\nThe durability of therapeutic expression depends primarily on whether the vector integrates into the genome\u2014allowing long-term expression in dividing cells\u2014or remains episomal, which leads to loss of expression in proliferating tissues. When genome editing is performed, however, the situation changes fundamentally: the genetic modification is maintained for as long as the edited cell population persists.", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "6. Comparative Evaluation: Ex Vivo and In Vivo Gene Modification Strategies", "text": "In ex vivo therapies, lentiviral vectors provide stable, long-term expression lasting many years, even after transplantation (e.g., Zynteglo, Casgevy, CAR-T therapies), as the modified hematopoietic stem cells or lymphocytes divide and transmit the integrated transgene to their progeny. In in vivo therapies, AAV vectors support durable expression in post-mitotic cells\u2014such as neurons, hepatocytes, muscle cells, and auditory sensory cells\u2014as demonstrated by therapies, including Zolgensma for spinal muscular atrophy, Luxturna, and otoferlin gene therapy [12].\n\nHowever, in proliferating tissues (e.g., hepatocytes in pediatric patients), expression declines over time due to dilution of episomal vectors. LNP-delivered mRNA enables transient expression lasting weeks to months, which is advantageous for in vivo CAR-T approaches by avoiding permanent genomic alterations [35], but represents a limitation for diseases that require sustained gene expression. Strategies to improve durability include the use of integrating lentiviral vectors in vivo (rarely employed because of safety concerns), episomal amplification systems within AAV vectors, or repeated low-dose administration of LNPs. As noted above, the situation differs when LNPs are used to deliver gene-editing systems, as the therapeutic effect can persist despite the transient presence of the editor. LNPs, similar to viral vectors, can be engaged in the targeted delivery [47].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "6. Comparative Evaluation: Ex Vivo and In Vivo Gene Modification Strategies", "text": "An instructive comparison in this context is provided by gene editing approaches targeting tRNA genes in the treatment of nonsense mutations [8]. Earlier strategies attempted to introduce a modified tRNA transgene capable of recognizing stop codons and inserting an amino acid during translation. In such cases, when delivered by AAV, the therapeutic effect was transient because episomal AAV genomes were gradually lost.\n\nBy contrast, current approaches use AAV to deliver a gene-editing system that modifies the endogenous tRNA gene. In this scenario, the AAV vector performs its function before being lost, while the corrected genomic sequence\u2014and thus the therapeutic effect\u2014persists. Lentiviral vectors carrying tRNA transgenes previously offered more durable expression; however, their random genomic integration. The associated risk of insertional mutagenesis was considered unacceptably high, and progress in eliminating those effects seems to be still inefficient [79].\n\nIn summary, AAV-mediated delivery of a tRNA transgene resulted in a transient effect, whereas lentiviral delivery provided durability at the cost of increased oncogenic risk. The newer strategy resolves this trade-off by using AAV not to introduce a tRNA transgene, but to install a permanent genomic modification of the endogenous tRNA gene\u2014so that even after the vector is lost, the therapeutic benefit remains.", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "6. Comparative Evaluation: Ex Vivo and In Vivo Gene Modification Strategies", "text": "Tissue and cellular tropism can be achieved through several distinct mechanisms and depend strongly on whether the therapy is performed ex vivo or in vivo. In ex vivo therapies, the concept of \u201ctropism\u201d is largely replaced by cell selection and directed delivery, as the genetic modification is performed outside the body. Typically, patient cells (hematopoietic stem cells (HSCs), keratinocytes, fibroblasts, or pancreatic cells) are isolated; then a therapeutic transgene is introduced, or a pathogenic mutation is edited, and the modified cells are returned to the patient. iPSCs are usually generated and expanded ex vivo, and clinically relevant, genetically modified derivatives are transplanted rather than iPSCs themselves.", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "6. Comparative Evaluation: Ex Vivo and In Vivo Gene Modification Strategies", "text": "For stable transgene expression in dividing cells, lentiviral or retroviral vectors are most commonly used because they integrate into the genome, supporting long-term expression. A representative example, discussed in Section 2, is therapy for sickle cell disease and \u03b2-thalassemia, in which HSCs are transduced with a lentiviral vector carrying a modified \u03b2-globin gene under specific regulatory control (\u03b2-globin promoter and locus control region), followed by transplantation after myeloablation. Similarly, in dystrophic epidermolysis bullosa, keratinocytes and fibroblasts are retrovirally modified with cDNA of the COL7A1 gene and applied as a graft directly to the wound site, which achieves anatomical targeting through the surgical procedure [34]. Sometimes, full transgene insertion is not required, but CRISPR-based editing, base editing, or prime editing could be performed directly in isolated cells to correct specific mutations (editing of NCF1 in chronic granulomatous disease) [9]. In different instances, in type 1 diabetes, hypoimmune donor-derived \u03b2-cells engineered by B2M and CIITA knockout combined with CD47 overexpression were transplanted into skeletal muscle to enable persistence and insulin production without immunosuppression [11]. Although this approach can be clinically meaningful, it should be underscored that anatomical delivery to the pancreas, despite being technically more challenging, may offer superior engraftment and functional integration [80,81].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "6. Comparative Evaluation: Ex Vivo and In Vivo Gene Modification Strategies", "text": "Overall, in ex vivo therapy, the critical determinants of \u201ctargeting\u201d are precise selection of the starting cell population and directed transplantation to the desired anatomical compartment (bone marrow, skin, muscle, or CNS), rather than vector tropism.", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "6. Comparative Evaluation: Ex Vivo and In Vivo Gene Modification Strategies", "text": "On the other hand, in in vivo therapies, tropism returns to its classical meaning: therapeutic transgenes or genome-editing systems are delivered directly into the body, requiring strong vector tropism and engineered targeting strategies. For many inherited liver diseases, in vivo delivery is more favorable than for CNS disorders because intravenously administered lipid nanoparticles (LNPs) naturally accumulate in the liver and are efficiently taken up by hepatocytes. This was used, for example, in CPS1 deficiency, where LNPs deliver mRNA encoding an adenine base editor together with a guide RNA to enable in vivo editing [82]. Despite the intrinsic liver tropism of LNPs, additional engineering can further improve selectivity. In hypercholesterolemia, in vivo editing of ANGPTL3 has been enhanced by LNP modifications that promote interactions with LDL receptor-related uptake pathways and scavenger receptors on hepatocytes. This strategy partially mimics physiological LDL clearance from blood, thereby increasing liver-selective delivery [47]. In contrast, in Huntington\u2019s disease, an AAV vector encoding an artificial microRNA targeting HTT (miHTT) was administered directly into the striatum using MRI-guided convection-enhanced delivery, leveraging neuronal tropism of AAV while still relying on local administration to overcome the blood\u2013brain barrier [51]. In congenital deafness due to OTOF mutations, a dual-AAV system carrying OTOF cDNA under a Myo15 promoter can drive expression in cochlear hair cells. However, the vector is still delivered locally to the inner ear via otolaryngologic surgery [12].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "6. Comparative Evaluation: Ex Vivo and In Vivo Gene Modification Strategies", "text": "In emerging in vivo CAR strategies, lipid nanoparticles with CAR mRNA exploit tropism for T or NK lymphocytes.\n\nTo sum up, in ex vivo therapies, \u201ctropism\u201d is effectively implemented through cell-type selection (HSCs, iPSC-derived products, keratinocytes) combined with targeted transplantation to the appropriate site (bone marrow, skin, muscle, or CNS). In in vivo therapies, the dominant determinants are natural or engineered vector tropism (AAV for neurons and hair cells; LNPs for hepatocytes, including additional LDLR/scavenger-mediated uptake), tissue-restricted promoters, and local administration strategies when systemic delivery is inadequate (MRI-guided intracranial infusion or inner-ear injection). Altogether, tropism, promoter specificity, delivery route, and in some settings hypoimmune engineering enable efficient delivery to selected tissues while minimizing dose requirements and safety risks (Table 3).", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "6. Comparative Evaluation: Ex Vivo and In Vivo Gene Modification Strategies", "text": "Tunability\u2014the control of expression level, timing, and tissue specificity\u2014is critical for minimizing toxicity and optimizing therapeutic efficacy, but the available strategies differ substantially between ex vivo and in vivo approaches. In ex vivo therapies, control is comparatively straightforward: strong constitutive promoters (e.g., EF1\u03b1, PGK, MSCV) can be used in combination with safety switches (such as tEGFR or iCasp9), or additional regulatory elements can be incorporated, including miRNA-responsive circuits in next-generation CAR designs. Representative examples include hypoimmunogenic CAR-T and CAR-NK therapies [10,58] (Table 4).\n\nIn in vivo therapies, tunability is more challenging. Using AAV or LNP platforms, it is difficult to introduce the number and complexity of regulatory elements required for features such as safety switches (e.g., tEGFR). In this context, tissue-specific promoters can provide partial solutions for certain patients by restricting expression to target cell types, such as Myo15 in otoferlin-related deafness [12], GfaABC1D in astrocytes [64], or thyroxine-binding globulin (TBG) in hepatocytes [83].\n\nThe transient activity of LNP-delivered mRNA provides an inherent form of temporal tunability through short-lived expression, which is advantageous in vivo CAR-T approaches [35] and personalized gene editing strategy [82] (Table 5).", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "6.1. Vector Safety: Ex Vivo vs. In Vivo Approaches", "text": "Vector safety remains one of the major challenges in gene and cell therapies, with risks and mitigation strategies differing substantially between ex vivo and in vivo approaches. In ex vivo therapies (e.g., modified hematopoietic stem cells, CAR-T/CAR-NK cells, and iPSC-derived cells), lentiviral and retroviral vectors\u2014which integrate into the genome\u2014are predominantly used. The principal risk associated with these vectors is insertional mutagenesis leading to proto-oncogene activation. This risk was particularly evident in early generations of gene therapies, such as the development of leukemia in ADA-SCID trials.\n\nContemporary solutions have markedly improved safety, most notably through the use of self-inactivating (SIN) lentiviral vectors. These improvements have been achieved, for example, by incorporating chromatin insulators (e.g., cHS4) [77], employing weaker promoters (such as PGK instead of MSCV), and limiting vector copy number. In addition, CAR-T therapies commonly incorporate so-called safety switches (e.g., tEGFR, iCasp9, RQR8), which enable selective elimination of the modified cells in the event of toxicity or oncogenic transformation [10,58].\n\nA different situation applies when induced pluripotent stem cells (iPSCs), rather than lymphocytes or HSCs, are used. In this case, initial reprogramming typically relies on episomal vectors, after which the introduction of therapeutic transgenes can be achieved using a variety of alternative delivery systems.", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "6.1. Vector Safety: Ex Vivo vs. In Vivo Approaches", "text": "In in vivo therapies, the risk of genomic integration must be minimized; therefore, episomal vectors such as adeno-associated viruses (AAVs) are commonly used. However, this approach introduces other challenges, including strong immune responses against the AAV capsid\u2014particularly the presence of neutralizing antibodies\u2014as well as potential immunogenicity of the newly expressed transgene product. Strategies to improve safety include AAV capsid detargeting, for example, through surface modifications that reduce immune recognition [78].\n\nLipid nanoparticle (LNP)\u2013delivered mRNA, which enables transient expression, offers an even lower risk of permanent genomic alteration and generally exhibits reduced immunogenicity. A remaining challenge for LNP-based approaches is precise control of tissue tropism; however, as demonstrated in the context of hypercholesterolemia, this is increasingly becoming feasible [47].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "6.2. Durability of Expression: Ex Vivo vs. In Vivo", "text": "The durability of therapeutic expression depends primarily on whether the vector integrates into the genome\u2014allowing long-term expression in dividing cells\u2014or remains episomal, which leads to loss of expression in proliferating tissues. When genome editing is performed, however, the situation changes fundamentally: the genetic modification is maintained for as long as the edited cell population persists.\n\nIn ex vivo therapies, lentiviral vectors provide stable, long-term expression lasting many years, even after transplantation (e.g., Zynteglo, Casgevy, CAR-T therapies), as the modified hematopoietic stem cells or lymphocytes divide and transmit the integrated transgene to their progeny. In in vivo therapies, AAV vectors support durable expression in post-mitotic cells\u2014such as neurons, hepatocytes, muscle cells, and auditory sensory cells\u2014as demonstrated by therapies, including Zolgensma for spinal muscular atrophy, Luxturna, and otoferlin gene therapy [12].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "6.2. Durability of Expression: Ex Vivo vs. In Vivo", "text": "However, in proliferating tissues (e.g., hepatocytes in pediatric patients), expression declines over time due to dilution of episomal vectors. LNP-delivered mRNA enables transient expression lasting weeks to months, which is advantageous for in vivo CAR-T approaches by avoiding permanent genomic alterations [35], but represents a limitation for diseases that require sustained gene expression. Strategies to improve durability include the use of integrating lentiviral vectors in vivo (rarely employed because of safety concerns), episomal amplification systems within AAV vectors, or repeated low-dose administration of LNPs. As noted above, the situation differs when LNPs are used to deliver gene-editing systems, as the therapeutic effect can persist despite the transient presence of the editor. LNPs, similar to viral vectors, can be engaged in the targeted delivery [47].\n\nAn instructive comparison in this context is provided by gene editing approaches targeting tRNA genes in the treatment of nonsense mutations [8]. Earlier strategies attempted to introduce a modified tRNA transgene capable of recognizing stop codons and inserting an amino acid during translation. In such cases, when delivered by AAV, the therapeutic effect was transient because episomal AAV genomes were gradually lost.", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "6.2. Durability of Expression: Ex Vivo vs. In Vivo", "text": "By contrast, current approaches use AAV to deliver a gene-editing system that modifies the endogenous tRNA gene. In this scenario, the AAV vector performs its function before being lost, while the corrected genomic sequence\u2014and thus the therapeutic effect\u2014persists. Lentiviral vectors carrying tRNA transgenes previously offered more durable expression; however, their random genomic integration. The associated risk of insertional mutagenesis was considered unacceptably high, and progress in eliminating those effects seems to be still inefficient [79].\n\nIn summary, AAV-mediated delivery of a tRNA transgene resulted in a transient effect, whereas lentiviral delivery provided durability at the cost of increased oncogenic risk. The newer strategy resolves this trade-off by using AAV not to introduce a tRNA transgene, but to install a permanent genomic modification of the endogenous tRNA gene\u2014so that even after the vector is lost, the therapeutic benefit remains.", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "6.3. Tissue and Cellular Tropism", "text": "Tissue and cellular tropism can be achieved through several distinct mechanisms and depend strongly on whether the therapy is performed ex vivo or in vivo. In ex vivo therapies, the concept of \u201ctropism\u201d is largely replaced by cell selection and directed delivery, as the genetic modification is performed outside the body. Typically, patient cells (hematopoietic stem cells (HSCs), keratinocytes, fibroblasts, or pancreatic cells) are isolated; then a therapeutic transgene is introduced, or a pathogenic mutation is edited, and the modified cells are returned to the patient. iPSCs are usually generated and expanded ex vivo, and clinically relevant, genetically modified derivatives are transplanted rather than iPSCs themselves.", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "6.3. Tissue and Cellular Tropism", "text": "For stable transgene expression in dividing cells, lentiviral or retroviral vectors are most commonly used because they integrate into the genome, supporting long-term expression. A representative example, discussed in Section 2, is therapy for sickle cell disease and \u03b2-thalassemia, in which HSCs are transduced with a lentiviral vector carrying a modified \u03b2-globin gene under specific regulatory control (\u03b2-globin promoter and locus control region), followed by transplantation after myeloablation. Similarly, in dystrophic epidermolysis bullosa, keratinocytes and fibroblasts are retrovirally modified with cDNA of the COL7A1 gene and applied as a graft directly to the wound site, which achieves anatomical targeting through the surgical procedure [34]. Sometimes, full transgene insertion is not required, but CRISPR-based editing, base editing, or prime editing could be performed directly in isolated cells to correct specific mutations (editing of NCF1 in chronic granulomatous disease) [9]. In different instances, in type 1 diabetes, hypoimmune donor-derived \u03b2-cells engineered by B2M and CIITA knockout combined with CD47 overexpression were transplanted into skeletal muscle to enable persistence and insulin production without immunosuppression [11]. Although this approach can be clinically meaningful, it should be underscored that anatomical delivery to the pancreas, despite being technically more challenging, may offer superior engraftment and functional integration [80,81].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "6.3. Tissue and Cellular Tropism", "text": "Overall, in ex vivo therapy, the critical determinants of \u201ctargeting\u201d are precise selection of the starting cell population and directed transplantation to the desired anatomical compartment (bone marrow, skin, muscle, or CNS), rather than vector tropism.", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "6.3. Tissue and Cellular Tropism", "text": "On the other hand, in in vivo therapies, tropism returns to its classical meaning: therapeutic transgenes or genome-editing systems are delivered directly into the body, requiring strong vector tropism and engineered targeting strategies. For many inherited liver diseases, in vivo delivery is more favorable than for CNS disorders because intravenously administered lipid nanoparticles (LNPs) naturally accumulate in the liver and are efficiently taken up by hepatocytes. This was used, for example, in CPS1 deficiency, where LNPs deliver mRNA encoding an adenine base editor together with a guide RNA to enable in vivo editing [82]. Despite the intrinsic liver tropism of LNPs, additional engineering can further improve selectivity. In hypercholesterolemia, in vivo editing of ANGPTL3 has been enhanced by LNP modifications that promote interactions with LDL receptor-related uptake pathways and scavenger receptors on hepatocytes. This strategy partially mimics physiological LDL clearance from blood, thereby increasing liver-selective delivery [47]. In contrast, in Huntington\u2019s disease, an AAV vector encoding an artificial microRNA targeting HTT (miHTT) was administered directly into the striatum using MRI-guided convection-enhanced delivery, leveraging neuronal tropism of AAV while still relying on local administration to overcome the blood\u2013brain barrier [51]. In congenital deafness due to OTOF mutations, a dual-AAV system carrying OTOF cDNA under a Myo15 promoter can drive expression in cochlear hair cells. However, the vector is still delivered locally to the inner ear via otolaryngologic surgery [12].", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "6.3. Tissue and Cellular Tropism", "text": "In emerging in vivo CAR strategies, lipid nanoparticles with CAR mRNA exploit tropism for T or NK lymphocytes.\n\nTo sum up, in ex vivo therapies, \u201ctropism\u201d is effectively implemented through cell-type selection (HSCs, iPSC-derived products, keratinocytes) combined with targeted transplantation to the appropriate site (bone marrow, skin, muscle, or CNS). In in vivo therapies, the dominant determinants are natural or engineered vector tropism (AAV for neurons and hair cells; LNPs for hepatocytes, including additional LDLR/scavenger-mediated uptake), tissue-restricted promoters, and local administration strategies when systemic delivery is inadequate (MRI-guided intracranial infusion or inner-ear injection). Altogether, tropism, promoter specificity, delivery route, and in some settings hypoimmune engineering enable efficient delivery to selected tissues while minimizing dose requirements and safety risks (Table 3).", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "6.4. Tunability of Expression: Ex Vivo vs. In Vivo", "text": "Tunability\u2014the control of expression level, timing, and tissue specificity\u2014is critical for minimizing toxicity and optimizing therapeutic efficacy, but the available strategies differ substantially between ex vivo and in vivo approaches. In ex vivo therapies, control is comparatively straightforward: strong constitutive promoters (e.g., EF1\u03b1, PGK, MSCV) can be used in combination with safety switches (such as tEGFR or iCasp9), or additional regulatory elements can be incorporated, including miRNA-responsive circuits in next-generation CAR designs. Representative examples include hypoimmunogenic CAR-T and CAR-NK therapies [10,58] (Table 4).\n\nIn in vivo therapies, tunability is more challenging. Using AAV or LNP platforms, it is difficult to introduce the number and complexity of regulatory elements required for features such as safety switches (e.g., tEGFR). In this context, tissue-specific promoters can provide partial solutions for certain patients by restricting expression to target cell types, such as Myo15 in otoferlin-related deafness [12], GfaABC1D in astrocytes [64], or thyroxine-binding globulin (TBG) in hepatocytes [83].\n\nThe transient activity of LNP-delivered mRNA provides an inherent form of temporal tunability through short-lived expression, which is advantageous in vivo CAR-T approaches [35] and personalized gene editing strategy [82] (Table 5).", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "7. Summary", "text": "In recent years, a breakthrough has occurred across several types of biological therapies that had previously faced significant challenges\u2014most notably in gene therapies and various forms of cell therapies.\n\nThis progress results from several key factors. Among them is the improvement of cell differentiation techniques, which has expanded the range of cell types that can be generated for therapeutic use. Equally important are the advances stemming from CRISPR technology and its derivatives, which have revolutionized the precision and scope of genetic modification.\n\nAnother crucial factor has been the enhancement of transgene delivery systems, driven by progress in both viral vector design and lipid nanoparticle (LNP) technology\u2014the latter accelerated by innovations developed during COVID-19 vaccination programs.", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "7. Summary", "text": "Nevertheless, therapies targeting blood cells and the liver remain the most accessible. The former are easier to modify ex vivo, whereas the latter are a natural target for LNPs. Researchers are demonstrating increasing ingenuity: a wide spectrum of genome-editing strategies is being explored, extending even to Cas13, which degrades aberrantly spliced mRNA. CRISPR-based genome editing is not always the optimal solution. In some cases, such as otoferlin deficiency, the introduction of a single transgene enables therapy across many different mutations. Conversely, editing the gene encoding tRNA may offer improvements in numerous rare diseases. Vectors do not necessarily need to deliver mRNA; they may instead deliver miRNA, as in Huntington\u2019s disease. Moreover, research teams are modifying promoters: for example, the Myo15 promoter for congenital deafness, the nestin promoter in viral therapies for glioblastoma, or GfaABC1D when introducing a CAR transgene in vivo into astrocytes.\n\nThe size of the gene is a very important issue when transgenes are applied. Taken together, these approaches show that current efforts fall increasingly within the domain of synthetic biology rather than traditional genetic engineering.\n\nEven more significant advancements can be observed in CAR-T therapy and its variants, where improved understanding and engineering have led to major therapeutic gains.", "title": "" }, { "pmid": 41827217, "pmcid": "PMC12986383", "section": "7. Summary", "text": "Across all these developments, the foundation of success lies in the deepening understanding of disease mechanisms and cellular biology, which continues to shape and guide the evolution of next-generation therapies.\n\nAn overview of the generation of engineered cellular therapies derived from autologous and allogeneic donors is presented in Table 6.\n\nThe examples of therapies described in this work are grouped according to the organs or structures they target. Table 7 illustrates, in the context of specific diseases, which organ-directed therapeutic strategies have progressed most rapidly in recent years. This does not imply that other approaches are not being developed; rather, it highlights where the fastest advances have occurred.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Introduction", "text": "Neurological disorders, characterized by progressive degeneration of the structure and function of the nervous system, are becoming an increasingly significant global health challenge and burden.1,2 The Global Burden of Disease (GBD) study in 2021\u00a0highlighted this phenomenon.3 With the ongoing global aging population, the prevalence of neurological diseases is expected to continue rising, placing an ever-growing strain on medical resources, family care, and public healthcare systems.4,5 The nervous system is a central network that controls sensory, motor, cognitive, and memory functions, and is composed of large numbers of neurons and glial cells.6 Neurofilaments form a cytoskeletal network within neuronal axons, maintaining axonal structural integrity and influencing the velocity of signal conduction.7 Any abnormalities in neuronal structure or signal transmission may lead to impairments in sensory, motor, or cognitive functions.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Introduction", "text": "Therapeutic strategies for neurological diseases face substantial challenges, and a wide range of approaches\u2014including conventional pharmaceuticals, antioxidants, neuroprotective agents, and natural bioactive compounds derived from traditional medicinal plants\u2014are being continuously developed and applied.8 For example, Shrivastava et al designed and synthesized a class of dual-target small-molecule compounds that simultaneously inhibit acetylcholinesterase (AChE) and \u03b2\u2011site amyloid precursor protein cleaving enzyme\u202f1 (BACE1), demonstrating the feasibility of multi-target intervention in Alzheimer\u2019s disease (AD).9", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Introduction", "text": "Disruption of copper (Cu) homeostasis has been identified as a key pathogenic mechanism in neurological disorders.10 Numerous studies have shown that dysregulation of copper homeostasis may impact the nervous system through mechanisms such as oxidative stress (OS), mitochondrial dysfunction, neuroinflammation, and protein misfolding.11 The premature aging model, copper sulfate\u2010induced stress\u2010induced premature senescence (CuSO4\u2011SIPS), has also confirmed copper\u2019s critical role in age-related functional decline and the development of age-associated diseases.12 As an essential trace element, copper is indispensable for the normal physiological activities of higher plants and animals.13 Human cells can only maintain optimal bioactivity within a narrow concentration range of copper ions.14 Copper homeostasis is delicately regulated through a complex network of copper-dependent proteins, ensuring precise intracellular distribution. Both copper deficiency and copper overload are detrimental to human health.15", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Introduction", "text": "In 2022, Tsvetkov et al first introduced the concept of cuproptosis.16 Unlike apoptosis, pyroptosis, or ferroptosis, cuproptosis is a newly identified form of programmed cell death triggered by copper ion overload.17 Its core mechanism involves excess copper binding to lipoylated proteins in the tricarboxylic acid (TCA) cycle, such as dihydrolipoamide S-acetyltransferase (DLAT) and dihydrolipoamide S-succinyltransferase (DLST), leading to abnormal aggregation of these lipoylated proteins. This disrupts the stability of iron\u2013sulfur (Fe-S) cluster proteins in oxidative phosphorylation (OXPHOS), leading to disulfide bond\u2013dependent aggregation of lipoylated proteins in the TCA cycle, resulting in destabilization or loss of iron-sulfur (Fe-S) cluster proteins and inducing proteotoxic stress.18 Notably, cuproptosis occurs only in cells with active mitochondrial OXPHOS, whereas glycolysis-dependent cells exhibit significant resistance.19", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Introduction", "text": "Lutsenko et al summarized and discussed the physiological functions, molecular mechanisms, and associated diseases of copper homeostasis in mammals.20 Meng et al reviewed the mechanisms of copper metabolism and regulation in the AD brain and discussed the involvement of cuproptosis in the pathological processes of AD.21 Peng et al focused on stroke, examining the physiological roles of copper in stroke and its relationships with pathological processes such as cuproptosis and OS.22 Gao et al summarized current knowledge on copper metabolism, mechanisms of cuproptosis, copper-related cell death, and copper-associated neurological diseases.23 Xu et al systematically summarized brain-specific copper homeostasis, detailing the roles of BBB transporters, glia-mediated copper buffering, and neuron-specific copper chaperones in maintaining cerebral copper balance.24 This article provides an in-depth investigation of the mechanisms underlying copper metabolism and cuproptosis, without limitation to a single disease or neurodegenerative disorders, and comprehensively reviews the associations between dysregulation of copper homeostasis and a wide range of neurological diseases, including WD and MD that can manifest with neurological symptoms.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Introduction", "text": "In addition to traditional chemical drugs and certain plant extracts that have shown beneficial effects in interventions for neurological diseases,25,26 novel therapeutics based on the regulation of copper homeostasis have increasingly become a research focus. This article also compiles potential therapeutic agents targeting copper homeostasis mechanisms, with the aim of providing new perspectives for the prevention and treatment of neurological diseases.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Systemic Copper Metabolism", "text": "Copper is an essential trace element for nearly all living organisms. According to previous studies, copper is extensively utilized by organisms in various physiological processes mediated by copper-dependent enzymes.27 Due to its potential toxicity, copper requires precise transport and homeostatic regulation28 (Figure 1).Figure 1Systemic copper metabolism. Dietary Cu(II) is reduced to Cu(I) in the small intestine, where it is absorbed into intestinal epithelial cells. It is then transported into the portal circulation via ATP7A. In the bloodstream, copper binds to soluble carriers such as ceruloplasmin (CP), \u03b12-macroglobulin, and human serum albumin (HSA), which deliver it to specific tissues or organs. Approximately 75% of copper entering the blood is taken up by the liver. In hepatocytes, copper is imported via CTR1. Within the liver, ATP7B has two major functions: it pumps copper into the Golgi apparatus for CP metalation and, under copper overload conditions, relocates to the apical membrane of bile canaliculi to excrete excess copper into bile, thereby maintaining copper homeostasis. Excess copper is mainly excreted via bile, with small amounts eliminated through urine, sweat, and other routes.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Systemic Copper Metabolism", "text": "Systemic copper metabolism. Dietary Cu(II) is reduced to Cu(I) in the small intestine, where it is absorbed into intestinal epithelial cells. It is then transported into the portal circulation via ATP7A. In the bloodstream, copper binds to soluble carriers such as ceruloplasmin (CP), \u03b12-macroglobulin, and human serum albumin (HSA), which deliver it to specific tissues or organs. Approximately 75% of copper entering the blood is taken up by the liver. In hepatocytes, copper is imported via CTR1. Within the liver, ATP7B has two major functions: it pumps copper into the Golgi apparatus for CP metalation and, under copper overload conditions, relocates to the apical membrane of bile canaliculi to excrete excess copper into bile, thereby maintaining copper homeostasis. Excess copper is mainly excreted via bile, with small amounts eliminated through urine, sweat, and other routes.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Systemic Copper Metabolism", "text": "Dietary copper is absorbed in the stomach, duodenum, and small intestine.29 In the intestinal lumen, Cu(II)(Cu2\u207a) is reduced to Cu(I)(Cu\u207a) by metalloreductases such as six-transmembrane epithelial antigen of the prostate (STEAP)30 and duodenal cytochrome b (DCYTB).31 Cu(I) is then transported into intestinal epithelial cells via the apical high-affinity copper transporter 1 (CTR1/SLC31A1).32 Inside the enterocytes, copper is shuttled by the antioxidant 1 copper chaperone (ATOX1) to the basolateral membrane, where it is exported into the portal circulation by ATPase copper-transporting alpha (ATP7A).33 In the bloodstream, copper binds to soluble carrier proteins such as albumin and \u03b12-macroglobulin for delivery to specific tissues and organs.34", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Systemic Copper Metabolism", "text": "Approximately 75% of the copper entering the bloodstream is taken up by the liver.35 Within hepatocytes, Cu(I) is internalized via CTR1. ATOX1 then delivers copper to ATP7A and ATPase copper-transporting beta (ATP7B) for incorporation into cuproenzymes. The copper chaperone for superoxide dismutase (CCS) transfers copper to superoxide dismutase 1 (SOD1), while the cytochrome c oxidase copper chaperone 17(COX17) delivers copper to cytochrome c oxidase (COX) in the mitochondria. Excess copper is stored by metallothionein (MT) or bound to glutathione (GSH).15,36 ATP7B plays a dual role: it pumps copper into the Golgi apparatus for incorporation into ceruloplasmin (CP), which is then secreted into the bloodstream for systemic distribution, and under conditions of copper excess, ATP7B translocates to the canalicular membrane to excrete surplus copper into the bile, thus regulating systemic copper homeostasis.37", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Systemic Copper Metabolism", "text": "The liver is the primary site for copper storage. Excess copper is mainly excreted via bile, with smaller amounts eliminated through urine, sweat, and desquamated tissues.38 When dietary copper intake is high, copper absorption decreases while excretion increases; the opposite occurs when intake is low.39 When copper levels in peripheral tissues and organs are insufficient to sustain normal physiological function, ATP7B located in the Golgi apparatus mobilizes stored copper from hepatocytes into the bloodstream, where it binds to plasma proteins and enters systemic circulation.40", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Cellular Copper Metabolism", "text": "Under normal physiological conditions, intracellular copper levels are tightly maintained within a narrow range by a complex network of proteins. This network includes cuproenzymes, copper chaperones, and membrane transporters, which work in coordination to regulate copper uptake, utilization, and efflux at the cellular level (Figure 2). Maintaining optimal copper concentrations is essential for cell survival\u2014copper deficiency disrupts cellular respiration and metabolism, while copper excess impairs cell viability and may lead to cell death.41 Unlike traditional mechanisms of cell death, cuproptosis exhibits a strong dependence on mitochondria, particularly the TCA cycle.42Figure 2Intracellular copper metabolism. Cu(II) is reduced to Cu(I) by STEAP on the cell membrane and then enters the cell via CTR1. SLC25A3 transports Cu(I) from the mitochondrial intermembrane space into the mitochondrial matrix and influences superoxide dismutase\u202f1 (SOD1) activity. COX17 transfers Cu(I) to SCO1 and COX11, where COX11 delivers copper to COX1 and SCO1 transfers it to COX2, facilitating the metal center assembly of cytochrome c oxidase (CCO). ATOX1 transports copper and targets it to ATP7A/ATP7B located on the trans-Golgi network (TGN) membrane. ATP7A/ATP7B use ATP hydrolysis to drive transmembrane copper transport, delivering copper into the TGN for the metalation of enzymes such as ceruloplasmin (CP) and extracellular superoxide dismutase\u202f3 (SOD3), which require copper as a cofactor. When cytosolic copper levels rise, ATP7A/ATP7B relocate from the TGN to the plasma membrane or intracellular secretory vesicles to export excess copper. Upon restoration of copper homeostasis, they return to the TGN to resume the transport cycle. ATOX1 can also shuttle copper into the nucleus, where it binds specific DNA promoter sequences to drive gene transcription. In the cytosol, CCS inserts Cu(I) into the copper-binding site of SOD1 and catalyzes the formation of the essential disulfide bond in SOD1.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Cellular Copper Metabolism", "text": "Intracellular copper metabolism. Cu(II) is reduced to Cu(I) by STEAP on the cell membrane and then enters the cell via CTR1. SLC25A3 transports Cu(I) from the mitochondrial intermembrane space into the mitochondrial matrix and influences superoxide dismutase\u202f1 (SOD1) activity. COX17 transfers Cu(I) to SCO1 and COX11, where COX11 delivers copper to COX1 and SCO1 transfers it to COX2, facilitating the metal center assembly of cytochrome c oxidase (CCO). ATOX1 transports copper and targets it to ATP7A/ATP7B located on the trans-Golgi network (TGN) membrane. ATP7A/ATP7B use ATP hydrolysis to drive transmembrane copper transport, delivering copper into the TGN for the metalation of enzymes such as ceruloplasmin (CP) and extracellular superoxide dismutase\u202f3 (SOD3), which require copper as a cofactor. When cytosolic copper levels rise, ATP7A/ATP7B relocate from the TGN to the plasma membrane or intracellular secretory vesicles to export excess copper. Upon restoration of copper homeostasis, they return to the TGN to resume the transport cycle. ATOX1 can also shuttle copper into the nucleus, where it binds specific DNA promoter sequences to drive gene transcription. In the cytosol, CCS inserts Cu(I) into the copper-binding site of SOD1 and catalyzes the formation of the essential disulfide bond in SOD1.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Cellular Copper Metabolism", "text": "Cellular copper uptake primarily depends on the CTR1. Under copper-deficient conditions, Specificity Protein 1 (Sp1) is activated and promotes the transcriptional upregulation of the CTR1 gene.43 Conversely, when copper is sufficient, copper ions suppress Sp1 activity and trigger the endocytosis and degradation of CTR1, thereby downregulating its expression and function.44 Mammalian copper homeostasis is regulated through the Cu-Sp1-CTR1 tripartite model.45\n\nHowever, dietary copper mainly exists in the oxidized form Cu(II), which must first be reduced to Cu(I) by membrane-bound metalloreductases such as STEAP2, STEAP3, and STEAP4,46,47 or by DCYTB,31 before it can be transported into the cell via CTR1. In any tissue expressing STEAP proteins (eg, intestine, liver, heart, prostate), Cu(II) can be reduced to Cu(I) and subsequently absorbed through the CTR1 channel. In intestinal epithelial cells, DCYTB and STEAP enzymes function in concert.48\n\nDivalent metal transporter 1 (DMT1), a member of the proton-coupled metal ion transporter family,49 can mediate Cu(II) uptake when CTR1 is limited or absent, such as in intestinal epithelial cells, though it is less efficient in transporting copper.50 In addition to CTR1 and DMT1, low-affinity Fe(II) transport protein (Fet4p)51,52 and divalent metal ion transporter SMF1 (Smf1p)53 can also transport Cu(II), but due to their low metal specificity, they are collectively referred to as low-affinity transporters.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Cellular Copper Metabolism", "text": "In summary, Cu(II) is first reduced to Cu(I) by STEAP family proteins or DCYTB at the cell membrane, and then transported into cells via CTR1. In contrast, copper ions entering through DMT1, Fet4p, or Smf1p bypass the redox conversion step and are directly taken up in the Cu(II) form.\n\nIn the cytoplasm, copper is transported by ATOX1 and targeted to the trans-Golgi network (TGN) membrane-bound transporters ATP7A and ATP7B.54,55 ATP7A and ATP7B are P-type ATPases that contain six metal-binding domains (MBDs) and utilize ATP hydrolysis to drive the transmembrane transport of copper into the TGN.56,57 This process facilitates the metallation of copper-dependent enzymes such as tyrosinase, CP, and superoxide dismutase 3 (SOD3).58 Under normal copper levels, ATP7A/ATP7B are localized to the TGN, where they function in enzyme loading.59 When cytoplasmic copper concentrations rise, ATP7A and ATP7B relocalize to the plasma membrane or intracellular secretory vesicles to export excess copper.58 Once copper levels normalize, they return to the TGN to resume another transport cycle.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Cellular Copper Metabolism", "text": "ATP7A primarily functions in tissues such as the intestine, placenta, and central nervous system (CNS), where it is responsible for delivering copper into the bloodstream.60 In contrast, ATP7B plays a dominant role in the liver, preventing copper accumulation by excreting it into bile.61 This tissue-specific division of labor ensures proper copper distribution and homeostasis. Mutations in the ATP7A or ATP7B genes result in two distinct genetic disorders: Menkes disease (MD), characterized by copper deficiency,62 and Wilson disease (WD), characterized by copper toxicity.63\n\nCytochrome c oxidase (COX, also known as complex IV) is a critical enzyme complex in the mitochondrial respiratory chain, essential for electron transfer and oxygen reduction, and plays a vital role in maintaining respiratory function.64 solute carrier family 25 member 3 (SLC25A3) transports copper ions from the mitochondrial intermembrane space (IMS) into the matrix, which is crucial for the metallation of COX and influences the activity of SOD1.65 The copper chaperone COX17 specifically transfers Cu(I) to synthesis of cytochrome c oxidase 1 (SCO1) and cytochrome c oxidase copper chaperone 11 (COX11), which assist in loading the copper centers of COX.66,67", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Cellular Copper Metabolism", "text": "Mitochondrially encoded cytochrome c oxidase subunit I (COX1) and subunit II (COX2) are core components of COX, each containing distinct copper-binding sites\u2014CuB and CuA, respectively.68 COX11 delivers copper to the CuB site on COX1, while SCO1 transfers copper to the CuA site on COX2.69 These two copper sites (CuA and CuB) form the catalytic core of COX,70 playing complementary and essential roles in oxygen reduction and electron transport. Under copper-deficient conditions, cells prioritize mitochondrial copper homeostasis, underscoring its primary importance in overall cellular copper balance.71\n\nIn the cytoplasm, CCS receives copper from CTR1 and inserts Cu(I) into the copper-binding site of SOD1, catalyzing the formation of essential disulfide bonds to activate the enzyme.72 SOD1 is an antioxidant enzyme located in both the cytosol and IMS, and CCS also facilitates the maturation of SOD1 within the IMS.73,74\n\nIn the nucleus, ATOX1 can also deliver copper ions, where it binds to specific DNA promoter sequences and promotes gene transcription.75\n\nWhen intracellular copper levels become excessive, ATP7A and ATP7B relocate from the TGN to vesicular compartments, which subsequently fuse with the plasma membrane to excrete copper ions via exocytosis into the bile. This represents a major pathway for endogenous copper excretion.37 Once copper concentrations return to physiological levels, these proteins relocalize back to the TGN.76", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Cellular Copper Metabolism", "text": "According to a study by Gioilli et al, small copper carriers (SCCs) also play an important role in copper transport and homeostasis. SCCs can serve as an alternative copper efflux pathway, transferring copper from the liver and intestinal epithelial cells to other cells, and under conditions of copper overload, they may be excreted into the urine.77", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Uptake", "text": "Cellular copper uptake primarily depends on the CTR1. Under copper-deficient conditions, Specificity Protein 1 (Sp1) is activated and promotes the transcriptional upregulation of the CTR1 gene.43 Conversely, when copper is sufficient, copper ions suppress Sp1 activity and trigger the endocytosis and degradation of CTR1, thereby downregulating its expression and function.44 Mammalian copper homeostasis is regulated through the Cu-Sp1-CTR1 tripartite model.45\n\nHowever, dietary copper mainly exists in the oxidized form Cu(II), which must first be reduced to Cu(I) by membrane-bound metalloreductases such as STEAP2, STEAP3, and STEAP4,46,47 or by DCYTB,31 before it can be transported into the cell via CTR1. In any tissue expressing STEAP proteins (eg, intestine, liver, heart, prostate), Cu(II) can be reduced to Cu(I) and subsequently absorbed through the CTR1 channel. In intestinal epithelial cells, DCYTB and STEAP enzymes function in concert.48\n\nDivalent metal transporter 1 (DMT1), a member of the proton-coupled metal ion transporter family,49 can mediate Cu(II) uptake when CTR1 is limited or absent, such as in intestinal epithelial cells, though it is less efficient in transporting copper.50 In addition to CTR1 and DMT1, low-affinity Fe(II) transport protein (Fet4p)51,52 and divalent metal ion transporter SMF1 (Smf1p)53 can also transport Cu(II), but due to their low metal specificity, they are collectively referred to as low-affinity transporters.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Uptake", "text": "In summary, Cu(II) is first reduced to Cu(I) by STEAP family proteins or DCYTB at the cell membrane, and then transported into cells via CTR1. In contrast, copper ions entering through DMT1, Fet4p, or Smf1p bypass the redox conversion step and are directly taken up in the Cu(II) form.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Utilization", "text": "In the cytoplasm, copper is transported by ATOX1 and targeted to the trans-Golgi network (TGN) membrane-bound transporters ATP7A and ATP7B.54,55 ATP7A and ATP7B are P-type ATPases that contain six metal-binding domains (MBDs) and utilize ATP hydrolysis to drive the transmembrane transport of copper into the TGN.56,57 This process facilitates the metallation of copper-dependent enzymes such as tyrosinase, CP, and superoxide dismutase 3 (SOD3).58 Under normal copper levels, ATP7A/ATP7B are localized to the TGN, where they function in enzyme loading.59 When cytoplasmic copper concentrations rise, ATP7A and ATP7B relocalize to the plasma membrane or intracellular secretory vesicles to export excess copper.58 Once copper levels normalize, they return to the TGN to resume another transport cycle.\n\nATP7A primarily functions in tissues such as the intestine, placenta, and central nervous system (CNS), where it is responsible for delivering copper into the bloodstream.60 In contrast, ATP7B plays a dominant role in the liver, preventing copper accumulation by excreting it into bile.61 This tissue-specific division of labor ensures proper copper distribution and homeostasis. Mutations in the ATP7A or ATP7B genes result in two distinct genetic disorders: Menkes disease (MD), characterized by copper deficiency,62 and Wilson disease (WD), characterized by copper toxicity.63", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Utilization", "text": "Cytochrome c oxidase (COX, also known as complex IV) is a critical enzyme complex in the mitochondrial respiratory chain, essential for electron transfer and oxygen reduction, and plays a vital role in maintaining respiratory function.64 solute carrier family 25 member 3 (SLC25A3) transports copper ions from the mitochondrial intermembrane space (IMS) into the matrix, which is crucial for the metallation of COX and influences the activity of SOD1.65 The copper chaperone COX17 specifically transfers Cu(I) to synthesis of cytochrome c oxidase 1 (SCO1) and cytochrome c oxidase copper chaperone 11 (COX11), which assist in loading the copper centers of COX.66,67\n\nMitochondrially encoded cytochrome c oxidase subunit I (COX1) and subunit II (COX2) are core components of COX, each containing distinct copper-binding sites\u2014CuB and CuA, respectively.68 COX11 delivers copper to the CuB site on COX1, while SCO1 transfers copper to the CuA site on COX2.69 These two copper sites (CuA and CuB) form the catalytic core of COX,70 playing complementary and essential roles in oxygen reduction and electron transport. Under copper-deficient conditions, cells prioritize mitochondrial copper homeostasis, underscoring its primary importance in overall cellular copper balance.71", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Utilization", "text": "In the cytoplasm, CCS receives copper from CTR1 and inserts Cu(I) into the copper-binding site of SOD1, catalyzing the formation of essential disulfide bonds to activate the enzyme.72 SOD1 is an antioxidant enzyme located in both the cytosol and IMS, and CCS also facilitates the maturation of SOD1 within the IMS.73,74\n\nIn the nucleus, ATOX1 can also deliver copper ions, where it binds to specific DNA promoter sequences and promotes gene transcription.75", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Efflux", "text": "When intracellular copper levels become excessive, ATP7A and ATP7B relocate from the TGN to vesicular compartments, which subsequently fuse with the plasma membrane to excrete copper ions via exocytosis into the bile. This represents a major pathway for endogenous copper excretion.37 Once copper concentrations return to physiological levels, these proteins relocalize back to the TGN.76\n\nAccording to a study by Gioilli et al, small copper carriers (SCCs) also play an important role in copper transport and homeostasis. SCCs can serve as an alternative copper efflux pathway, transferring copper from the liver and intestinal epithelial cells to other cells, and under conditions of copper overload, they may be excreted into the urine.77", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Copper Homeostasis Dysregulation", "text": "Reactive oxygen species (ROS) are byproducts of aerobic metabolism. When intracellular ROS levels exceed the cellular antioxidant defense capacity, they induce oxidative stress (OS), leading to oxidative damage of biomolecules and ultimately resulting in cell death.78 Excess copper can increase ROS production, causing OS, while also promoting structural and functional damage to mitochondria. Mitochondrial structural and functional damage, in turn, can feedback to further enhance ROS production, exacerbating oxidative injury.79\n\nDysregulation of copper homeostasis can exert neurotoxicity by inducing OS. In cell models, part of the mechanism by which copper ion carriers induce astrocyte death involves excessive ROS generation, suggesting a role for OS in copper toxicity.80 The use of the compound copper pyrithione (CPT) can also induce OS-mediated cytotoxicity, inhibiting neurite outgrowth.81 Copper deficiency increases the brain\u2019s susceptibility to OS.82 Additionally, copper may enhance the ability of 6-hydroxydopamine (6-OHDA) to induce OS, further damaging dopaminergic neurons.83", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Copper Homeostasis Dysregulation", "text": "Nuclear factor\u202fkappa\u2011light\u2011chain\u2011enhancer of activated B cells (NF-\u03baB) is a nuclear transcription factor that normally exists in the cytoplasm in an inactive state by binding to inhibitor of NF-\u03baB (I\u03baB) proteins.84 Excess copper can activate the I\u03baB kinase complex, promoting phosphorylation and degradation of I\u03baB proteins and thereby releasing the NF-\u03baB complex. The released p65 subunit is phosphorylated and translocates to the nucleus, where it binds specific DNA sequences to initiate transcription of inflammatory factors such as tumor necrosis factor-alpha (TNF-\u03b1), interleukin-1 (IL-1), and interleukin-6 (IL-6).85,86\n\nIn BV2 microglial cells, copper can increase ROS levels, activate the NF-\u03baB pathway, induce secretion of inflammatory factors by microglia, and impair mitophagy, ultimately leading to ferroptosis of dopaminergic neurons.85 Copper-binding peptides can mitigate microglial inflammation by inhibiting the NF-\u03baB pathway.87 In addition, ROS-induced OS can activate the NLRP3 inflammasome, promoting caspase-1\u2013dependent secretion of proinflammatory cytokines and inducing pyroptosis.88,89\n\nMitochondria are double-membrane organelles present in all eukaryotic cells, responsible for aerobic respiration, and cooperate with the cytoskeleton to maintain cell morphology and function.90 They also serve as critical sites for multiple metabolic pathways, including key reactions such as metal ion metabolism.91", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Copper Homeostasis Dysregulation", "text": "In neurons and glial cells, excess copper can accumulate in the mitochondrial matrix, disrupting mitochondrial membrane potential and inhibiting alpha\u2011ketoglutarate dehydrogenase (KGDH) and pyruvate dehydrogenase (PDH) activities, leading to reduced mitochondrial pyruvate production.92 The use of antioxidants such as dihydrolipoic acid can attenuate this process and the subsequent cell death.92\n\nAdditionally, copper can inhibit BNIP3-mediated mitophagy by downregulating the mitochondrial autophagy regulator BCL-2/adenovirus E1B 19kDa interacting protein 3 (BNIP3), which in traumatic brain injury (TBI) models manifests as exacerbated synaptic damage.93 AMP-activated protein kinase (AMPK), a cellular energy sensor, is activated during mitochondrial inhibition to help maintain ATP levels. In the cerebellum of copper-deficient mice, mitochondrial dysfunction leads to energy deficiency, activating AMPK and promoting its phosphorylation, which subsequently phosphorylates acetyl\u2011CoA carboxylase (ACC) to inhibit fatty acid synthesis and conserve energy.94", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Copper Homeostasis Dysregulation", "text": "When copper is present in the synaptic cleft, it can directly or indirectly modulate the activity of neurotransmitter receptors, thereby affecting excitability.95 In addition, copper may participate in the binding of synaptic vesicles to the cell membrane by regulating the interaction between \u03b1-synuclein (\u03b1-Syn) and synaptic vesicles.96 Nam et al proposed that the synaptic cleft may contain three elements: Cu(I)/Cu(II), \u03b2-amyloid protein (A\u03b2), and neurotransmitters. Cu(I)/Cu(II) and neurotransmitters released during neuronal excitation are key components for modulating neurotransmitter receptor activation and maintaining signal transduction in the synapse, whereas under pathological conditions, A\u03b2 may interact with these components to contribute to neurodegeneration.97", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Copper Homeostasis and Oxidative Stress", "text": "Reactive oxygen species (ROS) are byproducts of aerobic metabolism. When intracellular ROS levels exceed the cellular antioxidant defense capacity, they induce oxidative stress (OS), leading to oxidative damage of biomolecules and ultimately resulting in cell death.78 Excess copper can increase ROS production, causing OS, while also promoting structural and functional damage to mitochondria. Mitochondrial structural and functional damage, in turn, can feedback to further enhance ROS production, exacerbating oxidative injury.79\n\nDysregulation of copper homeostasis can exert neurotoxicity by inducing OS. In cell models, part of the mechanism by which copper ion carriers induce astrocyte death involves excessive ROS generation, suggesting a role for OS in copper toxicity.80 The use of the compound copper pyrithione (CPT) can also induce OS-mediated cytotoxicity, inhibiting neurite outgrowth.81 Copper deficiency increases the brain\u2019s susceptibility to OS.82 Additionally, copper may enhance the ability of 6-hydroxydopamine (6-OHDA) to induce OS, further damaging dopaminergic neurons.83", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Copper Homeostasis and Inflammatory Response", "text": "Nuclear factor\u202fkappa\u2011light\u2011chain\u2011enhancer of activated B cells (NF-\u03baB) is a nuclear transcription factor that normally exists in the cytoplasm in an inactive state by binding to inhibitor of NF-\u03baB (I\u03baB) proteins.84 Excess copper can activate the I\u03baB kinase complex, promoting phosphorylation and degradation of I\u03baB proteins and thereby releasing the NF-\u03baB complex. The released p65 subunit is phosphorylated and translocates to the nucleus, where it binds specific DNA sequences to initiate transcription of inflammatory factors such as tumor necrosis factor-alpha (TNF-\u03b1), interleukin-1 (IL-1), and interleukin-6 (IL-6).85,86\n\nIn BV2 microglial cells, copper can increase ROS levels, activate the NF-\u03baB pathway, induce secretion of inflammatory factors by microglia, and impair mitophagy, ultimately leading to ferroptosis of dopaminergic neurons.85 Copper-binding peptides can mitigate microglial inflammation by inhibiting the NF-\u03baB pathway.87 In addition, ROS-induced OS can activate the NLRP3 inflammasome, promoting caspase-1\u2013dependent secretion of proinflammatory cytokines and inducing pyroptosis.88,89", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Copper Homeostasis and Mitochondrial Dysfunction", "text": "Mitochondria are double-membrane organelles present in all eukaryotic cells, responsible for aerobic respiration, and cooperate with the cytoskeleton to maintain cell morphology and function.90 They also serve as critical sites for multiple metabolic pathways, including key reactions such as metal ion metabolism.91\n\nIn neurons and glial cells, excess copper can accumulate in the mitochondrial matrix, disrupting mitochondrial membrane potential and inhibiting alpha\u2011ketoglutarate dehydrogenase (KGDH) and pyruvate dehydrogenase (PDH) activities, leading to reduced mitochondrial pyruvate production.92 The use of antioxidants such as dihydrolipoic acid can attenuate this process and the subsequent cell death.92\n\nAdditionally, copper can inhibit BNIP3-mediated mitophagy by downregulating the mitochondrial autophagy regulator BCL-2/adenovirus E1B 19kDa interacting protein 3 (BNIP3), which in traumatic brain injury (TBI) models manifests as exacerbated synaptic damage.93 AMP-activated protein kinase (AMPK), a cellular energy sensor, is activated during mitochondrial inhibition to help maintain ATP levels. In the cerebellum of copper-deficient mice, mitochondrial dysfunction leads to energy deficiency, activating AMPK and promoting its phosphorylation, which subsequently phosphorylates acetyl\u2011CoA carboxylase (ACC) to inhibit fatty acid synthesis and conserve energy.94", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Copper Homeostasis and Synaptic Function", "text": "When copper is present in the synaptic cleft, it can directly or indirectly modulate the activity of neurotransmitter receptors, thereby affecting excitability.95 In addition, copper may participate in the binding of synaptic vesicles to the cell membrane by regulating the interaction between \u03b1-synuclein (\u03b1-Syn) and synaptic vesicles.96 Nam et al proposed that the synaptic cleft may contain three elements: Cu(I)/Cu(II), \u03b2-amyloid protein (A\u03b2), and neurotransmitters. Cu(I)/Cu(II) and neurotransmitters released during neuronal excitation are key components for modulating neurotransmitter receptor activation and maintaining signal transduction in the synapse, whereas under pathological conditions, A\u03b2 may interact with these components to contribute to neurodegeneration.97", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Mechanism of Cuproptosis", "text": "Copper is an essential trace element in organisms, playing irreplaceable roles in maintaining mitochondrial respiration, enzyme activity, and protein function.98 Cells can maintain normal activity only within an extremely narrow range of copper ion concentrations. Experiments using the MC3T3-E1 cell line (derived from the calvaria of a newborn mouse) showed that cells retain high activity only at copper concentrations of approximately 10\u22127-10\u22126\u202fM, and even slight increases approach the toxicity threshold.99,100 Dysregulation of copper homeostasis can induce mitochondrial morphological and functional abnormalities, leading to metabolic disorders and irreversible damage to the organism101 (Figure 3).Figure 3Mechanism of cuproptosis. Cu(II) is directly imported into the cell and mitochondria by the copper ionophore ELC. In mitochondria, FDX1 catalyzes the reduction of Cu(II) to Cu(I), accompanied by the generation of reactive oxygen species (ROS). FDX1 also interacts with LIAS to promote the lipoylation of DLAT. The lipoylated DLAT proteins interact with Cu(I) and subsequently aggregate. Additionally, Cu(I) disrupts iron\u2013sulfur (Fe\u2013S) cluster proteins, leading to their loss. Together, these effects induce cuproptosis. Cuproptosis causes the accumulation of key tricarboxylic acid (TCA) cycle components such as DLAT, DLST, NADH, and FADH2, reducing their activity and suppressing the electron transport chain (ETC), thereby decreasing ATP production.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Mechanism of Cuproptosis", "text": "Mechanism of cuproptosis. Cu(II) is directly imported into the cell and mitochondria by the copper ionophore ELC. In mitochondria, FDX1 catalyzes the reduction of Cu(II) to Cu(I), accompanied by the generation of reactive oxygen species (ROS). FDX1 also interacts with LIAS to promote the lipoylation of DLAT. The lipoylated DLAT proteins interact with Cu(I) and subsequently aggregate. Additionally, Cu(I) disrupts iron\u2013sulfur (Fe\u2013S) cluster proteins, leading to their loss. Together, these effects induce cuproptosis. Cuproptosis causes the accumulation of key tricarboxylic acid (TCA) cycle components such as DLAT, DLST, NADH, and FADH2, reducing their activity and suppressing the electron transport chain (ETC), thereby decreasing ATP production.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Mechanism of Cuproptosis", "text": "Unlike OS induced by copper overload, cuproptosis occurs primarily in mitochondria and depends on the interaction between copper ions and mitochondrial metabolism.102 Moreover, multiple studies have shown that cuproptosis can proceed even in the presence of antioxidants such as N-acetylcysteine (NAC) or ferrostatin\u20111, indicating that ROS are a concomitant rather than an essential driving factor.79,103 ROS accumulation may also activate other programmed cell death pathways, including apoptosis and ferroptosis, but these are only concurrent phenomena and not central mechanisms of cuproptosis.104 Furthermore, classical inhibitors of apoptosis, necrosis, pyroptosis, and ferroptosis (eg, caspase inhibitors, ferrostatin\u20111) are ineffective against cuproptosis; only copper chelators or knockout of key genes such as FDX1 and lipoyltransferases can reverse cell fate.105 These findings indicate that cuproptosis is a novel cell death mechanism completely independent of known programmed cell death pathways.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Mechanism of Cuproptosis", "text": "Under normal physiological conditions, copper uptake is mediated by CTR1, and copper efflux is regulated by ATP7A/ATP7B; however, under high copper load or in the presence of copper ionophores, these regulatory mechanisms fail to maintain copper homeostasis. Excess copper or exogenous copper ion carriers such as Elesclomol (ELC) can induce cuproptosis. ELC forms an Elesclomol-Cu(II) complex that transports Cu(II) from the extracellular space into the cytoplasm and mitochondria. Upon dissociation from the complex, Cu(II) undergoes repeated cycles of binding and release, leading to progressive accumulation of Cu(II) within mitochondria.106,107", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Mechanism of Cuproptosis", "text": "In the mitochondrial matrix, Ferredoxin 1 (FDX1), an Fe-S cluster reductase, serves as a critical regulator of cuproptosis and plays a dual role. FDX1 reduces Cu(II) to the more toxic Cu(I), a process that is particularly pronounced in cells highly dependent on OXPHOS.108 FDX1 is an upstream regulator of protein lipoylation. By directly interacting with lipoic acid synthetase (LIAS), FDX1 promotes LIAS-mediated lipoylation of target proteins, increasing the lipoylation levels of TCA cycle-related enzymes such as DLAT, thereby providing a basis for the interaction between copper and these modified proteins and ultimately contributing to the initiation of the cuproptosis signaling pathway.109 Knockout of FDX1 confers resistance to copper ionophore (such as ELC)-induced cell death, indicating that FDX1 is a key upstream regulator of cuproptosis.108 A 2024 study on the drug disulfiram (DSF) showed that inhibition of FDX1 reduces copper ion accumulation and alleviates cuproptosis. Additionally, it mitigates ischemia-reperfusion-induced neuroinflammation through modulation of the heat shock protein 70 (HSP70)/toll-like receptor 4 (TLR4)/NLR family pyrin domain containing 3 (NLRP3) pathway, contributing to brain tissue protection.110", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Mechanism of Cuproptosis", "text": "Lipoylation is a highly conserved post-translational modification in which lipoic acid is covalently attached to lysine residues of proteins, primarily occurring on key complexes of the mitochondrial TCA cycle and related metabolic pathways, including DLAT, DLST, dihydrolipoamide branched\u2011chain transacylase E2 (DBT), and glycine cleavage system protein\u202fH (GCSH).111 Excess Cu(I) can bind to certain critical mitochondrial metabolic enzymes, causing their misfolding and aggregation into clumps, thereby obstructing energy metabolism and activating the cellular proteostasis machinery. Ultimately, the inability to clear these damaged proteins leads to mitochondrial instability, collapse of cellular function, and the occurrence of cuproptosis.112,113\n\nIn in vitro experiments, FDX1 deletion causes mitochondrial proteins to lose lipoylation modifications and their Cu(II) binding sites, thereby reducing copper toxicity. However, because lipoylation is essential for the function of TCA cycle enzymes, FDX1 deletion impairs TCA metabolism, diminishes mitochondrial respiration and OXPHOS function, and markedly restricts cell growth under low-glucose conditions.109", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Mechanism of Cuproptosis", "text": "In cuproptosis, when a large amount of copper binds to lipoylated proteins, it not only induces aggregation of these proteins but also disrupts the stability of Fe-S cluster proteins, leading to their depletion and inactivation. Since Fe-S cluster proteins are extensively involved in the TCA cycle, electron transport, and nucleic acid repair, their loss exacerbates OXPHOS inefficiency, disrupts electron transport, increases ROS production, and causes widespread metabolic dysfunction.16,114 The loss of Fe-S cluster proteins, together with lipoylated protein aggregation, synergistically triggers mitochondrial proteotoxic stress, further promoting cell death.115", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Mitochondrial Lipoylated Protein Aggregation", "text": "Under normal physiological conditions, copper uptake is mediated by CTR1, and copper efflux is regulated by ATP7A/ATP7B; however, under high copper load or in the presence of copper ionophores, these regulatory mechanisms fail to maintain copper homeostasis. Excess copper or exogenous copper ion carriers such as Elesclomol (ELC) can induce cuproptosis. ELC forms an Elesclomol-Cu(II) complex that transports Cu(II) from the extracellular space into the cytoplasm and mitochondria. Upon dissociation from the complex, Cu(II) undergoes repeated cycles of binding and release, leading to progressive accumulation of Cu(II) within mitochondria.106,107", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Mitochondrial Lipoylated Protein Aggregation", "text": "In the mitochondrial matrix, Ferredoxin 1 (FDX1), an Fe-S cluster reductase, serves as a critical regulator of cuproptosis and plays a dual role. FDX1 reduces Cu(II) to the more toxic Cu(I), a process that is particularly pronounced in cells highly dependent on OXPHOS.108 FDX1 is an upstream regulator of protein lipoylation. By directly interacting with lipoic acid synthetase (LIAS), FDX1 promotes LIAS-mediated lipoylation of target proteins, increasing the lipoylation levels of TCA cycle-related enzymes such as DLAT, thereby providing a basis for the interaction between copper and these modified proteins and ultimately contributing to the initiation of the cuproptosis signaling pathway.109 Knockout of FDX1 confers resistance to copper ionophore (such as ELC)-induced cell death, indicating that FDX1 is a key upstream regulator of cuproptosis.108 A 2024 study on the drug disulfiram (DSF) showed that inhibition of FDX1 reduces copper ion accumulation and alleviates cuproptosis. Additionally, it mitigates ischemia-reperfusion-induced neuroinflammation through modulation of the heat shock protein 70 (HSP70)/toll-like receptor 4 (TLR4)/NLR family pyrin domain containing 3 (NLRP3) pathway, contributing to brain tissue protection.110", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Mitochondrial Lipoylated Protein Aggregation", "text": "Lipoylation is a highly conserved post-translational modification in which lipoic acid is covalently attached to lysine residues of proteins, primarily occurring on key complexes of the mitochondrial TCA cycle and related metabolic pathways, including DLAT, DLST, dihydrolipoamide branched\u2011chain transacylase E2 (DBT), and glycine cleavage system protein\u202fH (GCSH).111 Excess Cu(I) can bind to certain critical mitochondrial metabolic enzymes, causing their misfolding and aggregation into clumps, thereby obstructing energy metabolism and activating the cellular proteostasis machinery. Ultimately, the inability to clear these damaged proteins leads to mitochondrial instability, collapse of cellular function, and the occurrence of cuproptosis.112,113\n\nIn in vitro experiments, FDX1 deletion causes mitochondrial proteins to lose lipoylation modifications and their Cu(II) binding sites, thereby reducing copper toxicity. However, because lipoylation is essential for the function of TCA cycle enzymes, FDX1 deletion impairs TCA metabolism, diminishes mitochondrial respiration and OXPHOS function, and markedly restricts cell growth under low-glucose conditions.109", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Loss of Iron-Sulfur Cluster Proteins", "text": "In cuproptosis, when a large amount of copper binds to lipoylated proteins, it not only induces aggregation of these proteins but also disrupts the stability of Fe-S cluster proteins, leading to their depletion and inactivation. Since Fe-S cluster proteins are extensively involved in the TCA cycle, electron transport, and nucleic acid repair, their loss exacerbates OXPHOS inefficiency, disrupts electron transport, increases ROS production, and causes widespread metabolic dysfunction.16,114 The loss of Fe-S cluster proteins, together with lipoylated protein aggregation, synergistically triggers mitochondrial proteotoxic stress, further promoting cell death.115", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Alzheimer\u2019s disease (AD) is one of the most common neurological disorders, clinically characterized by progressive cognitive impairments in visual, language, executive, behavioral, and motor domains.116,117 The formation of extracellular A\u03b2 plaques and neurofibrillary tangles (NFTs) composed of hyperphosphorylated tau protein in the hippocampus and cerebral cortex are central to AD pathogenesis.118 According to the amyloid cascade hypothesis, abnormal processing of amyloid precursor protein (APP) during pathology leads to extracellular accumulation of A\u03b2, which triggers tau hyperphosphorylation and ultimately NFT formation.119,120 Concurrently, A\u03b2 aggregation induces neuroinflammation, oxidative stress (OS), and synaptic loss, culminating in neuronal death and cognitive decline.121", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "In AD patients, free copper levels (particularly in serum) are significantly elevated, resulting in diminished antioxidant defenses and mitochondrial dysfunction. Moreover, increased copper levels correlate with the rate of cognitive decline,122,123 and serum copper levels positively associate with AD risk124 (Figure 4).Figure 4Copper homeostasis and cuproptosis in Alzheimer\u2019s disease (AD). IFN-\u03b3 disrupts copper homeostasis. Excess copper increases free radicals via the Fenton reaction, leading to oxidative stress (OS) and cell death. These radicals interact with A\u03b2 and APP, promoting A\u03b2 aggregation and ROS production. Copper accumulation reduces synaptic proteins (SYP, PSD-95) and neurotransmitters (5-HT, GABA), and inhibits BDNF production through the CREB/BDNF pathway, impairing synaptic plasticity. APP cleavage by \u03b1- and \u03b3-secretase produces non-toxic P3, while BACE1 and \u03b3-secretase generate A\u03b240 and A\u03b242. Copper promotes \u03b1-cleavage, whereas copper deficiency favors \u03b2-cleavage, worsening AD pathology. Microglia can clear A\u03b2 or A\u03b2-antibody complexes, but excess copper impairs this function. In the hippocampus, copper elevates NO and LPO levels, damages antioxidant defenses, and disrupts neuronal function. Copper also induces tau hyperphosphorylation and aggregation.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Copper homeostasis and cuproptosis in Alzheimer\u2019s disease (AD). IFN-\u03b3 disrupts copper homeostasis. Excess copper increases free radicals via the Fenton reaction, leading to oxidative stress (OS) and cell death. These radicals interact with A\u03b2 and APP, promoting A\u03b2 aggregation and ROS production. Copper accumulation reduces synaptic proteins (SYP, PSD-95) and neurotransmitters (5-HT, GABA), and inhibits BDNF production through the CREB/BDNF pathway, impairing synaptic plasticity. APP cleavage by \u03b1- and \u03b3-secretase produces non-toxic P3, while BACE1 and \u03b3-secretase generate A\u03b240 and A\u03b242. Copper promotes \u03b1-cleavage, whereas copper deficiency favors \u03b2-cleavage, worsening AD pathology. Microglia can clear A\u03b2 or A\u03b2-antibody complexes, but excess copper impairs this function. In the hippocampus, copper elevates NO and LPO levels, damages antioxidant defenses, and disrupts neuronal function. Copper also induces tau hyperphosphorylation and aggregation.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Copper-induced OS may be one of the primary mechanisms underlying its neurotoxicity. Since copper in the brain can cycle between Cu(II) and Cu(I) oxidation states, it possesses redox activity and can readily catalyze the production of ROS, thereby exacerbating OS in the brain.125 Excess copper generates large amounts of free radicals through the Fenton reaction, inducing oxidative damage and cell death, which leads to neurodegeneration.126 The free radicals produced by the Fenton reaction are not only closely related to OS but can also directly interact with A\u03b2 plaques and APP, promoting A\u03b2 synthesis and aggregation.127 Copper can also bind directly to A\u03b2 plaques, catalyzing ROS generation; these ROS may cause oxidative damage to the A\u03b2 peptides themselves as well as surrounding molecules such as proteins and lipids.128 In the hippocampus, copper treatment significantly elevates levels of nitric oxide (NO) and lipid peroxidation products (LPO), impairs the antioxidant defense system, and consequently damages neural function.129 Copper overload can disrupt critical mitochondrial enzymes such as COX, causing electron transport chain blockage, sharply increasing ROS production, activating cuproptosis, and ultimately inducing neuronal apoptosis or necrosis.130 As the main energy supplier of cells, mitochondrial dysfunction not only increases neurotoxicity but also leads to synaptic loss and memory impairment.131", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Hippocampal dysfunction typically occurs early in AD, accompanied by inhibition of long\u2011term potentiation (LTP) and synaptic dysfunction in patients\u2019 brains. Excessive copper accumulation can downregulate presynaptic protein synaptophysin (SYP), postsynaptic density protein 95 (PSD-95), and neurotransmitters such as 5-hydroxytryptamine (5-HT) and gamma-aminobutyric acid (GABA) in mice.132 SYP expression correlates closely with the number of presynaptic vesicles, while PSD-95 supports and anchors postsynaptic receptors and regulates receptor-associated protein organization.133 5-HT and GABA are key neurotransmitters in the synaptic cleft involved in learning and memory processes.134,135 These four molecules are all involved in higher brain functions within the synaptic system, indicating that copper can simultaneously affect presynaptic regulation, postsynaptic modulation, and neurotransmitter release, leading to synaptic transmission impairments. Moreover, brain-derived neurotrophic factor (BDNF), an important target of cAMP response element-binding protein (CREB),136 regulates synaptic plasticity and participates in neuroprotection and neural regeneration.137 Copper may inhibit the CREB/BDNF signaling pathway, reducing BDNF expression and thereby suppressing synaptic plasticity, which hinders memory formation and consolidation.132", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Persistent activation of microglia is closely related to abnormal copper homeostasis and is involved in AD pathology.138 Microglia can phagocytose A\u03b2 or A\u03b2-antibody complexes,139 but excessive copper ions inhibit this phagocytic function. The inflammatory cytokine interferon-\u03b3 (IFN-\u03b3) not only triggers and exacerbates inflammation but also disrupts copper homeostasis by altering the cytoplasmic relocalization of copper transporter ATP7A, increasing copper uptake and upregulating CTR1 expression.140 This process may explain the fluctuating copper homeostasis observed in AD patients.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Furthermore, APP and tau proteins are critical in AD pathogenesis. APP acts as a receptor in the CNS, involved in synaptic growth, neuronal adhesion, axonal transport, and development.141 Under the action of \u03b1-secretase and \u03b3-secretase, APP produces a non-toxic P3 fragment. However, cleavage by \u03b2-secretase (BACE1) and \u03b3-secretase generates the longer peptides A\u03b240 and A\u03b242, which form amyloid plaques in the AD brain.142,143 Increased copper levels promote the \u03b1-secretase-mediated non-toxic pathway, whereas copper deficiency promotes the BACE1-mediated toxic pathway, accelerating A\u03b2 formation.144 Tau protein, a neuron-specific microtubule-associated protein, normally maintains axonal microtubule stability. In AD, tau becomes hyperphosphorylated and dissociates from microtubules, causing axonal transport deficits and neuronal structural damage, ultimately triggering cell death.145 ROS generation is an important trigger of tau hyperphosphorylation and filament aggregation. Excess copper promotes tau phosphorylation, while treatment with copper chelators can inhibit copper levels and markedly reduce tau phosphorylation.146,147 Studies also suggest that reducing brain copper by limiting dietary copper intake may be a feasible strategy to regulate tau pathology in AD.146", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Parkinson\u2019s disease (PD) is the second most common neurodegenerative disorder after AD, with rapidly increasing global prevalence and mortality.148 Clinically, it is characterized by bradykinesia, resting tremor, muscle rigidity, postural and gait abnormalities, often accompanied by cognitive impairment.149 Pathologically, PD is marked by progressive loss of dopaminergic neurons in the substantia nigra-striatum region and the presence of \u03b1-Syn inclusions in neuronal cytoplasm, forming Lewy bodies and Lewy neurites.150 \u03b1-Syn undergoes oligomerization and converts into \u03b2-sheet-rich protofibrils, forming cytotoxic amyloid aggregates.151 These aggregates coexist with mitochondrial dysfunction and decreased copper and GSH levels in brain regions, implicating OS and metabolic imbalance in PD pathology progression.152", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Studies show that copper levels are decreased in multiple brain regions of PD patients post-mortem, especially in the substantia nigra, as measured by inductively coupled plasma mass spectrometry (ICP-MS).153 Synchrotron X-ray fluorescence microscopy and particle-induced X-ray emission analyses indicate a 55\u201365% reduction in copper in the substantia nigra and locus coeruleus of PD brains154 (Figure 5). CP, the main copper carrier and ferroxidase in plasma, shows dysregulated function in PD patients. Reduced CP-bound copper may cause iron homeostasis disturbance, worsening neuronal injury.Figure 5Copper homeostasis and cuproptosis in Parkinson\u2019s disease (PD). In PD, copper directly interacts with \u03b1-synuclein (\u03b1-Syn), promoting its aggregation into toxic oligomers. These complexes generate ROS and dopamine oxidation products, impair energy metabolism, and exacerbate neuronal damage. Intracellular GSH chelates copper and reduces its toxicity. When GSH is depleted, free copper increases, leading to elevated DLAT levels and loss of Fe\u2013S cluster proteins, ultimately triggering copper-induced cell death. Copper bound to ceruloplasmin (CP) helps maintain iron homeostasis, which may protect neurons. CTR1 promotes \u03b1-Syn phosphorylation, contributing to dopaminergic neuron loss and PD progression. Overexpression of CTR1 may also deplete GSH and induce caspase-3-dependent cell death. Additionally, GSTM1-null and GSTT1-null genotypes may increase PD susceptibility.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Copper homeostasis and cuproptosis in Parkinson\u2019s disease (PD). In PD, copper directly interacts with \u03b1-synuclein (\u03b1-Syn), promoting its aggregation into toxic oligomers. These complexes generate ROS and dopamine oxidation products, impair energy metabolism, and exacerbate neuronal damage. Intracellular GSH chelates copper and reduces its toxicity. When GSH is depleted, free copper increases, leading to elevated DLAT levels and loss of Fe\u2013S cluster proteins, ultimately triggering copper-induced cell death. Copper bound to ceruloplasmin (CP) helps maintain iron homeostasis, which may protect neurons. CTR1 promotes \u03b1-Syn phosphorylation, contributing to dopaminergic neuron loss and PD progression. Overexpression of CTR1 may also deplete GSH and induce caspase-3-dependent cell death. Additionally, GSTM1-null and GSTT1-null genotypes may increase PD susceptibility.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "In vitro studies demonstrate that Cu(II) effectively induces spontaneous oligomerization of \u03b1-Syn at micromolar concentrations, promoting formation of oligomers and protofibrils.155,156 Binding of \u03b1-Syn with Cu(II) produces an N-terminal acetylated \u03b1-Syn-Cu(II) complex, exhibiting catechol oxidase and redox activity, catalyzing dopamine oxidation and ROS generation; this activity is further enhanced when membrane-bound.157,158 Additionally, metabolic energy decline is closely linked to disrupted catecholamine metabolism in highly active catecholaminergic neurons (eg, dopaminergic neurons in the substantia nigra pars compacta and noradrenergic neurons in the locus coeruleus), which disrupts redox-active metals such as copper and iron homeostasis.159 These mechanisms suggest \u03b1-Syn-Cu(II) exacerbates dopaminergic/noradrenergic neuronal dysfunction via OS and metabolic imbalance, contributing to PD pathology.\n\nCTR1 is the primary high-affinity copper importer. In PD animal models, CTR1 deficiency significantly reduces \u03b1-Syn phosphorylation at S129, decreases dopaminergic neuron loss, and improves motor dysfunction.160 Conversely, CTR1 overexpression or excess causes intracellular GSH depletion and caspase-3-dependent cell death, indicating that excessive copper can also induce neuronal injury.161 Decoppering treatment partially restores apoptosis levels and expression of copper death-related proteins,132 supporting a role for copper death mechanisms in PD.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Copper homeostasis is closely linked with GSH metabolism. Zhang et al used the high-efficiency GSH fluorescent probe R13 to find reduced GSH levels in PD mouse brains.162 Clinically, GSH levels are significantly decreased in the substantia nigra of PD patients.163 GSH chelates Cu(I), blocking copper toxicity; however, GSH deficiency increases free Cu(I), disrupting Fe-S cluster proteins and inducing OS and cell death.16 From a genetic perspective, deletion polymorphisms in glutathione S-transferase Mu1 (GSTM1) and Theta1 (GSTT1) are associated with increased PD susceptibility,164 emphasizing the importance of GSH metabolism in defending against copper toxicity. Supplementation with cysteine precursors (eg, 6-OHDA/xanthine induction schemes) enhances cellular GSH levels and restores neuronal vitality.165,166 GSH deficiency may underlie copper-induced neurodegeneration, making modulation of GSH metabolism a potential PD therapeutic strategy.\n\nRecent research links amorphous SOD1 aggregates with PD progression; these new SOD1-containing aggregates are amorphous and spherical, largely devoid of \u03b1-Syn but ubiquitin-positive, with copper as a structural and functional cofactor of SOD1, implicating copper deficiency in abnormal SOD1 aggregation.167 Indeed, SOD1 immunoreactivity has been detected in Lewy bodies and Lewy neurites of the substantia nigra and locus coeruleus in PD brains.168", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "In summary, copper homeostasis imbalance in PD manifests as localized copper deficiency rather than widespread copper toxicity, involving multiple mechanisms including \u03b1-Syn aggregation, catecholamine oxidation, GSH deficiency-induced metabolic imbalance, Fe-S protein dysfunction, and SOD1 aggregation. A complex network comprising CTR1, CP, GST, SOD1, and Fe\u2013S cluster proteins underlies these processes. These findings provide a theoretical foundation for PD early diagnosis and therapeutic strategies targeting copper metabolism.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Huntington\u2019s disease (HD) is a progressive neurodegenerative disorder with a genetic origin. The most common symptoms include loss of energy and initiative, poor perseverance and work quality, impaired judgment, decreased self-care ability, and emotional blunting; about half of studied patients exhibit emotional symptoms such as depression, anxiety, and irritability.169 HD is an autosomal dominant hereditary disease caused by CAG repeat expansion mutations in the gene encoding huntingtin (HTT) protein located on the short arm of chromosome 4.170 The N-terminus of the HTT protein contains a polyglutamine (polyQ) tract, and when the CAG repeats expand, a structurally defective mutated huntingtin (mHTT) protein is produced.171 This mHTT protein is prone to misfolding and forms \u03b2-sheet-rich aggregates and inclusions in the neuronal cytoplasm and nucleus.172 These abnormal aggregates disrupt cellular protein homeostasis, axonal transport, transcription, translation, mitochondrial, and synaptic functions.173,174 Diagnosis is typically confirmed by detecting the number of CAG repeats (\u226536 repeats, with \u226540 repeats being fully pathogenic) in the HTT gene of patients exhibiting clinical features.175", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Researchers have found abnormally high concentrations of copper in the striatum of both HD patients and HD mouse models, suggesting that copper may be involved in HD pathogenesis.176 In Drosophila models, copper ions increase the formation of mHTT aggregates and enhance their neurotoxicity,177 while copper chelators inhibit mHTT aggregate formation, further supporting the link between copper and HD pathogenesis.178 Copper ions promote mHTT aggregate formation by binding to mHTT monomers and oligomers, lowering the nucleation energy barrier for aggregation.176 In vitro studies show that copper ions selectively bind to the N-terminal fragment of HTT protein containing 17\u201368 glutamine residues, while other metal ions such as Fe3\u207a and Zn2\u207a show no significant binding affinity.178 Excess copper in the brain inhibits key mitochondrial and energy metabolism dehydrogenases, including lactate dehydrogenase (LDH)179 and succinate dehydrogenase (SDH),92 leading to energy metabolism imbalance. Neurons highly depend on lactate released by astrocytes as their primary energy substrate;180 thus, copper inhibition of LDH directly disrupts lactate metabolism, further weakening neuronal energy supply and exacerbating HD neurodegeneration. Additionally, in Drosophila HD models, copper ions increase mHTT aggregate formation and enhance accumulation of sulfur-positive A\u03b2 structures within mHTT aggregates, representing the toxic aggregation of mHTT and altering autophagy in the brain.177 Studies also show that copper ions directly bind mHTT, increasing its aggregation propensity and \u03b2-sheet structure, thereby enhancing its cytotoxicity in neurons.176", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Mutant HTT disrupts nuclear pore proteins in the nuclear pore complex (NPC), impairing nucleocytoplasmic transport and causing cellular toxicity.181 Meanwhile, mHTT interacts with the mitochondrial protein dynamin-related protein 1 (DRP1), promoting mitochondrial fission, resulting in mitochondrial dysfunction and synaptic protein sequestration.182 Moreover, compared to healthy controls, levels of several ubiquitin-proteasome system (UPS) enzymes, including ubiquitin-activating enzyme (E1), ubiquitin-conjugating enzyme (E2), and ubiquitin ligase (E3), are altered in HD patients. Increased expression of these enzymes may affect proteasome-dependent degradation of mHTT or influence its solubility, aggregation, or assembly, thereby directly or indirectly impacting cellular pathways and stress responses involved in HD pathogenesis.183 The UPS plays a crucial role in clearing aggregated mHTT protein.184 Additionally, UPS participates in the translation and regulation of copper transport proteins and copper chaperones; however, since UPS contains copper-dependent enzyme components, it is itself influenced by copper levels.184,185", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Amyotrophic lateral sclerosis (ALS) is a common neurological disorder characterized by the simultaneous degeneration of upper motor neurons in the motor cortex and lower motor neurons in the spinal cord and brainstem.186 Most ALS cases (~90%) are sporadic without a family history, while a minority (~10%) represent rare autosomal dominant familial amyotrophic lateral sclerosis (FALS).187 The exact etiology of sporadic amyotrophic lateral sclerosis (SALS) remains unclear; however, current consensus suggests it is a complex multistep disease triggered by a combination of genetic susceptibility, environmental exposures, and aging factors.188 The primary cause of FALS is mutations in the SOD1 gene, which encodes a metalloprotein that forms a highly stable homodimer by binding copper and zinc ions. SOD1 mutations cause OS, mitochondrial dysfunction, electron transport disruption, and protein aggregation.189,190", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "It is widely accepted that accumulation of misfolded conformers, oligomers, and aggregates of SOD1 in motor neurons is a core pathological mechanism of ALS. Due to disulfide bond disruption, SOD1 oligomers exhibit enhanced pro-oxidant activity and toxicity.191 Mutant SOD1 reduces protein folding stability by impairing metal binding and disulfide bond formation, leading to misfolding, aggregation, and eventual cytotoxicity.192 Copper is not only central to SOD1 enzymatic activity but is also crucial for its structural integrity and stability; abnormal copper metabolism may contribute to ALS pathogenesis by affecting SOD1 function or aggregation.193,194 One of the earliest pathological features in mutant SOD1 is loss of metal at copper-binding sites.195 Copper deficiency is common in mutant SOD1 cells, and aggregates are generally copper-deficient.196 Moreover, copper levels in cerebrospinal fluid (CSF) from ALS patients are lower than controls, especially in spinal-onset ALS, where copper concentration significantly decreases.197", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Overexpression of COX in adult mice carrying mutant SOD1 results in impaired copper transport in the late stages of life, suggesting that age-related copper deficiency may be an important factor in the onset of SOD1-induced ALS.198 Deficiency of metals such as copper and zinc promotes the exposure of hydrophobic residues in SOD1, which facilitates the formation and stabilization of toxic SOD1 oligomers. These oligomers exhibit high toxicity and pro-oxidant activity, representing a key pathological process leading to neuronal damage.199 In addition, the disruption of copper binding and disulfide bond formation in SOD1 may contribute to the onset and progression of ALS.200 When SOD1 forms disulfide bonds before binding metal ions, it loses the ability to be recognized and activated by CCS, thereby blocking the copper supply from CCS and hindering the maturation of the disulfide bond in SOD1.201", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Conversely, abnormally elevated free copper levels may act as immune stimuli. Misallocated copper ions generate large amounts of free radicals via Fenton reactions, directly causing oxidative damage and inducing microglial release of pro-inflammatory cytokines such as tumor necrosis factor-alpha (TNF-\u03b1) and interleukin-1 beta (IL-1\u03b2).194 Excess copper also activates NF-\u03baB in astrocytes, amplifying inflammatory responses and further damaging motor neurons.194 Indeed, in SOD1-ALS models, both copper deficiency (leading to protein dysfunction) and copper overload (triggering toxic inflammation) coexist and synergistically drive disease progression.194\n\nMitochondrial dysfunction has been observed in ALS mouse models and patients,202 representing a key hallmark of ALS neuropathology. Aberrant interaction between CCS and mutant SOD1 results in increased copper binding to SOD1, reducing copper transfer to mitochondria and causing accumulation of unstable SOD1, limiting ROS clearance, and ultimately leading to motor neuron toxicity.203 Overexpression of CCS accelerates disease progression in SOD1(G93A) transgenic mice (Gly93Ala mutation in the copper/zinc superoxide dismutase 1 gene), significantly reducing COX activity in late-stage disease.204 Furthermore, soluble misfolded mutant SOD1 deposits on mitochondrial intermembrane and outer membranes in ALS animals, which is a key mechanism underlying mitochondrial dysfunction.205", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Multiple sclerosis (MS) is an autoimmune inflammatory disease of the CNS and a leading cause of neurological disability in young adults.206 MS is characterized by neuronal lesions formed in the brain and spinal cord, resulting in a range of clinical symptoms including visual impairment, limb weakness, sensory abnormalities, and ataxia.207 Key factors triggering MS include neuroinflammation, oligodendrocyte death, subsequent OS, and axonal demyelination.208 Cuprizone, a copper chelator, is used to induce an MS mouse model in which microglial activation, demyelination, and mitochondrial damage have been observed.209", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Significantly elevated copper concentrations have been detected in the serum and CSF of MS patients. Upregulation of copper transport proteins in the CNS may contribute to astrocyte-mediated demyelination. Moreover, under neuroinflammatory conditions, copper homeostasis alterations mediated by tropomyosin receptor kinase B (TrkB) and upregulation of copper transporter CTR1 play critical roles in MS. These processes affect astrocyte copper uptake and release, ultimately inducing demyelination.210 Additionally, increased expression of transient receptor potential melastatin 2 (TRPM2) has been observed in the cuprizone model. TRPM2 gene promotes cuprizone-induced demyelination, synaptic loss, microglial NLRP3 inflammasome activation, and production of pro-inflammatory cytokines, indicating a key role for TRPM2 in MS-associated neuroinflammation.211 However, excessively low divalent copper ion concentrations inhibit TRPM2 channel activity irreversibly,212 thereby contributing to MS pathological features.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "SOD1 is an antioxidant and anti-inflammatory enzyme. Studies by Mezzaroba et al, Arakawa et al, and Rasoul et al have shown that OS induced by SOD1 deficiency plays an important role in tissue damage during MS progression.213\u2013215 However, excessive copper accumulation in vivo can cause copper toxicity and generate ROS under Fenton reaction conditions, leading to OS.216 Therefore, copper is currently believed to contribute to MS by inducing oxidative damage, which plays a key role in the pathogenesis of demyelination and neurodegeneration.215\n\nWilson disease (WD) is a non-traditional neurological disorder caused by various mutations in the ATP7B gene. It is characterized by impaired copper excretion into bile, leading to copper accumulation in the liver, while copper buildup in the brain may cause neuropsychiatric symptoms.217 The most common neurological manifestations of WD are movement disorders, including tremor, ataxia, dystonia, parkinsonism, and chorea.218,219 To date, over 600 pathogenic variants in ATP7B have been identified, with the most common being single nucleotide missense and nonsense mutations, followed by insertions/deletions, and more rarely, splice site mutations.220", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "ATP7B plays a dual critical role in the liver: on one hand, it transports copper to the TGN, where copper binds to CP and is excreted via bile; on the other hand, its dysfunction directly causes impaired copper excretion, leading to abnormal copper deposition in target organs.221 When ATP7B is inactivated in WD patients, excess copper cannot enter bile and instead overflows into the bloodstream, depositing in various tissues including the brain. The accumulation of free copper triggers neurological lesions, manifesting as neurological symptoms and psychiatric disorders.222,223 Simultaneously, ATP7B dysfunction causes loosely bound non-CP copper in the blood to bind to albumin and attack erythrocyte membranes, leading to coombs-negative hemolysis and potential hemolytic crises, ultimately resulting in anemia.224", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Copper toxicity is considered the main cause of organ damage in WD patients. When cellular copper levels reach a critical threshold and exceed the binding capacity of MT and GSH, OS is induced, resulting in free radical damage to proteins, lipids, and nucleic acid structures. This ultimately causes cellular damage and membrane rupture.126 Additionally, mitochondrial dysfunction has been observed in the livers of WD patients.225 Protein thiols are key targets of mitochondrial copper toxicity; copper accumulation in the liver attacks protein thiols, leading to mitochondrial dysfunction and subsequent hepatocellular injury in WD patients.226 GSH is a critical intracellular molecule protecting protein thiols from oxidation. Compared with liver mitochondria, brain mitochondria have lower total GSH levels, rendering them more sensitive to copper.226 Moreover, significantly elevated copper levels can be detected in nearly all brain regions of WD patients.227 Therefore, excessive copper accumulation may be a central pathological mechanism underlying the neurological damage and psychiatric abnormalities observed in WD.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Structural brain magnetic resonance imaging (MRI) scans of WD patients reveal marked changes in normal brain structures, including widespread lesions in the midbrain, globus pallidus, putamen, pons, cerebellum, and thalamus, as well as cortical atrophy.228 In the early disease stages, Alzheimer-type I and II astrocytes, along with morphologically abnormal astrocytes, can be observed in the basal ganglia of WD patients, representing typical neuropathological features of WD.229 Studies in WD mouse models have shown activated microglia and astrocytes in the striatum and corpus callosum regions, closely associated with demyelination and neuronal loss.230 Furthermore, astrocytes protect neurons from copper toxicity by sequestering excess copper ions.229 However, as the disease progresses, brain parenchymal copper levels increase significantly, astrocyte damage worsens, weakening their protective function, which further impairs neuronal physiology and ultimately leads to neuronal death.231", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Menkes disease (MD) is an X-linked recessive genetic disorder of copper metabolism caused by mutations in the copper transporter gene ATP7A. It primarily manifests as growth retardation and severe neurological impairments during infancy.232 ATP7A plays a critical role in the TGN, facilitating intracellular copper transport.233 In MD, mutations in ATP7A impair copper export function on the basolateral membrane of intestinal epithelial cells, leading to defective copper transport in the intestine, resulting in abnormally low copper levels in blood, kidney, liver, and brain, while copper accumulates abnormally within intestinal epithelial cells.126 Furthermore, COX deficiency has been observed in the brains of MD patients, which may be related to mutations in the SCO1 and SCO2 genes, essential for copper metallation of the CuA site.234 Therefore, mitochondrial dysfunction caused by reduced COX activity may contribute to neuronal degeneration in MD patients.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "The normal physiological functions of copper-dependent enzymes are impaired in MD patients. Dopamine-\u03b2-hydroxylase (DBH), a key enzyme in noradrenergic neurons responsible for converting dopamine to norepinephrine, shows decreased activity.235 This reduction leads to a characteristic neurochemical pattern in patients\u2019 plasma and CSF, namely elevated dopamine and its metabolites and decreased norepinephrine and its metabolites.235\u2013237 Clinically, this pattern can be used for newborn screening and early diagnosis of MD.235 In infants with MD, significantly increased levels of L-DOPA (L-3,4-dihydroxyphenylalanine, a precursor of catecholamines) have been detected in blood and CSF, indicating reduced DBH activity.238 Since DBH catalyzes the copper-dependent conversion of dopamine to norepinephrine, its dysfunction further reflects the copper deficiency-associated neurochemical disturbances in MD patients.239\n\nLysyl oxidase (LOX) is another copper-dependent enzyme catalyzing the initial oxidative deamination step in the crosslinking of collagen and elastin.240 In MD patients, LOX enzymatic activity is significantly decreased due to defective copper transport, leading to connective tissue abnormalities.241 Related studies show that reduced plasma copper levels and copper transport defects in MD patients not only decrease LOX catalytic activity but also downregulate the mRNA expression of LOX and its substrate pro-elastin.242,243", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Occipital horn syndrome (OHS) is a mild variant of WD caused by ATP7A gene defects, characterized mainly by connective tissue abnormalities, including characteristic bony projections (occipital horns) at the attachment of the occipital muscles.244 This phenotype is closely associated with decreased copper-dependent LOX activity and can result in typical kinky hair, vascular tortuosity, and peripheral arterial aneurysms.245 Due to residual ATP7A function (~30%) that retains partial copper transport, neurological symptoms in OHS patients are relatively mild.246 Their serum copper and CP levels are lower than normal but higher than in classic WD, while elevated catecholamine levels in CSF indicate insufficient DBH activity consistent with the neurochemical abnormalities of classic MD.236", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Stroke is a common neurological disorder caused by acute focal injury to the CNS. It ranks as the second leading cause of death worldwide among adults, second only to ischemic heart disease, and is the third leading cause of disability.247 Stroke is typically classified as ischemic or hemorrhagic, with ischemic stroke being more prevalent.248 Elevated plasma copper levels increase susceptibility to stroke.249 Plasma copper is positively associated with a higher risk of ischemic stroke but inversely associated with hemorrhagic stroke risk.250 Another meta-analysis showed that serum copper levels were significantly higher in ischemic stroke patients compared to controls.251 These studies collectively indicate that copper homeostasis imbalance is a potential risk factor for stroke.\n\nCopper imbalance influences stroke through pathways including vascular regeneration, hyperlipidemia, OS, maturation of neuroprotective peptides, and inflammatory responses.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Endothelial progenitor cells (EPCs) are bone marrow\u2013derived cells circulating in the blood that can differentiate into endothelial cells and participate in angiogenesis and tissue repair. Studies have shown that EPCs play a crucial role in the recovery process after ischemic brain injury by promoting vascular reperfusion, inhibiting apoptosis and inflammation in the ischemic penumbra, thereby improving neurological function.252 However, copper overload significantly impairs EPC migration, adhesion, and tube formation by upregulating thrombospondin-1 (TSP-1) expression, while also weakening their antioxidant defenses. This leads to suppressed cerebral microvascular angiogenesis post-ischemia and exacerbates brain injury after stroke.253,254\n\nAdditionally, a retrospective study based on the 2011\u20132016 NHANES database found that in women, serum copper levels positively correlated with total cholesterol and low-density lipoprotein low\u2011density lipoprotein (LDL). Each 1\u202f\u00b5g/dL increase in copper was associated with approximately 0.11 mg/dL and 0.09 mg/dL increases in total cholesterol and LDL, respectively, suggesting copper may indirectly elevate stroke risk by increasing blood lipid levels. However, as this was a cross-sectional study, prospective research is needed to validate this mechanism.255", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "The Relationship Among Copper Homeostasis, Cuproptosis, and Neurological Diseases", "text": "Copper also plays a critical role as an essential cofactor of SOD1, maintaining its antioxidant function. Dysregulated copper homeostasis may lead to reduced SOD1 activity or abnormal aggregation, enhancing OS and exacerbating neural damage.193,194 Conversely, SOD1 overexpression in ischemia-reperfusion animal models significantly reduces superoxide anion production and overall ROS levels, promoting neural stem cell survival and partially mitigating ischemic stroke damage.256\n\nFurthermore, copper is vital for the maturation of multiple neuropeptides, especially through the peptidylglycine \u03b1-amidating monooxygenase (PAM)-catalyzed C-terminal amidation pathway. Copper deficiency reduces PAM enzyme activity, resulting in neuropeptides such as neuropeptide Y (NPY) and corticotropin-releasing hormone (CRH) failing to form structurally mature C-terminal amidated forms.257,258 These peptides are crucial in neuroprotection and inflammation regulation following stroke. NPY has been shown to exert anti-inflammatory effects, reduce infarct volume, and promote neural regeneration,259,260 whereas CRH mediates stress-related inflammatory responses, modulating neuroinflammation during post-stroke pathology.261", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Alzheimer\u2019s Disease", "text": "Alzheimer\u2019s disease (AD) is one of the most common neurological disorders, clinically characterized by progressive cognitive impairments in visual, language, executive, behavioral, and motor domains.116,117 The formation of extracellular A\u03b2 plaques and neurofibrillary tangles (NFTs) composed of hyperphosphorylated tau protein in the hippocampus and cerebral cortex are central to AD pathogenesis.118 According to the amyloid cascade hypothesis, abnormal processing of amyloid precursor protein (APP) during pathology leads to extracellular accumulation of A\u03b2, which triggers tau hyperphosphorylation and ultimately NFT formation.119,120 Concurrently, A\u03b2 aggregation induces neuroinflammation, oxidative stress (OS), and synaptic loss, culminating in neuronal death and cognitive decline.121", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Alzheimer\u2019s Disease", "text": "In AD patients, free copper levels (particularly in serum) are significantly elevated, resulting in diminished antioxidant defenses and mitochondrial dysfunction. Moreover, increased copper levels correlate with the rate of cognitive decline,122,123 and serum copper levels positively associate with AD risk124 (Figure 4).Figure 4Copper homeostasis and cuproptosis in Alzheimer\u2019s disease (AD). IFN-\u03b3 disrupts copper homeostasis. Excess copper increases free radicals via the Fenton reaction, leading to oxidative stress (OS) and cell death. These radicals interact with A\u03b2 and APP, promoting A\u03b2 aggregation and ROS production. Copper accumulation reduces synaptic proteins (SYP, PSD-95) and neurotransmitters (5-HT, GABA), and inhibits BDNF production through the CREB/BDNF pathway, impairing synaptic plasticity. APP cleavage by \u03b1- and \u03b3-secretase produces non-toxic P3, while BACE1 and \u03b3-secretase generate A\u03b240 and A\u03b242. Copper promotes \u03b1-cleavage, whereas copper deficiency favors \u03b2-cleavage, worsening AD pathology. Microglia can clear A\u03b2 or A\u03b2-antibody complexes, but excess copper impairs this function. In the hippocampus, copper elevates NO and LPO levels, damages antioxidant defenses, and disrupts neuronal function. Copper also induces tau hyperphosphorylation and aggregation.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Alzheimer\u2019s Disease", "text": "Copper homeostasis and cuproptosis in Alzheimer\u2019s disease (AD). IFN-\u03b3 disrupts copper homeostasis. Excess copper increases free radicals via the Fenton reaction, leading to oxidative stress (OS) and cell death. These radicals interact with A\u03b2 and APP, promoting A\u03b2 aggregation and ROS production. Copper accumulation reduces synaptic proteins (SYP, PSD-95) and neurotransmitters (5-HT, GABA), and inhibits BDNF production through the CREB/BDNF pathway, impairing synaptic plasticity. APP cleavage by \u03b1- and \u03b3-secretase produces non-toxic P3, while BACE1 and \u03b3-secretase generate A\u03b240 and A\u03b242. Copper promotes \u03b1-cleavage, whereas copper deficiency favors \u03b2-cleavage, worsening AD pathology. Microglia can clear A\u03b2 or A\u03b2-antibody complexes, but excess copper impairs this function. In the hippocampus, copper elevates NO and LPO levels, damages antioxidant defenses, and disrupts neuronal function. Copper also induces tau hyperphosphorylation and aggregation.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Alzheimer\u2019s Disease", "text": "Copper-induced OS may be one of the primary mechanisms underlying its neurotoxicity. Since copper in the brain can cycle between Cu(II) and Cu(I) oxidation states, it possesses redox activity and can readily catalyze the production of ROS, thereby exacerbating OS in the brain.125 Excess copper generates large amounts of free radicals through the Fenton reaction, inducing oxidative damage and cell death, which leads to neurodegeneration.126 The free radicals produced by the Fenton reaction are not only closely related to OS but can also directly interact with A\u03b2 plaques and APP, promoting A\u03b2 synthesis and aggregation.127 Copper can also bind directly to A\u03b2 plaques, catalyzing ROS generation; these ROS may cause oxidative damage to the A\u03b2 peptides themselves as well as surrounding molecules such as proteins and lipids.128 In the hippocampus, copper treatment significantly elevates levels of nitric oxide (NO) and lipid peroxidation products (LPO), impairs the antioxidant defense system, and consequently damages neural function.129 Copper overload can disrupt critical mitochondrial enzymes such as COX, causing electron transport chain blockage, sharply increasing ROS production, activating cuproptosis, and ultimately inducing neuronal apoptosis or necrosis.130 As the main energy supplier of cells, mitochondrial dysfunction not only increases neurotoxicity but also leads to synaptic loss and memory impairment.131", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Alzheimer\u2019s Disease", "text": "Hippocampal dysfunction typically occurs early in AD, accompanied by inhibition of long\u2011term potentiation (LTP) and synaptic dysfunction in patients\u2019 brains. Excessive copper accumulation can downregulate presynaptic protein synaptophysin (SYP), postsynaptic density protein 95 (PSD-95), and neurotransmitters such as 5-hydroxytryptamine (5-HT) and gamma-aminobutyric acid (GABA) in mice.132 SYP expression correlates closely with the number of presynaptic vesicles, while PSD-95 supports and anchors postsynaptic receptors and regulates receptor-associated protein organization.133 5-HT and GABA are key neurotransmitters in the synaptic cleft involved in learning and memory processes.134,135 These four molecules are all involved in higher brain functions within the synaptic system, indicating that copper can simultaneously affect presynaptic regulation, postsynaptic modulation, and neurotransmitter release, leading to synaptic transmission impairments. Moreover, brain-derived neurotrophic factor (BDNF), an important target of cAMP response element-binding protein (CREB),136 regulates synaptic plasticity and participates in neuroprotection and neural regeneration.137 Copper may inhibit the CREB/BDNF signaling pathway, reducing BDNF expression and thereby suppressing synaptic plasticity, which hinders memory formation and consolidation.132", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Alzheimer\u2019s Disease", "text": "Persistent activation of microglia is closely related to abnormal copper homeostasis and is involved in AD pathology.138 Microglia can phagocytose A\u03b2 or A\u03b2-antibody complexes,139 but excessive copper ions inhibit this phagocytic function. The inflammatory cytokine interferon-\u03b3 (IFN-\u03b3) not only triggers and exacerbates inflammation but also disrupts copper homeostasis by altering the cytoplasmic relocalization of copper transporter ATP7A, increasing copper uptake and upregulating CTR1 expression.140 This process may explain the fluctuating copper homeostasis observed in AD patients.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Alzheimer\u2019s Disease", "text": "Furthermore, APP and tau proteins are critical in AD pathogenesis. APP acts as a receptor in the CNS, involved in synaptic growth, neuronal adhesion, axonal transport, and development.141 Under the action of \u03b1-secretase and \u03b3-secretase, APP produces a non-toxic P3 fragment. However, cleavage by \u03b2-secretase (BACE1) and \u03b3-secretase generates the longer peptides A\u03b240 and A\u03b242, which form amyloid plaques in the AD brain.142,143 Increased copper levels promote the \u03b1-secretase-mediated non-toxic pathway, whereas copper deficiency promotes the BACE1-mediated toxic pathway, accelerating A\u03b2 formation.144 Tau protein, a neuron-specific microtubule-associated protein, normally maintains axonal microtubule stability. In AD, tau becomes hyperphosphorylated and dissociates from microtubules, causing axonal transport deficits and neuronal structural damage, ultimately triggering cell death.145 ROS generation is an important trigger of tau hyperphosphorylation and filament aggregation. Excess copper promotes tau phosphorylation, while treatment with copper chelators can inhibit copper levels and markedly reduce tau phosphorylation.146,147 Studies also suggest that reducing brain copper by limiting dietary copper intake may be a feasible strategy to regulate tau pathology in AD.146", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Parkinson\u2019s Disease", "text": "Parkinson\u2019s disease (PD) is the second most common neurodegenerative disorder after AD, with rapidly increasing global prevalence and mortality.148 Clinically, it is characterized by bradykinesia, resting tremor, muscle rigidity, postural and gait abnormalities, often accompanied by cognitive impairment.149 Pathologically, PD is marked by progressive loss of dopaminergic neurons in the substantia nigra-striatum region and the presence of \u03b1-Syn inclusions in neuronal cytoplasm, forming Lewy bodies and Lewy neurites.150 \u03b1-Syn undergoes oligomerization and converts into \u03b2-sheet-rich protofibrils, forming cytotoxic amyloid aggregates.151 These aggregates coexist with mitochondrial dysfunction and decreased copper and GSH levels in brain regions, implicating OS and metabolic imbalance in PD pathology progression.152", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Parkinson\u2019s Disease", "text": "Studies show that copper levels are decreased in multiple brain regions of PD patients post-mortem, especially in the substantia nigra, as measured by inductively coupled plasma mass spectrometry (ICP-MS).153 Synchrotron X-ray fluorescence microscopy and particle-induced X-ray emission analyses indicate a 55\u201365% reduction in copper in the substantia nigra and locus coeruleus of PD brains154 (Figure 5). CP, the main copper carrier and ferroxidase in plasma, shows dysregulated function in PD patients. Reduced CP-bound copper may cause iron homeostasis disturbance, worsening neuronal injury.Figure 5Copper homeostasis and cuproptosis in Parkinson\u2019s disease (PD). In PD, copper directly interacts with \u03b1-synuclein (\u03b1-Syn), promoting its aggregation into toxic oligomers. These complexes generate ROS and dopamine oxidation products, impair energy metabolism, and exacerbate neuronal damage. Intracellular GSH chelates copper and reduces its toxicity. When GSH is depleted, free copper increases, leading to elevated DLAT levels and loss of Fe\u2013S cluster proteins, ultimately triggering copper-induced cell death. Copper bound to ceruloplasmin (CP) helps maintain iron homeostasis, which may protect neurons. CTR1 promotes \u03b1-Syn phosphorylation, contributing to dopaminergic neuron loss and PD progression. Overexpression of CTR1 may also deplete GSH and induce caspase-3-dependent cell death. Additionally, GSTM1-null and GSTT1-null genotypes may increase PD susceptibility.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Parkinson\u2019s Disease", "text": "Copper homeostasis and cuproptosis in Parkinson\u2019s disease (PD). In PD, copper directly interacts with \u03b1-synuclein (\u03b1-Syn), promoting its aggregation into toxic oligomers. These complexes generate ROS and dopamine oxidation products, impair energy metabolism, and exacerbate neuronal damage. Intracellular GSH chelates copper and reduces its toxicity. When GSH is depleted, free copper increases, leading to elevated DLAT levels and loss of Fe\u2013S cluster proteins, ultimately triggering copper-induced cell death. Copper bound to ceruloplasmin (CP) helps maintain iron homeostasis, which may protect neurons. CTR1 promotes \u03b1-Syn phosphorylation, contributing to dopaminergic neuron loss and PD progression. Overexpression of CTR1 may also deplete GSH and induce caspase-3-dependent cell death. Additionally, GSTM1-null and GSTT1-null genotypes may increase PD susceptibility.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Parkinson\u2019s Disease", "text": "In vitro studies demonstrate that Cu(II) effectively induces spontaneous oligomerization of \u03b1-Syn at micromolar concentrations, promoting formation of oligomers and protofibrils.155,156 Binding of \u03b1-Syn with Cu(II) produces an N-terminal acetylated \u03b1-Syn-Cu(II) complex, exhibiting catechol oxidase and redox activity, catalyzing dopamine oxidation and ROS generation; this activity is further enhanced when membrane-bound.157,158 Additionally, metabolic energy decline is closely linked to disrupted catecholamine metabolism in highly active catecholaminergic neurons (eg, dopaminergic neurons in the substantia nigra pars compacta and noradrenergic neurons in the locus coeruleus), which disrupts redox-active metals such as copper and iron homeostasis.159 These mechanisms suggest \u03b1-Syn-Cu(II) exacerbates dopaminergic/noradrenergic neuronal dysfunction via OS and metabolic imbalance, contributing to PD pathology.\n\nCTR1 is the primary high-affinity copper importer. In PD animal models, CTR1 deficiency significantly reduces \u03b1-Syn phosphorylation at S129, decreases dopaminergic neuron loss, and improves motor dysfunction.160 Conversely, CTR1 overexpression or excess causes intracellular GSH depletion and caspase-3-dependent cell death, indicating that excessive copper can also induce neuronal injury.161 Decoppering treatment partially restores apoptosis levels and expression of copper death-related proteins,132 supporting a role for copper death mechanisms in PD.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Parkinson\u2019s Disease", "text": "Copper homeostasis is closely linked with GSH metabolism. Zhang et al used the high-efficiency GSH fluorescent probe R13 to find reduced GSH levels in PD mouse brains.162 Clinically, GSH levels are significantly decreased in the substantia nigra of PD patients.163 GSH chelates Cu(I), blocking copper toxicity; however, GSH deficiency increases free Cu(I), disrupting Fe-S cluster proteins and inducing OS and cell death.16 From a genetic perspective, deletion polymorphisms in glutathione S-transferase Mu1 (GSTM1) and Theta1 (GSTT1) are associated with increased PD susceptibility,164 emphasizing the importance of GSH metabolism in defending against copper toxicity. Supplementation with cysteine precursors (eg, 6-OHDA/xanthine induction schemes) enhances cellular GSH levels and restores neuronal vitality.165,166 GSH deficiency may underlie copper-induced neurodegeneration, making modulation of GSH metabolism a potential PD therapeutic strategy.\n\nRecent research links amorphous SOD1 aggregates with PD progression; these new SOD1-containing aggregates are amorphous and spherical, largely devoid of \u03b1-Syn but ubiquitin-positive, with copper as a structural and functional cofactor of SOD1, implicating copper deficiency in abnormal SOD1 aggregation.167 Indeed, SOD1 immunoreactivity has been detected in Lewy bodies and Lewy neurites of the substantia nigra and locus coeruleus in PD brains.168", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Parkinson\u2019s Disease", "text": "In summary, copper homeostasis imbalance in PD manifests as localized copper deficiency rather than widespread copper toxicity, involving multiple mechanisms including \u03b1-Syn aggregation, catecholamine oxidation, GSH deficiency-induced metabolic imbalance, Fe-S protein dysfunction, and SOD1 aggregation. A complex network comprising CTR1, CP, GST, SOD1, and Fe\u2013S cluster proteins underlies these processes. These findings provide a theoretical foundation for PD early diagnosis and therapeutic strategies targeting copper metabolism.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Huntington\u2019s Disease", "text": "Huntington\u2019s disease (HD) is a progressive neurodegenerative disorder with a genetic origin. The most common symptoms include loss of energy and initiative, poor perseverance and work quality, impaired judgment, decreased self-care ability, and emotional blunting; about half of studied patients exhibit emotional symptoms such as depression, anxiety, and irritability.169 HD is an autosomal dominant hereditary disease caused by CAG repeat expansion mutations in the gene encoding huntingtin (HTT) protein located on the short arm of chromosome 4.170 The N-terminus of the HTT protein contains a polyglutamine (polyQ) tract, and when the CAG repeats expand, a structurally defective mutated huntingtin (mHTT) protein is produced.171 This mHTT protein is prone to misfolding and forms \u03b2-sheet-rich aggregates and inclusions in the neuronal cytoplasm and nucleus.172 These abnormal aggregates disrupt cellular protein homeostasis, axonal transport, transcription, translation, mitochondrial, and synaptic functions.173,174 Diagnosis is typically confirmed by detecting the number of CAG repeats (\u226536 repeats, with \u226540 repeats being fully pathogenic) in the HTT gene of patients exhibiting clinical features.175", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Huntington\u2019s Disease", "text": "Researchers have found abnormally high concentrations of copper in the striatum of both HD patients and HD mouse models, suggesting that copper may be involved in HD pathogenesis.176 In Drosophila models, copper ions increase the formation of mHTT aggregates and enhance their neurotoxicity,177 while copper chelators inhibit mHTT aggregate formation, further supporting the link between copper and HD pathogenesis.178 Copper ions promote mHTT aggregate formation by binding to mHTT monomers and oligomers, lowering the nucleation energy barrier for aggregation.176 In vitro studies show that copper ions selectively bind to the N-terminal fragment of HTT protein containing 17\u201368 glutamine residues, while other metal ions such as Fe3\u207a and Zn2\u207a show no significant binding affinity.178 Excess copper in the brain inhibits key mitochondrial and energy metabolism dehydrogenases, including lactate dehydrogenase (LDH)179 and succinate dehydrogenase (SDH),92 leading to energy metabolism imbalance. Neurons highly depend on lactate released by astrocytes as their primary energy substrate;180 thus, copper inhibition of LDH directly disrupts lactate metabolism, further weakening neuronal energy supply and exacerbating HD neurodegeneration. Additionally, in Drosophila HD models, copper ions increase mHTT aggregate formation and enhance accumulation of sulfur-positive A\u03b2 structures within mHTT aggregates, representing the toxic aggregation of mHTT and altering autophagy in the brain.177 Studies also show that copper ions directly bind mHTT, increasing its aggregation propensity and \u03b2-sheet structure, thereby enhancing its cytotoxicity in neurons.176", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Huntington\u2019s Disease", "text": "Mutant HTT disrupts nuclear pore proteins in the nuclear pore complex (NPC), impairing nucleocytoplasmic transport and causing cellular toxicity.181 Meanwhile, mHTT interacts with the mitochondrial protein dynamin-related protein 1 (DRP1), promoting mitochondrial fission, resulting in mitochondrial dysfunction and synaptic protein sequestration.182 Moreover, compared to healthy controls, levels of several ubiquitin-proteasome system (UPS) enzymes, including ubiquitin-activating enzyme (E1), ubiquitin-conjugating enzyme (E2), and ubiquitin ligase (E3), are altered in HD patients. Increased expression of these enzymes may affect proteasome-dependent degradation of mHTT or influence its solubility, aggregation, or assembly, thereby directly or indirectly impacting cellular pathways and stress responses involved in HD pathogenesis.183 The UPS plays a crucial role in clearing aggregated mHTT protein.184 Additionally, UPS participates in the translation and regulation of copper transport proteins and copper chaperones; however, since UPS contains copper-dependent enzyme components, it is itself influenced by copper levels.184,185", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Amyotrophic Lateral Sclerosis", "text": "Amyotrophic lateral sclerosis (ALS) is a common neurological disorder characterized by the simultaneous degeneration of upper motor neurons in the motor cortex and lower motor neurons in the spinal cord and brainstem.186 Most ALS cases (~90%) are sporadic without a family history, while a minority (~10%) represent rare autosomal dominant familial amyotrophic lateral sclerosis (FALS).187 The exact etiology of sporadic amyotrophic lateral sclerosis (SALS) remains unclear; however, current consensus suggests it is a complex multistep disease triggered by a combination of genetic susceptibility, environmental exposures, and aging factors.188 The primary cause of FALS is mutations in the SOD1 gene, which encodes a metalloprotein that forms a highly stable homodimer by binding copper and zinc ions. SOD1 mutations cause OS, mitochondrial dysfunction, electron transport disruption, and protein aggregation.189,190", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Amyotrophic Lateral Sclerosis", "text": "It is widely accepted that accumulation of misfolded conformers, oligomers, and aggregates of SOD1 in motor neurons is a core pathological mechanism of ALS. Due to disulfide bond disruption, SOD1 oligomers exhibit enhanced pro-oxidant activity and toxicity.191 Mutant SOD1 reduces protein folding stability by impairing metal binding and disulfide bond formation, leading to misfolding, aggregation, and eventual cytotoxicity.192 Copper is not only central to SOD1 enzymatic activity but is also crucial for its structural integrity and stability; abnormal copper metabolism may contribute to ALS pathogenesis by affecting SOD1 function or aggregation.193,194 One of the earliest pathological features in mutant SOD1 is loss of metal at copper-binding sites.195 Copper deficiency is common in mutant SOD1 cells, and aggregates are generally copper-deficient.196 Moreover, copper levels in cerebrospinal fluid (CSF) from ALS patients are lower than controls, especially in spinal-onset ALS, where copper concentration significantly decreases.197", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Amyotrophic Lateral Sclerosis", "text": "Overexpression of COX in adult mice carrying mutant SOD1 results in impaired copper transport in the late stages of life, suggesting that age-related copper deficiency may be an important factor in the onset of SOD1-induced ALS.198 Deficiency of metals such as copper and zinc promotes the exposure of hydrophobic residues in SOD1, which facilitates the formation and stabilization of toxic SOD1 oligomers. These oligomers exhibit high toxicity and pro-oxidant activity, representing a key pathological process leading to neuronal damage.199 In addition, the disruption of copper binding and disulfide bond formation in SOD1 may contribute to the onset and progression of ALS.200 When SOD1 forms disulfide bonds before binding metal ions, it loses the ability to be recognized and activated by CCS, thereby blocking the copper supply from CCS and hindering the maturation of the disulfide bond in SOD1.201", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Amyotrophic Lateral Sclerosis", "text": "Conversely, abnormally elevated free copper levels may act as immune stimuli. Misallocated copper ions generate large amounts of free radicals via Fenton reactions, directly causing oxidative damage and inducing microglial release of pro-inflammatory cytokines such as tumor necrosis factor-alpha (TNF-\u03b1) and interleukin-1 beta (IL-1\u03b2).194 Excess copper also activates NF-\u03baB in astrocytes, amplifying inflammatory responses and further damaging motor neurons.194 Indeed, in SOD1-ALS models, both copper deficiency (leading to protein dysfunction) and copper overload (triggering toxic inflammation) coexist and synergistically drive disease progression.194\n\nMitochondrial dysfunction has been observed in ALS mouse models and patients,202 representing a key hallmark of ALS neuropathology. Aberrant interaction between CCS and mutant SOD1 results in increased copper binding to SOD1, reducing copper transfer to mitochondria and causing accumulation of unstable SOD1, limiting ROS clearance, and ultimately leading to motor neuron toxicity.203 Overexpression of CCS accelerates disease progression in SOD1(G93A) transgenic mice (Gly93Ala mutation in the copper/zinc superoxide dismutase 1 gene), significantly reducing COX activity in late-stage disease.204 Furthermore, soluble misfolded mutant SOD1 deposits on mitochondrial intermembrane and outer membranes in ALS animals, which is a key mechanism underlying mitochondrial dysfunction.205", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Multiple Sclerosis", "text": "Multiple sclerosis (MS) is an autoimmune inflammatory disease of the CNS and a leading cause of neurological disability in young adults.206 MS is characterized by neuronal lesions formed in the brain and spinal cord, resulting in a range of clinical symptoms including visual impairment, limb weakness, sensory abnormalities, and ataxia.207 Key factors triggering MS include neuroinflammation, oligodendrocyte death, subsequent OS, and axonal demyelination.208 Cuprizone, a copper chelator, is used to induce an MS mouse model in which microglial activation, demyelination, and mitochondrial damage have been observed.209", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Multiple Sclerosis", "text": "Significantly elevated copper concentrations have been detected in the serum and CSF of MS patients. Upregulation of copper transport proteins in the CNS may contribute to astrocyte-mediated demyelination. Moreover, under neuroinflammatory conditions, copper homeostasis alterations mediated by tropomyosin receptor kinase B (TrkB) and upregulation of copper transporter CTR1 play critical roles in MS. These processes affect astrocyte copper uptake and release, ultimately inducing demyelination.210 Additionally, increased expression of transient receptor potential melastatin 2 (TRPM2) has been observed in the cuprizone model. TRPM2 gene promotes cuprizone-induced demyelination, synaptic loss, microglial NLRP3 inflammasome activation, and production of pro-inflammatory cytokines, indicating a key role for TRPM2 in MS-associated neuroinflammation.211 However, excessively low divalent copper ion concentrations inhibit TRPM2 channel activity irreversibly,212 thereby contributing to MS pathological features.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Multiple Sclerosis", "text": "SOD1 is an antioxidant and anti-inflammatory enzyme. Studies by Mezzaroba et al, Arakawa et al, and Rasoul et al have shown that OS induced by SOD1 deficiency plays an important role in tissue damage during MS progression.213\u2013215 However, excessive copper accumulation in vivo can cause copper toxicity and generate ROS under Fenton reaction conditions, leading to OS.216 Therefore, copper is currently believed to contribute to MS by inducing oxidative damage, which plays a key role in the pathogenesis of demyelination and neurodegeneration.215", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Wilson Disease", "text": "Wilson disease (WD) is a non-traditional neurological disorder caused by various mutations in the ATP7B gene. It is characterized by impaired copper excretion into bile, leading to copper accumulation in the liver, while copper buildup in the brain may cause neuropsychiatric symptoms.217 The most common neurological manifestations of WD are movement disorders, including tremor, ataxia, dystonia, parkinsonism, and chorea.218,219 To date, over 600 pathogenic variants in ATP7B have been identified, with the most common being single nucleotide missense and nonsense mutations, followed by insertions/deletions, and more rarely, splice site mutations.220\n\nATP7B plays a dual critical role in the liver: on one hand, it transports copper to the TGN, where copper binds to CP and is excreted via bile; on the other hand, its dysfunction directly causes impaired copper excretion, leading to abnormal copper deposition in target organs.221 When ATP7B is inactivated in WD patients, excess copper cannot enter bile and instead overflows into the bloodstream, depositing in various tissues including the brain. The accumulation of free copper triggers neurological lesions, manifesting as neurological symptoms and psychiatric disorders.222,223 Simultaneously, ATP7B dysfunction causes loosely bound non-CP copper in the blood to bind to albumin and attack erythrocyte membranes, leading to coombs-negative hemolysis and potential hemolytic crises, ultimately resulting in anemia.224", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Wilson Disease", "text": "Copper toxicity is considered the main cause of organ damage in WD patients. When cellular copper levels reach a critical threshold and exceed the binding capacity of MT and GSH, OS is induced, resulting in free radical damage to proteins, lipids, and nucleic acid structures. This ultimately causes cellular damage and membrane rupture.126 Additionally, mitochondrial dysfunction has been observed in the livers of WD patients.225 Protein thiols are key targets of mitochondrial copper toxicity; copper accumulation in the liver attacks protein thiols, leading to mitochondrial dysfunction and subsequent hepatocellular injury in WD patients.226 GSH is a critical intracellular molecule protecting protein thiols from oxidation. Compared with liver mitochondria, brain mitochondria have lower total GSH levels, rendering them more sensitive to copper.226 Moreover, significantly elevated copper levels can be detected in nearly all brain regions of WD patients.227 Therefore, excessive copper accumulation may be a central pathological mechanism underlying the neurological damage and psychiatric abnormalities observed in WD.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Wilson Disease", "text": "Structural brain magnetic resonance imaging (MRI) scans of WD patients reveal marked changes in normal brain structures, including widespread lesions in the midbrain, globus pallidus, putamen, pons, cerebellum, and thalamus, as well as cortical atrophy.228 In the early disease stages, Alzheimer-type I and II astrocytes, along with morphologically abnormal astrocytes, can be observed in the basal ganglia of WD patients, representing typical neuropathological features of WD.229 Studies in WD mouse models have shown activated microglia and astrocytes in the striatum and corpus callosum regions, closely associated with demyelination and neuronal loss.230 Furthermore, astrocytes protect neurons from copper toxicity by sequestering excess copper ions.229 However, as the disease progresses, brain parenchymal copper levels increase significantly, astrocyte damage worsens, weakening their protective function, which further impairs neuronal physiology and ultimately leads to neuronal death.231", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Menkes Disease", "text": "Menkes disease (MD) is an X-linked recessive genetic disorder of copper metabolism caused by mutations in the copper transporter gene ATP7A. It primarily manifests as growth retardation and severe neurological impairments during infancy.232 ATP7A plays a critical role in the TGN, facilitating intracellular copper transport.233 In MD, mutations in ATP7A impair copper export function on the basolateral membrane of intestinal epithelial cells, leading to defective copper transport in the intestine, resulting in abnormally low copper levels in blood, kidney, liver, and brain, while copper accumulates abnormally within intestinal epithelial cells.126 Furthermore, COX deficiency has been observed in the brains of MD patients, which may be related to mutations in the SCO1 and SCO2 genes, essential for copper metallation of the CuA site.234 Therefore, mitochondrial dysfunction caused by reduced COX activity may contribute to neuronal degeneration in MD patients.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Menkes Disease", "text": "The normal physiological functions of copper-dependent enzymes are impaired in MD patients. Dopamine-\u03b2-hydroxylase (DBH), a key enzyme in noradrenergic neurons responsible for converting dopamine to norepinephrine, shows decreased activity.235 This reduction leads to a characteristic neurochemical pattern in patients\u2019 plasma and CSF, namely elevated dopamine and its metabolites and decreased norepinephrine and its metabolites.235\u2013237 Clinically, this pattern can be used for newborn screening and early diagnosis of MD.235 In infants with MD, significantly increased levels of L-DOPA (L-3,4-dihydroxyphenylalanine, a precursor of catecholamines) have been detected in blood and CSF, indicating reduced DBH activity.238 Since DBH catalyzes the copper-dependent conversion of dopamine to norepinephrine, its dysfunction further reflects the copper deficiency-associated neurochemical disturbances in MD patients.239\n\nLysyl oxidase (LOX) is another copper-dependent enzyme catalyzing the initial oxidative deamination step in the crosslinking of collagen and elastin.240 In MD patients, LOX enzymatic activity is significantly decreased due to defective copper transport, leading to connective tissue abnormalities.241 Related studies show that reduced plasma copper levels and copper transport defects in MD patients not only decrease LOX catalytic activity but also downregulate the mRNA expression of LOX and its substrate pro-elastin.242,243", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Menkes Disease", "text": "Occipital horn syndrome (OHS) is a mild variant of WD caused by ATP7A gene defects, characterized mainly by connective tissue abnormalities, including characteristic bony projections (occipital horns) at the attachment of the occipital muscles.244 This phenotype is closely associated with decreased copper-dependent LOX activity and can result in typical kinky hair, vascular tortuosity, and peripheral arterial aneurysms.245 Due to residual ATP7A function (~30%) that retains partial copper transport, neurological symptoms in OHS patients are relatively mild.246 Their serum copper and CP levels are lower than normal but higher than in classic WD, while elevated catecholamine levels in CSF indicate insufficient DBH activity consistent with the neurochemical abnormalities of classic MD.236", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Stroke", "text": "Stroke is a common neurological disorder caused by acute focal injury to the CNS. It ranks as the second leading cause of death worldwide among adults, second only to ischemic heart disease, and is the third leading cause of disability.247 Stroke is typically classified as ischemic or hemorrhagic, with ischemic stroke being more prevalent.248 Elevated plasma copper levels increase susceptibility to stroke.249 Plasma copper is positively associated with a higher risk of ischemic stroke but inversely associated with hemorrhagic stroke risk.250 Another meta-analysis showed that serum copper levels were significantly higher in ischemic stroke patients compared to controls.251 These studies collectively indicate that copper homeostasis imbalance is a potential risk factor for stroke.\n\nCopper imbalance influences stroke through pathways including vascular regeneration, hyperlipidemia, OS, maturation of neuroprotective peptides, and inflammatory responses.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Stroke", "text": "Endothelial progenitor cells (EPCs) are bone marrow\u2013derived cells circulating in the blood that can differentiate into endothelial cells and participate in angiogenesis and tissue repair. Studies have shown that EPCs play a crucial role in the recovery process after ischemic brain injury by promoting vascular reperfusion, inhibiting apoptosis and inflammation in the ischemic penumbra, thereby improving neurological function.252 However, copper overload significantly impairs EPC migration, adhesion, and tube formation by upregulating thrombospondin-1 (TSP-1) expression, while also weakening their antioxidant defenses. This leads to suppressed cerebral microvascular angiogenesis post-ischemia and exacerbates brain injury after stroke.253,254\n\nAdditionally, a retrospective study based on the 2011\u20132016 NHANES database found that in women, serum copper levels positively correlated with total cholesterol and low-density lipoprotein low\u2011density lipoprotein (LDL). Each 1\u202f\u00b5g/dL increase in copper was associated with approximately 0.11 mg/dL and 0.09 mg/dL increases in total cholesterol and LDL, respectively, suggesting copper may indirectly elevate stroke risk by increasing blood lipid levels. However, as this was a cross-sectional study, prospective research is needed to validate this mechanism.255", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Stroke", "text": "Copper also plays a critical role as an essential cofactor of SOD1, maintaining its antioxidant function. Dysregulated copper homeostasis may lead to reduced SOD1 activity or abnormal aggregation, enhancing OS and exacerbating neural damage.193,194 Conversely, SOD1 overexpression in ischemia-reperfusion animal models significantly reduces superoxide anion production and overall ROS levels, promoting neural stem cell survival and partially mitigating ischemic stroke damage.256\n\nFurthermore, copper is vital for the maturation of multiple neuropeptides, especially through the peptidylglycine \u03b1-amidating monooxygenase (PAM)-catalyzed C-terminal amidation pathway. Copper deficiency reduces PAM enzyme activity, resulting in neuropeptides such as neuropeptide Y (NPY) and corticotropin-releasing hormone (CRH) failing to form structurally mature C-terminal amidated forms.257,258 These peptides are crucial in neuroprotection and inflammation regulation following stroke. NPY has been shown to exert anti-inflammatory effects, reduce infarct volume, and promote neural regeneration,259,260 whereas CRH mediates stress-related inflammatory responses, modulating neuroinflammation during post-stroke pathology.261", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Drugs Related to Neurological Diseases", "text": "For human copper homeostasis and cuproptosis, therapeutic drugs for neurological diseases are continuously being developed, and several agents with potential efficacy require further investigation. These drugs mainly include copper chelators, copper ionophores, copper nanoparticles, and natural compounds (Table 1).\nTable 1Drugs Related to Nervous System DiseasesCategoryDrug/CompoundMechanismAssociated DiseaseClinical StatusReferencesCopper chelatorsD-penicillamine (DPA)Promote urinary copper excretion and reduce systemic copper accumulation.WDClinical use[262\u2013264]ADPreclinical[265,266]Triethylenetetramine (trientine/TETA)Promote urinary copper excretion and reduce systemic copper accumulation.WDClinical use[267\u2013270]Zinc saltsInduces metallothionein in intestinal cells, inhibits copper absorption, reduces systemic copper.WDClinical use[267,268]Tetrathiomolybdate (TTM)Forms stable ternary complexes with copper and proteins, reduces serum free copper.WDClinical use[271\u2013273]ALSPreclinical[274,275]Clioquinol (CQ)Chelates copper and zinc, affects pathological metal metabolism in MS.MSPreclinical[276]Chelates copper and zinc, may reduce A\u03b2 plaque formation and tau pathology.AD[277\u2013279]Diacetyl-bis(4-methylthiosemicarbazonato)copper(II)(Cu(II)(atsm))Increases copper delivery to apo-SOD1, promotes conversion to more stable, less toxic holo-SOD1 form.ALSPreclinical[280\u2013282]Reduces oxidative stress, decreases \u03b1-synuclein aggregation.PD[283]Bis-choline tetrathiomolybdate (WTX101)Forms stable ternary complexes with copper and serum proteins, promotes biliary copper excretion.WDClinical trials[284,285]Tetradentate monoquinoline 20 (TDMQ20)Inhibits Cu-A\u03b2 complex-catalyzed oxidative stress.ADPreclinical[286\u2013288]Copper ion carrierElesclomol (ELC)Increases COX levels in the brain.MDPreclinical[289]5,7-Dichloro-2-[(dimethylamino)methyl]quinolin-8-ol (PBT2)Targets metal-induced A\u03b2 aggregation.ADClinical trials completed, not approved[290\u2013293]5,7-dichloro-2-[(ethylamino)methyl]-8-hydroxy-3-methylquinazolin-4(3H)-one (PBT434/ ATH434)Prevents \u03b1-Syn accumulation, protects nigrostriatal dopaminergic circuits and motor function in PD models.PDPreclinical[294]Copper nanoparticlesCopper nanoparticlesIncreases AChE and LRP1 levels, reduces A\u03b2 in brain and intestine, and decreases Tau in rat brain.ADPreclinical[295]Natural compoundsLuteolinDownregulates amyloid beta precursor protein (A\u03b2PP), inhibits A\u03b21-42 secretion, suppresses apoptosis, and regulates redox imbalance.ADPreclinical[296]ResveratrolRegulates plasma copper and zinc levels, affects oxidative stress and antioxidant status in copper-deficient rats; antioxidant, anti-inflammatory, reduces A\u03b2-induced lipid peroxidation, and modulates cellular signaling pathways.ADPreclinical[297,298]ApigeninCounteracts copper-mediated A\u03b2 neurotoxicity, reduces oxidative stress, inhibits ROS-induced signaling, prevents apoptosis, and protects mitochondrial function.ADPreclinical[299]RutinAlleviates oxidative stress (OS) and neuroinflammation to improve copper-induced brain injury.Cerebral injury.Preclinical[300]", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Drugs Related to Neurological Diseases", "text": "Copper chelators can bind excess copper ions, blocking their participation in redox reactions and thereby reducing ROS production and protein aggregation.301 Determining the optimal timing for intervention with copper chelators is a major challenge, as the development and permeability of the blood-brain barrier (BBB) directly determine whether copper ions are sequestered in the bloodstream or delivered into neurons.302 Most classical chelators have limited BBB penetration, preventing them from reaching brain tissue to exert their effects.303 For example, D-penicillamine (DPA) poorly crosses the BBB, cannot directly modulate intracellular copper homeostasis, and its therapeutic efficacy is therefore limited.304 Efforts to enhance BBB penetration by increasing compound hydrophobicity may lead to increased toxicity or altered clearance, creating a trade-off between hydrophilicity and hydrophobicity.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Drugs Related to Neurological Diseases", "text": "In addition, clioquinol (CQ) and its derivative 5,7-Dichloro-2-[(dimethylamino)methyl]quinolin-8-ol (PBT2), although able to penetrate the BBB due to their chemical properties, lack selectivity for copper ions. These chelators cannot distinguish between copper bound to essential metalloproteins and toxic copper associated with A\u03b2, resulting in uneven copper distribution in the nervous system and potentially exacerbating neural damage. Consequently, CQ and PBT2\u00a0have not achieved the expected efficacy in clinical trials.305 Therefore, developing copper chelators with high selectivity and good BBB permeability remains a significant challenge in the treatment of neurological diseases. Future research should focus on optimizing the molecular structures of chelators to enhance copper ion selectivity and BBB penetration, thereby achieving more effective therapeutic outcomes.\n\nD-penicillamine (DPA) is a thiol-containing \u03b1\u2011amino thiol and a degradation product of penicillin with high affinity for copper ions. In Drosophila melanogaster experiments, DPA significantly extended the survival of flies exposed to Cu(II), reversed Cu(II)-induced changes in oxidative stress (OS) markers\u2014such as restoring total thiol (T-SH) and GSH levels\u2014and reduced copper-induced alterations in AChE activity, demonstrating certain neuroprotective effects.306", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Drugs Related to Neurological Diseases", "text": "DPA is an effective clinical treatment for WD. Both DPA and triethylenetetramine (trientine/TETA) mobilize copper in the liver and other tissues and promote its urinary excretion, reversing hepatic, neurological, and psychiatric symptoms in most WD patients.263,267 However, a systematic meta-analysis found that although DPA is effective, it is associated with a relatively higher incidence of adverse reactions compared with zinc therapy. In clinical practice, trientine or zinc salts may be considered as alternative or adjunct therapies.268\n\nDPA may also have potential effects in AD. A small-scale clinical trial showed that DPA could reduce OS markers in the serum of AD patients but did not significantly improve cognitive function or slow clinical progression.265 In animal models, intranasally administered DPA carriers demonstrated possible effects on A\u03b2 deposition and improved cognitive performance, although extensive further validation is still required.266", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Drugs Related to Neurological Diseases", "text": "Triethylenetetramine (trientine/TETA) is a polyamine-structured copper chelator that forms stable complexes with copper ions through its four nitrogen atoms, effectively reducing systemic copper levels. It is an oral drug used for the treatment of WD.269 Trientine acts similarly to DPA but has fewer side effects and a lower risk of neurological deterioration. It promotes urinary copper excretion and inhibits intestinal copper absorption.270,307 Compared with DPA for restoring copper balance, trientine is particularly suitable for patients who are intolerant to penicillamine.308 Additionally, trientine can suppress in vivo inflammatory responses induced by hepatic radiofrequency ablation (RFA).309\n\nTetrathiomolybdate (TTM) has potent anti-copper properties. TTM forms stable ternary complexes with copper and proteins, reducing serum free copper levels.271 By rapidly lowering serum copper, TTM can restore normal copper homeostasis within weeks without causing abnormal elevations in serum copper levels.271 TTM is an effective clinical treatment for WD, significantly improving neurological symptoms, and the risk of neurological deterioration in WD patients treated with TTM is markedly lower than with TETA.272,273 TTM may cause adverse effects such as anemia, leukopenia, and elevated transaminase levels, but these are generally reversible with dose adjustment or temporary discontinuation.10", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Drugs Related to Neurological Diseases", "text": "There is experimental evidence that TTM has therapeutic effects in ALS animal models. TTM can remove copper ions from copper-thiolate clusters such as SOD1, reducing the amount of SOD1 aggregates. It prolongs survival in familial ALS mouse models (SOD1(G93A)), mitigates loss of motor neurons and axons, and prevents skeletal muscle atrophy.274,275 In addition, TTM effectively lowers spinal cord copper levels and inhibits lipid peroxidation (LPO), thereby suppressing SOD1 enzymatic activity in SOD1(G93A) mice.275 However, there is currently insufficient clinical data to demonstrate clear efficacy in ALS patients.\n\nClioquinol (CQ) is a hydrophobic small molecule capable of crossing the BBB and exhibiting moderate affinity for copper and zinc ions, showing therapeutic potential in several neurological disease models.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Drugs Related to Neurological Diseases", "text": "CQ\u2019s chelation of copper and zinc effectively reduces spinal white matter damage and behavioral deficits in MS mouse models, markedly decreases microglial activation in experimental autoimmune encephalomyelitis mice, improves clinical symptoms, and lowers the incidence of spinal demyelination.276 In AD studies, CQ reduces ROS generation and LPO in yeast cells, decreases the toxicity of A\u03b242 (an isoform of A\u03b2), and restores GSH homeostasis disrupted by A\u03b242 through regulation of Yes-associated protein 1 (YAP1), thereby protecting cells from OS.277 In AD mouse models, CQ treatment increases soluble A\u03b2 levels and reduces A\u03b2 deposition in the brain.278 Intracellular Cu(I) accumulation is significantly reduced, and CQ also lowers endogenous phosphorylated tau in primary cortical neurons of mice.279\n\nDiacetyl-bis(4-methylthiosemicarbazonato)copper(II)(Cu(II)(atsm)) is an inflammation-modulating compound with potential therapeutic effects for ALS and PD.282,310", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Drugs Related to Neurological Diseases", "text": "In ALS, treatment with Cu(II)(atsm) after symptom onset is less effective than pre-symptomatic administration, but it can still improve motor function and survival.281 In SOD1(G37R) mice (where glycine at residue 37 in SOD1 is replaced with arginine), Cu(II)(atsm) enhances copper delivery to metal-free (apo) SOD1, promoting its conversion to a more stable and less toxic holo-SOD1 form,280 inhibiting SOD1 activity and reducing aggregation of mutant SOD1 protein. Improvements in motor function and survival in SOD1(G37R) mice support Cu(II)(atsm) as a potential ALS therapeutic.281\n\nIn PD models, Cu(II)(atsm) exhibits neuroprotective effects associated with reduced OS and decreased \u03b1-Syn aggregation.283 Cu(II)(atsm) also reduces A\u03b2 levels in APP-CHO cells (Chinese hamster ovary cells overexpressing APP).311 Additionally, Cu(II)(atsm) serves as a tracer for positron emission tomography (PET) imaging. In animal models, Cu(II)(atsm) demonstrates neuroprotective potential, possibly by improving mitochondrial function and reducing OS to protect dopaminergic neurons.312 However, its clinical efficacy requires further validation.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Drugs Related to Neurological Diseases", "text": "Bis-choline tetrathiomolybdate (WTX101) is a first-in-class oral copper-protein chelator for the treatment of WD that is currently undergoing clinical trials. WTX101 can directly remove excess copper from hepatic copper stores and form stable ternary complexes with copper and proteins such as serum albumin in the circulation, promoting biliary copper excretion.284\n\nIn a Phase II open-label study, WTX101 demonstrated the potential to rapidly reduce serum non\u2011ceruloplasmin\u2011bound copper (NCC) and improve certain pathological markers.285\n\nTetradentate monoquinoline 20 (TDMQ20) acts on the cholinergic system to regulate copper homeostasis in the brains of AD patients. TDMQ20 improves memory and cognitive function in AD rat models by inhibiting OS catalyzed by Cu\u2011A\u03b2 complexes.286,287 TDMQ20 efficiently crosses the BBB and is a potential therapeutic candidate for AD.288\n\nElesclomol (ELC) is a mitochondria-targeting copper ionophore and an inducer of OS. ELC targets mitochondria by forming a complex with copper and transporting it into the organelle, where ELC-Cu(II) is reduced to Cu(I), triggering ROS production.313 Studies have shown that ELC has potential to improve MD, as it delivers copper to mitochondria and increases COX levels in the brain, mitigating detrimental neurological lesions and improving survival in mottled brindled mice, a severe MD mouse model.289", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Drugs Related to Neurological Diseases", "text": "The CQ derivative 5,7-Dichloro-2-[(dimethylamino)methyl]quinolin-8-ol (PBT2) is similar to CQ, targeting metal-induced A\u03b2 aggregation, but is more effective as a zinc/copper ionophore and has improved BBB permeability and solubility,290 showing potential benefits in AD treatment. In AD mouse models, PBT2 inhibits copper-induced A\u03b2 accumulation, reduces interstitial A\u03b2 levels and tau phosphorylation, and restores cognitive function.291,292 PBT2\u00a0has undergone Phase II clinical trials in both AD and HD, demonstrating good safety but without clear clinical efficacy, indicating the need for further investigation.293,305\n\nAdditionally, in PD animal models, 5,7-dichloro-2-[(ethylamino)methyl]-8-hydroxy-3-methylquinazolin-4(3H)-one (PBT434/ATH434) prevents \u03b1-Syn accumulation and protects the nigrostriatal dopaminergic circuitry and motor function.294,314\n\nNanoparticle delivery systems can cross the BBB and have the potential to transport DPA into the brain, preventing A\u03b21-42 accumulation and reducing metal ion buildup in CNS diseases.315 Copper nanoparticles increase AChE and low-density lipoprotein (LDL) receptor-related protein 1 (LRP1) levels, reduce A\u03b2 in the brain and gut, and decrease tau protein in the rat brain.295", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Drugs Related to Neurological Diseases", "text": "However, in animal studies, high intake of copper nanoparticles severely affects hepatic drug metabolism in rats by inhibiting the expression of multiple cytochrome P450 (CYP450) enzymes, increasing the risk of drug-drug interactions.316 The combination of copper oxide nanoparticles (CuONPs) and furan synergistically enhances cardiovascular toxicity in zebrafish embryos.317\n\nFor neurological diseases, the optimal type of copper nanoparticles and their mechanisms of action remain unclear. Copper nanoparticle-based drugs for clinical use in neurological disorders are yet to be developed.\n\nNatural compounds are a treasure trove for humans, with diverse and often mysterious biological effects. Many medicinal plants and their natural constituents possess antioxidant, free radical\u2013scavenging, and neuroprotective properties, demonstrating significant efficacy in preventing copper-induced neurotoxicity.\n\nLuteolin, a plant flavonoid extracted from Artemisia species, exhibits neuroprotective effects in neurological diseases and traumatic brain injury (TBI).296 Luteolin exerts its effects by downregulating amyloid beta precursor protein (A\u03b2PP) expression, inhibiting A\u03b21-42 secretion, suppressing apoptosis, and modulating redox imbalance.318 Another study showed that coordination and transfer of divalent copper ions significantly enhance luteolin\u2019s free radical scavenging efficiency and antioxidant activity.319", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Drugs Related to Neurological Diseases", "text": "Apigenin, a low-toxicity flavonoid, antagonizes copper-mediated A\u03b2 neurotoxicity and provides neuroprotection by alleviating OS, inhibiting ROS-induced signaling pathways, preventing apoptosis, and preserving mitochondrial function.299\n\nRutin, a glycoside and polyphenol, can cross the BBB and exhibits antioxidant properties. Rutin mitigates copper-induced brain injury\u2014including cortical perforation layers and neuronal degeneration\u2014by reducing OS and neuroinflammation.300\n\nResveratrol, a natural polyphenol found in various plants, modulates plasma copper and zinc levels and influences OS and antioxidant status in copper-deficient rats.320 Resveratrol improves memory and neuroinflammation through antioxidant, anti-inflammatory effects, reduction of A\u03b2-induced LPO, and regulation of cellular signaling pathways.297 Supplementation with resveratrol has shown some effects on cognitive and functional decline in AD patients.298 Additionally, resveratrol regulates cellular proteostasis by upregulating autophagy, thereby attenuating CuSO4-induced senescence.12 Another study reported that resveratrol effectively alleviates OS and hepatic-renal injury induced by CuONPs.321", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Copper Chelators", "text": "Copper chelators can bind excess copper ions, blocking their participation in redox reactions and thereby reducing ROS production and protein aggregation.301 Determining the optimal timing for intervention with copper chelators is a major challenge, as the development and permeability of the blood-brain barrier (BBB) directly determine whether copper ions are sequestered in the bloodstream or delivered into neurons.302 Most classical chelators have limited BBB penetration, preventing them from reaching brain tissue to exert their effects.303 For example, D-penicillamine (DPA) poorly crosses the BBB, cannot directly modulate intracellular copper homeostasis, and its therapeutic efficacy is therefore limited.304 Efforts to enhance BBB penetration by increasing compound hydrophobicity may lead to increased toxicity or altered clearance, creating a trade-off between hydrophilicity and hydrophobicity.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Copper Chelators", "text": "In addition, clioquinol (CQ) and its derivative 5,7-Dichloro-2-[(dimethylamino)methyl]quinolin-8-ol (PBT2), although able to penetrate the BBB due to their chemical properties, lack selectivity for copper ions. These chelators cannot distinguish between copper bound to essential metalloproteins and toxic copper associated with A\u03b2, resulting in uneven copper distribution in the nervous system and potentially exacerbating neural damage. Consequently, CQ and PBT2\u00a0have not achieved the expected efficacy in clinical trials.305 Therefore, developing copper chelators with high selectivity and good BBB permeability remains a significant challenge in the treatment of neurological diseases. Future research should focus on optimizing the molecular structures of chelators to enhance copper ion selectivity and BBB penetration, thereby achieving more effective therapeutic outcomes.\n\nD-penicillamine (DPA) is a thiol-containing \u03b1\u2011amino thiol and a degradation product of penicillin with high affinity for copper ions. In Drosophila melanogaster experiments, DPA significantly extended the survival of flies exposed to Cu(II), reversed Cu(II)-induced changes in oxidative stress (OS) markers\u2014such as restoring total thiol (T-SH) and GSH levels\u2014and reduced copper-induced alterations in AChE activity, demonstrating certain neuroprotective effects.306", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Copper Chelators", "text": "DPA is an effective clinical treatment for WD. Both DPA and triethylenetetramine (trientine/TETA) mobilize copper in the liver and other tissues and promote its urinary excretion, reversing hepatic, neurological, and psychiatric symptoms in most WD patients.263,267 However, a systematic meta-analysis found that although DPA is effective, it is associated with a relatively higher incidence of adverse reactions compared with zinc therapy. In clinical practice, trientine or zinc salts may be considered as alternative or adjunct therapies.268\n\nDPA may also have potential effects in AD. A small-scale clinical trial showed that DPA could reduce OS markers in the serum of AD patients but did not significantly improve cognitive function or slow clinical progression.265 In animal models, intranasally administered DPA carriers demonstrated possible effects on A\u03b2 deposition and improved cognitive performance, although extensive further validation is still required.266", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Copper Chelators", "text": "Triethylenetetramine (trientine/TETA) is a polyamine-structured copper chelator that forms stable complexes with copper ions through its four nitrogen atoms, effectively reducing systemic copper levels. It is an oral drug used for the treatment of WD.269 Trientine acts similarly to DPA but has fewer side effects and a lower risk of neurological deterioration. It promotes urinary copper excretion and inhibits intestinal copper absorption.270,307 Compared with DPA for restoring copper balance, trientine is particularly suitable for patients who are intolerant to penicillamine.308 Additionally, trientine can suppress in vivo inflammatory responses induced by hepatic radiofrequency ablation (RFA).309\n\nTetrathiomolybdate (TTM) has potent anti-copper properties. TTM forms stable ternary complexes with copper and proteins, reducing serum free copper levels.271 By rapidly lowering serum copper, TTM can restore normal copper homeostasis within weeks without causing abnormal elevations in serum copper levels.271 TTM is an effective clinical treatment for WD, significantly improving neurological symptoms, and the risk of neurological deterioration in WD patients treated with TTM is markedly lower than with TETA.272,273 TTM may cause adverse effects such as anemia, leukopenia, and elevated transaminase levels, but these are generally reversible with dose adjustment or temporary discontinuation.10", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Copper Chelators", "text": "There is experimental evidence that TTM has therapeutic effects in ALS animal models. TTM can remove copper ions from copper-thiolate clusters such as SOD1, reducing the amount of SOD1 aggregates. It prolongs survival in familial ALS mouse models (SOD1(G93A)), mitigates loss of motor neurons and axons, and prevents skeletal muscle atrophy.274,275 In addition, TTM effectively lowers spinal cord copper levels and inhibits lipid peroxidation (LPO), thereby suppressing SOD1 enzymatic activity in SOD1(G93A) mice.275 However, there is currently insufficient clinical data to demonstrate clear efficacy in ALS patients.\n\nClioquinol (CQ) is a hydrophobic small molecule capable of crossing the BBB and exhibiting moderate affinity for copper and zinc ions, showing therapeutic potential in several neurological disease models.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Copper Chelators", "text": "CQ\u2019s chelation of copper and zinc effectively reduces spinal white matter damage and behavioral deficits in MS mouse models, markedly decreases microglial activation in experimental autoimmune encephalomyelitis mice, improves clinical symptoms, and lowers the incidence of spinal demyelination.276 In AD studies, CQ reduces ROS generation and LPO in yeast cells, decreases the toxicity of A\u03b242 (an isoform of A\u03b2), and restores GSH homeostasis disrupted by A\u03b242 through regulation of Yes-associated protein 1 (YAP1), thereby protecting cells from OS.277 In AD mouse models, CQ treatment increases soluble A\u03b2 levels and reduces A\u03b2 deposition in the brain.278 Intracellular Cu(I) accumulation is significantly reduced, and CQ also lowers endogenous phosphorylated tau in primary cortical neurons of mice.279\n\nDiacetyl-bis(4-methylthiosemicarbazonato)copper(II)(Cu(II)(atsm)) is an inflammation-modulating compound with potential therapeutic effects for ALS and PD.282,310", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Copper Chelators", "text": "In ALS, treatment with Cu(II)(atsm) after symptom onset is less effective than pre-symptomatic administration, but it can still improve motor function and survival.281 In SOD1(G37R) mice (where glycine at residue 37 in SOD1 is replaced with arginine), Cu(II)(atsm) enhances copper delivery to metal-free (apo) SOD1, promoting its conversion to a more stable and less toxic holo-SOD1 form,280 inhibiting SOD1 activity and reducing aggregation of mutant SOD1 protein. Improvements in motor function and survival in SOD1(G37R) mice support Cu(II)(atsm) as a potential ALS therapeutic.281\n\nIn PD models, Cu(II)(atsm) exhibits neuroprotective effects associated with reduced OS and decreased \u03b1-Syn aggregation.283 Cu(II)(atsm) also reduces A\u03b2 levels in APP-CHO cells (Chinese hamster ovary cells overexpressing APP).311 Additionally, Cu(II)(atsm) serves as a tracer for positron emission tomography (PET) imaging. In animal models, Cu(II)(atsm) demonstrates neuroprotective potential, possibly by improving mitochondrial function and reducing OS to protect dopaminergic neurons.312 However, its clinical efficacy requires further validation.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Copper Chelators", "text": "Bis-choline tetrathiomolybdate (WTX101) is a first-in-class oral copper-protein chelator for the treatment of WD that is currently undergoing clinical trials. WTX101 can directly remove excess copper from hepatic copper stores and form stable ternary complexes with copper and proteins such as serum albumin in the circulation, promoting biliary copper excretion.284\n\nIn a Phase II open-label study, WTX101 demonstrated the potential to rapidly reduce serum non\u2011ceruloplasmin\u2011bound copper (NCC) and improve certain pathological markers.285\n\nTetradentate monoquinoline 20 (TDMQ20) acts on the cholinergic system to regulate copper homeostasis in the brains of AD patients. TDMQ20 improves memory and cognitive function in AD rat models by inhibiting OS catalyzed by Cu\u2011A\u03b2 complexes.286,287 TDMQ20 efficiently crosses the BBB and is a potential therapeutic candidate for AD.288", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "D-Penicillamine", "text": "D-penicillamine (DPA) is a thiol-containing \u03b1\u2011amino thiol and a degradation product of penicillin with high affinity for copper ions. In Drosophila melanogaster experiments, DPA significantly extended the survival of flies exposed to Cu(II), reversed Cu(II)-induced changes in oxidative stress (OS) markers\u2014such as restoring total thiol (T-SH) and GSH levels\u2014and reduced copper-induced alterations in AChE activity, demonstrating certain neuroprotective effects.306\n\nDPA is an effective clinical treatment for WD. Both DPA and triethylenetetramine (trientine/TETA) mobilize copper in the liver and other tissues and promote its urinary excretion, reversing hepatic, neurological, and psychiatric symptoms in most WD patients.263,267 However, a systematic meta-analysis found that although DPA is effective, it is associated with a relatively higher incidence of adverse reactions compared with zinc therapy. In clinical practice, trientine or zinc salts may be considered as alternative or adjunct therapies.268\n\nDPA may also have potential effects in AD. A small-scale clinical trial showed that DPA could reduce OS markers in the serum of AD patients but did not significantly improve cognitive function or slow clinical progression.265 In animal models, intranasally administered DPA carriers demonstrated possible effects on A\u03b2 deposition and improved cognitive performance, although extensive further validation is still required.266", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Triethylenetetramine", "text": "Triethylenetetramine (trientine/TETA) is a polyamine-structured copper chelator that forms stable complexes with copper ions through its four nitrogen atoms, effectively reducing systemic copper levels. It is an oral drug used for the treatment of WD.269 Trientine acts similarly to DPA but has fewer side effects and a lower risk of neurological deterioration. It promotes urinary copper excretion and inhibits intestinal copper absorption.270,307 Compared with DPA for restoring copper balance, trientine is particularly suitable for patients who are intolerant to penicillamine.308 Additionally, trientine can suppress in vivo inflammatory responses induced by hepatic radiofrequency ablation (RFA).309", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Tetrathiomolybdate", "text": "Tetrathiomolybdate (TTM) has potent anti-copper properties. TTM forms stable ternary complexes with copper and proteins, reducing serum free copper levels.271 By rapidly lowering serum copper, TTM can restore normal copper homeostasis within weeks without causing abnormal elevations in serum copper levels.271 TTM is an effective clinical treatment for WD, significantly improving neurological symptoms, and the risk of neurological deterioration in WD patients treated with TTM is markedly lower than with TETA.272,273 TTM may cause adverse effects such as anemia, leukopenia, and elevated transaminase levels, but these are generally reversible with dose adjustment or temporary discontinuation.10\n\nThere is experimental evidence that TTM has therapeutic effects in ALS animal models. TTM can remove copper ions from copper-thiolate clusters such as SOD1, reducing the amount of SOD1 aggregates. It prolongs survival in familial ALS mouse models (SOD1(G93A)), mitigates loss of motor neurons and axons, and prevents skeletal muscle atrophy.274,275 In addition, TTM effectively lowers spinal cord copper levels and inhibits lipid peroxidation (LPO), thereby suppressing SOD1 enzymatic activity in SOD1(G93A) mice.275 However, there is currently insufficient clinical data to demonstrate clear efficacy in ALS patients.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Clioquinol", "text": "Clioquinol (CQ) is a hydrophobic small molecule capable of crossing the BBB and exhibiting moderate affinity for copper and zinc ions, showing therapeutic potential in several neurological disease models.\n\nCQ\u2019s chelation of copper and zinc effectively reduces spinal white matter damage and behavioral deficits in MS mouse models, markedly decreases microglial activation in experimental autoimmune encephalomyelitis mice, improves clinical symptoms, and lowers the incidence of spinal demyelination.276 In AD studies, CQ reduces ROS generation and LPO in yeast cells, decreases the toxicity of A\u03b242 (an isoform of A\u03b2), and restores GSH homeostasis disrupted by A\u03b242 through regulation of Yes-associated protein 1 (YAP1), thereby protecting cells from OS.277 In AD mouse models, CQ treatment increases soluble A\u03b2 levels and reduces A\u03b2 deposition in the brain.278 Intracellular Cu(I) accumulation is significantly reduced, and CQ also lowers endogenous phosphorylated tau in primary cortical neurons of mice.279", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Diacetyl-Bis(4-Methylthiosemicarbazonato)copper(II)", "text": "Diacetyl-bis(4-methylthiosemicarbazonato)copper(II)(Cu(II)(atsm)) is an inflammation-modulating compound with potential therapeutic effects for ALS and PD.282,310\n\nIn ALS, treatment with Cu(II)(atsm) after symptom onset is less effective than pre-symptomatic administration, but it can still improve motor function and survival.281 In SOD1(G37R) mice (where glycine at residue 37 in SOD1 is replaced with arginine), Cu(II)(atsm) enhances copper delivery to metal-free (apo) SOD1, promoting its conversion to a more stable and less toxic holo-SOD1 form,280 inhibiting SOD1 activity and reducing aggregation of mutant SOD1 protein. Improvements in motor function and survival in SOD1(G37R) mice support Cu(II)(atsm) as a potential ALS therapeutic.281\n\nIn PD models, Cu(II)(atsm) exhibits neuroprotective effects associated with reduced OS and decreased \u03b1-Syn aggregation.283 Cu(II)(atsm) also reduces A\u03b2 levels in APP-CHO cells (Chinese hamster ovary cells overexpressing APP).311 Additionally, Cu(II)(atsm) serves as a tracer for positron emission tomography (PET) imaging. In animal models, Cu(II)(atsm) demonstrates neuroprotective potential, possibly by improving mitochondrial function and reducing OS to protect dopaminergic neurons.312 However, its clinical efficacy requires further validation.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Bis-Choline Tetrathiomolybdate", "text": "Bis-choline tetrathiomolybdate (WTX101) is a first-in-class oral copper-protein chelator for the treatment of WD that is currently undergoing clinical trials. WTX101 can directly remove excess copper from hepatic copper stores and form stable ternary complexes with copper and proteins such as serum albumin in the circulation, promoting biliary copper excretion.284\n\nIn a Phase II open-label study, WTX101 demonstrated the potential to rapidly reduce serum non\u2011ceruloplasmin\u2011bound copper (NCC) and improve certain pathological markers.285", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Tetradentate Monoquinolines", "text": "Tetradentate monoquinoline 20 (TDMQ20) acts on the cholinergic system to regulate copper homeostasis in the brains of AD patients. TDMQ20 improves memory and cognitive function in AD rat models by inhibiting OS catalyzed by Cu\u2011A\u03b2 complexes.286,287 TDMQ20 efficiently crosses the BBB and is a potential therapeutic candidate for AD.288", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Copper Ion Carriers", "text": "Elesclomol (ELC) is a mitochondria-targeting copper ionophore and an inducer of OS. ELC targets mitochondria by forming a complex with copper and transporting it into the organelle, where ELC-Cu(II) is reduced to Cu(I), triggering ROS production.313 Studies have shown that ELC has potential to improve MD, as it delivers copper to mitochondria and increases COX levels in the brain, mitigating detrimental neurological lesions and improving survival in mottled brindled mice, a severe MD mouse model.289\n\nThe CQ derivative 5,7-Dichloro-2-[(dimethylamino)methyl]quinolin-8-ol (PBT2) is similar to CQ, targeting metal-induced A\u03b2 aggregation, but is more effective as a zinc/copper ionophore and has improved BBB permeability and solubility,290 showing potential benefits in AD treatment. In AD mouse models, PBT2 inhibits copper-induced A\u03b2 accumulation, reduces interstitial A\u03b2 levels and tau phosphorylation, and restores cognitive function.291,292 PBT2\u00a0has undergone Phase II clinical trials in both AD and HD, demonstrating good safety but without clear clinical efficacy, indicating the need for further investigation.293,305\n\nAdditionally, in PD animal models, 5,7-dichloro-2-[(ethylamino)methyl]-8-hydroxy-3-methylquinazolin-4(3H)-one (PBT434/ATH434) prevents \u03b1-Syn accumulation and protects the nigrostriatal dopaminergic circuitry and motor function.294,314", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Elesclomol", "text": "Elesclomol (ELC) is a mitochondria-targeting copper ionophore and an inducer of OS. ELC targets mitochondria by forming a complex with copper and transporting it into the organelle, where ELC-Cu(II) is reduced to Cu(I), triggering ROS production.313 Studies have shown that ELC has potential to improve MD, as it delivers copper to mitochondria and increases COX levels in the brain, mitigating detrimental neurological lesions and improving survival in mottled brindled mice, a severe MD mouse model.289", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "5,7-Dichloro-2-[(Dimethylamino)methyl]-8-Hydroxyquinoline", "text": "The CQ derivative 5,7-Dichloro-2-[(dimethylamino)methyl]quinolin-8-ol (PBT2) is similar to CQ, targeting metal-induced A\u03b2 aggregation, but is more effective as a zinc/copper ionophore and has improved BBB permeability and solubility,290 showing potential benefits in AD treatment. In AD mouse models, PBT2 inhibits copper-induced A\u03b2 accumulation, reduces interstitial A\u03b2 levels and tau phosphorylation, and restores cognitive function.291,292 PBT2\u00a0has undergone Phase II clinical trials in both AD and HD, demonstrating good safety but without clear clinical efficacy, indicating the need for further investigation.293,305\n\nAdditionally, in PD animal models, 5,7-dichloro-2-[(ethylamino)methyl]-8-hydroxy-3-methylquinazolin-4(3H)-one (PBT434/ATH434) prevents \u03b1-Syn accumulation and protects the nigrostriatal dopaminergic circuitry and motor function.294,314", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Copper Nanoparticles", "text": "Nanoparticle delivery systems can cross the BBB and have the potential to transport DPA into the brain, preventing A\u03b21-42 accumulation and reducing metal ion buildup in CNS diseases.315 Copper nanoparticles increase AChE and low-density lipoprotein (LDL) receptor-related protein 1 (LRP1) levels, reduce A\u03b2 in the brain and gut, and decrease tau protein in the rat brain.295\n\nHowever, in animal studies, high intake of copper nanoparticles severely affects hepatic drug metabolism in rats by inhibiting the expression of multiple cytochrome P450 (CYP450) enzymes, increasing the risk of drug-drug interactions.316 The combination of copper oxide nanoparticles (CuONPs) and furan synergistically enhances cardiovascular toxicity in zebrafish embryos.317\n\nFor neurological diseases, the optimal type of copper nanoparticles and their mechanisms of action remain unclear. Copper nanoparticle-based drugs for clinical use in neurological disorders are yet to be developed.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Natural Compounds", "text": "Natural compounds are a treasure trove for humans, with diverse and often mysterious biological effects. Many medicinal plants and their natural constituents possess antioxidant, free radical\u2013scavenging, and neuroprotective properties, demonstrating significant efficacy in preventing copper-induced neurotoxicity.\n\nLuteolin, a plant flavonoid extracted from Artemisia species, exhibits neuroprotective effects in neurological diseases and traumatic brain injury (TBI).296 Luteolin exerts its effects by downregulating amyloid beta precursor protein (A\u03b2PP) expression, inhibiting A\u03b21-42 secretion, suppressing apoptosis, and modulating redox imbalance.318 Another study showed that coordination and transfer of divalent copper ions significantly enhance luteolin\u2019s free radical scavenging efficiency and antioxidant activity.319\n\nApigenin, a low-toxicity flavonoid, antagonizes copper-mediated A\u03b2 neurotoxicity and provides neuroprotection by alleviating OS, inhibiting ROS-induced signaling pathways, preventing apoptosis, and preserving mitochondrial function.299\n\nRutin, a glycoside and polyphenol, can cross the BBB and exhibits antioxidant properties. Rutin mitigates copper-induced brain injury\u2014including cortical perforation layers and neuronal degeneration\u2014by reducing OS and neuroinflammation.300", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Natural Compounds", "text": "Resveratrol, a natural polyphenol found in various plants, modulates plasma copper and zinc levels and influences OS and antioxidant status in copper-deficient rats.320 Resveratrol improves memory and neuroinflammation through antioxidant, anti-inflammatory effects, reduction of A\u03b2-induced LPO, and regulation of cellular signaling pathways.297 Supplementation with resveratrol has shown some effects on cognitive and functional decline in AD patients.298 Additionally, resveratrol regulates cellular proteostasis by upregulating autophagy, thereby attenuating CuSO4-induced senescence.12 Another study reported that resveratrol effectively alleviates OS and hepatic-renal injury induced by CuONPs.321", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Conclusion and Future Perspectives", "text": "Copper is an essential element involved in oxidative metabolism and is indispensable for brain cell function. The storage and utilization of copper affect tissues and organs throughout the body. Copper-dependent protein networks\u2014including copper reductases, copper chaperones, and membrane transporters\u2014play critical regulatory roles in maintaining intracellular copper homeostasis. Disruption of copper homeostasis contributes significantly to the pathogenesis of various neurological disorders. Both copper overload and copper deficiency can affect the nervous system through multiple mechanisms, including the induction of oxidative stress (OS), mitochondrial dysfunction, interference with protein folding, and disturbance of metabolic balance, ultimately leading to neuronal dysfunction and death.\n\nIn recent years, the discovery of cuproptosis has expanded our understanding of copper-related pathological mechanisms. Unlike classical forms of cell death such as apoptosis, necrosis, or ferroptosis, cuproptosis is characterized by the aberrant binding of copper ions to lipoylated proteins, resulting in protein aggregation and loss of iron\u2013sulfur cluster proteins, which trigger mitochondrial destabilization and metabolic collapse.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Conclusion and Future Perspectives", "text": "Copper dyshomeostasis is closely associated with numerous neurological diseases, including Alzheimer\u2019s disease (AD), Parkinson\u2019s disease (PD), Huntington\u2019s disease (HD), amyotrophic lateral sclerosis (ALS), multiple sclerosis (MS), Wilson disease (WD), Menkes disease (MD), and stroke. The pathogenic mechanisms of copper metabolism abnormalities vary considerably among different diseases, encompassing both the toxic effects of copper overload and enzyme dysfunction or metabolic imbalance caused by copper deficiency. These two opposing states often coexist in different stages or regions of the same disease.", "title": "" }, { "pmid": 41783572, "pmcid": "PMC12955603", "section": "Conclusion and Future Perspectives", "text": "Copper death (cuproptosis) modulators hold potential for alleviating neurotoxicity and protecting neurons. Developing more precise molecular-targeted modulators\u2014such as drugs affecting copper transporters, copper chaperones, or mitochondrial copper utilization\u2014could provide neuroprotection by intervening in the core regulatory mechanisms of cuproptosis. This approach may form part of future therapeutic strategies but faces challenges, including poor BBB penetration, insufficient targeting, and the risk of inhibiting essential metalloprotein enzymes. Combining modulators with targeted delivery systems, such as brain-targeted nanoparticles, may enhance their efficacy and safety in the central nervous system (CNS), enabling more precise treatment. In the context of neurological drug development, all preclinical experiments on cuproptosis modulators should be conducted according to GLP standards to ensure scientific rigor and reliability of experimental design and data analysis. The research workflow should encompass key steps, including cell and animal models, drug administration, assessment of copper homeostasis, and measurement of cuproptosis activity, thereby providing a high-quality evidence base for the subsequent development of cuproptosis-targeted therapeutics.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Results", "text": "The K845N variant of LIG1 substitutes a positively charged lysine with a polar asparagine within the conserved oligonucleotide binding-fold domain (OBD) that is essential for DNA binding and positioning (Fig. 1 A and B). K845 participates in interdomain interactions between the OBD and the adenylylation domain (AdD) (SI Appendix, Fig. S1). To investigate the impact of the K845N substitution on DNA ligase activity, we assessed the in vitro ligase activity of LIG1 with lysine at position 845 [LIG1 wild-type (WT)] and LIG1 with asparagine at position 845 (LIG1 K845N) using gel-based DNA ligation assays (Fig. 1C). These assays were performed using full-length LIG1 WT and K845N to evaluate ligation activity in the context of the intact enzyme. Full-length LIG1 WT and K845N proteins were purified in their adenylylated forms which allowed us to measure the concentration of active enzyme using single turnover ligation in the absence of ATP (i.e., active site titrations; SI Appendix, Fig. S2C). LIG1 WT and K845N showed similar active concentrations of 89% and 92%, demonstrating that the K845N substitution does not impact the purification of the adenylylated enzyme.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Results", "text": "LIG1 K845N shows reduced ligase activity toward mismatched substrates. (A) Domain architecture of LIG1. The DNA binding domain (DBD), AdD, and OBD comprise the catalytic core. (B) Multiple sequence alignment of LIG1 orthologs by Clustal Omega. The position of K845 is indicated by the red box. (C) Illustration of LIG1 reaction. LIG1 is first adenylylated by ATP at K568, forming a LIG1-AMP intermediate. The second step, adenyl transfer, involves the transfer of AMP from LIG1 to the 5\u2019 phosphate end of the DNA nick, resulting in a DNA-AMP intermediate. Finally, in the nick sealing step, the 3\u2019 hydroxyl group at the DNA nick attacks the 5\u2019 phosphodiester bond, displacing AMP and sealing the DNA backbone. LIG1 may also release premature AMP-DNA intermediates, known as abortive ligation. Our in vitro ligation reaction uses nicked substrate generated by annealing three oligonucleotides, one of which is labeled with FAM (green star), allowing substrate, aborted DNA-AMP, and ligated product to be distinguished. X:Y = canonical or mismatched base pair. (D\u2013G) Representative gels and quantified fraction of ligation product showing the ligase activity of full-length LIG1 WT and K845N in the context of nucleotide pairs at the 3\u2019 nick position. A:T (D), C:G (E), G:T (F), 8-oxoG:A (G) containing nicked 34mer DNA (300 nM) was incubated with increasing concentrations of LIG1 proteins for 5 min at 37 \u00b0C. Reactions were performed with 1 mM ATP, 10 mM MgCl2, 50 mM MOPS pH 7.5 and 150 mM NaCl. Reactants were analyzed on 15% TBE-urea polyacrylamide mini gels (note that in these gels substrate and AMP-DNA are not resolved). The line graphs show the quantification of the fraction of ligated product from three independent experiments (Mean \u00b1 SD).", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Results", "text": "We compared ligase activity of full-length LIG1 K845N to WT in a standard ligase assay buffer containing a saturating amount of Mg2+ (~10 mM) and varying concentrations of enzyme (Fig. 1 D\u2013G). We used 34mer canonical nicked DNA substrates with different nucleotides at the 3\u2019 end of the nick (Fig. 1C and SI Appendix, Table S1). Ligase activity was quantified as the ratio of the intensity of the ligated DNA product relative to the summed intensities of the ligated DNA, aborted AMP-DNA, and unreacted substrate. Using a dA:dT (hereafter simplified to A:T) substrate, both LIG1 WT and K845N were catalytically active (Fig. 1D). In contrast, there was a modest, but significant decrease in ligase activity for K845N when a 3\u2019 C:G substrate was tested (Fig. 1E). This pattern was also observed with T:A and G:C substrates, where complementary base pairing was reversed (SI Appendix, Fig. S3). From these data it is apparent that K845N is highly active as a DNA ligase and has context-dependent effects that differ from those of LIG1 WT.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Results", "text": "LIG1-catalyzed nick sealing completes DNA replication and repair pathways, and therefore the ability to distinguish mismatches from correctly paired bases helps to maintain genomic integrity. We next tested the promutagenic contexts that would result from polymerase incorporation errors using all 12 possible mismatches at the 3\u2019-OH side of the nicked DNA (Fig. 1F and SI Appendix, Figs. S4 and S5). Under our experimental conditions, the K845N variant showed no detectable ligation activity on any of the mismatched substrates, whereas LIG1 WT exhibited weak but detectable activity on the G:T, C:T, T:C, T:G, C:A, and A:C mismatches (Fig. 1F and SI Appendix, Fig. S4). For the remaining six base pair combinations, WT showed little or no ligase activity and K845N also showed no detectable ligase activity (SI Appendix, Fig. S5). Thus, while K845N displays reduced ligase activity for all substrates tested, this systematic screen demonstrates that the defect is disproportionately greater for mismatched substrates, resulting in near-complete suppression of ligation at promutagenic nicks.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Results", "text": "LIG1 is known to be challenged in distinguishing an 8-oxoG:A mispair relative to other types of mismatches due to its stable Hoogsteen base pairing (25, 32\u201334) and yet this is a common promutagenic substrate formed under oxidative conditions. We therefore compared the activity of LIG1 K845N and WT in the 3\u2019 8-oxoG:A ligation context. As was observed for the nonoxidized mismatches, K845N showed reduced ligase activity with 8-oxoG:A compared to LIG1 WT (Fig. 1G). Reduced ligation activity toward replication errors such as mismatches or incorporation of 8-oxo-dGTP from the oxidized nucleotide pool could elevate the overall fidelity of genome maintenance and support a protective function of K845N.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Results", "text": "To elucidate how the K845N substitution alters the biochemical properties of LIG1, we performed a thorough kinetic characterization of \u2206232 LIG1 under the same conditions that have previously been used to study other rare variants of LIG1 (26, 29, 30). This truncation removes the disordered and poorly conserved N terminus of LIG1 which contains the nuclear localization sequence (24, 35). Active site titrations were used to determine the concentration of active \u2206232 LIG1 (SI Appendix, Fig. S6). Differential scanning fluorimetry showed that \u2206232 LIG1 K845N has the same global stability as the WT protein (SI Appendix, Fig. S7). We next measured DNA binding using a previously described assay that employs a nicked DNA substrate in the absence of Mg2+ to prevent ligation (30). We found that \u2206232 LIG1 K845N binds to a nicked DNA substrate with the same affinity as LIG1 WT (SI Appendix, Fig. S7D). We then investigated the catalytic activity of \u2206232 LIG1 variants using steady-state kinetics. Representative time courses for DNA ligation are shown in Fig. 2A comparing LIG1 WT and K845N with an undamaged nicked DNA substrate (C:G) and a substrate containing an 8-oxoG:A at the 3\u2019 end of the nick. Similar to what was observed for the full-length LIG1 at higher Mg2+ concentration with a different sequence context (Fig. 1 E and G), K845N exhibited reduced activity on the undamaged substrate and substantially less activity on the 8-oxoG:A substrate. Using high-resolution sequencing gels (denaturing polyacrylamide gel electrophoresis (PAGE)), the abortive AMP-DNA species that is particularly prevalent in reaction of the 8-oxoG:A substrate is resolved from unreacted substrate. We next measured the initial rates of ligation at saturating concentration of DNA (1,000 nM). Representative initial rates are shown for the canonical substrate in Fig. 2B. These data show that the reduced production of sealed DNA product in reactions catalyzed by K845N (solid lines) is accompanied by a detectable increase in the amount of AMP-DNA release (dashed lines; Fig. 2B). Analogous reactions performed with the 8-oxoG:A substrate show that AMP-DNA release by LIG1 WT is roughly on par with the amount of ligated DNA, whereas LIG1 K845N releases mostly AMP-DNA and fails to complete the ligation reaction (Fig. 2C).", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Results", "text": "K845N is a higher fidelity, but lower activity LIG1 variant as determined by Michaelis\u2013Menten kinetics. (A) Representative denaturing PAGE showing in vitro ligation reaction with WT or K845N \u2206232 LIG1 (1 nM) with C:G or 8-oxoG:A containing nicked DNA (500 nM), resolving ligated product, aborted product, and substrate. Reactions were performed at 37 \u00b0C with 1 mM ATP, 2 mM MgCl2 (1.0 mM free Mg2). Time courses for ligation of 1,000 nM C:G (B) and 8-oxoG:A (C) 28mer substrate were plotted as number of turnovers with 1 and 5 nM LIG1 WT and 1.5 and 10 nM LIG1 K845N, respectively. Michaelis\u2013Menten dependence for C:G (D) and 8-oxoG:A (E) 28mer nicked DNA was determined with 0.2 mM ATP and 1.0 mM free Mg2+. (F) The fraction of abortive ligation determined from initial rates of ligation. The catalytic efficiency for ligation (G) and the discrimination against 8-oxoG:A (H) were calculated from the kinetic parameters in panels D and E. Data reported are mean \u00b1 SD (N \u2265 3) and are summarized in SI Appendix, Table S2.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Results", "text": "We next measured initial rates of reaction for the canonical nicked DNA substrate for \u2206232 LIG1 WT and K845N across a range of substrate concentrations and fit the Michaelis\u2013Menten equation to the data (Fig. 2D). The derived steady-state rate constants are tabulated in SI Appendix, Table S2. The maximal rate constant (kcat) values were 0.52 \u00b1 0.01 and 0.20 \u00b1 0.01 s\u22121 for LIG1 WT and K845N, respectively (a 2.6-fold decrease; Fig. 2D and SI Appendix, Table S2). The K\u039c value is 1.9-fold higher for K845N (SI Appendix, Table S2), resulting in a 4.9-fold decrease in the catalytic efficiency (kcat/KM). These kcat/KM values are plotted in Fig. 2G. Analogous experiments were performed to determine steady-state kinetic parameters for the 8-oxoG:A substrate (Fig. 2E). There is a stark difference between the ligase activity of K845N and WT with this substrate, with most of the difference in the kcat values (25-fold decrease) and a modest increase in KM (2.3-fold; SI Appendix, Table S2), which corresponds to a 58-fold defect in kcat/KM (Fig. 2G). We quantified the fraction of abortive ligation for each enzyme with each substrate (Fig. 2F), revealing that K845N increases the amount of abortive ligation for both substrates, with a more pronounced reduction in ligated product for the 8-oxoG:A substrate than for the undamaged substrate. The fraction of abortive ligation increases by 1.8-fold, corresponding to a 7.0-fold reduction in the fraction ligated (Flig) of the 8-oxoG:A substrate relative to the WT enzyme (FligWT/FligK845N = (1 \u2212 0.51)/(1 \u2212 0.93) = 0.49/0.07 = 7.0). Thus, the enhanced ability of LIG1 K845N to reject the 8-oxoG:A-containing nick is attributed to both a lowered rate of nick sealing and a higher rate of abortive release of AMP-DNA.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Results", "text": "To quantify the degree to which the K845N substitution increases the fidelity of LIG1, we calculated the discrimination factor by which \u2206232 LIG1 WT and K845N distinguish 8-oxoG:A from undamaged nicks (SI Appendix, Table S2). Consistent with previous reports, LIG1 WT struggles to discriminate an 8-oxoG:A pair from other Watson\u2013Crick paired nicks (25, 32, 34). This is due to the ability of 8-oxoG:A to form a Hoogsteen basepair that is further stabilized by hydrogen bonding in the LIG1 active site (25). Our data show only a 5.1-fold discrimination for LIG1 WT in ligation of 8-oxoG:A versus a canonical paired nick (Fig. 2H and SI Appendix, Table S2). In contrast, K845N discriminates against 8-oxoG:A by a factor of 60-fold, which is 12-fold higher fidelity than the LIG1 WT enzyme (Fig. 2H and SI Appendix, Table S2). K845 is highly conserved among mammalian LIG1 homologs; however, these data suggest that the K845N substitution confers unique biochemical properties. With minimal deleterious effect on ligation of canonical substrates, K845N shows an enhanced fidelity in preventing ligation of an upstream mismatch.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Results", "text": "Based on these biochemical studies, we hypothesized that the enhanced fidelity of LIG1 K845N leads to a reduction in cellular DNA mutation rate, providing a survival advantage under cellular stress conditions. Given previous data implicating oxidative stress as a potential contributor to HD (36), we tested whether K845N might protect against oxidative DNA damage-induced stress, first using human embryonic kidney (HEK) 293 cells stably overexpressing either full-length LIG1 WT or K845N at similar levels (SI Appendix, Fig. S9A). Cells were treated with 8 \u00b5M menadione that induces oxidative DNA damage, and viability was assessed immediately after treatment and after 24 or 48 h of recovery (Fig. 3A). Immediately after treatment, viabilities (~66%) were not different among HEK 293 cells transfected with empty vector (EV), and those expressing LIG1 WT or K845N. After 24 h of recovery, both LIG1 WT and K845N expressing cells showed slightly increased viabilities compared to EV-transfected cells. Interestingly, after 48 h of recovery, cells expressing K845N showed significantly higher viability than EV-transfected (P < 0.0001) or LIG1 WT-expressing cells (P < 0.0001). Furthermore, when assessed after 48 h of recovery at various concentrations of menadione, cells expressing K845N showed consistently higher viability across all treatment conditions (SI Appendix, Fig. S8A), implying that the K845N variant confers a recovery-dependent protective effect against oxidative stress. To test whether the recovery-dependent phenotype could reflect compensatory changes in the abundance of DNA ligases, we quantified LIG1, LIG3, and LIG4 protein levels in untreated and menadione-treated cells, with or without recovery but did not observe any significant differences, indicating that the observed phenotype is not attributable to altered DNA ligase expression (SI Appendix, Fig. S9 A\u2013D). Further, LIG1 WT or K845N expressing cells showed no significant difference in cellular growth rate, indicating that intrinsic growth rates do not influence the differential cell viabilities (SI Appendix, Fig. S9E). To determine whether reduced ligase activity alone drives the K845N survival phenotype, we tested the catalytic-dead LIG1 mutant K568A (24, 37). Unlike K845N, K568A did not increase viability following menadione treatment and resembled WT cells (SI Appendix, Fig. S8 C and D), indicating that enhanced survival is not due to loss of catalytic activity but instead reflects a fidelity-dependent effect.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Results", "text": "Cells harboring the LIG1 K845N variant exhibit increased viability and decreased DNA mutation under conditions that induce DNA damage. (A) HEK 293 cells [EV-transfected or expressing LIG1 WT or LIG1 K845N (K845N)] were treated with 8 \u00b5M menadione for 4 h. Cells were washed in fresh medium and further incubated for the indicated times. Cellular viability was measured by CTG. Bar graph shows mean of three biological replicates \u00b1 SD. Statistical significance was determined from two-way ANOVA with Tukey\u2019s multiple comparison test (****P < 0.0001, **P < 0.01, *P < 0.05). (B and C) HEK 293 cells (WT or K845N) were treated with menadione for 4 h at the indicated concentrations and washed in fresh medium. After 48 h recovery, the cells were harvested. (B) A portion of the harvested cells was used to measure cell viability by CTG. Bar graph shows mean % viability relative to baseline (0% menadione) of two biological replicates \u00b1 SD. (C) DNA was extracted from the remaining cells and mutation frequencies were determined by duplex sequencing. Bar graph shows mean % MF relative to baseline (0% menadione) of two biological replicates \u00b1 SD. Statistical significance in B and C was determined from two-way ANOVA with Tukey\u2019s multiple comparison test (*P < 0.05, **P < 0.01, ***P < 0.001). (D) LCLs derived from HD patients were treated with menadione for 4 h at the indicated concentrations. LCLs were either homozygous for the LIG1 rs145821638 (GRCh38\u2013Chr19:48117686) reference C-allele (LIG1+/+) or heterozygous for the 19AM3 modifier variant A-allele (LIG1K845N/+). Cells were washed in fresh medium and incubated for 24 h. Viability of each LCL was determined using CTG from the mean value of three biological replicates. Bar graph shows mean % viability relative to baseline (0% menadione) \u00b1 SD (n = 4 LIG1+/+ independent LCLs, n = 3 independent LIG1K845N/+ LCLs). Statistical significance indicated in figure was determined by two-way ANOVA with Tukey\u2019s multiple comparison test. The table on the Right shows the HTT CAG repeat lengths for both alleles (CAG1 and CAG2) in HD patient derived LCLs carrying either the LIG1 reference allele (LIG1+/+) or the heterozygous LIG1 K845N variant (LIG1 K845N/+).", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Results", "text": "To directly assess genomic DNA mutation rates, we performed duplex sequencing on HEK 293 cells expressing either LIG1 WT or K845N. We reoptimized conditions for a scaled-up assay that would provide sufficient cells for duplex sequencing, establishing that under these conditions, cells expressing LIG1 K845N exhibited significantly higher viability when treated with 8 \u00b5M or 20 \u00b5M menadione followed by 48 h of recovery (Fig. 3B). We partitioned the cells for viability measurement and extracted DNA from the remaining cells. Duplex sequencing was performed to determine mutation frequency (MF). In both LIG1 WT- and LIG1 K845N-expressing cells MF increased in a menadione concentration-dependent manner (SI Appendix, Fig. S8B). While cells expressing LIG1 K845N exhibited a slightly higher baseline MF in the absence of treatment, they exhibited slightly lower MF in the presence of menadione. Relative to baseline (0 \u00b5M menadione), there were significant main effects (two-way ANOVA) of both menadione concentration (P = 0.0069) and LIG1 variant (P = 0.0377) on MF. Correcting for multiple comparisons revealed that the relative MF in K845N-expressing cells was significantly lower than in WT-expressing cells at 20 \u00b5M menadione (P = 0.0279), with a similar trend observed at 8 \u00b5M (Fig. 3C). These results suggest that the increased cellular viability conferred by K845N may be contributed by a decrease in mutation burden, consistent with the enhanced fidelity of K845N.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Results", "text": "To extend these findings to HD patient-derived cells, we examined lymphoblastoid cell lines (LCLs) harboring only the LIG1 rs145821638 (GRCh38 - Chr19:48117686) reference C-allele (LIG1+/+) or heterozygous HD LCLs harboring the 19AM3 modifier variant A-allele (LIG1K845N/+). We treated HD patient LIG1+/+ and LIG1K845N/+ LCLs with varying concentrations of menadione, and as with HEK 293 cells, viability was measured immediately, and after 24 and 48 h of recovery. While LIG1+/+ LCLs exhibited increased viability immediately following menadione treatment without recovery (P = 0.0402) (SI Appendix, Fig. S10A), LIG1K845N/+ LCLs showed significantly higher viability than LIG1+/+ LCLs after a 24-h recovery period postmenadione treatment (Fig. 3D). Relative to baseline (0 \u00b5M menadione), two-way ANOVA identified significant main effects of both menadione concentration (P < 0.0001) and K845N genotype (P = 0.0417) on cell viability, with no significant interaction between the two factors (P = 0.7225). This indicates that the effects of menadione treatment and genotype were additive, and that both genotypes exhibited similar viability responses across increasing menadione concentrations. No individual genotype comparison remained significant after multiple testing correction, though viability increased to the greatest extent in LIG1K845N/+ vs. LIG1+/+ LCLs treated with 21 \u00b5M menadione (P = 0.0589). After 48 h, differences were similar to those observed at 24 h but were less distinct (SI Appendix, Fig. S10B). Given the small number of available HD LCLs with the K845N variant\u2014limited to heterozygous carriers of this rare variant, and the genetic background heterogeneity of the patient samples, the statistical power in the LCLs is limited. Nevertheless, consistent with findings in HEK 293 cells, where the protective effect of LIG1 K845N emerged after a recovery period, HD LCLs expressing the endogenous LIG1 K845N variant also exhibited a recovery-dependent protective phenotype. As in the HEK 293 overexpression system, we observed no difference in basal proliferation rate between LIG1+/+ and LIG1K845N/+ cells (SI Appendix, Fig. S10C) and no significant differences in LIG1, LIG3, and LIG4 protein levels in untreated and menadione-treated cells, with or without recovery. (SI Appendix, Fig. S11 A\u2013D).", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Results", "text": "The somatic expansion of the HTT CAG repeat drives the onset of clinical phenotypes in HD. Age-dependent somatic expansion is readily measured in HttQ111 CAG knock-in mice where it occurs at a high rate in both the striatum and the liver (18, 38). We have previously shown that somatic expansion in both these tissues can be modified to different degrees by MMR gene and various other gene knockouts and can also be modified by missense mutation impacting protein function (18, 39\u201342). To test whether the LIG1 missense variant associated with delayed HD impacts somatic CAG expansion, we used CRISPR-Cas9-mediated homology-directed repair to introduce the same mutation as the human modifier (19AM3, rs145821638 C->A) into the mouse Lig1 gene, changing the LIG1 lysine residue at amino acid 843 to an asparagine (Fig. 4A). As highlighted in Fig. 1B, K845 is a conserved residue, indicating its functional importance across species, including the mouse. We derived a Lig1K843N line, which we crossed with HttQ111 mice. The Lig1K843N variant did not impact embryonic development, with Lig1+/+, Lig1K843N/+, or Lig1K843N/K843N born at expected Mendelian ratios (SI Appendix, Fig. S12), contrasting with Lig1 null alleles where homozygous null embryos do not survive beyond E15.5/E16.5 (19, 20). Consistent with the viability of Lig1K843N/K843N mice, the K843N change did not alter Lig1 expression levels (SI Appendix, Fig. S13). We also did not observe any overt phenotype in Lig1K843N/+ or Lig1K843N/K843N mice. The mice did not develop tumors, in contrast to mice harboring the LIG1 Syndrome-associated R771W point mutation that in patient cells (46BR1 cell line) reduces ligase activity 20-fold (43). HttQ111/+ mice with Lig1+/+, Lig1K843N/+, and Lig1K843N/K843N genotypes were aged for analyses of somatic expansion. The two most unstable tissues, striatum, and liver (Fig. 4B and SI Appendix, Fig. S15) revealed significantly reduced somatic expansion in Lig1K843N/K843N striata compared to Lig1+/+ striata at 3, 6, and 10 mo of age (3 mo P < 0.001, 6 and 10 mo P < 0.0001) and significantly reduced somatic expansion in Lig1K843N/K843N liver at 6 and 10 mo of age (6 mo P < 0.001, 10 mo P < 0.0001). Expansion was also significantly reduced in Lig1K843N/K843N cortex at 3 mo (P < 0.001) and 6 mo of age (P < 0.0001) and in the hippocampus at 6 mo of age (P < 0.001) (SI Appendix, Figs. S14 and S15). There were no significant differences in the inherited repeat lengths in the cohorts of mice with different Lig1 genotypes (SI Appendix, Fig. S16). Further, expansion was not impacted by the presence of a silent protospacer adjacent motif (PAM) mutation cointroduced with K843N, as determined in an independent line of mice harboring the PAM mutation in the absence of the K843N change (Materials and Methods and SI Appendix, Fig. S17). Of the tissues analyzed, LIG1K843N appeared to have the most robust impact on expansion in the striatum. Although we observed Lig1K843N allele dose-dependent trends, expansion was not significantly impacted in heterozygous Lig1K843N/+ mice relative to Lig1+/+ mice. The impact of this variant appears to be moderate, with one variant allele reducing striatal expansion by 3 to 5% and two mutant alleles reducing expansion by ~15 to 16% in 6 to 10-mo mice. Overall, these data support reduced somatic expansion as a consequence of a LIG1 functional alteration elicited by the K843N substitution.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Results", "text": "The orthologous K843N mutation suppresses somatic CAG expansion in mice. (A) CRISPR-Cas9-mediated generation of the Lig1K843N knock-in allele. Top: The sequence is part of mouse Lig1 exon 26 (canonical transcript ENSMUST00000177588.10) with amino acid translation showing lysine (K) at amino acid 843 (orthologous to human K845). The targeting sgRNA and PAM are shown. A 150 nt single-stranded oligonucleotide (not depicted in the Figure; refer to Materials and Methods for detail) was used for homology-directed repair to mutate G > A changing K843 to asparagine (N) (pink box). This also introduced a silent C > T change within the PAM (green box). Bottom: Sanger sequence from a Lig1 WT mouse, and heterozygous and homozygous K843N knock-in mice. (B) Somatic CAG expansion indices (Min to Max box-whisker plots) from striatum and liver of HttQ111/+ mice with different Lig1 genotypes. 3 mo: Lig1+/+ N = 2, Lig1K843N/+ N = 2, Lig1K843N/K843N N = 8. 6 mo: Lig1+/+ N = 4, Lig1K843N/+ N = 9, Lig1K843N/K843N N = 7. 10 mo: Lig1+/+ N = 9, Lig1K843N/+ N = 9, Lig1K843N/K843N N = 10. **P < 0.01; ***P < 0.001, ****P < 0.0001 (one-way ANOVA, comparing all genotypes for each age and tissue, with Tukey\u2019s multiple comparison test).", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "LIG1 K845N Shows Reduced Ligase Activity In Vitro That Is Sensitive to the Ligation Context, Particularly Promutagenic 3\u2019 Mismatches.", "text": "The K845N variant of LIG1 substitutes a positively charged lysine with a polar asparagine within the conserved oligonucleotide binding-fold domain (OBD) that is essential for DNA binding and positioning (Fig. 1 A and B). K845 participates in interdomain interactions between the OBD and the adenylylation domain (AdD) (SI Appendix, Fig. S1). To investigate the impact of the K845N substitution on DNA ligase activity, we assessed the in vitro ligase activity of LIG1 with lysine at position 845 [LIG1 wild-type (WT)] and LIG1 with asparagine at position 845 (LIG1 K845N) using gel-based DNA ligation assays (Fig. 1C). These assays were performed using full-length LIG1 WT and K845N to evaluate ligation activity in the context of the intact enzyme. Full-length LIG1 WT and K845N proteins were purified in their adenylylated forms which allowed us to measure the concentration of active enzyme using single turnover ligation in the absence of ATP (i.e., active site titrations; SI Appendix, Fig. S2C). LIG1 WT and K845N showed similar active concentrations of 89% and 92%, demonstrating that the K845N substitution does not impact the purification of the adenylylated enzyme.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "LIG1 K845N Shows Reduced Ligase Activity In Vitro That Is Sensitive to the Ligation Context, Particularly Promutagenic 3\u2019 Mismatches.", "text": "LIG1 K845N shows reduced ligase activity toward mismatched substrates. (A) Domain architecture of LIG1. The DNA binding domain (DBD), AdD, and OBD comprise the catalytic core. (B) Multiple sequence alignment of LIG1 orthologs by Clustal Omega. The position of K845 is indicated by the red box. (C) Illustration of LIG1 reaction. LIG1 is first adenylylated by ATP at K568, forming a LIG1-AMP intermediate. The second step, adenyl transfer, involves the transfer of AMP from LIG1 to the 5\u2019 phosphate end of the DNA nick, resulting in a DNA-AMP intermediate. Finally, in the nick sealing step, the 3\u2019 hydroxyl group at the DNA nick attacks the 5\u2019 phosphodiester bond, displacing AMP and sealing the DNA backbone. LIG1 may also release premature AMP-DNA intermediates, known as abortive ligation. Our in vitro ligation reaction uses nicked substrate generated by annealing three oligonucleotides, one of which is labeled with FAM (green star), allowing substrate, aborted DNA-AMP, and ligated product to be distinguished. X:Y = canonical or mismatched base pair. (D\u2013G) Representative gels and quantified fraction of ligation product showing the ligase activity of full-length LIG1 WT and K845N in the context of nucleotide pairs at the 3\u2019 nick position. A:T (D), C:G (E), G:T (F), 8-oxoG:A (G) containing nicked 34mer DNA (300 nM) was incubated with increasing concentrations of LIG1 proteins for 5 min at 37 \u00b0C. Reactions were performed with 1 mM ATP, 10 mM MgCl2, 50 mM MOPS pH 7.5 and 150 mM NaCl. Reactants were analyzed on 15% TBE-urea polyacrylamide mini gels (note that in these gels substrate and AMP-DNA are not resolved). The line graphs show the quantification of the fraction of ligated product from three independent experiments (Mean \u00b1 SD).", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "LIG1 K845N Shows Reduced Ligase Activity In Vitro That Is Sensitive to the Ligation Context, Particularly Promutagenic 3\u2019 Mismatches.", "text": "We compared ligase activity of full-length LIG1 K845N to WT in a standard ligase assay buffer containing a saturating amount of Mg2+ (~10 mM) and varying concentrations of enzyme (Fig. 1 D\u2013G). We used 34mer canonical nicked DNA substrates with different nucleotides at the 3\u2019 end of the nick (Fig. 1C and SI Appendix, Table S1). Ligase activity was quantified as the ratio of the intensity of the ligated DNA product relative to the summed intensities of the ligated DNA, aborted AMP-DNA, and unreacted substrate. Using a dA:dT (hereafter simplified to A:T) substrate, both LIG1 WT and K845N were catalytically active (Fig. 1D). In contrast, there was a modest, but significant decrease in ligase activity for K845N when a 3\u2019 C:G substrate was tested (Fig. 1E). This pattern was also observed with T:A and G:C substrates, where complementary base pairing was reversed (SI Appendix, Fig. S3). From these data it is apparent that K845N is highly active as a DNA ligase and has context-dependent effects that differ from those of LIG1 WT.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "LIG1 K845N Shows Reduced Ligase Activity In Vitro That Is Sensitive to the Ligation Context, Particularly Promutagenic 3\u2019 Mismatches.", "text": "LIG1-catalyzed nick sealing completes DNA replication and repair pathways, and therefore the ability to distinguish mismatches from correctly paired bases helps to maintain genomic integrity. We next tested the promutagenic contexts that would result from polymerase incorporation errors using all 12 possible mismatches at the 3\u2019-OH side of the nicked DNA (Fig. 1F and SI Appendix, Figs. S4 and S5). Under our experimental conditions, the K845N variant showed no detectable ligation activity on any of the mismatched substrates, whereas LIG1 WT exhibited weak but detectable activity on the G:T, C:T, T:C, T:G, C:A, and A:C mismatches (Fig. 1F and SI Appendix, Fig. S4). For the remaining six base pair combinations, WT showed little or no ligase activity and K845N also showed no detectable ligase activity (SI Appendix, Fig. S5). Thus, while K845N displays reduced ligase activity for all substrates tested, this systematic screen demonstrates that the defect is disproportionately greater for mismatched substrates, resulting in near-complete suppression of ligation at promutagenic nicks.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "LIG1 K845N Shows Reduced Ligase Activity In Vitro That Is Sensitive to the Ligation Context, Particularly Promutagenic 3\u2019 Mismatches.", "text": "LIG1 is known to be challenged in distinguishing an 8-oxoG:A mispair relative to other types of mismatches due to its stable Hoogsteen base pairing (25, 32\u201334) and yet this is a common promutagenic substrate formed under oxidative conditions. We therefore compared the activity of LIG1 K845N and WT in the 3\u2019 8-oxoG:A ligation context. As was observed for the nonoxidized mismatches, K845N showed reduced ligase activity with 8-oxoG:A compared to LIG1 WT (Fig. 1G). Reduced ligation activity toward replication errors such as mismatches or incorporation of 8-oxo-dGTP from the oxidized nucleotide pool could elevate the overall fidelity of genome maintenance and support a protective function of K845N.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "K845N Selectively Aborts Ligation of an 8-oxoG:A Mismatch to Enhance the Fidelity of Ligation.", "text": "To elucidate how the K845N substitution alters the biochemical properties of LIG1, we performed a thorough kinetic characterization of \u2206232 LIG1 under the same conditions that have previously been used to study other rare variants of LIG1 (26, 29, 30). This truncation removes the disordered and poorly conserved N terminus of LIG1 which contains the nuclear localization sequence (24, 35). Active site titrations were used to determine the concentration of active \u2206232 LIG1 (SI Appendix, Fig. S6). Differential scanning fluorimetry showed that \u2206232 LIG1 K845N has the same global stability as the WT protein (SI Appendix, Fig. S7). We next measured DNA binding using a previously described assay that employs a nicked DNA substrate in the absence of Mg2+ to prevent ligation (30). We found that \u2206232 LIG1 K845N binds to a nicked DNA substrate with the same affinity as LIG1 WT (SI Appendix, Fig. S7D). We then investigated the catalytic activity of \u2206232 LIG1 variants using steady-state kinetics. Representative time courses for DNA ligation are shown in Fig. 2A comparing LIG1 WT and K845N with an undamaged nicked DNA substrate (C:G) and a substrate containing an 8-oxoG:A at the 3\u2019 end of the nick. Similar to what was observed for the full-length LIG1 at higher Mg2+ concentration with a different sequence context (Fig. 1 E and G), K845N exhibited reduced activity on the undamaged substrate and substantially less activity on the 8-oxoG:A substrate. Using high-resolution sequencing gels (denaturing polyacrylamide gel electrophoresis (PAGE)), the abortive AMP-DNA species that is particularly prevalent in reaction of the 8-oxoG:A substrate is resolved from unreacted substrate. We next measured the initial rates of ligation at saturating concentration of DNA (1,000 nM). Representative initial rates are shown for the canonical substrate in Fig. 2B. These data show that the reduced production of sealed DNA product in reactions catalyzed by K845N (solid lines) is accompanied by a detectable increase in the amount of AMP-DNA release (dashed lines; Fig. 2B). Analogous reactions performed with the 8-oxoG:A substrate show that AMP-DNA release by LIG1 WT is roughly on par with the amount of ligated DNA, whereas LIG1 K845N releases mostly AMP-DNA and fails to complete the ligation reaction (Fig. 2C).", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "K845N Selectively Aborts Ligation of an 8-oxoG:A Mismatch to Enhance the Fidelity of Ligation.", "text": "K845N is a higher fidelity, but lower activity LIG1 variant as determined by Michaelis\u2013Menten kinetics. (A) Representative denaturing PAGE showing in vitro ligation reaction with WT or K845N \u2206232 LIG1 (1 nM) with C:G or 8-oxoG:A containing nicked DNA (500 nM), resolving ligated product, aborted product, and substrate. Reactions were performed at 37 \u00b0C with 1 mM ATP, 2 mM MgCl2 (1.0 mM free Mg2). Time courses for ligation of 1,000 nM C:G (B) and 8-oxoG:A (C) 28mer substrate were plotted as number of turnovers with 1 and 5 nM LIG1 WT and 1.5 and 10 nM LIG1 K845N, respectively. Michaelis\u2013Menten dependence for C:G (D) and 8-oxoG:A (E) 28mer nicked DNA was determined with 0.2 mM ATP and 1.0 mM free Mg2+. (F) The fraction of abortive ligation determined from initial rates of ligation. The catalytic efficiency for ligation (G) and the discrimination against 8-oxoG:A (H) were calculated from the kinetic parameters in panels D and E. Data reported are mean \u00b1 SD (N \u2265 3) and are summarized in SI Appendix, Table S2.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "K845N Selectively Aborts Ligation of an 8-oxoG:A Mismatch to Enhance the Fidelity of Ligation.", "text": "We next measured initial rates of reaction for the canonical nicked DNA substrate for \u2206232 LIG1 WT and K845N across a range of substrate concentrations and fit the Michaelis\u2013Menten equation to the data (Fig. 2D). The derived steady-state rate constants are tabulated in SI Appendix, Table S2. The maximal rate constant (kcat) values were 0.52 \u00b1 0.01 and 0.20 \u00b1 0.01 s\u22121 for LIG1 WT and K845N, respectively (a 2.6-fold decrease; Fig. 2D and SI Appendix, Table S2). The K\u039c value is 1.9-fold higher for K845N (SI Appendix, Table S2), resulting in a 4.9-fold decrease in the catalytic efficiency (kcat/KM). These kcat/KM values are plotted in Fig. 2G. Analogous experiments were performed to determine steady-state kinetic parameters for the 8-oxoG:A substrate (Fig. 2E). There is a stark difference between the ligase activity of K845N and WT with this substrate, with most of the difference in the kcat values (25-fold decrease) and a modest increase in KM (2.3-fold; SI Appendix, Table S2), which corresponds to a 58-fold defect in kcat/KM (Fig. 2G). We quantified the fraction of abortive ligation for each enzyme with each substrate (Fig. 2F), revealing that K845N increases the amount of abortive ligation for both substrates, with a more pronounced reduction in ligated product for the 8-oxoG:A substrate than for the undamaged substrate. The fraction of abortive ligation increases by 1.8-fold, corresponding to a 7.0-fold reduction in the fraction ligated (Flig) of the 8-oxoG:A substrate relative to the WT enzyme (FligWT/FligK845N = (1 \u2212 0.51)/(1 \u2212 0.93) = 0.49/0.07 = 7.0). Thus, the enhanced ability of LIG1 K845N to reject the 8-oxoG:A-containing nick is attributed to both a lowered rate of nick sealing and a higher rate of abortive release of AMP-DNA.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "K845N Selectively Aborts Ligation of an 8-oxoG:A Mismatch to Enhance the Fidelity of Ligation.", "text": "To quantify the degree to which the K845N substitution increases the fidelity of LIG1, we calculated the discrimination factor by which \u2206232 LIG1 WT and K845N distinguish 8-oxoG:A from undamaged nicks (SI Appendix, Table S2). Consistent with previous reports, LIG1 WT struggles to discriminate an 8-oxoG:A pair from other Watson\u2013Crick paired nicks (25, 32, 34). This is due to the ability of 8-oxoG:A to form a Hoogsteen basepair that is further stabilized by hydrogen bonding in the LIG1 active site (25). Our data show only a 5.1-fold discrimination for LIG1 WT in ligation of 8-oxoG:A versus a canonical paired nick (Fig. 2H and SI Appendix, Table S2). In contrast, K845N discriminates against 8-oxoG:A by a factor of 60-fold, which is 12-fold higher fidelity than the LIG1 WT enzyme (Fig. 2H and SI Appendix, Table S2). K845 is highly conserved among mammalian LIG1 homologs; however, these data suggest that the K845N substitution confers unique biochemical properties. With minimal deleterious effect on ligation of canonical substrates, K845N shows an enhanced fidelity in preventing ligation of an upstream mismatch.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Cells Harboring the LIG1 K845N Variant Exhibit Increased Viability and Decreased Genomic DNA Mutation under Oxidative Stress.", "text": "Based on these biochemical studies, we hypothesized that the enhanced fidelity of LIG1 K845N leads to a reduction in cellular DNA mutation rate, providing a survival advantage under cellular stress conditions. Given previous data implicating oxidative stress as a potential contributor to HD (36), we tested whether K845N might protect against oxidative DNA damage-induced stress, first using human embryonic kidney (HEK) 293 cells stably overexpressing either full-length LIG1 WT or K845N at similar levels (SI Appendix, Fig. S9A). Cells were treated with 8 \u00b5M menadione that induces oxidative DNA damage, and viability was assessed immediately after treatment and after 24 or 48 h of recovery (Fig. 3A). Immediately after treatment, viabilities (~66%) were not different among HEK 293 cells transfected with empty vector (EV), and those expressing LIG1 WT or K845N. After 24 h of recovery, both LIG1 WT and K845N expressing cells showed slightly increased viabilities compared to EV-transfected cells. Interestingly, after 48 h of recovery, cells expressing K845N showed significantly higher viability than EV-transfected (P < 0.0001) or LIG1 WT-expressing cells (P < 0.0001). Furthermore, when assessed after 48 h of recovery at various concentrations of menadione, cells expressing K845N showed consistently higher viability across all treatment conditions (SI Appendix, Fig. S8A), implying that the K845N variant confers a recovery-dependent protective effect against oxidative stress. To test whether the recovery-dependent phenotype could reflect compensatory changes in the abundance of DNA ligases, we quantified LIG1, LIG3, and LIG4 protein levels in untreated and menadione-treated cells, with or without recovery but did not observe any significant differences, indicating that the observed phenotype is not attributable to altered DNA ligase expression (SI Appendix, Fig. S9 A\u2013D). Further, LIG1 WT or K845N expressing cells showed no significant difference in cellular growth rate, indicating that intrinsic growth rates do not influence the differential cell viabilities (SI Appendix, Fig. S9E). To determine whether reduced ligase activity alone drives the K845N survival phenotype, we tested the catalytic-dead LIG1 mutant K568A (24, 37). Unlike K845N, K568A did not increase viability following menadione treatment and resembled WT cells (SI Appendix, Fig. S8 C and D), indicating that enhanced survival is not due to loss of catalytic activity but instead reflects a fidelity-dependent effect.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Cells Harboring the LIG1 K845N Variant Exhibit Increased Viability and Decreased Genomic DNA Mutation under Oxidative Stress.", "text": "Cells harboring the LIG1 K845N variant exhibit increased viability and decreased DNA mutation under conditions that induce DNA damage. (A) HEK 293 cells [EV-transfected or expressing LIG1 WT or LIG1 K845N (K845N)] were treated with 8 \u00b5M menadione for 4 h. Cells were washed in fresh medium and further incubated for the indicated times. Cellular viability was measured by CTG. Bar graph shows mean of three biological replicates \u00b1 SD. Statistical significance was determined from two-way ANOVA with Tukey\u2019s multiple comparison test (****P < 0.0001, **P < 0.01, *P < 0.05). (B and C) HEK 293 cells (WT or K845N) were treated with menadione for 4 h at the indicated concentrations and washed in fresh medium. After 48 h recovery, the cells were harvested. (B) A portion of the harvested cells was used to measure cell viability by CTG. Bar graph shows mean % viability relative to baseline (0% menadione) of two biological replicates \u00b1 SD. (C) DNA was extracted from the remaining cells and mutation frequencies were determined by duplex sequencing. Bar graph shows mean % MF relative to baseline (0% menadione) of two biological replicates \u00b1 SD. Statistical significance in B and C was determined from two-way ANOVA with Tukey\u2019s multiple comparison test (*P < 0.05, **P < 0.01, ***P < 0.001). (D) LCLs derived from HD patients were treated with menadione for 4 h at the indicated concentrations. LCLs were either homozygous for the LIG1 rs145821638 (GRCh38\u2013Chr19:48117686) reference C-allele (LIG1+/+) or heterozygous for the 19AM3 modifier variant A-allele (LIG1K845N/+). Cells were washed in fresh medium and incubated for 24 h. Viability of each LCL was determined using CTG from the mean value of three biological replicates. Bar graph shows mean % viability relative to baseline (0% menadione) \u00b1 SD (n = 4 LIG1+/+ independent LCLs, n = 3 independent LIG1K845N/+ LCLs). Statistical significance indicated in figure was determined by two-way ANOVA with Tukey\u2019s multiple comparison test. The table on the Right shows the HTT CAG repeat lengths for both alleles (CAG1 and CAG2) in HD patient derived LCLs carrying either the LIG1 reference allele (LIG1+/+) or the heterozygous LIG1 K845N variant (LIG1 K845N/+).", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Cells Harboring the LIG1 K845N Variant Exhibit Increased Viability and Decreased Genomic DNA Mutation under Oxidative Stress.", "text": "To directly assess genomic DNA mutation rates, we performed duplex sequencing on HEK 293 cells expressing either LIG1 WT or K845N. We reoptimized conditions for a scaled-up assay that would provide sufficient cells for duplex sequencing, establishing that under these conditions, cells expressing LIG1 K845N exhibited significantly higher viability when treated with 8 \u00b5M or 20 \u00b5M menadione followed by 48 h of recovery (Fig. 3B). We partitioned the cells for viability measurement and extracted DNA from the remaining cells. Duplex sequencing was performed to determine mutation frequency (MF). In both LIG1 WT- and LIG1 K845N-expressing cells MF increased in a menadione concentration-dependent manner (SI Appendix, Fig. S8B). While cells expressing LIG1 K845N exhibited a slightly higher baseline MF in the absence of treatment, they exhibited slightly lower MF in the presence of menadione. Relative to baseline (0 \u00b5M menadione), there were significant main effects (two-way ANOVA) of both menadione concentration (P = 0.0069) and LIG1 variant (P = 0.0377) on MF. Correcting for multiple comparisons revealed that the relative MF in K845N-expressing cells was significantly lower than in WT-expressing cells at 20 \u00b5M menadione (P = 0.0279), with a similar trend observed at 8 \u00b5M (Fig. 3C). These results suggest that the increased cellular viability conferred by K845N may be contributed by a decrease in mutation burden, consistent with the enhanced fidelity of K845N.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Cells Harboring the LIG1 K845N Variant Exhibit Increased Viability and Decreased Genomic DNA Mutation under Oxidative Stress.", "text": "To extend these findings to HD patient-derived cells, we examined lymphoblastoid cell lines (LCLs) harboring only the LIG1 rs145821638 (GRCh38 - Chr19:48117686) reference C-allele (LIG1+/+) or heterozygous HD LCLs harboring the 19AM3 modifier variant A-allele (LIG1K845N/+). We treated HD patient LIG1+/+ and LIG1K845N/+ LCLs with varying concentrations of menadione, and as with HEK 293 cells, viability was measured immediately, and after 24 and 48 h of recovery. While LIG1+/+ LCLs exhibited increased viability immediately following menadione treatment without recovery (P = 0.0402) (SI Appendix, Fig. S10A), LIG1K845N/+ LCLs showed significantly higher viability than LIG1+/+ LCLs after a 24-h recovery period postmenadione treatment (Fig. 3D). Relative to baseline (0 \u00b5M menadione), two-way ANOVA identified significant main effects of both menadione concentration (P < 0.0001) and K845N genotype (P = 0.0417) on cell viability, with no significant interaction between the two factors (P = 0.7225). This indicates that the effects of menadione treatment and genotype were additive, and that both genotypes exhibited similar viability responses across increasing menadione concentrations. No individual genotype comparison remained significant after multiple testing correction, though viability increased to the greatest extent in LIG1K845N/+ vs. LIG1+/+ LCLs treated with 21 \u00b5M menadione (P = 0.0589). After 48 h, differences were similar to those observed at 24 h but were less distinct (SI Appendix, Fig. S10B). Given the small number of available HD LCLs with the K845N variant\u2014limited to heterozygous carriers of this rare variant, and the genetic background heterogeneity of the patient samples, the statistical power in the LCLs is limited. Nevertheless, consistent with findings in HEK 293 cells, where the protective effect of LIG1 K845N emerged after a recovery period, HD LCLs expressing the endogenous LIG1 K845N variant also exhibited a recovery-dependent protective phenotype. As in the HEK 293 overexpression system, we observed no difference in basal proliferation rate between LIG1+/+ and LIG1K845N/+ cells (SI Appendix, Fig. S10C) and no significant differences in LIG1, LIG3, and LIG4 protein levels in untreated and menadione-treated cells, with or without recovery. (SI Appendix, Fig. S11 A\u2013D).", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "The Mouse K845N Orthologous Variant Suppresses Somatic CAG Expansion in ", "text": "The somatic expansion of the HTT CAG repeat drives the onset of clinical phenotypes in HD. Age-dependent somatic expansion is readily measured in HttQ111 CAG knock-in mice where it occurs at a high rate in both the striatum and the liver (18, 38). We have previously shown that somatic expansion in both these tissues can be modified to different degrees by MMR gene and various other gene knockouts and can also be modified by missense mutation impacting protein function (18, 39\u201342). To test whether the LIG1 missense variant associated with delayed HD impacts somatic CAG expansion, we used CRISPR-Cas9-mediated homology-directed repair to introduce the same mutation as the human modifier (19AM3, rs145821638 C->A) into the mouse Lig1 gene, changing the LIG1 lysine residue at amino acid 843 to an asparagine (Fig. 4A). As highlighted in Fig. 1B, K845 is a conserved residue, indicating its functional importance across species, including the mouse. We derived a Lig1K843N line, which we crossed with HttQ111 mice. The Lig1K843N variant did not impact embryonic development, with Lig1+/+, Lig1K843N/+, or Lig1K843N/K843N born at expected Mendelian ratios (SI Appendix, Fig. S12), contrasting with Lig1 null alleles where homozygous null embryos do not survive beyond E15.5/E16.5 (19, 20). Consistent with the viability of Lig1K843N/K843N mice, the K843N change did not alter Lig1 expression levels (SI Appendix, Fig. S13). We also did not observe any overt phenotype in Lig1K843N/+ or Lig1K843N/K843N mice. The mice did not develop tumors, in contrast to mice harboring the LIG1 Syndrome-associated R771W point mutation that in patient cells (46BR1 cell line) reduces ligase activity 20-fold (43). HttQ111/+ mice with Lig1+/+, Lig1K843N/+, and Lig1K843N/K843N genotypes were aged for analyses of somatic expansion. The two most unstable tissues, striatum, and liver (Fig. 4B and SI Appendix, Fig. S15) revealed significantly reduced somatic expansion in Lig1K843N/K843N striata compared to Lig1+/+ striata at 3, 6, and 10 mo of age (3 mo P < 0.001, 6 and 10 mo P < 0.0001) and significantly reduced somatic expansion in Lig1K843N/K843N liver at 6 and 10 mo of age (6 mo P < 0.001, 10 mo P < 0.0001). Expansion was also significantly reduced in Lig1K843N/K843N cortex at 3 mo (P < 0.001) and 6 mo of age (P < 0.0001) and in the hippocampus at 6 mo of age (P < 0.001) (SI Appendix, Figs. S14 and S15). There were no significant differences in the inherited repeat lengths in the cohorts of mice with different Lig1 genotypes (SI Appendix, Fig. S16). Further, expansion was not impacted by the presence of a silent protospacer adjacent motif (PAM) mutation cointroduced with K843N, as determined in an independent line of mice harboring the PAM mutation in the absence of the K843N change (Materials and Methods and SI Appendix, Fig. S17). Of the tissues analyzed, LIG1K843N appeared to have the most robust impact on expansion in the striatum. Although we observed Lig1K843N allele dose-dependent trends, expansion was not significantly impacted in heterozygous Lig1K843N/+ mice relative to Lig1+/+ mice. The impact of this variant appears to be moderate, with one variant allele reducing striatal expansion by 3 to 5% and two mutant alleles reducing expansion by ~15 to 16% in 6 to 10-mo mice. Overall, these data support reduced somatic expansion as a consequence of a LIG1 functional alteration elicited by the K843N substitution.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "The Mouse K845N Orthologous Variant Suppresses Somatic CAG Expansion in ", "text": "The orthologous K843N mutation suppresses somatic CAG expansion in mice. (A) CRISPR-Cas9-mediated generation of the Lig1K843N knock-in allele. Top: The sequence is part of mouse Lig1 exon 26 (canonical transcript ENSMUST00000177588.10) with amino acid translation showing lysine (K) at amino acid 843 (orthologous to human K845). The targeting sgRNA and PAM are shown. A 150 nt single-stranded oligonucleotide (not depicted in the Figure; refer to Materials and Methods for detail) was used for homology-directed repair to mutate G > A changing K843 to asparagine (N) (pink box). This also introduced a silent C > T change within the PAM (green box). Bottom: Sanger sequence from a Lig1 WT mouse, and heterozygous and homozygous K843N knock-in mice. (B) Somatic CAG expansion indices (Min to Max box-whisker plots) from striatum and liver of HttQ111/+ mice with different Lig1 genotypes. 3 mo: Lig1+/+ N = 2, Lig1K843N/+ N = 2, Lig1K843N/K843N N = 8. 6 mo: Lig1+/+ N = 4, Lig1K843N/+ N = 9, Lig1K843N/K843N N = 7. 10 mo: Lig1+/+ N = 9, Lig1K843N/+ N = 9, Lig1K843N/K843N N = 10. **P < 0.01; ***P < 0.001, ****P < 0.0001 (one-way ANOVA, comparing all genotypes for each age and tissue, with Tukey\u2019s multiple comparison test).", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Discussion", "text": "The minor allele of rs145821638, coding for the LIG1 K845N variant, was identified in GWAS as the top SNV capturing a modifier effect associated with delayed HD clinical phenotypes (13, 17). The impact of this SNV is substantial, delaying motor onset by 7 to 8 y. This rare variant, present in approximately 1 in 700 individuals (across all populations in gnomAD v.4.1.0), substitutes a conserved lysine within the OBD with a polar asparagine predicted to disrupt hydrogen bonding interactions at the domain interface between the OBD and the AdD. In this study, we evaluated the hypothesis that the HD modifier effect reflects mechanisms directly impacted by this LIG1 K845N missense change by studying its biochemical, cellular, and molecular consequences.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Discussion", "text": "Biallelic loss of function LIG1 missense mutations associated with LIG1 Syndrome exhibit decreases in the maximal rate constant and in the catalytic efficiency for DNA ligation (Table 1). In comparison, our biochemical analyses of full-length and \u2206232 truncated forms of LIG1 reveal the impact of K845N to be distinct from LIG1 alleles associated with immunodeficiency syndrome (Table 1). We find that K845N maintains near-normal ligation activity on undamaged nicked DNA substrates but exhibits reduced activity toward mismatched substrates. This enhanced substrate discrimination, including against 8-oxoG:A base pairs, highlights K845N as a higher-fidelity variant of LIG1. Kinetic studies demonstrate a 59-fold reduction in catalytic efficiency (kcat/KM) on 8-oxoG:A substrates compared to WT, accompanied by an increase in abortive ligation events. Notably, the K845N variant exhibits 60-fold discrimination between mismatched and correctly paired substrates, far exceeding the fivefold discrimination for WT.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Discussion", "text": "Importantly, much of this discrimination manifests as increased abortive ligation of the improper nick which will prevent binding by another ligase and ensure additional time for other repair processes to take place. Thus, the enhanced ligation fidelity of K845N is predicted to manifest in a heterozygous background (i.e., even if cells express both LIG1 WT and K845N). This is relevant in HD patients where profound clinical modification is seen in individuals heterozygous for this LIG1 variant. It should also be noted that AMP-DNA intermediates are toxic to certain primary neurons in the absence of aprataxin, a hydrolase that cleaves AMP-DNA linkages to regenerate the 5\u2019 phosphate (44). Loss of function mutations in aprataxin cause Ataxia with Oculomotor Apraxia type 1 (45, 46). Given the prevalence of the K845N variant it is unlikely that abortive ligation overwhelms the ability of aprataxin to repair AMP-DNA sites in the brain. Nevertheless, how AMP\u2013DNA intermediates are handled across different cellular contexts, and how this balance influences survival during recovery from oxidative stress, will require future investigation. For example, delayed ligation could influence DNA damage signaling during the recovery phase. LIG1 has been reported to interact with checkpoint-associated complexes such as hRad17\u2013RFC, which promotes loading of the 9-1-1 clamp at sites of DNA damage (47). Slower or more selective ligation at damaged nicks could therefore extend engagement of checkpoint or stress-response pathways, providing additional time for repair.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Discussion", "text": "The K845N substitution is predicted to alter the interface between the AdD and the OBD which is expected to stabilize the closed conformation of LIG1 when it encircles the DNA substrate (SI Appendix, Fig. S1). We hypothesize that disruption of the interactions between K845 and the backbone amides of L750 and V753 will lead to increased flexibility and shift the bound conformation toward a more open conformation that allows for dissociation from the AMP-DNA intermediate. Opening of the OB-fold domain (OBD) in the LIG1 DNA complex has been observed in molecular dynamics simulations (48). This concept is illustrated in Fig. 5 (Bottom) whereby the increased dissociation from the AMP-DNA intermediate for a mismatched substrate provides the basis for enhanced ligase fidelity. Under the conditions tested, K845N retains a sufficiently stabilized closed complex that dissociation from the matched substrate is minimal and only a modest reduction in catalytic efficiency is observed (Fig. 5, Top).", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Discussion", "text": "Schematic illustrating increased fidelity of ligation due to the K845N substitution. Top panel: Ligation of canonical base-paired substrate (black X:Y pair). (i) LIG1 WT produces mostly fully ligated DNA product (thick arrow and black depiction), with a small amount of aborted AMP-DNA (thin arrow and gray depiction). (ii) LIG1 K845N does not have a large impact on the proportions of ligated DNA and AMP-DNA. Bottom panel: Ligation of mismatched substrate (blue X:Y). (iii) LIG1 WT produces both fully ligated DNA product and aborted AMP-DNA (intermediate arrows). (iv) In contrast, LIG1 K845N produces less ligated product (thin arrow and gray depiction) and releases more aborted AMP-DNA (thick arrow and black depiction).", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Discussion", "text": "Functionally, this fidelity enhancement translated into a measurable benefit in cellular assays. HEK 293 cells expressing LIG1 K845N exhibited significantly greater viability than LIG1 WT expressing cells following treatment with the oxidative damaging drug menadione. Notably, this protective effect was most pronounced after an extended recovery period, suggesting that the benefits conferred by K845N may arise gradually, potentially through enhanced DNA repair processes that require time to manifest. Consistent with this, K845N-expressing cells also demonstrated a reduced MF. Supporting the relevance of this observation in a disease context, HD patient-derived LCLs heterozygous for the K845N variant showed a modest increase in viability following menadione-induced stress after recovery. Overall, our findings provide evidence that K845N can enhance genome maintenance presumably through increased ligation fidelity.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Discussion", "text": "Importantly, HttQ111 knock-in mice carrying the orthologous Lig1 missense variant (encoding LIG1 K843N) show a significant reduction in CAG repeat expansion in multiple tissues including the striatum and cortex that are particularly vulnerable in HD. The impact on expansion was modest; expansion was significantly suppressed in homozygous Lig1K843N/K843N mice and trended in the same direction in heterozygous Lig1K843N/+ mice. Thus, on the face of it, the degree of expansion suppression in the mice may seem insufficient to account for the strong disease-delaying effect in patients heterozygous for this variant. However, as clinical onset is driven by somatic expansion that occurs over decades of life (8), the cumulative impact of small modifying effects may be substantial. Indeed, a profound reduction of somatic expansion, such as that elicited by MMR gene knockouts (18), would be unexpected over a 10-mo period in mice. Our study has only examined the impact of the LIG1 variant on the somatic expansion of ~100 CAG repeats, and an interesting possibility is that its effect may depend on CAG length, for example having different relative impacts on the recently described slow and rapid phases of somatic expansion (8). Interestingly, though LIG1 variants modify clinical HD phenotypes, a recent GWAS of HD CAG repeat expansion in the blood did not identify significant signals at this locus. This suggests that, like several other loci that impact on either clinical phenotypes or blood CAG expansion but not the other, the LIG1 locus can involve cell-type specific effects (17). Analyses of single cell gene expression data in human and mouse indicate that DNA LIG1, as well as DNA ligases 3 and 4, seem to be expressed at higher levels in stable striatal cell types, i.e., not medium spiny neurons (SI Appendix, Fig. S18). The impact of the Lig1K843N variant on transcriptional dysregulation and on pathological markers (e.g., mutant HTT aggregation) in individual cell types will be a subject of future research. Overall, our data in mouse provide evidence that the mechanism underlying the 19AM3 disease-delaying effect may be contributed, at least in part, by an impact of the LIG1 K845N missense variant slowing the rate of somatic expansion in the brain.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Discussion", "text": "It is of interest that neither knockout of Lig1 nor the presence of the orthologous Lig1 R771W variant that reduces ligase activity 20-fold had any impact on the somatic expansion of CAG/CTG repeats in mouse models of HD and myotonic dystrophy type 1 respectively, though LIG1 levels modified CAG instability in an in vitro assay and the R771W variant promoted maternal germline contractions (18, 43, 49, 50). Thus, our data provide evidence for a role of LIG1 in the instability of CAG repeats in somatic tissues. These observations suggest that K845N may have unique biochemical properties that suppress somatic CAG expansion. While the precise mechanisms of somatic repeat expansion are not fully delineated, human and mouse genetic studies as well as biochemical data support a model in which expansions are driven by the recognition and processing of extruded CAG/CTG loops/structures by the MMR machinery (17, 18, 42, 51, 52). In this model (depicted in ref. 17) CAG/CTG loop-outs are substrates for MutS\u03b2 (MSH2-MSH3) binding followed by recruitment of MutL\u03b3 (MLH1-MLH3) endonuclease, DNA incision on the non-loop-containing strand, DNA excision, gap-filling, and ligation. Other factors, notably FAN1, counteract this expansion-prone mechanism by nicking DNA adjacent to the loop out. The presumed ligase concluding this process is LIG1 due to its involvement in MMR (53). We also have no direct evidence that Lig3 modifies expansion, while knockout of Lig4 promotes expansion in HttQ111 mice (17). It is currently unclear how the LIG1 K845N variant might suppress expansion, and whether it acts at the final ligation step of a MMR-directed mechanism, or alternatively, acts at an earlier step in the process. One hypothesis is that the elevated fidelity that this variant confers might mitigate expansion by rejecting certain substrates that support further expansion (e.g., mismatched bases resulting from oxidative damage, previously implicated in CAG expansion (54, 55). Repeat-containing structures or flaps may themselves mimic mismatched DNA, and an alternative hypothesis is that failure to ligate certain substrates might directly promote pathway(s) that suppress expansion. e.g., LIG1 K845N may limit religation of FAN1-induced nicks adjacent to a repeat slip-out. Further mechanistic and biochemical studies to understand how K845N suppresses expansion will be a subject of future study.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Discussion", "text": "In summary, our data demonstrate that the K845N variant of LIG1 confers enhanced substrate discrimination and increased repair fidelity, suppresses somatic CAG expansion and is associated with increased cell survival and decreased genome DNA MF independent of repeat expansion. As somatic expansion is a key disease driver, the suppression of somatic expansion by LIG1 K845N provides a plausible mechanistic underpinning for the HD-delaying 19AM3 modifier effect. Our findings also argue that the 19AM3 modifier effect may reflect, in part, increased cell survival due to genome-wide protection against DNA damage under conditions of genotoxic stress. These two potential (and nonmutually exclusive) mechanisms merit further investigation. Taken together, our results provide a mechanistic foundation for considering DNA ligase fidelity as a therapeutic target in HD and potentially in other trinucleotide repeat disorders.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Materials and Methods", "text": "Full-length and \u0394232 LIG1 WT and K845N proteins were expressed and purified using established protocols to generate catalytically active enzyme for biochemical assays (see the details in SI Appendix, Materials and Methods). Protein purity and integrity were assessed by SDS-PAGE followed by Coomassie Blue staining. In addition, protein quality was further verified by western blotting using three independent LIG1 antibodies recognizing distinct epitopes, confirming comparable integrity of WT and K845N proteins (SI Appendix, Fig. S2 A and B).\n\nThe DNA oligonucleotides were synthesized by Integrated DNA Technologies (IDT) and gel purified. Additional experimental details are provided in SI Appendix. The sequences of oligonucleotide substrates are listed in SI Appendix, Table S1.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Materials and Methods", "text": "To determine the active concentration of LIG1 in solution, active site titration assays were utilized (35). Briefly, full-length (12.5 nM to 400 nM) and \u0394232 (20 nM to 300 nM) LIG1 variants were combined with 100 nM annealed assay substrate in the absence of ATP to prevent multiple turnover. This reaction was conducted in standard reaction buffer containing 50 mM MOPS pH 7.5, 1 mM DTT, 0.5 mg/mL BSA, 10 mM MgCl2, with the ionic strength adjusted to 150 mM using NaCl. After incubation at 37 \u00b0C for 1 h, reactions were quenched with standard loading buffer (90% formamide, 50 mM EDTA, 0.01% bromophenol blue, 0.01% xylene cyanol) and were heated to 95 \u00b0C for 5 min. The sealed product was separated from the substrate by denaturing PAGE on a 15% polyacrylamide/8 M urea gel. Active enzyme concentration was determined by segmental linear regression analysis in GraphPad Prism software.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Materials and Methods", "text": "The ligation assays for full-length LIG1 were performed in 150 mM NaCl, 50 mM MOPS (pH 7.5), 1.0 mM DTT, 0.5 mM ATP, 10 mM MgCl2, 300 nM nicked DNA substrate. Purified LIG1 (final 5 to 40 nM) was added to ligation buffer and the mixture was incubated for 5 min at 37 \u00b0C. Ligation reactions were quenched by cooling to 4 \u00b0C followed by addition of loading buffer (90% formamide, 50 mM EDTA), heated at 95 \u00b0C for 5 min. The samples were separated on 15% denaturing urea acrylamide mini gels and the gel was scanned with iBright imaging system (Thermo Fisher Scientific). Fluorescence intensity was analyzed using image Lab software (Bio-Rad). The fraction of observed ligation product was determined by dividing band intensity of product by the total band intensity.\n\nSteady-state kinetics experiments were performed with \u2206232 LIG1 at 37 \u00b0C in standard reaction buffer with 1.2 mM MgCl2 (1 mM free Mg2+), 0.2 mM ATP, 50 mM MOPS pH 7.5, ionic strength adjusted to 150 mM using NaCl, and varying LIG1 and DNA concentrations. Each reaction time point was quenched in standard loading buffer and heated to 95 \u00b0C before gel loading. Sealed and abortive products were separated from substrate by denaturing PAGE on 15% polyacrylamide/8 M urea sequencing gels. Fluorescent bands were detected using an Amersham Typhoon 5 imager and quantified with ImageQuant TL software (Cytiva).", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Materials and Methods", "text": "Initial rates were determined by fitting the fraction product (<0.20) to straight lines in GraphPad Prism, calculating reaction velocity from the slope. The fraction of abortive ligation is defined by Eq. 1. The initial rates of sealed product formation were fit with the Michaelis\u2013Menten equation (Eq. 2) to determine kcat and KM values, as well as the catalytic efficiency (kcat/KM). The discrimination between 8-oxoG:A-containing substrate and undamaged (C:G) substrate is given by Eq. 3. In some cases, the time courses are expressed as the number of turnovers, which is calculated by dividing the concentration of ligated product by the concentration of enzyme.[1]Abortive\u00a0Product=[Intermediate]Intermediate+[Sealed\u00a0Product],[2]v0[Enz]=kcat[S]KM+[S],", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Materials and Methods", "text": "We used a panel of LCLs from seven previously collected and genotyped HD individuals participating in HD research. Of these, three were heterozygous for the 19AM3 modifier variant rs145821638 (GRCh38 - Chr19:48117686-C-A), all with HTT CAG repeat lengths 46/17, and four lacked the minor (A) allele, with HTT CAG repeat lengths 42/15, 41/17, 43/17, and 47/17. We also analyzed LCLs from two unaffected individuals lacking the 19AM3 minor allele (HTT CAG repeat lengths 17/15 and 20/19). LCLs were obtained from the MGB IRB-approved CHGR Neurodegenerative Repository. Genotyping of the normal and expanded CAG repeats was performed using a standardized PCR assay (56). LIG1 SNV rs145821638 (GRCh38 - Chr19:48117686-C-A) genotype was determined using a TaqMan SNP genotyping assay from Life Technologies. The reaction contained 1\u00d7 TaqMan Genotyping Master mix (Life Technologies, part # 4371355) 1\u00d7 of custom TaqMan SNP Genotyping assay mix [Assay mix comes at 40\u00d7 is diluted to 20\u00d7 in TE buffer pH8.0 (Life Technologies part #AM9849)] and 10 ng of DNA in total reaction volume of 12.5 \u03bcL. The reaction was run on the Roche LightCycler 480 II with the following conditions: 95 \u00b0C for 10 min, followed by 35 cycles of 92 \u00b0C for 15 s, 60 \u00b0C for 1 min. Data were collected at the end of each amplification cycle and known homozygous WT and heterozygous controls were run with every plate to assign alleles.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Materials and Methods", "text": "To generate HEK 293 cells stably overexpressing either full-length LIG1 WT, K845N or K568A, HEK 293 cells were transfected in 24-well plates with pcDNA3.1 vectors encoding N-terminal FLAG-tagged full-length LIG1 WT, FLAG-tagged full-length LIG1 K845N, FLAG-tagged full-length LIG1 K568A or an EV using Lipofectamine 3000 (Life Technologies). The K845N and K568A construct was generated by site-directed mutagenesis (New England Biolabs) of the pcDNA3.1-LIG1 WT plasmid. After transfection, cells were subjected to G418 (Geneticin, Gibco) selection to enrich transfected cells. Following antibiotic selection, heterogeneous pooled populations were used for all experiments without further single-clone isolation. Independently maintained cultures of these heterogeneous populations, which were passaged and handled separately, were treated as biological replicates.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Materials and Methods", "text": "For cell viability assays, cells were seeded into 96-well plates at a density of 2 \u00d7 104 cells/well in DMEM medium, and incubated overnight at 37 \u00b0C. The cells were treated with multiple dilutions of menadione (MP biomedicals) for 4 h, washed in fresh medium and incubated for indicated durations at 37 \u00b0C with 5% CO2. The number of viable cells was measured using the CellTiter-Glo (CTG) cell viability assay (Promega) according to the manufacturer\u2019s instructions. Three technical replicate measurements were performed for each biological replicate. The mean of the technical replicate measurements was used for statistical analyses across the biological replicate samples. The data were expressed as the percentage of the amount of ATP measured relative to untreated controls.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Materials and Methods", "text": "For duplex sequencing, HEK 293 cells were seeded into six-well plates at a density of 8 \u00d7 105 cells/well in DMEM medium, and incubated overnight at 37 \u00b0C. The cells were treated with 8 and 20 \u03bcM of menadione (MP biomedicals) for 4 h, washed in fresh medium and incubated for 48 h under normal conditions. After incubation, cells were harvested. A portion of the harvested cells was taken and cell viability was measured using CTG as above. The remaining cells were pelleted by centrifuge and stored at \u221280 \u00b0C until analysis. The cell pellets were sent to inotiv (https://www.inotiv.com) for genomic DNA extraction, library preparation, duplex sequencing, and MF analysis. Additional experimental details are provided in SI Appendix.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Materials and Methods", "text": "LCLs were grown in a humidified incubator in RPMI 1640 medium (Sigma-Aldrich) supplemented with 10% fetal bovine serum and 1% Penicillin-Streptomycin-Glutamine at 37 \u00b0C and 5% CO2. LCLs were seeded into 24-well plates at a density of 5 \u00d7 105 cells/well in RPMI 1640 medium and treated with multiple dilution of menadione (The highest concentration was 100 \u03bcM and serial dilution of 4/5 were made) for 4 h, after which the medium was replaced with fresh medium for indicated recovery periods. The number of viable cells was determined using CTG as above. Three technical replicate measurements were performed for each biological replicate (different LCLs). The mean of the technical replicate measurements was used for statistical analyses across the biological replicate samples. The viability was expressed as the percentage of the amount of ATP measured relative to untreated controls.\n\nThis study was carried out in accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the NIH under an approved protocol of the Massachusetts General Hospital Institutional Animal Care and Use Committees (IACUCs) (MGH: protocol 2009N000216). All animal procedures were carried out to minimize pain and discomfort, under the approved IACUC protocol. Animal husbandry was performed under controlled temperature, humidity, and light/dark cycles.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Materials and Methods", "text": "CRISPR/Cas9-mediated homology-directed repair was carried out essentially as reported (57) (Genome Modification Facility of Harvard University). Candidate single guide RNAs (sgRNAs) were chosen using the Broad Institute design tool (58, 59), four sgRNAs screened in embryos using Sanger sequencing to test guide efficiency, from which one was picked based on having the highest estimated editing efficiency. This gRNA (5\u2019>3\u2019 5\u2019 TCACCTCCCACACAATGCTT; GRCm39 chr7:13,043,112-13,043,131 [- strand]) was coinjected into C57BL/6 J zygotes with Cas9 protein (Alt-R\u00ae S.p. Cas9 Nuclease V3, IDT) and the following 150 nt single-stranded oligonucleotide donor (ssODN) template (Genewiz) complementary to the guide sequence.\n\n5\u2019 TCTGCTCACTGCAGGCCCTGGTATTGCCTACCCCACGCCCCTATGTGAGGATTGATGGGGCAGTGGCCCCAGACCATTGGCTGGATCCAAGCATTGTGTGGGAGGTGAATTGTGCTGATCTTTCCCT", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Materials and Methods", "text": "The ssODN harbored the coding G > T mutation (bold font) specifying asparagine at position 843 in the mouse LIG1 protein (in Lig1 exon 26, canonical transcript ENSMUST00000177588.10) and a silent G > T mutation (italic font) within the S.p. Cas9 PAM sequence (converting the GGG PAM on the minus strand to GGA) to prevent endonuclease cleavage of the recombined sequence. Injected zygotes were implanted into pseudopregnant females and pups screened at weaning by Sanger sequencing for the presence of the targeted allele. We established germline transmission from a pup that exhibited ~100% of targeted sequence and further verified the presence of the targeted mutations (K843N and silent PAM) and absence of other sequence variants using MiSeq (primers F: 5\u2019-CAGGCCCTGGTATTGCCTAC; R: 5\u2019- CCATCACCTCTGCCTTCCTT). The line was subsequently maintained in crosses with C57BL/6J and sperm frozen at the Jackson laboratories. We also serendipitously obtained a line harboring only the silent PAM mutation, which we maintained in the same way and used as an additional control line. Mice were genotyped using custom designed TaqMan SNP genotyping assays from Life Technologies, specific for either the K843N variant or the PAM sequence variant. Reactions contained 1\u00d7 TaqMan Genotyping Master mix (Life Technologies, part # 4371355) 1\u00d7 of custom TaqMan SNP Genotyping assay mix (Assay mix comes at 40\u00d7 is diluted to 20\u00d7 in TE buffer pH 8.0 (Life Technologies part #AM9849) and 10 ng of DNA in total reaction volume of 12.5 \u00b5L. The reaction was run on the Roche LightCycler 480 II with the following conditions: 95 \u00b0C for 10 min, followed by 35 cycles of 92 \u00b0C for 15 s, 60 \u00b0C for 1 min, and data collected at the end of each amplification cycle. Known WT, heterozygous, and homozygous mutant controls were run with every plate to assign alleles.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Materials and Methods", "text": "HttQ111 mice on the C57BL/6J genetic background (60) were crossed with Lig1K843N mice to generate HttQ111/+ mice with Lig1+/+, Lig1K843N/+, and Lig1K843N/K843N genotypes. Separate crosses were set up between HttQ111/+ mice and the \u201cPAM only\u201d mutant line to generate a small cohort of HttQ111/+ PAM/PAM control mice. Lig1 alleles were genotyped as above and the HttQ111 allele genotyped according to Kovalenko et al. (61). Genomic DNA from mice at 3, 6, or 10 mo of age was isolated from tissues using the DNeasy DNA blood and tissue kit (Qiagen) and somatic CAG instability analysis was performed using a human-specific PCR assay that amplifies the HTT CAG repeat from the knock-in allele (41). The forward primer was fluorescently labeled with 6-FAM (Thermo Fisher Scientific) and products were resolved using the ABI 3730xl DNA analyzer (Thermo Fisher Scientific) with GeneScan 500 LIZ as internal size standard (Thermo Fisher Scientific). GeneMapper v5 (Thermo Fisher Scientific) was used to generate CAG repeat size distribution traces. CAG expansion indices were calculated from GeneMapper peak height data as previously described, using a 5% relative peak height threshold cut-off (62). Expansion indices in all tissues were determined relative to the modal allele in the liver trace from the respective mouse (representing the modal repeat length in a population of liver cells that is very stable over time and distinguished as mice age from the unstable hepatocyte population (38).", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Materials and Methods", "text": "Total RNA was extracted from 25 to 30 mg of mouse liver using a hybrid Trizol-Qiagen method (modified from https://imageslab.fiu.edu/wp-content/uploads/2020/02/RNA-extraction_Trizol_RNeasy_hybrid.pdf). The tissue was homogenized in Trizol reagent followed by phase separation according to Trizol manufacturer\u2019s protocol. The top (aqueous) phase was mixed with two parts of RLT Plus buffer and applied to genomic DNA Eliminator spin columns (both buffer and columns were from the RNeasy Plus Mini kit, Qiagen, 74134). All the following steps were performed according to the RNeasy Plus Mini kit protocol. Quality control of resulting RNA was done with Agilent TapeStation 3000. The cDNA was synthesized with the Superscript III kit (Invitrogen, 18080-051) using oligo(dT) primer, according to the manufacturer\u2019s instructions. The relative expression level of Lig1 was quantified by PCR, using 2 \u00b5L of undiluted cDNA, 18 \u00b5L Taqman FastAdvanced Master Mix (Invitrogen, 4444556), and 1 \u00b5L of a ready-to-use Taqman assay from Invitrogen (Lig1: Mm00495331_m1 mapping to exons 21 to 22 and Mm01199942_m1 mapping to exons 3 to 4). PCR conditions were as follows: UNG incubation: 50 \u00b0C (2 min), initial denaturation and polymerase activation: 95 \u00b0C (20 s), 40 cycles of 95 \u00b0C (3 s), 60 \u00b0C (30 s), final extension 72 \u00b0C (10 min). Relative expression values (DCt) of Lig1 for each mouse were calculated using geometric mean of the Ct values of three housekeeping genes (Ppia, Mm02342430_g1; Actb, Mm00607939_s1; Gusb, Mm00446953_m1). Expression fold change (2\u2212DDCt) of Lig1 for each mouse was calculated relative to the geomean of relative expression values (DCt) of the control group (Lig1+/+) mice.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Materials and Methods", "text": "Statistical analyses were performed using GraphPad Prism v10. For comparisons of menadione-induced effects between LIG1 WT and LIG1 K845N-expressing cells, two-way ANOVA followed by Tukey\u2019s multiple comparison test were conducted. This analysis was applied to both cell viability assays and MF measurements obtained from duplex sequencing. Proliferation rates of HEK 293 EV, LIG1 WT, and LIG1 K845N were analyzed by linear regression, whereas proliferation rates of LCLs were assessed using a mixed effect model to account for interindividual variability. For quantification of LIG1, LIG3, and LIG4 expression, one-way ANOVA with Tukey\u2019s multiple comparison test was used for HEK 293 EV, LIG1 WT, and LIG1 K845N, while comparisons between the two LCL groups were evaluated using the unpaired t test with Welch\u2019s correction. Comparisons of somatic expansion and Lig1 expression levels in mouse tissues were carried out using one-way ANOVA with Tukey\u2019s multiple comparison test. Mendelian ratios of pups with different Lig1 genotypes were assessed with chi-squared tests. Significance was represented using a symbol-based threshold system: ****P < 0.0001, **P < 0.001, **P < 0.01, and *P < 0.05.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Protein Expression and Purification.", "text": "Full-length and \u0394232 LIG1 WT and K845N proteins were expressed and purified using established protocols to generate catalytically active enzyme for biochemical assays (see the details in SI Appendix, Materials and Methods). Protein purity and integrity were assessed by SDS-PAGE followed by Coomassie Blue staining. In addition, protein quality was further verified by western blotting using three independent LIG1 antibodies recognizing distinct epitopes, confirming comparable integrity of WT and K845N proteins (SI Appendix, Fig. S2 A and B).", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Preparation of DNA Substrates.", "text": "The DNA oligonucleotides were synthesized by Integrated DNA Technologies (IDT) and gel purified. Additional experimental details are provided in SI Appendix. The sequences of oligonucleotide substrates are listed in SI Appendix, Table S1.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Active Site Titration.", "text": "To determine the active concentration of LIG1 in solution, active site titration assays were utilized (35). Briefly, full-length (12.5 nM to 400 nM) and \u0394232 (20 nM to 300 nM) LIG1 variants were combined with 100 nM annealed assay substrate in the absence of ATP to prevent multiple turnover. This reaction was conducted in standard reaction buffer containing 50 mM MOPS pH 7.5, 1 mM DTT, 0.5 mg/mL BSA, 10 mM MgCl2, with the ionic strength adjusted to 150 mM using NaCl. After incubation at 37 \u00b0C for 1 h, reactions were quenched with standard loading buffer (90% formamide, 50 mM EDTA, 0.01% bromophenol blue, 0.01% xylene cyanol) and were heated to 95 \u00b0C for 5 min. The sealed product was separated from the substrate by denaturing PAGE on a 15% polyacrylamide/8 M urea gel. Active enzyme concentration was determined by segmental linear regression analysis in GraphPad Prism software.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Gel-Based Ligation Assays for Full-Length LIG1.", "text": "The ligation assays for full-length LIG1 were performed in 150 mM NaCl, 50 mM MOPS (pH 7.5), 1.0 mM DTT, 0.5 mM ATP, 10 mM MgCl2, 300 nM nicked DNA substrate. Purified LIG1 (final 5 to 40 nM) was added to ligation buffer and the mixture was incubated for 5 min at 37 \u00b0C. Ligation reactions were quenched by cooling to 4 \u00b0C followed by addition of loading buffer (90% formamide, 50 mM EDTA), heated at 95 \u00b0C for 5 min. The samples were separated on 15% denaturing urea acrylamide mini gels and the gel was scanned with iBright imaging system (Thermo Fisher Scientific). Fluorescence intensity was analyzed using image Lab software (Bio-Rad). The fraction of observed ligation product was determined by dividing band intensity of product by the total band intensity.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Steady-State Kinetics with \u2206232 LIG1.", "text": "Steady-state kinetics experiments were performed with \u2206232 LIG1 at 37 \u00b0C in standard reaction buffer with 1.2 mM MgCl2 (1 mM free Mg2+), 0.2 mM ATP, 50 mM MOPS pH 7.5, ionic strength adjusted to 150 mM using NaCl, and varying LIG1 and DNA concentrations. Each reaction time point was quenched in standard loading buffer and heated to 95 \u00b0C before gel loading. Sealed and abortive products were separated from substrate by denaturing PAGE on 15% polyacrylamide/8 M urea sequencing gels. Fluorescent bands were detected using an Amersham Typhoon 5 imager and quantified with ImageQuant TL software (Cytiva).\n\nInitial rates were determined by fitting the fraction product (<0.20) to straight lines in GraphPad Prism, calculating reaction velocity from the slope. The fraction of abortive ligation is defined by Eq. 1. The initial rates of sealed product formation were fit with the Michaelis\u2013Menten equation (Eq. 2) to determine kcat and KM values, as well as the catalytic efficiency (kcat/KM). The discrimination between 8-oxoG:A-containing substrate and undamaged (C:G) substrate is given by Eq. 3. In some cases, the time courses are expressed as the number of turnovers, which is calculated by dividing the concentration of ligated product by the concentration of enzyme.[1]Abortive\u00a0Product=[Intermediate]Intermediate+[Sealed\u00a0Product],[2]v0[Enz]=kcat[S]KM+[S],", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "HD Patient and Control LCLs.", "text": "We used a panel of LCLs from seven previously collected and genotyped HD individuals participating in HD research. Of these, three were heterozygous for the 19AM3 modifier variant rs145821638 (GRCh38 - Chr19:48117686-C-A), all with HTT CAG repeat lengths 46/17, and four lacked the minor (A) allele, with HTT CAG repeat lengths 42/15, 41/17, 43/17, and 47/17. We also analyzed LCLs from two unaffected individuals lacking the 19AM3 minor allele (HTT CAG repeat lengths 17/15 and 20/19). LCLs were obtained from the MGB IRB-approved CHGR Neurodegenerative Repository. Genotyping of the normal and expanded CAG repeats was performed using a standardized PCR assay (56). LIG1 SNV rs145821638 (GRCh38 - Chr19:48117686-C-A) genotype was determined using a TaqMan SNP genotyping assay from Life Technologies. The reaction contained 1\u00d7 TaqMan Genotyping Master mix (Life Technologies, part # 4371355) 1\u00d7 of custom TaqMan SNP Genotyping assay mix [Assay mix comes at 40\u00d7 is diluted to 20\u00d7 in TE buffer pH8.0 (Life Technologies part #AM9849)] and 10 ng of DNA in total reaction volume of 12.5 \u03bcL. The reaction was run on the Roche LightCycler 480 II with the following conditions: 95 \u00b0C for 10 min, followed by 35 cycles of 92 \u00b0C for 15 s, 60 \u00b0C for 1 min. Data were collected at the end of each amplification cycle and known homozygous WT and heterozygous controls were run with every plate to assign alleles.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Cell Viability Assay and Duplex Sequencing.", "text": "To generate HEK 293 cells stably overexpressing either full-length LIG1 WT, K845N or K568A, HEK 293 cells were transfected in 24-well plates with pcDNA3.1 vectors encoding N-terminal FLAG-tagged full-length LIG1 WT, FLAG-tagged full-length LIG1 K845N, FLAG-tagged full-length LIG1 K568A or an EV using Lipofectamine 3000 (Life Technologies). The K845N and K568A construct was generated by site-directed mutagenesis (New England Biolabs) of the pcDNA3.1-LIG1 WT plasmid. After transfection, cells were subjected to G418 (Geneticin, Gibco) selection to enrich transfected cells. Following antibiotic selection, heterogeneous pooled populations were used for all experiments without further single-clone isolation. Independently maintained cultures of these heterogeneous populations, which were passaged and handled separately, were treated as biological replicates.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Cell Viability Assay and Duplex Sequencing.", "text": "For cell viability assays, cells were seeded into 96-well plates at a density of 2 \u00d7 104 cells/well in DMEM medium, and incubated overnight at 37 \u00b0C. The cells were treated with multiple dilutions of menadione (MP biomedicals) for 4 h, washed in fresh medium and incubated for indicated durations at 37 \u00b0C with 5% CO2. The number of viable cells was measured using the CellTiter-Glo (CTG) cell viability assay (Promega) according to the manufacturer\u2019s instructions. Three technical replicate measurements were performed for each biological replicate. The mean of the technical replicate measurements was used for statistical analyses across the biological replicate samples. The data were expressed as the percentage of the amount of ATP measured relative to untreated controls.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Cell Viability Assay and Duplex Sequencing.", "text": "For duplex sequencing, HEK 293 cells were seeded into six-well plates at a density of 8 \u00d7 105 cells/well in DMEM medium, and incubated overnight at 37 \u00b0C. The cells were treated with 8 and 20 \u03bcM of menadione (MP biomedicals) for 4 h, washed in fresh medium and incubated for 48 h under normal conditions. After incubation, cells were harvested. A portion of the harvested cells was taken and cell viability was measured using CTG as above. The remaining cells were pelleted by centrifuge and stored at \u221280 \u00b0C until analysis. The cell pellets were sent to inotiv (https://www.inotiv.com) for genomic DNA extraction, library preparation, duplex sequencing, and MF analysis. Additional experimental details are provided in SI Appendix.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Cell Viability Assay and Duplex Sequencing.", "text": "LCLs were grown in a humidified incubator in RPMI 1640 medium (Sigma-Aldrich) supplemented with 10% fetal bovine serum and 1% Penicillin-Streptomycin-Glutamine at 37 \u00b0C and 5% CO2. LCLs were seeded into 24-well plates at a density of 5 \u00d7 105 cells/well in RPMI 1640 medium and treated with multiple dilution of menadione (The highest concentration was 100 \u03bcM and serial dilution of 4/5 were made) for 4 h, after which the medium was replaced with fresh medium for indicated recovery periods. The number of viable cells was determined using CTG as above. Three technical replicate measurements were performed for each biological replicate (different LCLs). The mean of the technical replicate measurements was used for statistical analyses across the biological replicate samples. The viability was expressed as the percentage of the amount of ATP measured relative to untreated controls.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Generation of ", "text": "This study was carried out in accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the NIH under an approved protocol of the Massachusetts General Hospital Institutional Animal Care and Use Committees (IACUCs) (MGH: protocol 2009N000216). All animal procedures were carried out to minimize pain and discomfort, under the approved IACUC protocol. Animal husbandry was performed under controlled temperature, humidity, and light/dark cycles.\n\nCRISPR/Cas9-mediated homology-directed repair was carried out essentially as reported (57) (Genome Modification Facility of Harvard University). Candidate single guide RNAs (sgRNAs) were chosen using the Broad Institute design tool (58, 59), four sgRNAs screened in embryos using Sanger sequencing to test guide efficiency, from which one was picked based on having the highest estimated editing efficiency. This gRNA (5\u2019>3\u2019 5\u2019 TCACCTCCCACACAATGCTT; GRCm39 chr7:13,043,112-13,043,131 [- strand]) was coinjected into C57BL/6 J zygotes with Cas9 protein (Alt-R\u00ae S.p. Cas9 Nuclease V3, IDT) and the following 150 nt single-stranded oligonucleotide donor (ssODN) template (Genewiz) complementary to the guide sequence.\n\n5\u2019 TCTGCTCACTGCAGGCCCTGGTATTGCCTACCCCACGCCCCTATGTGAGGATTGATGGGGCAGTGGCCCCAGACCATTGGCTGGATCCAAGCATTGTGTGGGAGGTGAATTGTGCTGATCTTTCCCT", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Generation of ", "text": "The ssODN harbored the coding G > T mutation (bold font) specifying asparagine at position 843 in the mouse LIG1 protein (in Lig1 exon 26, canonical transcript ENSMUST00000177588.10) and a silent G > T mutation (italic font) within the S.p. Cas9 PAM sequence (converting the GGG PAM on the minus strand to GGA) to prevent endonuclease cleavage of the recombined sequence. Injected zygotes were implanted into pseudopregnant females and pups screened at weaning by Sanger sequencing for the presence of the targeted allele. We established germline transmission from a pup that exhibited ~100% of targeted sequence and further verified the presence of the targeted mutations (K843N and silent PAM) and absence of other sequence variants using MiSeq (primers F: 5\u2019-CAGGCCCTGGTATTGCCTAC; R: 5\u2019- CCATCACCTCTGCCTTCCTT). The line was subsequently maintained in crosses with C57BL/6J and sperm frozen at the Jackson laboratories. We also serendipitously obtained a line harboring only the silent PAM mutation, which we maintained in the same way and used as an additional control line. Mice were genotyped using custom designed TaqMan SNP genotyping assays from Life Technologies, specific for either the K843N variant or the PAM sequence variant. Reactions contained 1\u00d7 TaqMan Genotyping Master mix (Life Technologies, part # 4371355) 1\u00d7 of custom TaqMan SNP Genotyping assay mix (Assay mix comes at 40\u00d7 is diluted to 20\u00d7 in TE buffer pH 8.0 (Life Technologies part #AM9849) and 10 ng of DNA in total reaction volume of 12.5 \u00b5L. The reaction was run on the Roche LightCycler 480 II with the following conditions: 95 \u00b0C for 10 min, followed by 35 cycles of 92 \u00b0C for 15 s, 60 \u00b0C for 1 min, and data collected at the end of each amplification cycle. Known WT, heterozygous, and homozygous mutant controls were run with every plate to assign alleles.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Mouse Crosses with ", "text": "HttQ111 mice on the C57BL/6J genetic background (60) were crossed with Lig1K843N mice to generate HttQ111/+ mice with Lig1+/+, Lig1K843N/+, and Lig1K843N/K843N genotypes. Separate crosses were set up between HttQ111/+ mice and the \u201cPAM only\u201d mutant line to generate a small cohort of HttQ111/+ PAM/PAM control mice. Lig1 alleles were genotyped as above and the HttQ111 allele genotyped according to Kovalenko et al. (61). Genomic DNA from mice at 3, 6, or 10 mo of age was isolated from tissues using the DNeasy DNA blood and tissue kit (Qiagen) and somatic CAG instability analysis was performed using a human-specific PCR assay that amplifies the HTT CAG repeat from the knock-in allele (41). The forward primer was fluorescently labeled with 6-FAM (Thermo Fisher Scientific) and products were resolved using the ABI 3730xl DNA analyzer (Thermo Fisher Scientific) with GeneScan 500 LIZ as internal size standard (Thermo Fisher Scientific). GeneMapper v5 (Thermo Fisher Scientific) was used to generate CAG repeat size distribution traces. CAG expansion indices were calculated from GeneMapper peak height data as previously described, using a 5% relative peak height threshold cut-off (62). Expansion indices in all tissues were determined relative to the modal allele in the liver trace from the respective mouse (representing the modal repeat length in a population of liver cells that is very stable over time and distinguished as mice age from the unstable hepatocyte population (38).", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Assaying ", "text": "Total RNA was extracted from 25 to 30 mg of mouse liver using a hybrid Trizol-Qiagen method (modified from https://imageslab.fiu.edu/wp-content/uploads/2020/02/RNA-extraction_Trizol_RNeasy_hybrid.pdf). The tissue was homogenized in Trizol reagent followed by phase separation according to Trizol manufacturer\u2019s protocol. The top (aqueous) phase was mixed with two parts of RLT Plus buffer and applied to genomic DNA Eliminator spin columns (both buffer and columns were from the RNeasy Plus Mini kit, Qiagen, 74134). All the following steps were performed according to the RNeasy Plus Mini kit protocol. Quality control of resulting RNA was done with Agilent TapeStation 3000. The cDNA was synthesized with the Superscript III kit (Invitrogen, 18080-051) using oligo(dT) primer, according to the manufacturer\u2019s instructions. The relative expression level of Lig1 was quantified by PCR, using 2 \u00b5L of undiluted cDNA, 18 \u00b5L Taqman FastAdvanced Master Mix (Invitrogen, 4444556), and 1 \u00b5L of a ready-to-use Taqman assay from Invitrogen (Lig1: Mm00495331_m1 mapping to exons 21 to 22 and Mm01199942_m1 mapping to exons 3 to 4). PCR conditions were as follows: UNG incubation: 50 \u00b0C (2 min), initial denaturation and polymerase activation: 95 \u00b0C (20 s), 40 cycles of 95 \u00b0C (3 s), 60 \u00b0C (30 s), final extension 72 \u00b0C (10 min). Relative expression values (DCt) of Lig1 for each mouse were calculated using geometric mean of the Ct values of three housekeeping genes (Ppia, Mm02342430_g1; Actb, Mm00607939_s1; Gusb, Mm00446953_m1). Expression fold change (2\u2212DDCt) of Lig1 for each mouse was calculated relative to the geomean of relative expression values (DCt) of the control group (Lig1+/+) mice.", "title": "" }, { "pmid": 41770933, "pmcid": "PMC12974472", "section": "Statistical Analyses.", "text": "Statistical analyses were performed using GraphPad Prism v10. For comparisons of menadione-induced effects between LIG1 WT and LIG1 K845N-expressing cells, two-way ANOVA followed by Tukey\u2019s multiple comparison test were conducted. This analysis was applied to both cell viability assays and MF measurements obtained from duplex sequencing. Proliferation rates of HEK 293 EV, LIG1 WT, and LIG1 K845N were analyzed by linear regression, whereas proliferation rates of LCLs were assessed using a mixed effect model to account for interindividual variability. For quantification of LIG1, LIG3, and LIG4 expression, one-way ANOVA with Tukey\u2019s multiple comparison test was used for HEK 293 EV, LIG1 WT, and LIG1 K845N, while comparisons between the two LCL groups were evaluated using the unpaired t test with Welch\u2019s correction. Comparisons of somatic expansion and Lig1 expression levels in mouse tissues were carried out using one-way ANOVA with Tukey\u2019s multiple comparison test. Mendelian ratios of pups with different Lig1 genotypes were assessed with chi-squared tests. Significance was represented using a symbol-based threshold system: ****P < 0.0001, **P < 0.001, **P < 0.01, and *P < 0.05.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Introduction", "text": "Drug repurposing has the potential to find molecules that have previously been overlooked [1], to the benefit of rare diseases (RDs). RDs are seldom the target of drug research due to low prevalence among individuals [2]. Research and development of therapies against RDs can be a more financially viable option thanks to the introduction of drug repurposing \u2013especially when considering the use of Machine Learning (ML) techniques, which allow for high computational power and lower R&D costs. Such approaches, in fact, can narrow down the number of candidates ahead of clinical trials. Yet, this approach comes with its own challenges: data from external sources (e.g. drug databases, biological interaction databases) must be organised into structures that allow the model to be trained and predictions to be made; and most databases for life science fields of study focus only on positive associations (e.g. known interactions between a gene and a drug). In fact, the lack of information on negative associations hinders the accuracy of ML predictions. Throughout the paper we will explain how we applied ML techniques to a case study disease (i.e. Huntington\u2019s disease) to solve the above-mentioned challenges.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Introduction", "text": "Modern technologies involving computational approaches (e.g. ML, Knowledge Graph (KG)) are increasingly being applied to biological studies due to how available data have grown exponentially in size: more data requires more analytical capabilities. Continuing improvements in omics techniques have resulted in the production of larger datasets in shorter time spans, thus requiring more advanced computational infrastructure. For instance, powerful and complex analysis tools developments in Deep Learning (DL) have been recently applied to RNA thanks to the availability of huge amounts of data via high-throughput experiments [3].\n\nMoreover, advancements in DL led in recent years to more insights into RDs, which are diseases with a high impact on individuals while having limited investments from pharmaceutical companies [1, 4].\n\nCharacterised by low prevalence, RDs result from specific genetic mutations (i.e. Mendelian diseases) [5] or harmful environmental effects [6]. Despite being individually categorised as \u2018rare\u2019, they all together affect a significant portion of the population: about 10% in Europe and America in 2020 [7]. More recent global estimates suggest that RDs collectively affect 3\u20136% of the global population as of 2024 [2], thus representing as a whole approximately 250\u2013450\u00a0M people worldwide. Yet most of them lack approved treatments, with fewer than 50% of patients reaching a diagnosis with current testing methods [2].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Introduction", "text": "Traditional drug discovery takes several years (i.e. 10\u201315) and billions of dollars in investment (i.e. $2.6B on average) [8] so that one active molecule can possibly reach the market and be commercially profitable within the expiration of patents. Drugs have to be proven to be safe and with an efficiency-to-side-effects ratio that is an improvement compared to existing treatments. Moreover, only 10% of candidate drugs reach the market [8]. These financial restrictions and the low drug-to-market turnover lead to RDs being seldom targeted for drug development due to the low number of patients, and therefore the low income that would derive: in 2020 the Food and Drug Administration (FDA) approved just 31 Orphan Drugs out of a total of 53 drugs approved that year [9] (58%). Recently this trend has shown no major improvement or decline, with the FDA approving 26 new molecules out of 50 in 2024 (52%) [2].\n\nDrug repurposing on the other hand is more cost-efficient. Bringing a repurposed drug to the market is estimated to cost only $300\u00a0M [1]. Repurposed drugs also require a shorter development time, since at least pre-clinical and safety steps have been already performed, and the risk of failure is much lower [1]. Thus, investigating ML techniques that apply to drug repurposing is promising.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Introduction", "text": "The landscape of computational drug repurposing has expanded significantly in recent years, with 181 phenotypic drug discovery and repurposing resources now available [10]. These resources encompass diverse computational approaches including structure-based methods, network-based approaches, and ML techniques applied to various data types (e.g. omics data, chemical structures) [11]. Recent comprehensive reviews highlight the importance of selecting appropriate computational resources based on a specific project\u2019s requirements, evaluation criteria, and disease context [10, 12].\n\nHuntington\u2019s disease (HD) is one of the most common case studies in research for new drugs against RDs due to several genetic signature associated to this disease [13] resulting from its fairly more common occurrence. Associations between genes and drugs have been thoroughly mapped since 1993, when HD\u2019s cause was identified [13]. Mutations in the \u2018huntingtin\u2019 genes leading to increased number of CAG repeats are involved in selective degeneration of GABAergic basal ganglia [14] leading to neurodegenerative and hyperkinetic movement disorder characterised by chorea (i.e. involuntary movements) and other symptoms [15]. Symptoms of neurodegenerative diseases usually are rather easy to identify and track over time, making it an ideal case study [15].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Introduction", "text": "As further discussed in detail in Section\u00a0\u201cBackground\u201d, recent drug repurposing research approaches mostly involved traditional genetics and modern genomics [16], yet the advent of ML and Big Data introduced far more powerful approaches. Ambesi-Impiombato et al. recently developed a systematic -yet not automated- ML approach to determine newly repurposed drugs against HD, according to the Gene Set Enrichment Analysis model by Subramanian et al. [17]. Oldham et al. investigated the correlation between several samples in microarray experiments involving HD [18], and Mohan et al. recently developed a progression model for HD [19].\n\nSuch computational techniques therefore lead to novel approaches against HD, yet the reproducible and scalable nature of pipelines and software does not restrict them to a specific disease, nor it bounds them to being outdated with time. In fact, they allow for the approaches to be used against other diseases or be updated over time with few changes [20].\n\nThe FAIR Principles state that research assets should be Findable, Accessible, Interoperable and Reusable [21]. Since their introduction, these principles have been widely adopted in biomedical and natural sciences [22], with organizations like NIH requiring FAIR-aligned data management in funded research projects [23]. Implementation of such principles has been associated with improved research efficiency and reduced costs [24] when introducing novel solutions to challenges [25].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Introduction", "text": "Principles for data management and software development (e.g. FAIR4RS) [21, 26] promote best practices in computational research. For software, this involves maintaining comprehensive documentation of datasets and external sources, annotating code with clear explanations, reporting all random seeds used, and making source code publicly available [27]. Version control on code repositories (e.g. Github) from the start of development allows for full understanding of all coding choices by the author whenever a third party wants to review or improve the work [26]. The desired result is a computational tool that can be independently validated, composed of reusable modules, and scaled according to specific requirements.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Introduction", "text": "This paper makes the following contributions to the field of computational drug repurposing for RDs:Methodological novelty Development of a semantically-valid negative sampling strategy combining network centrality and molecular similarity metrics to address the lack of negative associations in databases for biological associations;Empirical validation Systematic evaluation using multiple metrics (i.e. AUROC, AUPRC, Precision@K, Recall@K) across different filtering strategies and iteration-based consensus predictions, demonstrating the predictive performance of our pipeline;Practical scalability Implementation of an automated, reproducible pipeline leveraging open APIs (i.e. Monarch Initiative, DGIdb) and established ML frameworks (i.e. Node2Vec, XGBoost, GCN) that can be applied to any disease\u2014specifically RDs.\n\nMethodological novelty Development of a semantically-valid negative sampling strategy combining network centrality and molecular similarity metrics to address the lack of negative associations in databases for biological associations;\n\nEmpirical validation Systematic evaluation using multiple metrics (i.e. AUROC, AUPRC, Precision@K, Recall@K) across different filtering strategies and iteration-based consensus predictions, demonstrating the predictive performance of our pipeline;", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Introduction", "text": "Practical scalability Implementation of an automated, reproducible pipeline leveraging open APIs (i.e. Monarch Initiative, DGIdb) and established ML frameworks (i.e. Node2Vec, XGBoost, GCN) that can be applied to any disease\u2014specifically RDs.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Background", "text": "Several approaches have been applied to identify molecules for drug discovery and drug repurposing, ranging from the use of only human-curated datasets [28, 29] to the semi-automated fetching of data from online databases [30, 31]. Bio-semantics\u2014as the field of studies in Semantics for biological associations between semantic entities\u2013 has advanced our understanding in representing complex biological relationships. Within this field, semantic web technologies (e.g. Resource Description Framework (RDF) and Web Ontology Language (OWL)) have proven to be the best data structuring formats to organise biological systems in networks of associations. This is particularly useful for ML training, since several methods require structured data [32]. KGs specifically provide information-rich networks [20, 33] that allow for powerful ML predictions.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Background", "text": "To systematically position our approach within the computational drug repurposing landscape, we examined several methodologies catalogued by Tanoli et al. in their survey of 181 in silico resources [10]. The computational approaches specifically for drug repurposing encompass several different strategies\u2014each of them with their own advantages and limitations. Structure-based approaches leverage molecular docking and protein analysis (e.g. AutoDock Vina for binding affinity-based docking predictions [34], AlphaFold for homology modelling [35, 36], and Chimera for binding site visualisation [37]). Despite being relatively fast, such structure-based strategies require having identified molecular targets with accurate 3D structures, and they often lack the accuracy of standalone approaches [10]. Knowledge-based methods instead aggregate curated associations between drugs and diseases through databases and ontologies (e.g. DrugBank for drug\u2013target interactions [38], ChEMBL and PubChem for biochemical bioactivity data [39, 40], Reactome for pathway information [41]), thus providing comprehensive evidence-based scoring\u2014but requiring extensive manual curation [10]. Network-based approaches exploit specifically protein-protein interactions and biological pathways to generate network-derived hypotheses (e.g. STRING database for protein interactions [42], MSigDB for gene signatures [43], KEGG and IntAct for pathway data [44, 45]). Even though network-based methods facilitate identification of disease mechanisms, they require careful selection of relevant biological contexts [10]. Finally, ML platforms are designed to integrate diverse data modalities, including gradient boosting methods (e.g. XGBoost [46]), Graph Neural Network (GNN) (e.g. Graph Convolutional Networks (GCN) via PyTorch-Geometric (PyG) [47, 48]), and DL frameworks. The main advantage of ML approaches is that they offer time-efficient predictions\u2014yet they also require high-quality large-scale training data and may struggle to generalise across understudied disease domains [10]. Recent data-driven strategies have demonstrated the value of harmonising drug-target interaction data from multiple curated resources to enhance reproducibility and enable systematic pathway-based predictions [12], while supervised ML methods that leverage comprehensive target activity profiles have shown particular promise in cancer drug repurposing [11].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Background", "text": "Prior of the introduction of computational means to scientific research, drug repurposing was a slow and not systematic approach based on serendipitous observations [32]: it was heavily dependent on clinical observations and luck, thus also being expensive. Examples of repurposed drugs range from toxic compounds (e.g. Arsenic Trioxide as a cancer treatment, [49]), to drugs that were first abandoned for adverse effects (e.g. Thalidomide also as a cancer treatment, [49]) or that were initially developed to treat other conditions (e.g. Sildenafil as an erectile dysfunction treatment, [32]; or recently as a candidate treatment against Alzheimer\u2019s disease [50]). Recently Xia et al. developed a non-computational approach based on testing libraries of known drugs against tumoroids (i.e. tissues derived from stem cells meant to mimic tumours) [49]. Such an approach is extremely promising, although it lacks the convenience and inexpensiveness of in-silico methods for early stage identification of candidate repurposed drugs.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Background", "text": "In contrast to clinical approaches, computational approaches are flexible and scalable solutions for early state identification of repurposable drugs. Especially bio-semantics-based computational approaches that focus on data structures that can easily handle networks of associations between biological entities (e.g. disease and gene nodes) [20, 33]. As previously mentioned, KGs have proven to be a powerful method to organise complex biomedical data, due to being graphs where both nodes (a.k.a vertices) and edges (a.k.a. links) are labelled. Labels are storing information on biological entities in the nodes and on the nature of their interaction in the edges [51]. They have successfully been implemented in non-automated drug repurposing experiments via human-curated datasets against Covid-19 [51] and in novel drug R&D [20].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Background", "text": "KGs provide an effective framework for managing research data [52] because they integrate ontological structures (e.g. OMIM, OWL) that define semantic relationships and hierarchies between entities, thus enabling meaningful knowledge representation throughout the network [20]. \u2018bioknowledge-reviewer\u2019 is one of such tools for structured and partially automated knowledge-drive research [30]. It builds a complex network of biological associations to NGLY1 deficiency (i.e. developmental delay) used as case study from several databases (e.g. Monarch Initiative, DGIdb), curated.csv files for key nodes, transcriptomic RNA-seq datasets in Drosophila and dataframes on Transcription Factors\u2019 (TFs) regulation on genes from the Molecular Signature Database [30]. The resulting network is then preserved on a neo4j server, and data retrieval is performed by Cypher\u2014an SQL-based querying language optimised for KGs. Each query returns a hypothesis that can then be tested experimentally. Such an approach allows for knowledge-driven research that narrows down candidates and reduces R&D costs compared to traditional approaches.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Background", "text": "Monarch Initiative\u2019s database is a KG of biological associations focusing mostly on diseases, genes and phenotypes\u2014codified as an RDF via proprietary identifiers (i.e. Uniform Resource Identifies, URIs; e.g. MONDO) [53]. It is key in building computational tools for knowledge-driven association networks thanks to public APIs for automated retrieval of associations to any given node [28].\n\nOther computational approaches instead focus on pre-computed datasets. The \u2018DrugRepo\u2019 platform provides information on drug similarity, drug-target interactions and disease-gene associations for more than 650 disease [29]. Despite the final networks not being structured as KGs, it uses KGs and RDFs to compute them [29].\n\nThe advantage of the increasing computational power available to researchers is that queries can be computed in real time in relatively short time frames [28]. Reep extended \u2018bioknowledge-reviewer\u2019s semi-automated approach [30] to applications in drug repurposing: after fetching all the genes associated to one of the symptoms of the disease of interest from Monarch Initiative, their tool retrieves all the drugs associated to the network from DGIdb\u2014which focuses on gene-drug associations [31]. Compared to Monarch Initiative, DGIdb is a relational SQL database that uses Entrez, HGNC and GO identifiers for genes, and mostly ChEMBL and Wikidata identifiers for drugs [38].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Background", "text": "The resulting Reep network of genes and drugs is used to train a ML model and predict newly repurposed drugs to the disease of interest [31].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Background", "text": "Negative samples are an extremely relevant field of studies in Semantics and Network Biology [54]. Specifically, valid negative samples are all the missing links in a network that could be positive given certain conditions. For example if \"Macron - president of - France\" is a positive samples, \"Obama - president of - France\" would be a valid negative sample\u2014while \"Bread - president of - France\" would be invalid, as it is semantically nonsensical. Such an approach is also defined as Contrastive Learning, and Liu et al. made a custom drug repurposing pipeline against Alzheimer\u2019s disease and breast neoplasm by generating negative samples accordingly [55]. Hubert et al. focused on discriminating invalid negative samples from valid ones to weight them differently in a prediction [56]. Out of several approaches they took into consideration, semantic loss of function\u2014as the degradation or failure of a system to maintain semantic validity,1\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\begin{aligned} L_{semantic}(h,r,t) = -\\log P(semantically\\_valid(h,r,t) | \\theta ) \\end{aligned}$$\\end{document}where \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\theta $$\\end{document} represents the model parameters and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$semantically\\_valid$$\\end{document} evaluates domain/range constraint satisfaction, which quantifies the degree to which a model violates semantic coherence during training\u2014was the one that returned the best enrichment score according to the Sem@K metric (i.e. Semantic Similarity at K,2\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\begin{aligned} Sem@K = \\frac{1}{|Q|} \\sum _{q \\in Q} \\frac{|\\{r \\in top\\text {-}K(q) : sim(r, ground\\_truth) > \\tau \\}|}{K} \\end{aligned}$$\\end{document}where Q is the query set, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$top\\text {-}K(q)$$\\end{document} represents the top-K results for query q, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$sim(\\cdot ,\\cdot )$$\\end{document} computes semantic similarity, and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\tau $$\\end{document} is the similarity threshold) [56].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Background", "text": "The generation of artificial negative samples that are also semantically valid is especially relevant in biological studies and bio-semantics [57]: very few databases (e.g. EBI\u2019s IntAct, Negatome Database) store information on known lack of associations between biological entities\u2014and, even when they do, negative data might not be fully integrated despite its importance [58].\n\nThe advantage of real-time pipelines for specific queries compared to pre-computed datasets are the specificity of the results and the currency of source data. The disadvantages instead are that real-time pipelines take time, yet with current computational approaches and hardware capabilities the time frames to run comparative pipelines reduce\u2014also allowing to run more complex pipelines in comparative time frames.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Background", "text": "The key drivers in developing automated\u2014or semi-automated\u2013 pipelines are the ML models and the infrastructures that run the pipelines themselves. In fact, ML models will be provided with training data in specific formats. In the context of KGs and association networks, information for each node (i.e. biological entity) and the nodes associated to it via edges are converted into node embeddings\u2014dense low-dimensional vector representations of structural and semantic relationships in the biological association network\u2014for ML training [59]. Node embedding is performed by dedicated algorithms\u2014such as Node2Vec, a flexible graph embedding method that captures both local neighbourhood structures and broader network topology [60]. Then, ML models return predictions in vectors for new associations between nodes. These steps can take long time since each biological entity in the network has to be taken into consideration, thus parallelisation techniques are advantageous to reduce training and prediction time frames [61].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Background", "text": "To convert the heterogeneous biological association network into a format suitable for ML, we employed \u2018node2vec\u2019 [62], a python implementation of the Node2Vec algorithmic framework that learns continuous feature representations for nodes in networks. The Node2Vec algorithm proceeds in two stages: first, it generates a corpus of node sequences through biased random walks on the graph; second, it applies a learning approach that maximises the likelihood of observing a node\u2019s network neighbourhood [60]. This method is particularly well-suited for biological networks where labelled training data is scarce and the network structure itself contains rich semantic information about gene-drug relationships.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Background", "text": "Node2Vec excels over alternative KG embedding methods in real-time automated pipelines because it is optimised for general network structures without extensive ontological specifications\u2014unlike others that are designed primarly for typed, relational KG with explicit semantic relationships (e.g. TransE [63], DistMult [64], ComplEx [65]). Such extensive ontological specifications would in fact be hard to achieve in a semi-automated tool that is populated from databased APIs, where data schemas evolve independently across source databases. For this reason, we did not include these methods in our baseline comparisons. RDF2Vec (via the \u2018PyRDF2Vec\u2019 package) [66] would also fit such a scenario, but it has not been supported since v0.2.3 in June 2021. Node2Vec also offers higher computational efficiency compared to other methods \u2013resulting in shorter run times\u2013 and it has demonstrated effectiveness specifically for biological network analysis and gene-drug interaction predictions [67, 68].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Background", "text": "Drug repurposing approaches involving graph-based ML methods implement similar tools and techniques found in other applications of the same nature in other fields of applications\u2014tools such as gradient boosting ML models. Biological associations can be organised in a bio-semantics network, and fed as training data for ML models. XGBoost is a gradient boosting model that has been used in biological studies both with tabular data to reduce the number of medical tests required to make a diagnosis [46, 59], and with graph-based data to make a drug repurposing prediction [31]. PyG instead is a library to build GNN models [47]: Wallett et al. made predictions from a bipartite network of non-biological associations using this package [69], while Wang et al. used to generated the pre-computed data for the \u2018DrugRepo\u2019 platform [29]. Such results are promising and could be tested with a network of gene-drug associations. Both XGBoost and PyG use unsupervised methods, and they are available as python packages.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Background", "text": "Large datasets require processing and only a portion can be handled with current hardware and techniques. Nowadays there is a major discrepancy between scientists\u2019 ability to collect data, and the availability of means to process them: individual datasets coming from Next-Generation Sequencing experiments addressing Single Nucleotide Polimorphisms are now extremely easy to obtain in any lab within 24\u201348\u00a0h thanks to commercial products such as Oxford Nanopore\u2019s MinION [70]\u2014which sells for under \u20ac3000.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Background", "text": "The average cost to then compute that same dataset within the same time-frame is instead much higher and personal computers are increasingly less capable of handling extensive data analysis on such datasets [71]. In fact, Courneya et al. pointed out that the lack of adequate computational resources would undermine the potential discoveries coming from modern Omics and inhibit further advancement of scientific research [61]. Their solution was to develop a remote service for their university\u2019s departments where scientists could perform traditional bioinformatics data processing (e.g. sequence alignment via BWA or STAR respectively for DNA or RNA sequences) with all the computational advantages of server racks. Bioinformatics tools available via web-apps can perform even more specific operations: Petrini et al. developed a genomics variant calling tool that takes advantage of parallelisation, a technique that allows to perform concurrent operations on different computational units (e.g. kernel instances, CPU and GPU cores) at the same time [72]. The result is a drastic optimisation of time-frames.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Background", "text": "Huntington\u2019s disease (HD) serves as an exemplary model for RD drug repurposing research due to being well-characterised. HD is a mendelian disease with a monogenic cause, a well-established gene-drug association network, and traceable clinical phenotypes. The disease originates from a single genetic mutation in the huntingtin gene (HTT) located on chromosome 4, specifically involving an abnormal expansion of CAG trinucleotide repeats [15]. In healthy individuals, the HTT gene typically contains 10\u201335 CAG repeats. However, individuals with HD can carry up to 180 repeats in most extreme case\u2014as the repeat length directly correlates to disease severity and age of symptom onset: 60 repeats already lead to early onset [14]. This quantitative genotype-phenotype relationship makes HD particularly suitable for computational drug repurposing, as the genetic mechanism is precisely defined and well-documented across diverse biological databases and clinical cohorts.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Background", "text": "The molecular pathogenesis of HD involves the mutant HTT protein triggering selective neurodegeneration, particularly affecting the GABAergic neurons of the basal ganglia striatum [14]. The loss of these inhibitory GABAergic neurons leads to dysregulation of motor circuits, manifesting as characteristic hyperkinetic movement disorders including chorea (i.e. involuntary, jerky movements), rigidity, and progressive cognitive decline [15]. The selectivity of GABAergic neuronal vulnerability has been extensively mapped at the molecular level [13], providing a rich foundation to target genetic mechanisms for therapy.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Background", "text": "From a clinical perspective, HD presents several advantages for drug discovery research. Disease symptoms are objectively measurable and relatively easy to track over time compared to many other neurodegenerative diseases, enabling robust assessment of therapeutic interventions [15]. Associations between genes and drugs relevant to HD have been systematically documented in biological databases since 1993 [13], accumulating decades of both experimental and clinical evidence that are especially relevant for drug repurposing. Since the disease disproportionately affects working-age adults (i.e. typical symptom onset between 30 and 50 years of age) with profound impact on quality of life, there is strong clinical motivation for novel treatments despite the relatively small patient population [2]. These characteristics combine to make HD an ideal test case for validating scalable, disease-specific drug repurposing pipelines that can subsequently be adapted to other rare neurodegenerative conditions with similar molecular mechanisms or comparable knowledge base maturity.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Background", "text": "Moreover, existing computational approaches have demonstrated the feasibility of knowledge-driven drug repurposing for HD. Ambesi-Impiombato et al. developed a systematic ML approach that leverages Gene Set Enrichment Analysis to identify candidate repurposed drugs for HD [17]. Similarly, microarray-based studies have characterised transcriptomic correlations in HD-relevant tissue samples [18], and ML models have been recently developed to predict HD disease trajectories [19]. These precedents demonstrate both the feasibility and the clinical relevance of computational approaches to HD drug discovery\u2014while also highlighting opportunities for automation, scalability, and integration of diverse data sources as in our pipeline addresses.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Current landscape of computational drug repurposing", "text": "To systematically position our approach within the computational drug repurposing landscape, we examined several methodologies catalogued by Tanoli et al. in their survey of 181 in silico resources [10]. The computational approaches specifically for drug repurposing encompass several different strategies\u2014each of them with their own advantages and limitations. Structure-based approaches leverage molecular docking and protein analysis (e.g. AutoDock Vina for binding affinity-based docking predictions [34], AlphaFold for homology modelling [35, 36], and Chimera for binding site visualisation [37]). Despite being relatively fast, such structure-based strategies require having identified molecular targets with accurate 3D structures, and they often lack the accuracy of standalone approaches [10]. Knowledge-based methods instead aggregate curated associations between drugs and diseases through databases and ontologies (e.g. DrugBank for drug\u2013target interactions [38], ChEMBL and PubChem for biochemical bioactivity data [39, 40], Reactome for pathway information [41]), thus providing comprehensive evidence-based scoring\u2014but requiring extensive manual curation [10]. Network-based approaches exploit specifically protein-protein interactions and biological pathways to generate network-derived hypotheses (e.g. STRING database for protein interactions [42], MSigDB for gene signatures [43], KEGG and IntAct for pathway data [44, 45]). Even though network-based methods facilitate identification of disease mechanisms, they require careful selection of relevant biological contexts [10]. Finally, ML platforms are designed to integrate diverse data modalities, including gradient boosting methods (e.g. XGBoost [46]), Graph Neural Network (GNN) (e.g. Graph Convolutional Networks (GCN) via PyTorch-Geometric (PyG) [47, 48]), and DL frameworks. The main advantage of ML approaches is that they offer time-efficient predictions\u2014yet they also require high-quality large-scale training data and may struggle to generalise across understudied disease domains [10]. Recent data-driven strategies have demonstrated the value of harmonising drug-target interaction data from multiple curated resources to enhance reproducibility and enable systematic pathway-based predictions [12], while supervised ML methods that leverage comprehensive target activity profiles have shown particular promise in cancer drug repurposing [11].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Clinical approaches to drug repurposing", "text": "Prior of the introduction of computational means to scientific research, drug repurposing was a slow and not systematic approach based on serendipitous observations [32]: it was heavily dependent on clinical observations and luck, thus also being expensive. Examples of repurposed drugs range from toxic compounds (e.g. Arsenic Trioxide as a cancer treatment, [49]), to drugs that were first abandoned for adverse effects (e.g. Thalidomide also as a cancer treatment, [49]) or that were initially developed to treat other conditions (e.g. Sildenafil as an erectile dysfunction treatment, [32]; or recently as a candidate treatment against Alzheimer\u2019s disease [50]). Recently Xia et al. developed a non-computational approach based on testing libraries of known drugs against tumoroids (i.e. tissues derived from stem cells meant to mimic tumours) [49]. Such an approach is extremely promising, although it lacks the convenience and inexpensiveness of in-silico methods for early stage identification of candidate repurposed drugs.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Bio-semantics-based computational approaches to drug repurposing", "text": "In contrast to clinical approaches, computational approaches are flexible and scalable solutions for early state identification of repurposable drugs. Especially bio-semantics-based computational approaches that focus on data structures that can easily handle networks of associations between biological entities (e.g. disease and gene nodes) [20, 33]. As previously mentioned, KGs have proven to be a powerful method to organise complex biomedical data, due to being graphs where both nodes (a.k.a vertices) and edges (a.k.a. links) are labelled. Labels are storing information on biological entities in the nodes and on the nature of their interaction in the edges [51]. They have successfully been implemented in non-automated drug repurposing experiments via human-curated datasets against Covid-19 [51] and in novel drug R&D [20].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Bio-semantics-based computational approaches to drug repurposing", "text": "KGs provide an effective framework for managing research data [52] because they integrate ontological structures (e.g. OMIM, OWL) that define semantic relationships and hierarchies between entities, thus enabling meaningful knowledge representation throughout the network [20]. \u2018bioknowledge-reviewer\u2019 is one of such tools for structured and partially automated knowledge-drive research [30]. It builds a complex network of biological associations to NGLY1 deficiency (i.e. developmental delay) used as case study from several databases (e.g. Monarch Initiative, DGIdb), curated.csv files for key nodes, transcriptomic RNA-seq datasets in Drosophila and dataframes on Transcription Factors\u2019 (TFs) regulation on genes from the Molecular Signature Database [30]. The resulting network is then preserved on a neo4j server, and data retrieval is performed by Cypher\u2014an SQL-based querying language optimised for KGs. Each query returns a hypothesis that can then be tested experimentally. Such an approach allows for knowledge-driven research that narrows down candidates and reduces R&D costs compared to traditional approaches.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Bio-semantics-based computational approaches to drug repurposing", "text": "Monarch Initiative\u2019s database is a KG of biological associations focusing mostly on diseases, genes and phenotypes\u2014codified as an RDF via proprietary identifiers (i.e. Uniform Resource Identifies, URIs; e.g. MONDO) [53]. It is key in building computational tools for knowledge-driven association networks thanks to public APIs for automated retrieval of associations to any given node [28].\n\nOther computational approaches instead focus on pre-computed datasets. The \u2018DrugRepo\u2019 platform provides information on drug similarity, drug-target interactions and disease-gene associations for more than 650 disease [29]. Despite the final networks not being structured as KGs, it uses KGs and RDFs to compute them [29].\n\nThe advantage of the increasing computational power available to researchers is that queries can be computed in real time in relatively short time frames [28]. Reep extended \u2018bioknowledge-reviewer\u2019s semi-automated approach [30] to applications in drug repurposing: after fetching all the genes associated to one of the symptoms of the disease of interest from Monarch Initiative, their tool retrieves all the drugs associated to the network from DGIdb\u2014which focuses on gene-drug associations [31]. Compared to Monarch Initiative, DGIdb is a relational SQL database that uses Entrez, HGNC and GO identifiers for genes, and mostly ChEMBL and Wikidata identifiers for drugs [38].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Bio-semantics-based computational approaches to drug repurposing", "text": "The resulting Reep network of genes and drugs is used to train a ML model and predict newly repurposed drugs to the disease of interest [31].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Bio-semantics-based computational approaches to drug repurposing", "text": "Negative samples are an extremely relevant field of studies in Semantics and Network Biology [54]. Specifically, valid negative samples are all the missing links in a network that could be positive given certain conditions. For example if \"Macron - president of - France\" is a positive samples, \"Obama - president of - France\" would be a valid negative sample\u2014while \"Bread - president of - France\" would be invalid, as it is semantically nonsensical. Such an approach is also defined as Contrastive Learning, and Liu et al. made a custom drug repurposing pipeline against Alzheimer\u2019s disease and breast neoplasm by generating negative samples accordingly [55]. Hubert et al. focused on discriminating invalid negative samples from valid ones to weight them differently in a prediction [56]. Out of several approaches they took into consideration, semantic loss of function\u2014as the degradation or failure of a system to maintain semantic validity,1\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\begin{aligned} L_{semantic}(h,r,t) = -\\log P(semantically\\_valid(h,r,t) | \\theta ) \\end{aligned}$$\\end{document}where \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\theta $$\\end{document} represents the model parameters and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$semantically\\_valid$$\\end{document} evaluates domain/range constraint satisfaction, which quantifies the degree to which a model violates semantic coherence during training\u2014was the one that returned the best enrichment score according to the Sem@K metric (i.e. Semantic Similarity at K,2\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\begin{aligned} Sem@K = \\frac{1}{|Q|} \\sum _{q \\in Q} \\frac{|\\{r \\in top\\text {-}K(q) : sim(r, ground\\_truth) > \\tau \\}|}{K} \\end{aligned}$$\\end{document}where Q is the query set, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$top\\text {-}K(q)$$\\end{document} represents the top-K results for query q, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$sim(\\cdot ,\\cdot )$$\\end{document} computes semantic similarity, and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\tau $$\\end{document} is the similarity threshold) [56].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Bio-semantics-based computational approaches to drug repurposing", "text": "The generation of artificial negative samples that are also semantically valid is especially relevant in biological studies and bio-semantics [57]: very few databases (e.g. EBI\u2019s IntAct, Negatome Database) store information on known lack of associations between biological entities\u2014and, even when they do, negative data might not be fully integrated despite its importance [58].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Computational tools for constructing biological knowledge graphs", "text": "KGs provide an effective framework for managing research data [52] because they integrate ontological structures (e.g. OMIM, OWL) that define semantic relationships and hierarchies between entities, thus enabling meaningful knowledge representation throughout the network [20]. \u2018bioknowledge-reviewer\u2019 is one of such tools for structured and partially automated knowledge-drive research [30]. It builds a complex network of biological associations to NGLY1 deficiency (i.e. developmental delay) used as case study from several databases (e.g. Monarch Initiative, DGIdb), curated.csv files for key nodes, transcriptomic RNA-seq datasets in Drosophila and dataframes on Transcription Factors\u2019 (TFs) regulation on genes from the Molecular Signature Database [30]. The resulting network is then preserved on a neo4j server, and data retrieval is performed by Cypher\u2014an SQL-based querying language optimised for KGs. Each query returns a hypothesis that can then be tested experimentally. Such an approach allows for knowledge-driven research that narrows down candidates and reduces R&D costs compared to traditional approaches.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Computational tools for constructing biological knowledge graphs", "text": "Monarch Initiative\u2019s database is a KG of biological associations focusing mostly on diseases, genes and phenotypes\u2014codified as an RDF via proprietary identifiers (i.e. Uniform Resource Identifies, URIs; e.g. MONDO) [53]. It is key in building computational tools for knowledge-driven association networks thanks to public APIs for automated retrieval of associations to any given node [28].\n\nOther computational approaches instead focus on pre-computed datasets. The \u2018DrugRepo\u2019 platform provides information on drug similarity, drug-target interactions and disease-gene associations for more than 650 disease [29]. Despite the final networks not being structured as KGs, it uses KGs and RDFs to compute them [29].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Semi-automated pipeline for drug repurposing in biological networks", "text": "The advantage of the increasing computational power available to researchers is that queries can be computed in real time in relatively short time frames [28]. Reep extended \u2018bioknowledge-reviewer\u2019s semi-automated approach [30] to applications in drug repurposing: after fetching all the genes associated to one of the symptoms of the disease of interest from Monarch Initiative, their tool retrieves all the drugs associated to the network from DGIdb\u2014which focuses on gene-drug associations [31]. Compared to Monarch Initiative, DGIdb is a relational SQL database that uses Entrez, HGNC and GO identifiers for genes, and mostly ChEMBL and Wikidata identifiers for drugs [38].\n\nThe resulting Reep network of genes and drugs is used to train a ML model and predict newly repurposed drugs to the disease of interest [31].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Algorithm for semantic generation of negative samples", "text": "Negative samples are an extremely relevant field of studies in Semantics and Network Biology [54]. Specifically, valid negative samples are all the missing links in a network that could be positive given certain conditions. For example if \"Macron - president of - France\" is a positive samples, \"Obama - president of - France\" would be a valid negative sample\u2014while \"Bread - president of - France\" would be invalid, as it is semantically nonsensical. Such an approach is also defined as Contrastive Learning, and Liu et al. made a custom drug repurposing pipeline against Alzheimer\u2019s disease and breast neoplasm by generating negative samples accordingly [55]. Hubert et al. focused on discriminating invalid negative samples from valid ones to weight them differently in a prediction [56]. Out of several approaches they took into consideration, semantic loss of function\u2014as the degradation or failure of a system to maintain semantic validity,1\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\begin{aligned} L_{semantic}(h,r,t) = -\\log P(semantically\\_valid(h,r,t) | \\theta ) \\end{aligned}$$\\end{document}where \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\theta $$\\end{document} represents the model parameters and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$semantically\\_valid$$\\end{document} evaluates domain/range constraint satisfaction, which quantifies the degree to which a model violates semantic coherence during training\u2014was the one that returned the best enrichment score according to the Sem@K metric (i.e. Semantic Similarity at K,2\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\begin{aligned} Sem@K = \\frac{1}{|Q|} \\sum _{q \\in Q} \\frac{|\\{r \\in top\\text {-}K(q) : sim(r, ground\\_truth) > \\tau \\}|}{K} \\end{aligned}$$\\end{document}where Q is the query set, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$top\\text {-}K(q)$$\\end{document} represents the top-K results for query q, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$sim(\\cdot ,\\cdot )$$\\end{document} computes semantic similarity, and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\tau $$\\end{document} is the similarity threshold) [56].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Algorithm for semantic generation of negative samples", "text": "The generation of artificial negative samples that are also semantically valid is especially relevant in biological studies and bio-semantics [57]: very few databases (e.g. EBI\u2019s IntAct, Negatome Database) store information on known lack of associations between biological entities\u2014and, even when they do, negative data might not be fully integrated despite its importance [58].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Automating drug repurposing pipelines", "text": "The advantage of real-time pipelines for specific queries compared to pre-computed datasets are the specificity of the results and the currency of source data. The disadvantages instead are that real-time pipelines take time, yet with current computational approaches and hardware capabilities the time frames to run comparative pipelines reduce\u2014also allowing to run more complex pipelines in comparative time frames.\n\nThe key drivers in developing automated\u2014or semi-automated\u2013 pipelines are the ML models and the infrastructures that run the pipelines themselves. In fact, ML models will be provided with training data in specific formats. In the context of KGs and association networks, information for each node (i.e. biological entity) and the nodes associated to it via edges are converted into node embeddings\u2014dense low-dimensional vector representations of structural and semantic relationships in the biological association network\u2014for ML training [59]. Node embedding is performed by dedicated algorithms\u2014such as Node2Vec, a flexible graph embedding method that captures both local neighbourhood structures and broader network topology [60]. Then, ML models return predictions in vectors for new associations between nodes. These steps can take long time since each biological entity in the network has to be taken into consideration, thus parallelisation techniques are advantageous to reduce training and prediction time frames [61].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Automating drug repurposing pipelines", "text": "To convert the heterogeneous biological association network into a format suitable for ML, we employed \u2018node2vec\u2019 [62], a python implementation of the Node2Vec algorithmic framework that learns continuous feature representations for nodes in networks. The Node2Vec algorithm proceeds in two stages: first, it generates a corpus of node sequences through biased random walks on the graph; second, it applies a learning approach that maximises the likelihood of observing a node\u2019s network neighbourhood [60]. This method is particularly well-suited for biological networks where labelled training data is scarce and the network structure itself contains rich semantic information about gene-drug relationships.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Automating drug repurposing pipelines", "text": "Node2Vec excels over alternative KG embedding methods in real-time automated pipelines because it is optimised for general network structures without extensive ontological specifications\u2014unlike others that are designed primarly for typed, relational KG with explicit semantic relationships (e.g. TransE [63], DistMult [64], ComplEx [65]). Such extensive ontological specifications would in fact be hard to achieve in a semi-automated tool that is populated from databased APIs, where data schemas evolve independently across source databases. For this reason, we did not include these methods in our baseline comparisons. RDF2Vec (via the \u2018PyRDF2Vec\u2019 package) [66] would also fit such a scenario, but it has not been supported since v0.2.3 in June 2021. Node2Vec also offers higher computational efficiency compared to other methods \u2013resulting in shorter run times\u2013 and it has demonstrated effectiveness specifically for biological network analysis and gene-drug interaction predictions [67, 68].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Automating drug repurposing pipelines", "text": "Drug repurposing approaches involving graph-based ML methods implement similar tools and techniques found in other applications of the same nature in other fields of applications\u2014tools such as gradient boosting ML models. Biological associations can be organised in a bio-semantics network, and fed as training data for ML models. XGBoost is a gradient boosting model that has been used in biological studies both with tabular data to reduce the number of medical tests required to make a diagnosis [46, 59], and with graph-based data to make a drug repurposing prediction [31]. PyG instead is a library to build GNN models [47]: Wallett et al. made predictions from a bipartite network of non-biological associations using this package [69], while Wang et al. used to generated the pre-computed data for the \u2018DrugRepo\u2019 platform [29]. Such results are promising and could be tested with a network of gene-drug associations. Both XGBoost and PyG use unsupervised methods, and they are available as python packages.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Automating drug repurposing pipelines", "text": "Large datasets require processing and only a portion can be handled with current hardware and techniques. Nowadays there is a major discrepancy between scientists\u2019 ability to collect data, and the availability of means to process them: individual datasets coming from Next-Generation Sequencing experiments addressing Single Nucleotide Polimorphisms are now extremely easy to obtain in any lab within 24\u201348\u00a0h thanks to commercial products such as Oxford Nanopore\u2019s MinION [70]\u2014which sells for under \u20ac3000.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Automating drug repurposing pipelines", "text": "The average cost to then compute that same dataset within the same time-frame is instead much higher and personal computers are increasingly less capable of handling extensive data analysis on such datasets [71]. In fact, Courneya et al. pointed out that the lack of adequate computational resources would undermine the potential discoveries coming from modern Omics and inhibit further advancement of scientific research [61]. Their solution was to develop a remote service for their university\u2019s departments where scientists could perform traditional bioinformatics data processing (e.g. sequence alignment via BWA or STAR respectively for DNA or RNA sequences) with all the computational advantages of server racks. Bioinformatics tools available via web-apps can perform even more specific operations: Petrini et al. developed a genomics variant calling tool that takes advantage of parallelisation, a technique that allows to perform concurrent operations on different computational units (e.g. kernel instances, CPU and GPU cores) at the same time [72]. The result is a drastic optimisation of time-frames.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Node embeddings via Node2Vec", "text": "To convert the heterogeneous biological association network into a format suitable for ML, we employed \u2018node2vec\u2019 [62], a python implementation of the Node2Vec algorithmic framework that learns continuous feature representations for nodes in networks. The Node2Vec algorithm proceeds in two stages: first, it generates a corpus of node sequences through biased random walks on the graph; second, it applies a learning approach that maximises the likelihood of observing a node\u2019s network neighbourhood [60]. This method is particularly well-suited for biological networks where labelled training data is scarce and the network structure itself contains rich semantic information about gene-drug relationships.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Node embeddings via Node2Vec", "text": "Node2Vec excels over alternative KG embedding methods in real-time automated pipelines because it is optimised for general network structures without extensive ontological specifications\u2014unlike others that are designed primarly for typed, relational KG with explicit semantic relationships (e.g. TransE [63], DistMult [64], ComplEx [65]). Such extensive ontological specifications would in fact be hard to achieve in a semi-automated tool that is populated from databased APIs, where data schemas evolve independently across source databases. For this reason, we did not include these methods in our baseline comparisons. RDF2Vec (via the \u2018PyRDF2Vec\u2019 package) [66] would also fit such a scenario, but it has not been supported since v0.2.3 in June 2021. Node2Vec also offers higher computational efficiency compared to other methods \u2013resulting in shorter run times\u2013 and it has demonstrated effectiveness specifically for biological network analysis and gene-drug interaction predictions [67, 68].", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Graph-based ML prediction of network associations", "text": "Drug repurposing approaches involving graph-based ML methods implement similar tools and techniques found in other applications of the same nature in other fields of applications\u2014tools such as gradient boosting ML models. Biological associations can be organised in a bio-semantics network, and fed as training data for ML models. XGBoost is a gradient boosting model that has been used in biological studies both with tabular data to reduce the number of medical tests required to make a diagnosis [46, 59], and with graph-based data to make a drug repurposing prediction [31]. PyG instead is a library to build GNN models [47]: Wallett et al. made predictions from a bipartite network of non-biological associations using this package [69], while Wang et al. used to generated the pre-computed data for the \u2018DrugRepo\u2019 platform [29]. Such results are promising and could be tested with a network of gene-drug associations. Both XGBoost and PyG use unsupervised methods, and they are available as python packages.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Parallelisation in computing as a processing solution for big data", "text": "Large datasets require processing and only a portion can be handled with current hardware and techniques. Nowadays there is a major discrepancy between scientists\u2019 ability to collect data, and the availability of means to process them: individual datasets coming from Next-Generation Sequencing experiments addressing Single Nucleotide Polimorphisms are now extremely easy to obtain in any lab within 24\u201348\u00a0h thanks to commercial products such as Oxford Nanopore\u2019s MinION [70]\u2014which sells for under \u20ac3000.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Parallelisation in computing as a processing solution for big data", "text": "The average cost to then compute that same dataset within the same time-frame is instead much higher and personal computers are increasingly less capable of handling extensive data analysis on such datasets [71]. In fact, Courneya et al. pointed out that the lack of adequate computational resources would undermine the potential discoveries coming from modern Omics and inhibit further advancement of scientific research [61]. Their solution was to develop a remote service for their university\u2019s departments where scientists could perform traditional bioinformatics data processing (e.g. sequence alignment via BWA or STAR respectively for DNA or RNA sequences) with all the computational advantages of server racks. Bioinformatics tools available via web-apps can perform even more specific operations: Petrini et al. developed a genomics variant calling tool that takes advantage of parallelisation, a technique that allows to perform concurrent operations on different computational units (e.g. kernel instances, CPU and GPU cores) at the same time [72]. The result is a drastic optimisation of time-frames.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Huntington\u2019s disease as a model rare disease for drug repurposing", "text": "Huntington\u2019s disease (HD) serves as an exemplary model for RD drug repurposing research due to being well-characterised. HD is a mendelian disease with a monogenic cause, a well-established gene-drug association network, and traceable clinical phenotypes. The disease originates from a single genetic mutation in the huntingtin gene (HTT) located on chromosome 4, specifically involving an abnormal expansion of CAG trinucleotide repeats [15]. In healthy individuals, the HTT gene typically contains 10\u201335 CAG repeats. However, individuals with HD can carry up to 180 repeats in most extreme case\u2014as the repeat length directly correlates to disease severity and age of symptom onset: 60 repeats already lead to early onset [14]. This quantitative genotype-phenotype relationship makes HD particularly suitable for computational drug repurposing, as the genetic mechanism is precisely defined and well-documented across diverse biological databases and clinical cohorts.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Huntington\u2019s disease as a model rare disease for drug repurposing", "text": "The molecular pathogenesis of HD involves the mutant HTT protein triggering selective neurodegeneration, particularly affecting the GABAergic neurons of the basal ganglia striatum [14]. The loss of these inhibitory GABAergic neurons leads to dysregulation of motor circuits, manifesting as characteristic hyperkinetic movement disorders including chorea (i.e. involuntary, jerky movements), rigidity, and progressive cognitive decline [15]. The selectivity of GABAergic neuronal vulnerability has been extensively mapped at the molecular level [13], providing a rich foundation to target genetic mechanisms for therapy.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Huntington\u2019s disease as a model rare disease for drug repurposing", "text": "From a clinical perspective, HD presents several advantages for drug discovery research. Disease symptoms are objectively measurable and relatively easy to track over time compared to many other neurodegenerative diseases, enabling robust assessment of therapeutic interventions [15]. Associations between genes and drugs relevant to HD have been systematically documented in biological databases since 1993 [13], accumulating decades of both experimental and clinical evidence that are especially relevant for drug repurposing. Since the disease disproportionately affects working-age adults (i.e. typical symptom onset between 30 and 50 years of age) with profound impact on quality of life, there is strong clinical motivation for novel treatments despite the relatively small patient population [2]. These characteristics combine to make HD an ideal test case for validating scalable, disease-specific drug repurposing pipelines that can subsequently be adapted to other rare neurodegenerative conditions with similar molecular mechanisms or comparable knowledge base maturity.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Huntington\u2019s disease as a model rare disease for drug repurposing", "text": "Moreover, existing computational approaches have demonstrated the feasibility of knowledge-driven drug repurposing for HD. Ambesi-Impiombato et al. developed a systematic ML approach that leverages Gene Set Enrichment Analysis to identify candidate repurposed drugs for HD [17]. Similarly, microarray-based studies have characterised transcriptomic correlations in HD-relevant tissue samples [18], and ML models have been recently developed to predict HD disease trajectories [19]. These precedents demonstrate both the feasibility and the clinical relevance of computational approaches to HD drug discovery\u2014while also highlighting opportunities for automation, scalability, and integration of diverse data sources as in our pipeline addresses.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "The pipeline was tested against HD as a case study. The input for the pipeline is Monarch Initiative\u2019s proprietary URI for the disease: \u2018MONDO:0007739\u2019.\n\nMoreover, the generic call to DGIdb\u2019s API returned a total of 69,907 drugs present in their database.Table 1Node type distribution in the network of associations to HD: nodes are classified based on their identifier according to specific subsets of ontologies that are used in the pipeline for further enrichment; nodes labelled with \u2018Other\u2019 are so far ignored for the purpose of this studyNode typePrefixesCountOther-6920DiseaseMONDO1086Drugchembl, wikidata2002GeneHGNC, MGI, GO, NCBIgene, ZFIN, Xenbase3773PhenotypeHP1695Total15,476SubtotalWithout \u2018Other\u20198556Table 2Edge type distribution in the network of associations to HD: edges are classified based on their identifier according to specific descriptive labels that are for training and prediction; edges labelled with \u2018Other\u2019 are so far ignored for the purpose of this studyEdge typeCountOther10,392Disease-to-gene264Drug-to-drug4065Gene-to-drug16,459Gene-to-gene1340Total32,520Subtotal22,128\n\nNode type distribution in the network of associations to HD: nodes are classified based on their identifier according to specific subsets of ontologies that are used in the pipeline for further enrichment; nodes labelled with \u2018Other\u2019 are so far ignored for the purpose of this study", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "Edge type distribution in the network of associations to HD: edges are classified based on their identifier according to specific descriptive labels that are for training and prediction; edges labelled with \u2018Other\u2019 are so far ignored for the purpose of this study\n\nInformation from the network of biological associations to the disease of interest are stored as dictionaries (i.e. nodes) or list of dictionaries (i.e. edges, which store subject node, relationship, object node and notes). Here\u2019s an example:", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "Only a portion of the nodes are useful for the drug re-purposing: anything other then diseases, phenotypes, genes and drugs are not used at latter steps (Tables 1 and 2). Moreover, not all drugs can be used during the drugsimilarity.py step since not all the URIs can be mapped to known IDs to be converted into SMILES. Of the 69,907 drugs available in DGIdb, only 12,872 (18.4%) have ChEMBL or Wikidata identifiers that are potentially mappable to SMILES representations (Section\u00a0\u201cDatabase APIs for biological, gene-drug associations anddrug-drug similarity\u201d), and of these, 11,603 (16.6%) were successfully mapped (i.e. 11,596 from ChEMBL and 7 from Wikidata). This mapping limitation affects the drug similarity network construction, but it does not impact the primary gene-drug association predictions, as the drug similarity step serves to enhance negative sample selection rather than define the prediction space. Nodes and edges from the network of associations to HD that are used for drug repurposing are listed in Tables 1 and 2\u2014with counts from a reference run performed on October 7th 2025. This represents a substantial expansion compared to the network constructed by Reep, which comprised 2189 nodes and 17,467 edges using June 2022 API versions [31], demonstrating the value of updated data infrastructure and the inclusion of all DGIdb drugs\u2014rather than only those already associated with the disease of interest (Section\u00a0\u201cDatabase APIs for biological, gene-drug associations anddrug-drug similarity\u201d).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "Compared to the full network (Fig.\u00a01), the smaller network (Fig.\u00a02) of 1000 (26.50%) genes, 600 drugs (29.97%) and 970 other mappable nodes (10.00%) generated with the same distribution of centrality scores had much lower edge density: 267 gene-to-drug edges (1.62%), 131 gene-to-gene edges (4.81%), and 513 other edges (4.81%) (Table 9). Appendix A presents the validation of the subset network as a representation of the full network for the purpose of iteration-consensus testing.Fig. 1Full network of associations to HD: highly interconnected nodes (mostly diseases in dark blue, genes in light blue and phenotypes in blue) at the centre and less interconnected nodes (mostly drugs in red and other entities in orange) at the fringes joined by biological associations (blue lines) and gene-to-drug edges (red lines); the full network comprises 15,476 nodes and 32,520 edges from the October 7, 2025 data snapshot (Tables 1 and 2)Fig. 2Network made up of a subset of nodes from the full network in Figure 1: generated according to the random sampling method described in Section \u201cPipeline upgrades\u201d; nodes are sampled to retain the same proportions according to labels in Table 1; statistical validation (Appendix A) confirms preservation of betweenness centrality distributions (i.e. p>0.05, KS-test) and representative topology, enabling iteration-based evaluation while reducing computational requirements and time frame", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "Full network of associations to HD: highly interconnected nodes (mostly diseases in dark blue, genes in light blue and phenotypes in blue) at the centre and less interconnected nodes (mostly drugs in red and other entities in orange) at the fringes joined by biological associations (blue lines) and gene-to-drug edges (red lines); the full network comprises 15,476 nodes and 32,520 edges from the October 7, 2025 data snapshot (Tables 1 and 2)\n\nNetwork made up of a subset of nodes from the full network in Figure 1: generated according to the random sampling method described in Section \u201cPipeline upgrades\u201d; nodes are sampled to retain the same proportions according to labels in Table 1; statistical validation (Appendix A) confirms preservation of betweenness centrality distributions (i.e. p>0.05, KS-test) and representative topology, enabling iteration-based evaluation while reducing computational requirements and time frame\n\nThe results from the generation of negative samples (Section\u00a0\u201cGeneration of semantically-valid negative samples based on centrality measures\u201d) vary depending on the centrality measure used: nodes that score higher will lead to more edges containing it being deemed valid negative samples. Gene node \u2018HGNC:644\u2019 (AR, Androgen Receptor) for example is the top scoring according to the degree centrality measure (0.0479), yet it is not according to eigenvector and in-betweenness centrality measures (respectively 0.2930 with max=0.3143, and 0.0739 with max=0.0852).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "Moreover, this step still takes into consideration hub nodes that are not relevant for this purpose: genes such as \u2018HGNC:11998\u2019 (TP53) (i.e. a transcription factor for apoptosis that is silenced in tumors) are highly interconnected because of their nature of home-keeping genes. Yet they cannot be automatically ignored due to the lack of a database that annotates them accordingly and that uses an API for mapping call requests.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "To validate the contribution of our combined centrality and similarity-based approach, we performed an ablation study comparing four negative sampling strategies on the full HD network using GCN: random negatives as a baseline, similarity-only, centrality-only, and \u2018combined\u2019 (i.e. centrality+similarity). All strategies achieved high discrimination (i.e. AUROC 0.91\u22120.92, Table 3), with the combined approach demonstrating the best performance across metrics (i.e. AUROC=0.9175, AUPRC=0.6904, F1=0.5503, Table 3), representing modest \u2013yet consistent\u2014improvements over the other sampling methods (e.g. +0.5% AUROC, +1.8% AUPRC, +7.0% F1-score to baseline). The smaller-than-anticipated effect size (Table 3) is attributed to the design of negative samples generation: we only consider gene-to-drug edges for training and prediction, thus even randomly generated valid positive edges are informative enough to support prediction. Nonetheless, the \u2018combined\u2019 approach provides the best performance (Table 3). Our approach\u2014mediated also by Node2Vec\u2019s fit to the sparse nature of biological networks (Section\u00a0\u201cParallelisation in gene-drug embeddings\u201d)\u2014allows to reduce computational times in the prediction phase (Section\u00a0\u201cParallelisation in gene-drug embeddings\u201d). Detailed ablation methodology, comprehensive metrics from Table 3, and comparison plots are provided in Appendix D.Table 3Negative sampling ablation study summary: \u2018combined\u2019 (i.e. centrality+similarity) strategy outperforms random baseline with modest but consistent improvementsStrategyAUROCAUPRCF1-scoreRandom (baseline)0.91290.67820.5144Combined0.91750.69040.5503Improvement+0.5%+1.8%+7.0%Full results comparing all four strategies (i.e. random, similarity-only, centrality-only, combined) with comprehensive metrics and analysis are provided in Appendix D", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "Negative sampling ablation study summary: \u2018combined\u2019 (i.e. centrality+similarity) strategy outperforms random baseline with modest but consistent improvements\n\nFull results comparing all four strategies (i.e. random, similarity-only, centrality-only, combined) with comprehensive metrics and analysis are provided in Appendix D\n\nDue to the differences between the two ML methods (i.e. XGBoost and GCN) in the kind of embeddings to feed for training, the time required to building them changes quite a lot. Where the pipeline running the XGBoost method requires to compute any interaction (i.e. present as positive or valid negative, or missing as invalid negative) between gene and drug nodes (e.g. for Huntington\u2019s disease run: 3031 * 4582 = 13,888,042), the pipeline running the GCN method only requires to fill the matrix with the known positive and valid negative associations (i.e. 28,260 + 17,956 = 46,216). GCN embedding generation completes in approximately 48\u00a0s, requiring only 0.3% of the time needed for XGBoost embedding generation (i.e. approximately 4.4\u00a0h).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "Both methods also require the construction of an empty prediction matrix for all possible combinations between genes in the network and all drugs from DGIdb (i.e. 3031 * 69,907 = 211,888,117), but this step take considerably less time due to not having to iterate through edges to determine the presence or absence of an association. Nonetheless, the entire process still takes from half an hour to almost 24\u00a0h: for the purpose of saving time, the emb_l=1 parameter allows to load existing embeddings files instead of computing them provided that they exist (i.e. same disease, same date -although the date can be brute forced with the set_date() utility function at the beginning of the function for this specific purpose). On non-specialised computational hardware (e.g. laptop with 10-core CPU @3.2GHz in our case), the complete embedding generation process ranges from 30\u00a0min\u2014with GCN on the subset network\u2014to approximately 20\u00a0h\u2014with XGBoost on full network. Parallel processing reduces these times by a factor proportional to the number of available cores.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "Results from the ML step of the pipeline with the two different models (i.e. XGBoost and GCN) vary quite a lot\u2014both in runtime and prediction volume. Although the size of the network \u2013thus also the size of the node embedding training file\u2014has a role in how long and straining the training step will be, the parameter optimisation setting (i.e. ml_depth) is even more influential. The prediction itself does not take that long compared to the training. Both models have three incremental degrees of complexity for the parameter optimisation step (i.e. \u2018ultralight\u2019, \u2018light\u2019 and \u2018full\u2019). Yet, the GCN model at ml_depth=\u2018full\u2019 still takes significantly less even than the XGBoost model at ml_depth=\u2018ultralight\u2019.Table 4Quantitative performance metrics for single-run and consensus iteration predictions (Section\u00a0\u201cPrediction iterations as a solution for seed variability in ML methods\u201d): AUROC, AUPRC, and F1-score values with 95% bootstrap confidence intervals demonstrate GCN\u2019s superior discrimination ability across all configurationsModel configurationAUROC [95% CI]AUPRC [95% CI]F1P@10R@100Single-Run Performance (Full Network)GCN (raw)0.9313 [0.9222, 0.9407]0.7280 [0.6979, 0.7577]0.59221.00.1418GCN (filtered)0.9236 [0.9139, 0.9323]0.7077 [0.6752, 0.7405]0.55901.00.1400XGBoost (raw)0.3958 [0.3778, 0.4135]0.0706 [0.0649, 0.0769]0.16670.200.0069XGBoost (filtered)0.4152 [0.3950, 0.4364]0.0749 [0.0685, 0.0819]0.16670.000.0150Consensus Iterations (Subset Network)GCN (200 iterations)0.9865 [0.9764, 0.9946]0.9131 [0.8546, 0.9610]0.58991.00.9242XGBoost (10 iterations)0.4556 [0.3719, 0.5367]0.1023 [0.0678, 0.1429]0.16670.100.1667GCN achieves F1-scores comparable to Reep (F1=0.867), while Precision@10 (P@10) and Recall@100 (R@100) metrics show consensus GCN achieves perfect top-10 precision and recovers 92.42% of test edges within top-100 predictions. Filtered configurations use prob\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge $$\\end{document}0.65, cluster\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge $$\\end{document}0.8 thresholds (more details in Section\u00a0\u201cML training and prediction in enriched KG\u201d)Table 5Comparison of ML model computational performance by parameter optimisation setting: each model (i.e. GCN and XGBoost) is run with different parameter optimisation settings (\u2018ml_depth\u2019) to show the difference in increase of runtime between respectively a GNN and a boosting tree at maximum settingsModel\u2018ml_depth\u2019Avg runtimeAvg CPU usage (%)GCN (PyG)full0\u00a0h 2\u00a0m 47\u00a0s22.33GCN (PyG)light0\u00a0h 2\u00a0m 29\u00a0s14.75GCN (PyG)ultralight0\u00a0h 1\u00a0m 29\u00a0s11.00XGBoostfull3\u00a0h 19\u00a0m 43\u00a0s38.68XGBoostlight0\u00a0h 37\u00a0m 14\u00a0s12.61XGBoostultralight0\u00a0h 38\u00a0m 54\u00a0s11.91", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "Quantitative performance metrics for single-run and consensus iteration predictions (Section\u00a0\u201cPrediction iterations as a solution for seed variability in ML methods\u201d): AUROC, AUPRC, and F1-score values with 95% bootstrap confidence intervals demonstrate GCN\u2019s superior discrimination ability across all configurations\n\nGCN achieves F1-scores comparable to Reep (F1=0.867), while Precision@10 (P@10) and Recall@100 (R@100) metrics show consensus GCN achieves perfect top-10 precision and recovers 92.42% of test edges within top-100 predictions. Filtered configurations use prob\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge $$\\end{document}0.65, cluster\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge $$\\end{document}0.8 thresholds (more details in Section\u00a0\u201cML training and prediction in enriched KG\u201d)", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "Comparison of ML model computational performance by parameter optimisation setting: each model (i.e. GCN and XGBoost) is run with different parameter optimisation settings (\u2018ml_depth\u2019) to show the difference in increase of runtime between respectively a GNN and a boosting tree at maximum settings\n\nQuantitative performance metrics in Table 4 demonstrate GCN\u2019s superior discrimination ability across all model configurations, while Table 5 shows average runtimes and CPU usage for the full pipeline on the smaller artificial network with parameter emb_l=1 parameter, therefore the only steps that were performed are the ones involving the ML step. Results show that for the smaller network there is no big difference between using the \u2018light\u2019 or the \u2018ultralight\u2019 parameter optimisation degrees of complexity, and this is also true when considering just the training and prediction times with the full network.Fig. 3Performance evaluation of GCN single-run predictions on full network: Left panel shows receiver operating characteristic (ROC) curve comparing GCN performance (red) against random baseline (dotted black), demonstrating excellent discrimination ability with AUROC=0.9236. Right panel shows precision-recall curve with GCN achieving AUPRC=0.7077 (red) substantially outperforming the random baseline at 0.0909 (dotted black), confirming strong predictive performance for gene-to-drug edge prediction on the HD network", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "Performance evaluation of GCN single-run predictions on full network: Left panel shows receiver operating characteristic (ROC) curve comparing GCN performance (red) against random baseline (dotted black), demonstrating excellent discrimination ability with AUROC=0.9236. Right panel shows precision-recall curve with GCN achieving AUPRC=0.7077 (red) substantially outperforming the random baseline at 0.0909 (dotted black), confirming strong predictive performance for gene-to-drug edge prediction on the HD network", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "Single-run predictions on the full network demonstrate substantial differences between models in both prediction volume and discrimination ability. Quantitative evaluation using an 80/20 train/test split \u2013with held-out edges from the original network\u2013 reveals that GCN achieves excellent raw discrimination (i.e. AUROC=0.9313 [0.9222, 0.9407], AUPRC=0.7280 [0.6979, 0.7577], F1=0.5922) while XGBoost shows limited discrimination ability (i.e. AUROC=0.3958 [0.3778, 0.4135], AUPRC=0.0706 [0.0649, 0.0769], F1=0.1667) (Fig.\u00a03). When probability filtering is applied (i.e. prob>0.65, cluster>0.8), GCN maintains strong performance (i.e. AUROC=0.9236 [0.9139, 0.9323], AUPRC=0.7077 [0.6752, 0.7405], F1=0.5590) with perfect Precision@10, while XGBoost performance shows mixed results (AUROC=0.4152 [0.3950, 0.4364], AUPRC=0.0749 [0.0685, 0.0819], F1=0.1667) (Fig.\u00a03). Compared to Reep\u2019s reported F1-score of 0.867, our models return lower values due to architectural differences: in fact compared to their version that only predicted repurposed drugs from a smaller network only with drugs already associated to the disease of interest, our model takes into consideration a much larger and more sparse network where predictions are hard to be quantified with F1 scores. Moreover, our dual-model approach with consensus-based ranking provides additional robustness through model complementarity as demonstrated in Section\u00a0\u201cPrediction iterations as a solution for seed variability in ML methods\u201d and in Appendix C.Fig. 4XGBoost single run prediction: Full network of associations to HD from Figure 1 showing newly predicted gene-to-drug edges in green via the XGBoost model as described in Section \u201cPrediction of repurposed drugs in binary models with non-binary datasets\u201dFig. 5GCN single run prediction: Full network of associations to HD from Figure 1 showing newly predicted gene-to-drug edges in green via the GCN model as described in Section \u201cPrediction of repurposed drugs in binary models with non-binary datasets\u201d", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "XGBoost single run prediction: Full network of associations to HD from Figure 1 showing newly predicted gene-to-drug edges in green via the XGBoost model as described in Section \u201cPrediction of repurposed drugs in binary models with non-binary datasets\u201d\n\nGCN single run prediction: Full network of associations to HD from Figure 1 showing newly predicted gene-to-drug edges in green via the GCN model as described in Section \u201cPrediction of repurposed drugs in binary models with non-binary datasets\u201d\n\nIn terms of prediction volume, starting from a network with 251 positive gene-to-drug edges and approximately 5\u00a0M possible gene-to-drug edges, the two models exhibit markedly different behavior: XGBoost returns between 20 and 160 predicted edges respectively with 0.95 and 0.6 probability filter, while GCN returns approximately 120,000 (0.95 filter) to 500,000 (0.65 filter) edges. This indicates that GCN is substantially more lenient in positive prediction even at high probability thresholds.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "To evaluate whether these architecturally distinct models converge on similar drug candidates, we performed convergence analysis using Jaccard similarity across top-K rankings. Under standard filtering thresholds (i.e. prob\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge 0.65$$\\end{document}, cluster\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge 0.8$$\\end{document}; more details in Section \u201cML training and prediction in enriched KG\u201d), single-run predictions showed minimal overlap due to XGBoost\u2019s limited prediction volume of 3 drugs compared to GCN with 298 drugs, with Jaccard similarity near zero for top-K rankings. However, this divergence appears to be an artifact of filtering rather than fundamental model disagreement (detailed analysis in Appendix C).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "Individual runs with GCN return many predicted repurposed drugs (i.e. \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$> 1100$$\\end{document}) with little edges associated to them (i.e. \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\le 10$$\\end{document}) (Fig.\u00a05), while individual runs with XGBoost return fewer predicted repurposed drugs (i.e. 5\u201340) with many more edges associated to them (i.e. \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$> 200$$\\end{document}) (Fig.\u00a04). Among the latter, several are associated to genes involved in neurological diseases (e.g. RECQL gene involved in Alzheimer\u2019s disease to Koparin, SMN1 and SMN2 genes involved in spinal muscular atrophy to Methyl Robustone) according to validation via PubChem (National Center for Biotechnology Information, part of the USA National Institute of Health and National Library of Medicine). Other predicted repurposed drugs were only associated to unrelated genes.Fig. 6Prediction agreement between XGBoost and GCN on full network single runs: Venn diagrams showing overlap of drugs from unfiltered predicted gene-to-drug edges classified by both models. XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} and GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} represent drugs from edges predicted as positive (prob\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge $$\\end{document}0.5) by XGBoost and GCN respectively, while XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^-$$\\end{document} and GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^-$$\\end{document} represent drugs predicted from edges below these thresholds. The XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document}\n\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\cap $$\\end{document} GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} intersection (n=2002) indicates drugs with high confidence from both models, while XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document}\n\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\cap $$\\end{document} GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^-$$\\end{document} (n=4187) reveal model-specific predictions", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "Prediction agreement between XGBoost and GCN on full network single runs: Venn diagrams showing overlap of drugs from unfiltered predicted gene-to-drug edges classified by both models. XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} and GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} represent drugs from edges predicted as positive (prob\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge $$\\end{document}0.5) by XGBoost and GCN respectively, while XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^-$$\\end{document} and GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^-$$\\end{document} represent drugs predicted from edges below these thresholds. The XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document}\n\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\cap $$\\end{document} GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} intersection (n=2002) indicates drugs with high confidence from both models, while XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document}\n\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\cap $$\\end{document} GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^-$$\\end{document} (n=4187) reveal model-specific predictions", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "Convergence analysis across both single-run and iteration-based predictions (Section\u00a0\u201cPrediction iterations as a solution for seed variability in ML methods\u201d) identified several drugs predicted consistently by both models across different evaluation frameworks, suggesting robust cross-model agreement on high-priority candidates (Fig.\u00a06, detailed analysis in Appendix C).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "To assess prediction stability and enable consensus ranking across random seeds, we performed iteration-based evaluation on the validated subset network (Appendix A): 10 iterations for XGBoost (i.e. 7.8\u00a0h total, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\sim 47$$\\end{document} minutes per iteration) and 200 iterations for GCN (i.e. 5.5\u00a0h total, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\sim 35$$\\end{document} seconds per iteration). The occurrences of predicted gene-to-drug edges across iterations exhibit distinct distributional patterns (Figs.\u00a07 and 8): XGBoost predictions display a binomial distribution spanning from 0 to the maximum number of occurrences (Fig.\u00a09), while GCN predictions show no specific distribution with all predicted edges appearing in at least 182 out of 200 iterations (Fig.\u00a010), indicating substantially higher prediction stability.Fig. 7XGBoost 10-iteration run prediction: Subset network of associations to HD from Fig.\u00a03 showing newly predicted gene-to-drug edges in green via 10 iterations of the XGBoost model as described in Section \"Prediction iterations as a solution for seed variability in ML methods\" (detailed distribution in Fig.\u00a09)Fig. 8GCN 200-iteration run prediction: Subset network of associations to HD from Fig.\u00a03 showing newly predicted gene-to-drug edges in green via 200 iterations of the GCN model as described in Section \"Prediction iterations as a solution for seed variability in ML methods\" (detailed distribution in Fig.\u00a010)Fig. 9Distribution of occurrences for predicted edges in XGBoost 10-iteration run: predicted edges follow a binomial distribution, with few edges being predicted consistently in all 10 iterations; this distribution indicates substantial seed-dependent variability in XGBoost predictions, with most edges appearing in fewer iterations (see Appendix B for detailed stability analysis)Fig. 10Distribution of occurrences for predicted edges in GCN 200-iteration run: predicted edges do not follow a specific distribution with all edges at least being predicted in 182 off all 200 iterations (i.e. 91% minimum occurrence rate), and the majority of edges predicted in all 200 iterations. This highly skewed distribution (i.e. all edges being predicted in at least 182 runs) indicates substantially higher prediction stability for GCN compared to XGBoost (Fig.\u00a09) (see Appendix B for detailed stability analysis)", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "XGBoost 10-iteration run prediction: Subset network of associations to HD from Fig.\u00a03 showing newly predicted gene-to-drug edges in green via 10 iterations of the XGBoost model as described in Section \"Prediction iterations as a solution for seed variability in ML methods\" (detailed distribution in Fig.\u00a09)\n\nGCN 200-iteration run prediction: Subset network of associations to HD from Fig.\u00a03 showing newly predicted gene-to-drug edges in green via 200 iterations of the GCN model as described in Section \"Prediction iterations as a solution for seed variability in ML methods\" (detailed distribution in Fig.\u00a010)\n\nDistribution of occurrences for predicted edges in XGBoost 10-iteration run: predicted edges follow a binomial distribution, with few edges being predicted consistently in all 10 iterations; this distribution indicates substantial seed-dependent variability in XGBoost predictions, with most edges appearing in fewer iterations (see Appendix B for detailed stability analysis)", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "Distribution of occurrences for predicted edges in GCN 200-iteration run: predicted edges do not follow a specific distribution with all edges at least being predicted in 182 off all 200 iterations (i.e. 91% minimum occurrence rate), and the majority of edges predicted in all 200 iterations. This highly skewed distribution (i.e. all edges being predicted in at least 182 runs) indicates substantially higher prediction stability for GCN compared to XGBoost (Fig.\u00a09) (see Appendix B for detailed stability analysis)\n\nConsensus predictions from multiple iterations demonstrate dramatically different performance characteristics (Table 4). For GCN, consensus across 200 iterations achieves excellent discrimination (i.e. AUROC=0.9865 [0.9764, 0.9946], AUPRC=0.9131 [0.8546, 0.9610]) with outstanding top-K performance (i.e. Precision@10=1.0, Recall@100=0.9242), recovering 92.42% of test edges among the top 100 predictions. In contrast, XGBoost consensus across 10 iterations shows poor discrimination (i.e. AUROC=0.4556 [0.3719, 0.5367], AUPRC=0.1023 [0.0678, 0.1429]) with limited recall (i.e. Recall@100=0.1667), indicating that XGBoost suffers more from lower link prediction efficiency in 80/20 train/split conditions than GCN. We hypothise that the removal of that many positive gene-to-drug edges deeply affect the outcome of prediction, due to the sparce nature of biological networks\u2014especially in RDs.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "Figure\u00a011 presents a heatmap comparing prediction confidences for 20 selected drugs across different evaluation frameworks (single-run on full network, single-run on subset network, and iteration-based consensus on subset network for both XGBoost and GCN). The heatmap reveals systematic patterns in prediction confidence: drugs consistently predicted by both models across evaluation frameworks represent high-confidence candidates (e.g. Ergocristine, Fig.\u00a011), while drugs with framework-specific confidence patterns highlight the complementary nature of single-run and iteration-based approaches.Fig. 11Prediction confidence for 20 selected drugs across evaluation frameworks: Heatmap showing prediction probabilities for drugs identified through convergence analysis, evaluated across XGBoost and GCN using both the single-run approach on both full network and subset network, and the iteration-based consensus approach on the subset network. Colour represents prediction confidence (darker red = higher probability). Consistent high-confidence predictions across all six frameworks represent robust candidates, while framework-specific patterns reveal differences in model calibration and the stabilising effect of iteration-based consensus. Notably, Ergocristine has higher confidence in the GCN iteration consensus approach on the subset network than in the single run approach on the full network, in support of our thesis that iterations can lead to candidates that are overlooked over single runs. White cells indicate random predictions", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "Prediction confidence for 20 selected drugs across evaluation frameworks: Heatmap showing prediction probabilities for drugs identified through convergence analysis, evaluated across XGBoost and GCN using both the single-run approach on both full network and subset network, and the iteration-based consensus approach on the subset network. Colour represents prediction confidence (darker red = higher probability). Consistent high-confidence predictions across all six frameworks represent robust candidates, while framework-specific patterns reveal differences in model calibration and the stabilising effect of iteration-based consensus. Notably, Ergocristine has higher confidence in the GCN iteration consensus approach on the subset network than in the single run approach on the full network, in support of our thesis that iterations can lead to candidates that are overlooked over single runs. White cells indicate random predictions", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "The iteration-based consensus approach reveals an important pattern: some drugs without obvious disease-gene associations may still hold therapeutic relevance when evaluated through pharmacological mechanisms rather than network proximity alone. Ergocristine (CHEMBL1201247) exemplifies this pattern. While not being directly associated with HD-related genes in public databases, Ergocristine demonstrated higher and more stable prediction confidence in iteration-based consensus (i.e. 200 iterations on GCN) compared to single-run predictions on both full and subset networks (detailed probability distributions in Appendix C). Notably, Ergocristine emerged from the convergence analysis as one of the drugs predicted by both XGBoost and GCN across different evaluation frameworks: it appears in the overlap between models on both full network single runs and subset network iteration runs, demonstrating robust cross-model agreement. According to PubChem classification, Ergocristine functions as a Peripheral Nervous System Agent with mechanisms involving neurotransmitter modulation and motor pathway regulation\u2013pharmacologically relevant to HD given the disease\u2019s characteristic GABAergic medium spiny neuron degeneration and motor dysfunction [2]. While this higher prediction confidence may be coincidental, it also suggests that iteration-based consensus could distinguish genuine signal from stochastic noise\u2013and surface biologically plausible candidates that might be overlooked or underweighted in single-run predictions.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "Detailed iteration analysis\u2014including edge occurrence distributions, per-iteration variability, and filtering comparisons\u2013 is provided in Appendix B. Analysis of five hand-picked held-out edges involving key HD-associated genes (TUBB3, GSK3B, HTT, FASN) confirms this pattern: GCN recovers 4 out of 5 edges within its top predictions, while XGBoost recovers 1 out of 5. This demonstrates the substantial difference in recall capability between the models (detailed per-edge recovery analysis in Appendix B).Fig. 12Consensus prediction agreement between models on subset network: Venn diagram comparing drugs from unfiltered positive predictions from XGBoost (10 iterations) and all predictions from GCN (200 iterations) on the subset network. Only positive consensus predictions for XGBoost are stored in iterations. The overlap XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document}\n\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\cap $$\\end{document} GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} (n=591) represents 17.74% of XGBoost predicted drugs and 100% of GCN drugs from positive predictions, demonstrating moderate convergence on high-priority drug candidates despite different iteration counts and model architectures. Model-specific predictions (XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} only: n=2709) reflect differences in confidence calibration and prediction volume. Detailed top-K overlap analysis is provided in Appendix C (Fig.\u00a020)", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "Consensus prediction agreement between models on subset network: Venn diagram comparing drugs from unfiltered positive predictions from XGBoost (10 iterations) and all predictions from GCN (200 iterations) on the subset network. Only positive consensus predictions for XGBoost are stored in iterations. The overlap XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document}\n\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\cap $$\\end{document} GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} (n=591) represents 17.74% of XGBoost predicted drugs and 100% of GCN drugs from positive predictions, demonstrating moderate convergence on high-priority drug candidates despite different iteration counts and model architectures. Model-specific predictions (XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} only: n=2709) reflect differences in confidence calibration and prediction volume. Detailed top-K overlap analysis is provided in Appendix C (Fig.\u00a020)", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Results", "text": "Convergence analysis using relaxed thresholds (Appendix C) confirms that filtering creates artificial divergence: when analysed without probability filters, consensus predictions from both models demonstrate strong agreement at top ranks\u2013100% overlap at top-10 (i.e. Jaccard=1.0), 95% at top-20, 94% at top-50, and 96% at top-100 (Fig.\u00a012). This indicates that despite substantial differences in discrimination ability and confidence calibration, both models fundamentally identify similar high-priority drug candidates for HD.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Pipeline upgrades", "text": "Information from the network of biological associations to the disease of interest are stored as dictionaries (i.e. nodes) or list of dictionaries (i.e. edges, which store subject node, relationship, object node and notes). Here\u2019s an example:", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Pipeline upgrades", "text": "Only a portion of the nodes are useful for the drug re-purposing: anything other then diseases, phenotypes, genes and drugs are not used at latter steps (Tables 1 and 2). Moreover, not all drugs can be used during the drugsimilarity.py step since not all the URIs can be mapped to known IDs to be converted into SMILES. Of the 69,907 drugs available in DGIdb, only 12,872 (18.4%) have ChEMBL or Wikidata identifiers that are potentially mappable to SMILES representations (Section\u00a0\u201cDatabase APIs for biological, gene-drug associations anddrug-drug similarity\u201d), and of these, 11,603 (16.6%) were successfully mapped (i.e. 11,596 from ChEMBL and 7 from Wikidata). This mapping limitation affects the drug similarity network construction, but it does not impact the primary gene-drug association predictions, as the drug similarity step serves to enhance negative sample selection rather than define the prediction space. Nodes and edges from the network of associations to HD that are used for drug repurposing are listed in Tables 1 and 2\u2014with counts from a reference run performed on October 7th 2025. This represents a substantial expansion compared to the network constructed by Reep, which comprised 2189 nodes and 17,467 edges using June 2022 API versions [31], demonstrating the value of updated data infrastructure and the inclusion of all DGIdb drugs\u2014rather than only those already associated with the disease of interest (Section\u00a0\u201cDatabase APIs for biological, gene-drug associations anddrug-drug similarity\u201d).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Pipeline upgrades", "text": "Compared to the full network (Fig.\u00a01), the smaller network (Fig.\u00a02) of 1000 (26.50%) genes, 600 drugs (29.97%) and 970 other mappable nodes (10.00%) generated with the same distribution of centrality scores had much lower edge density: 267 gene-to-drug edges (1.62%), 131 gene-to-gene edges (4.81%), and 513 other edges (4.81%) (Table 9). Appendix A presents the validation of the subset network as a representation of the full network for the purpose of iteration-consensus testing.Fig. 1Full network of associations to HD: highly interconnected nodes (mostly diseases in dark blue, genes in light blue and phenotypes in blue) at the centre and less interconnected nodes (mostly drugs in red and other entities in orange) at the fringes joined by biological associations (blue lines) and gene-to-drug edges (red lines); the full network comprises 15,476 nodes and 32,520 edges from the October 7, 2025 data snapshot (Tables 1 and 2)Fig. 2Network made up of a subset of nodes from the full network in Figure 1: generated according to the random sampling method described in Section \u201cPipeline upgrades\u201d; nodes are sampled to retain the same proportions according to labels in Table 1; statistical validation (Appendix A) confirms preservation of betweenness centrality distributions (i.e. p>0.05, KS-test) and representative topology, enabling iteration-based evaluation while reducing computational requirements and time frame", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Pipeline upgrades", "text": "Full network of associations to HD: highly interconnected nodes (mostly diseases in dark blue, genes in light blue and phenotypes in blue) at the centre and less interconnected nodes (mostly drugs in red and other entities in orange) at the fringes joined by biological associations (blue lines) and gene-to-drug edges (red lines); the full network comprises 15,476 nodes and 32,520 edges from the October 7, 2025 data snapshot (Tables 1 and 2)\n\nNetwork made up of a subset of nodes from the full network in Figure 1: generated according to the random sampling method described in Section \u201cPipeline upgrades\u201d; nodes are sampled to retain the same proportions according to labels in Table 1; statistical validation (Appendix A) confirms preservation of betweenness centrality distributions (i.e. p>0.05, KS-test) and representative topology, enabling iteration-based evaluation while reducing computational requirements and time frame\n\nThe results from the generation of negative samples (Section\u00a0\u201cGeneration of semantically-valid negative samples based on centrality measures\u201d) vary depending on the centrality measure used: nodes that score higher will lead to more edges containing it being deemed valid negative samples. Gene node \u2018HGNC:644\u2019 (AR, Androgen Receptor) for example is the top scoring according to the degree centrality measure (0.0479), yet it is not according to eigenvector and in-betweenness centrality measures (respectively 0.2930 with max=0.3143, and 0.0739 with max=0.0852).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Pipeline upgrades", "text": "Moreover, this step still takes into consideration hub nodes that are not relevant for this purpose: genes such as \u2018HGNC:11998\u2019 (TP53) (i.e. a transcription factor for apoptosis that is silenced in tumors) are highly interconnected because of their nature of home-keeping genes. Yet they cannot be automatically ignored due to the lack of a database that annotates them accordingly and that uses an API for mapping call requests.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Pipeline upgrades", "text": "To validate the contribution of our combined centrality and similarity-based approach, we performed an ablation study comparing four negative sampling strategies on the full HD network using GCN: random negatives as a baseline, similarity-only, centrality-only, and \u2018combined\u2019 (i.e. centrality+similarity). All strategies achieved high discrimination (i.e. AUROC 0.91\u22120.92, Table 3), with the combined approach demonstrating the best performance across metrics (i.e. AUROC=0.9175, AUPRC=0.6904, F1=0.5503, Table 3), representing modest \u2013yet consistent\u2014improvements over the other sampling methods (e.g. +0.5% AUROC, +1.8% AUPRC, +7.0% F1-score to baseline). The smaller-than-anticipated effect size (Table 3) is attributed to the design of negative samples generation: we only consider gene-to-drug edges for training and prediction, thus even randomly generated valid positive edges are informative enough to support prediction. Nonetheless, the \u2018combined\u2019 approach provides the best performance (Table 3). Our approach\u2014mediated also by Node2Vec\u2019s fit to the sparse nature of biological networks (Section\u00a0\u201cParallelisation in gene-drug embeddings\u201d)\u2014allows to reduce computational times in the prediction phase (Section\u00a0\u201cParallelisation in gene-drug embeddings\u201d). Detailed ablation methodology, comprehensive metrics from Table 3, and comparison plots are provided in Appendix D.Table 3Negative sampling ablation study summary: \u2018combined\u2019 (i.e. centrality+similarity) strategy outperforms random baseline with modest but consistent improvementsStrategyAUROCAUPRCF1-scoreRandom (baseline)0.91290.67820.5144Combined0.91750.69040.5503Improvement+0.5%+1.8%+7.0%Full results comparing all four strategies (i.e. random, similarity-only, centrality-only, combined) with comprehensive metrics and analysis are provided in Appendix D", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Pipeline upgrades", "text": "Negative sampling ablation study summary: \u2018combined\u2019 (i.e. centrality+similarity) strategy outperforms random baseline with modest but consistent improvements\n\nFull results comparing all four strategies (i.e. random, similarity-only, centrality-only, combined) with comprehensive metrics and analysis are provided in Appendix D\n\nDue to the differences between the two ML methods (i.e. XGBoost and GCN) in the kind of embeddings to feed for training, the time required to building them changes quite a lot. Where the pipeline running the XGBoost method requires to compute any interaction (i.e. present as positive or valid negative, or missing as invalid negative) between gene and drug nodes (e.g. for Huntington\u2019s disease run: 3031 * 4582 = 13,888,042), the pipeline running the GCN method only requires to fill the matrix with the known positive and valid negative associations (i.e. 28,260 + 17,956 = 46,216). GCN embedding generation completes in approximately 48\u00a0s, requiring only 0.3% of the time needed for XGBoost embedding generation (i.e. approximately 4.4\u00a0h).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Pipeline upgrades", "text": "Both methods also require the construction of an empty prediction matrix for all possible combinations between genes in the network and all drugs from DGIdb (i.e. 3031 * 69,907 = 211,888,117), but this step take considerably less time due to not having to iterate through edges to determine the presence or absence of an association. Nonetheless, the entire process still takes from half an hour to almost 24\u00a0h: for the purpose of saving time, the emb_l=1 parameter allows to load existing embeddings files instead of computing them provided that they exist (i.e. same disease, same date -although the date can be brute forced with the set_date() utility function at the beginning of the function for this specific purpose). On non-specialised computational hardware (e.g. laptop with 10-core CPU @3.2GHz in our case), the complete embedding generation process ranges from 30\u00a0min\u2014with GCN on the subset network\u2014to approximately 20\u00a0h\u2014with XGBoost on full network. Parallel processing reduces these times by a factor proportional to the number of available cores.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Database APIs for biological, gene-drug associations and drug-drug similarity", "text": "Information from the network of biological associations to the disease of interest are stored as dictionaries (i.e. nodes) or list of dictionaries (i.e. edges, which store subject node, relationship, object node and notes). Here\u2019s an example:", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Database APIs for biological, gene-drug associations and drug-drug similarity", "text": "Only a portion of the nodes are useful for the drug re-purposing: anything other then diseases, phenotypes, genes and drugs are not used at latter steps (Tables 1 and 2). Moreover, not all drugs can be used during the drugsimilarity.py step since not all the URIs can be mapped to known IDs to be converted into SMILES. Of the 69,907 drugs available in DGIdb, only 12,872 (18.4%) have ChEMBL or Wikidata identifiers that are potentially mappable to SMILES representations (Section\u00a0\u201cDatabase APIs for biological, gene-drug associations anddrug-drug similarity\u201d), and of these, 11,603 (16.6%) were successfully mapped (i.e. 11,596 from ChEMBL and 7 from Wikidata). This mapping limitation affects the drug similarity network construction, but it does not impact the primary gene-drug association predictions, as the drug similarity step serves to enhance negative sample selection rather than define the prediction space. Nodes and edges from the network of associations to HD that are used for drug repurposing are listed in Tables 1 and 2\u2014with counts from a reference run performed on October 7th 2025. This represents a substantial expansion compared to the network constructed by Reep, which comprised 2189 nodes and 17,467 edges using June 2022 API versions [31], demonstrating the value of updated data infrastructure and the inclusion of all DGIdb drugs\u2014rather than only those already associated with the disease of interest (Section\u00a0\u201cDatabase APIs for biological, gene-drug associations anddrug-drug similarity\u201d).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Database APIs for biological, gene-drug associations and drug-drug similarity", "text": "Compared to the full network (Fig.\u00a01), the smaller network (Fig.\u00a02) of 1000 (26.50%) genes, 600 drugs (29.97%) and 970 other mappable nodes (10.00%) generated with the same distribution of centrality scores had much lower edge density: 267 gene-to-drug edges (1.62%), 131 gene-to-gene edges (4.81%), and 513 other edges (4.81%) (Table 9). Appendix A presents the validation of the subset network as a representation of the full network for the purpose of iteration-consensus testing.Fig. 1Full network of associations to HD: highly interconnected nodes (mostly diseases in dark blue, genes in light blue and phenotypes in blue) at the centre and less interconnected nodes (mostly drugs in red and other entities in orange) at the fringes joined by biological associations (blue lines) and gene-to-drug edges (red lines); the full network comprises 15,476 nodes and 32,520 edges from the October 7, 2025 data snapshot (Tables 1 and 2)Fig. 2Network made up of a subset of nodes from the full network in Figure 1: generated according to the random sampling method described in Section \u201cPipeline upgrades\u201d; nodes are sampled to retain the same proportions according to labels in Table 1; statistical validation (Appendix A) confirms preservation of betweenness centrality distributions (i.e. p>0.05, KS-test) and representative topology, enabling iteration-based evaluation while reducing computational requirements and time frame", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Database APIs for biological, gene-drug associations and drug-drug similarity", "text": "Full network of associations to HD: highly interconnected nodes (mostly diseases in dark blue, genes in light blue and phenotypes in blue) at the centre and less interconnected nodes (mostly drugs in red and other entities in orange) at the fringes joined by biological associations (blue lines) and gene-to-drug edges (red lines); the full network comprises 15,476 nodes and 32,520 edges from the October 7, 2025 data snapshot (Tables 1 and 2)\n\nNetwork made up of a subset of nodes from the full network in Figure 1: generated according to the random sampling method described in Section \u201cPipeline upgrades\u201d; nodes are sampled to retain the same proportions according to labels in Table 1; statistical validation (Appendix A) confirms preservation of betweenness centrality distributions (i.e. p>0.05, KS-test) and representative topology, enabling iteration-based evaluation while reducing computational requirements and time frame", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Generation of semantically-valid negative samples based on centrality measures", "text": "The results from the generation of negative samples (Section\u00a0\u201cGeneration of semantically-valid negative samples based on centrality measures\u201d) vary depending on the centrality measure used: nodes that score higher will lead to more edges containing it being deemed valid negative samples. Gene node \u2018HGNC:644\u2019 (AR, Androgen Receptor) for example is the top scoring according to the degree centrality measure (0.0479), yet it is not according to eigenvector and in-betweenness centrality measures (respectively 0.2930 with max=0.3143, and 0.0739 with max=0.0852).\n\nMoreover, this step still takes into consideration hub nodes that are not relevant for this purpose: genes such as \u2018HGNC:11998\u2019 (TP53) (i.e. a transcription factor for apoptosis that is silenced in tumors) are highly interconnected because of their nature of home-keeping genes. Yet they cannot be automatically ignored due to the lack of a database that annotates them accordingly and that uses an API for mapping call requests.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Generation of semantically-valid negative samples based on centrality measures", "text": "To validate the contribution of our combined centrality and similarity-based approach, we performed an ablation study comparing four negative sampling strategies on the full HD network using GCN: random negatives as a baseline, similarity-only, centrality-only, and \u2018combined\u2019 (i.e. centrality+similarity). All strategies achieved high discrimination (i.e. AUROC 0.91\u22120.92, Table 3), with the combined approach demonstrating the best performance across metrics (i.e. AUROC=0.9175, AUPRC=0.6904, F1=0.5503, Table 3), representing modest \u2013yet consistent\u2014improvements over the other sampling methods (e.g. +0.5% AUROC, +1.8% AUPRC, +7.0% F1-score to baseline). The smaller-than-anticipated effect size (Table 3) is attributed to the design of negative samples generation: we only consider gene-to-drug edges for training and prediction, thus even randomly generated valid positive edges are informative enough to support prediction. Nonetheless, the \u2018combined\u2019 approach provides the best performance (Table 3). Our approach\u2014mediated also by Node2Vec\u2019s fit to the sparse nature of biological networks (Section\u00a0\u201cParallelisation in gene-drug embeddings\u201d)\u2014allows to reduce computational times in the prediction phase (Section\u00a0\u201cParallelisation in gene-drug embeddings\u201d). Detailed ablation methodology, comprehensive metrics from Table 3, and comparison plots are provided in Appendix D.Table 3Negative sampling ablation study summary: \u2018combined\u2019 (i.e. centrality+similarity) strategy outperforms random baseline with modest but consistent improvementsStrategyAUROCAUPRCF1-scoreRandom (baseline)0.91290.67820.5144Combined0.91750.69040.5503Improvement+0.5%+1.8%+7.0%Full results comparing all four strategies (i.e. random, similarity-only, centrality-only, combined) with comprehensive metrics and analysis are provided in Appendix D", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Generation of semantically-valid negative samples based on centrality measures", "text": "Negative sampling ablation study summary: \u2018combined\u2019 (i.e. centrality+similarity) strategy outperforms random baseline with modest but consistent improvements\n\nFull results comparing all four strategies (i.e. random, similarity-only, centrality-only, combined) with comprehensive metrics and analysis are provided in Appendix D", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Parallelisation in gene-drug embeddings", "text": "Due to the differences between the two ML methods (i.e. XGBoost and GCN) in the kind of embeddings to feed for training, the time required to building them changes quite a lot. Where the pipeline running the XGBoost method requires to compute any interaction (i.e. present as positive or valid negative, or missing as invalid negative) between gene and drug nodes (e.g. for Huntington\u2019s disease run: 3031 * 4582 = 13,888,042), the pipeline running the GCN method only requires to fill the matrix with the known positive and valid negative associations (i.e. 28,260 + 17,956 = 46,216). GCN embedding generation completes in approximately 48\u00a0s, requiring only 0.3% of the time needed for XGBoost embedding generation (i.e. approximately 4.4\u00a0h).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Parallelisation in gene-drug embeddings", "text": "Both methods also require the construction of an empty prediction matrix for all possible combinations between genes in the network and all drugs from DGIdb (i.e. 3031 * 69,907 = 211,888,117), but this step take considerably less time due to not having to iterate through edges to determine the presence or absence of an association. Nonetheless, the entire process still takes from half an hour to almost 24\u00a0h: for the purpose of saving time, the emb_l=1 parameter allows to load existing embeddings files instead of computing them provided that they exist (i.e. same disease, same date -although the date can be brute forced with the set_date() utility function at the beginning of the function for this specific purpose). On non-specialised computational hardware (e.g. laptop with 10-core CPU @3.2GHz in our case), the complete embedding generation process ranges from 30\u00a0min\u2014with GCN on the subset network\u2014to approximately 20\u00a0h\u2014with XGBoost on full network. Parallel processing reduces these times by a factor proportional to the number of available cores.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "Results from the ML step of the pipeline with the two different models (i.e. XGBoost and GCN) vary quite a lot\u2014both in runtime and prediction volume. Although the size of the network \u2013thus also the size of the node embedding training file\u2014has a role in how long and straining the training step will be, the parameter optimisation setting (i.e. ml_depth) is even more influential. The prediction itself does not take that long compared to the training. Both models have three incremental degrees of complexity for the parameter optimisation step (i.e. \u2018ultralight\u2019, \u2018light\u2019 and \u2018full\u2019). Yet, the GCN model at ml_depth=\u2018full\u2019 still takes significantly less even than the XGBoost model at ml_depth=\u2018ultralight\u2019.Table 4Quantitative performance metrics for single-run and consensus iteration predictions (Section\u00a0\u201cPrediction iterations as a solution for seed variability in ML methods\u201d): AUROC, AUPRC, and F1-score values with 95% bootstrap confidence intervals demonstrate GCN\u2019s superior discrimination ability across all configurationsModel configurationAUROC [95% CI]AUPRC [95% CI]F1P@10R@100Single-Run Performance (Full Network)GCN (raw)0.9313 [0.9222, 0.9407]0.7280 [0.6979, 0.7577]0.59221.00.1418GCN (filtered)0.9236 [0.9139, 0.9323]0.7077 [0.6752, 0.7405]0.55901.00.1400XGBoost (raw)0.3958 [0.3778, 0.4135]0.0706 [0.0649, 0.0769]0.16670.200.0069XGBoost (filtered)0.4152 [0.3950, 0.4364]0.0749 [0.0685, 0.0819]0.16670.000.0150Consensus Iterations (Subset Network)GCN (200 iterations)0.9865 [0.9764, 0.9946]0.9131 [0.8546, 0.9610]0.58991.00.9242XGBoost (10 iterations)0.4556 [0.3719, 0.5367]0.1023 [0.0678, 0.1429]0.16670.100.1667GCN achieves F1-scores comparable to Reep (F1=0.867), while Precision@10 (P@10) and Recall@100 (R@100) metrics show consensus GCN achieves perfect top-10 precision and recovers 92.42% of test edges within top-100 predictions. Filtered configurations use prob\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge $$\\end{document}0.65, cluster\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge $$\\end{document}0.8 thresholds (more details in Section\u00a0\u201cML training and prediction in enriched KG\u201d)Table 5Comparison of ML model computational performance by parameter optimisation setting: each model (i.e. GCN and XGBoost) is run with different parameter optimisation settings (\u2018ml_depth\u2019) to show the difference in increase of runtime between respectively a GNN and a boosting tree at maximum settingsModel\u2018ml_depth\u2019Avg runtimeAvg CPU usage (%)GCN (PyG)full0\u00a0h 2\u00a0m 47\u00a0s22.33GCN (PyG)light0\u00a0h 2\u00a0m 29\u00a0s14.75GCN (PyG)ultralight0\u00a0h 1\u00a0m 29\u00a0s11.00XGBoostfull3\u00a0h 19\u00a0m 43\u00a0s38.68XGBoostlight0\u00a0h 37\u00a0m 14\u00a0s12.61XGBoostultralight0\u00a0h 38\u00a0m 54\u00a0s11.91", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "Quantitative performance metrics for single-run and consensus iteration predictions (Section\u00a0\u201cPrediction iterations as a solution for seed variability in ML methods\u201d): AUROC, AUPRC, and F1-score values with 95% bootstrap confidence intervals demonstrate GCN\u2019s superior discrimination ability across all configurations\n\nGCN achieves F1-scores comparable to Reep (F1=0.867), while Precision@10 (P@10) and Recall@100 (R@100) metrics show consensus GCN achieves perfect top-10 precision and recovers 92.42% of test edges within top-100 predictions. Filtered configurations use prob\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge $$\\end{document}0.65, cluster\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge $$\\end{document}0.8 thresholds (more details in Section\u00a0\u201cML training and prediction in enriched KG\u201d)", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "Comparison of ML model computational performance by parameter optimisation setting: each model (i.e. GCN and XGBoost) is run with different parameter optimisation settings (\u2018ml_depth\u2019) to show the difference in increase of runtime between respectively a GNN and a boosting tree at maximum settings\n\nQuantitative performance metrics in Table 4 demonstrate GCN\u2019s superior discrimination ability across all model configurations, while Table 5 shows average runtimes and CPU usage for the full pipeline on the smaller artificial network with parameter emb_l=1 parameter, therefore the only steps that were performed are the ones involving the ML step. Results show that for the smaller network there is no big difference between using the \u2018light\u2019 or the \u2018ultralight\u2019 parameter optimisation degrees of complexity, and this is also true when considering just the training and prediction times with the full network.Fig. 3Performance evaluation of GCN single-run predictions on full network: Left panel shows receiver operating characteristic (ROC) curve comparing GCN performance (red) against random baseline (dotted black), demonstrating excellent discrimination ability with AUROC=0.9236. Right panel shows precision-recall curve with GCN achieving AUPRC=0.7077 (red) substantially outperforming the random baseline at 0.0909 (dotted black), confirming strong predictive performance for gene-to-drug edge prediction on the HD network", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "Performance evaluation of GCN single-run predictions on full network: Left panel shows receiver operating characteristic (ROC) curve comparing GCN performance (red) against random baseline (dotted black), demonstrating excellent discrimination ability with AUROC=0.9236. Right panel shows precision-recall curve with GCN achieving AUPRC=0.7077 (red) substantially outperforming the random baseline at 0.0909 (dotted black), confirming strong predictive performance for gene-to-drug edge prediction on the HD network", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "Single-run predictions on the full network demonstrate substantial differences between models in both prediction volume and discrimination ability. Quantitative evaluation using an 80/20 train/test split \u2013with held-out edges from the original network\u2013 reveals that GCN achieves excellent raw discrimination (i.e. AUROC=0.9313 [0.9222, 0.9407], AUPRC=0.7280 [0.6979, 0.7577], F1=0.5922) while XGBoost shows limited discrimination ability (i.e. AUROC=0.3958 [0.3778, 0.4135], AUPRC=0.0706 [0.0649, 0.0769], F1=0.1667) (Fig.\u00a03). When probability filtering is applied (i.e. prob>0.65, cluster>0.8), GCN maintains strong performance (i.e. AUROC=0.9236 [0.9139, 0.9323], AUPRC=0.7077 [0.6752, 0.7405], F1=0.5590) with perfect Precision@10, while XGBoost performance shows mixed results (AUROC=0.4152 [0.3950, 0.4364], AUPRC=0.0749 [0.0685, 0.0819], F1=0.1667) (Fig.\u00a03). Compared to Reep\u2019s reported F1-score of 0.867, our models return lower values due to architectural differences: in fact compared to their version that only predicted repurposed drugs from a smaller network only with drugs already associated to the disease of interest, our model takes into consideration a much larger and more sparse network where predictions are hard to be quantified with F1 scores. Moreover, our dual-model approach with consensus-based ranking provides additional robustness through model complementarity as demonstrated in Section\u00a0\u201cPrediction iterations as a solution for seed variability in ML methods\u201d and in Appendix C.Fig. 4XGBoost single run prediction: Full network of associations to HD from Figure 1 showing newly predicted gene-to-drug edges in green via the XGBoost model as described in Section \u201cPrediction of repurposed drugs in binary models with non-binary datasets\u201dFig. 5GCN single run prediction: Full network of associations to HD from Figure 1 showing newly predicted gene-to-drug edges in green via the GCN model as described in Section \u201cPrediction of repurposed drugs in binary models with non-binary datasets\u201d", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "XGBoost single run prediction: Full network of associations to HD from Figure 1 showing newly predicted gene-to-drug edges in green via the XGBoost model as described in Section \u201cPrediction of repurposed drugs in binary models with non-binary datasets\u201d\n\nGCN single run prediction: Full network of associations to HD from Figure 1 showing newly predicted gene-to-drug edges in green via the GCN model as described in Section \u201cPrediction of repurposed drugs in binary models with non-binary datasets\u201d\n\nIn terms of prediction volume, starting from a network with 251 positive gene-to-drug edges and approximately 5\u00a0M possible gene-to-drug edges, the two models exhibit markedly different behavior: XGBoost returns between 20 and 160 predicted edges respectively with 0.95 and 0.6 probability filter, while GCN returns approximately 120,000 (0.95 filter) to 500,000 (0.65 filter) edges. This indicates that GCN is substantially more lenient in positive prediction even at high probability thresholds.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "To evaluate whether these architecturally distinct models converge on similar drug candidates, we performed convergence analysis using Jaccard similarity across top-K rankings. Under standard filtering thresholds (i.e. prob\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge 0.65$$\\end{document}, cluster\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge 0.8$$\\end{document}; more details in Section \u201cML training and prediction in enriched KG\u201d), single-run predictions showed minimal overlap due to XGBoost\u2019s limited prediction volume of 3 drugs compared to GCN with 298 drugs, with Jaccard similarity near zero for top-K rankings. However, this divergence appears to be an artifact of filtering rather than fundamental model disagreement (detailed analysis in Appendix C).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "Individual runs with GCN return many predicted repurposed drugs (i.e. \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$> 1100$$\\end{document}) with little edges associated to them (i.e. \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\le 10$$\\end{document}) (Fig.\u00a05), while individual runs with XGBoost return fewer predicted repurposed drugs (i.e. 5\u201340) with many more edges associated to them (i.e. \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$> 200$$\\end{document}) (Fig.\u00a04). Among the latter, several are associated to genes involved in neurological diseases (e.g. RECQL gene involved in Alzheimer\u2019s disease to Koparin, SMN1 and SMN2 genes involved in spinal muscular atrophy to Methyl Robustone) according to validation via PubChem (National Center for Biotechnology Information, part of the USA National Institute of Health and National Library of Medicine). Other predicted repurposed drugs were only associated to unrelated genes.Fig. 6Prediction agreement between XGBoost and GCN on full network single runs: Venn diagrams showing overlap of drugs from unfiltered predicted gene-to-drug edges classified by both models. XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} and GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} represent drugs from edges predicted as positive (prob\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge $$\\end{document}0.5) by XGBoost and GCN respectively, while XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^-$$\\end{document} and GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^-$$\\end{document} represent drugs predicted from edges below these thresholds. The XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document}\n\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\cap $$\\end{document} GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} intersection (n=2002) indicates drugs with high confidence from both models, while XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document}\n\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\cap $$\\end{document} GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^-$$\\end{document} (n=4187) reveal model-specific predictions", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "Prediction agreement between XGBoost and GCN on full network single runs: Venn diagrams showing overlap of drugs from unfiltered predicted gene-to-drug edges classified by both models. XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} and GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} represent drugs from edges predicted as positive (prob\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge $$\\end{document}0.5) by XGBoost and GCN respectively, while XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^-$$\\end{document} and GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^-$$\\end{document} represent drugs predicted from edges below these thresholds. The XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document}\n\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\cap $$\\end{document} GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} intersection (n=2002) indicates drugs with high confidence from both models, while XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document}\n\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\cap $$\\end{document} GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^-$$\\end{document} (n=4187) reveal model-specific predictions", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "Convergence analysis across both single-run and iteration-based predictions (Section\u00a0\u201cPrediction iterations as a solution for seed variability in ML methods\u201d) identified several drugs predicted consistently by both models across different evaluation frameworks, suggesting robust cross-model agreement on high-priority candidates (Fig.\u00a06, detailed analysis in Appendix C).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "To assess prediction stability and enable consensus ranking across random seeds, we performed iteration-based evaluation on the validated subset network (Appendix A): 10 iterations for XGBoost (i.e. 7.8\u00a0h total, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\sim 47$$\\end{document} minutes per iteration) and 200 iterations for GCN (i.e. 5.5\u00a0h total, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\sim 35$$\\end{document} seconds per iteration). The occurrences of predicted gene-to-drug edges across iterations exhibit distinct distributional patterns (Figs.\u00a07 and 8): XGBoost predictions display a binomial distribution spanning from 0 to the maximum number of occurrences (Fig.\u00a09), while GCN predictions show no specific distribution with all predicted edges appearing in at least 182 out of 200 iterations (Fig.\u00a010), indicating substantially higher prediction stability.Fig. 7XGBoost 10-iteration run prediction: Subset network of associations to HD from Fig.\u00a03 showing newly predicted gene-to-drug edges in green via 10 iterations of the XGBoost model as described in Section \"Prediction iterations as a solution for seed variability in ML methods\" (detailed distribution in Fig.\u00a09)Fig. 8GCN 200-iteration run prediction: Subset network of associations to HD from Fig.\u00a03 showing newly predicted gene-to-drug edges in green via 200 iterations of the GCN model as described in Section \"Prediction iterations as a solution for seed variability in ML methods\" (detailed distribution in Fig.\u00a010)Fig. 9Distribution of occurrences for predicted edges in XGBoost 10-iteration run: predicted edges follow a binomial distribution, with few edges being predicted consistently in all 10 iterations; this distribution indicates substantial seed-dependent variability in XGBoost predictions, with most edges appearing in fewer iterations (see Appendix B for detailed stability analysis)Fig. 10Distribution of occurrences for predicted edges in GCN 200-iteration run: predicted edges do not follow a specific distribution with all edges at least being predicted in 182 off all 200 iterations (i.e. 91% minimum occurrence rate), and the majority of edges predicted in all 200 iterations. This highly skewed distribution (i.e. all edges being predicted in at least 182 runs) indicates substantially higher prediction stability for GCN compared to XGBoost (Fig.\u00a09) (see Appendix B for detailed stability analysis)", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "XGBoost 10-iteration run prediction: Subset network of associations to HD from Fig.\u00a03 showing newly predicted gene-to-drug edges in green via 10 iterations of the XGBoost model as described in Section \"Prediction iterations as a solution for seed variability in ML methods\" (detailed distribution in Fig.\u00a09)\n\nGCN 200-iteration run prediction: Subset network of associations to HD from Fig.\u00a03 showing newly predicted gene-to-drug edges in green via 200 iterations of the GCN model as described in Section \"Prediction iterations as a solution for seed variability in ML methods\" (detailed distribution in Fig.\u00a010)\n\nDistribution of occurrences for predicted edges in XGBoost 10-iteration run: predicted edges follow a binomial distribution, with few edges being predicted consistently in all 10 iterations; this distribution indicates substantial seed-dependent variability in XGBoost predictions, with most edges appearing in fewer iterations (see Appendix B for detailed stability analysis)", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "Distribution of occurrences for predicted edges in GCN 200-iteration run: predicted edges do not follow a specific distribution with all edges at least being predicted in 182 off all 200 iterations (i.e. 91% minimum occurrence rate), and the majority of edges predicted in all 200 iterations. This highly skewed distribution (i.e. all edges being predicted in at least 182 runs) indicates substantially higher prediction stability for GCN compared to XGBoost (Fig.\u00a09) (see Appendix B for detailed stability analysis)\n\nConsensus predictions from multiple iterations demonstrate dramatically different performance characteristics (Table 4). For GCN, consensus across 200 iterations achieves excellent discrimination (i.e. AUROC=0.9865 [0.9764, 0.9946], AUPRC=0.9131 [0.8546, 0.9610]) with outstanding top-K performance (i.e. Precision@10=1.0, Recall@100=0.9242), recovering 92.42% of test edges among the top 100 predictions. In contrast, XGBoost consensus across 10 iterations shows poor discrimination (i.e. AUROC=0.4556 [0.3719, 0.5367], AUPRC=0.1023 [0.0678, 0.1429]) with limited recall (i.e. Recall@100=0.1667), indicating that XGBoost suffers more from lower link prediction efficiency in 80/20 train/split conditions than GCN. We hypothise that the removal of that many positive gene-to-drug edges deeply affect the outcome of prediction, due to the sparce nature of biological networks\u2014especially in RDs.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "Figure\u00a011 presents a heatmap comparing prediction confidences for 20 selected drugs across different evaluation frameworks (single-run on full network, single-run on subset network, and iteration-based consensus on subset network for both XGBoost and GCN). The heatmap reveals systematic patterns in prediction confidence: drugs consistently predicted by both models across evaluation frameworks represent high-confidence candidates (e.g. Ergocristine, Fig.\u00a011), while drugs with framework-specific confidence patterns highlight the complementary nature of single-run and iteration-based approaches.Fig. 11Prediction confidence for 20 selected drugs across evaluation frameworks: Heatmap showing prediction probabilities for drugs identified through convergence analysis, evaluated across XGBoost and GCN using both the single-run approach on both full network and subset network, and the iteration-based consensus approach on the subset network. Colour represents prediction confidence (darker red = higher probability). Consistent high-confidence predictions across all six frameworks represent robust candidates, while framework-specific patterns reveal differences in model calibration and the stabilising effect of iteration-based consensus. Notably, Ergocristine has higher confidence in the GCN iteration consensus approach on the subset network than in the single run approach on the full network, in support of our thesis that iterations can lead to candidates that are overlooked over single runs. White cells indicate random predictions", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "Prediction confidence for 20 selected drugs across evaluation frameworks: Heatmap showing prediction probabilities for drugs identified through convergence analysis, evaluated across XGBoost and GCN using both the single-run approach on both full network and subset network, and the iteration-based consensus approach on the subset network. Colour represents prediction confidence (darker red = higher probability). Consistent high-confidence predictions across all six frameworks represent robust candidates, while framework-specific patterns reveal differences in model calibration and the stabilising effect of iteration-based consensus. Notably, Ergocristine has higher confidence in the GCN iteration consensus approach on the subset network than in the single run approach on the full network, in support of our thesis that iterations can lead to candidates that are overlooked over single runs. White cells indicate random predictions", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "The iteration-based consensus approach reveals an important pattern: some drugs without obvious disease-gene associations may still hold therapeutic relevance when evaluated through pharmacological mechanisms rather than network proximity alone. Ergocristine (CHEMBL1201247) exemplifies this pattern. While not being directly associated with HD-related genes in public databases, Ergocristine demonstrated higher and more stable prediction confidence in iteration-based consensus (i.e. 200 iterations on GCN) compared to single-run predictions on both full and subset networks (detailed probability distributions in Appendix C). Notably, Ergocristine emerged from the convergence analysis as one of the drugs predicted by both XGBoost and GCN across different evaluation frameworks: it appears in the overlap between models on both full network single runs and subset network iteration runs, demonstrating robust cross-model agreement. According to PubChem classification, Ergocristine functions as a Peripheral Nervous System Agent with mechanisms involving neurotransmitter modulation and motor pathway regulation\u2013pharmacologically relevant to HD given the disease\u2019s characteristic GABAergic medium spiny neuron degeneration and motor dysfunction [2]. While this higher prediction confidence may be coincidental, it also suggests that iteration-based consensus could distinguish genuine signal from stochastic noise\u2013and surface biologically plausible candidates that might be overlooked or underweighted in single-run predictions.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "Detailed iteration analysis\u2014including edge occurrence distributions, per-iteration variability, and filtering comparisons\u2013 is provided in Appendix B. Analysis of five hand-picked held-out edges involving key HD-associated genes (TUBB3, GSK3B, HTT, FASN) confirms this pattern: GCN recovers 4 out of 5 edges within its top predictions, while XGBoost recovers 1 out of 5. This demonstrates the substantial difference in recall capability between the models (detailed per-edge recovery analysis in Appendix B).Fig. 12Consensus prediction agreement between models on subset network: Venn diagram comparing drugs from unfiltered positive predictions from XGBoost (10 iterations) and all predictions from GCN (200 iterations) on the subset network. Only positive consensus predictions for XGBoost are stored in iterations. The overlap XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document}\n\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\cap $$\\end{document} GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} (n=591) represents 17.74% of XGBoost predicted drugs and 100% of GCN drugs from positive predictions, demonstrating moderate convergence on high-priority drug candidates despite different iteration counts and model architectures. Model-specific predictions (XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} only: n=2709) reflect differences in confidence calibration and prediction volume. Detailed top-K overlap analysis is provided in Appendix C (Fig.\u00a020)", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "Consensus prediction agreement between models on subset network: Venn diagram comparing drugs from unfiltered positive predictions from XGBoost (10 iterations) and all predictions from GCN (200 iterations) on the subset network. Only positive consensus predictions for XGBoost are stored in iterations. The overlap XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document}\n\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\cap $$\\end{document} GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} (n=591) represents 17.74% of XGBoost predicted drugs and 100% of GCN drugs from positive predictions, demonstrating moderate convergence on high-priority drug candidates despite different iteration counts and model architectures. Model-specific predictions (XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} only: n=2709) reflect differences in confidence calibration and prediction volume. Detailed top-K overlap analysis is provided in Appendix C (Fig.\u00a020)", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "Convergence analysis using relaxed thresholds (Appendix C) confirms that filtering creates artificial divergence: when analysed without probability filters, consensus predictions from both models demonstrate strong agreement at top ranks\u2013100% overlap at top-10 (i.e. Jaccard=1.0), 95% at top-20, 94% at top-50, and 96% at top-100 (Fig.\u00a012). This indicates that despite substantial differences in discrimination ability and confidence calibration, both models fundamentally identify similar high-priority drug candidates for HD.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction of repurposed drugs in binary models with non-binary datasets", "text": "Results from the ML step of the pipeline with the two different models (i.e. XGBoost and GCN) vary quite a lot\u2014both in runtime and prediction volume. Although the size of the network \u2013thus also the size of the node embedding training file\u2014has a role in how long and straining the training step will be, the parameter optimisation setting (i.e. ml_depth) is even more influential. The prediction itself does not take that long compared to the training. Both models have three incremental degrees of complexity for the parameter optimisation step (i.e. \u2018ultralight\u2019, \u2018light\u2019 and \u2018full\u2019). Yet, the GCN model at ml_depth=\u2018full\u2019 still takes significantly less even than the XGBoost model at ml_depth=\u2018ultralight\u2019.Table 4Quantitative performance metrics for single-run and consensus iteration predictions (Section\u00a0\u201cPrediction iterations as a solution for seed variability in ML methods\u201d): AUROC, AUPRC, and F1-score values with 95% bootstrap confidence intervals demonstrate GCN\u2019s superior discrimination ability across all configurationsModel configurationAUROC [95% CI]AUPRC [95% CI]F1P@10R@100Single-Run Performance (Full Network)GCN (raw)0.9313 [0.9222, 0.9407]0.7280 [0.6979, 0.7577]0.59221.00.1418GCN (filtered)0.9236 [0.9139, 0.9323]0.7077 [0.6752, 0.7405]0.55901.00.1400XGBoost (raw)0.3958 [0.3778, 0.4135]0.0706 [0.0649, 0.0769]0.16670.200.0069XGBoost (filtered)0.4152 [0.3950, 0.4364]0.0749 [0.0685, 0.0819]0.16670.000.0150Consensus Iterations (Subset Network)GCN (200 iterations)0.9865 [0.9764, 0.9946]0.9131 [0.8546, 0.9610]0.58991.00.9242XGBoost (10 iterations)0.4556 [0.3719, 0.5367]0.1023 [0.0678, 0.1429]0.16670.100.1667GCN achieves F1-scores comparable to Reep (F1=0.867), while Precision@10 (P@10) and Recall@100 (R@100) metrics show consensus GCN achieves perfect top-10 precision and recovers 92.42% of test edges within top-100 predictions. Filtered configurations use prob\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge $$\\end{document}0.65, cluster\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge $$\\end{document}0.8 thresholds (more details in Section\u00a0\u201cML training and prediction in enriched KG\u201d)Table 5Comparison of ML model computational performance by parameter optimisation setting: each model (i.e. GCN and XGBoost) is run with different parameter optimisation settings (\u2018ml_depth\u2019) to show the difference in increase of runtime between respectively a GNN and a boosting tree at maximum settingsModel\u2018ml_depth\u2019Avg runtimeAvg CPU usage (%)GCN (PyG)full0\u00a0h 2\u00a0m 47\u00a0s22.33GCN (PyG)light0\u00a0h 2\u00a0m 29\u00a0s14.75GCN (PyG)ultralight0\u00a0h 1\u00a0m 29\u00a0s11.00XGBoostfull3\u00a0h 19\u00a0m 43\u00a0s38.68XGBoostlight0\u00a0h 37\u00a0m 14\u00a0s12.61XGBoostultralight0\u00a0h 38\u00a0m 54\u00a0s11.91", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction of repurposed drugs in binary models with non-binary datasets", "text": "Quantitative performance metrics for single-run and consensus iteration predictions (Section\u00a0\u201cPrediction iterations as a solution for seed variability in ML methods\u201d): AUROC, AUPRC, and F1-score values with 95% bootstrap confidence intervals demonstrate GCN\u2019s superior discrimination ability across all configurations\n\nGCN achieves F1-scores comparable to Reep (F1=0.867), while Precision@10 (P@10) and Recall@100 (R@100) metrics show consensus GCN achieves perfect top-10 precision and recovers 92.42% of test edges within top-100 predictions. Filtered configurations use prob\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge $$\\end{document}0.65, cluster\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge $$\\end{document}0.8 thresholds (more details in Section\u00a0\u201cML training and prediction in enriched KG\u201d)", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction of repurposed drugs in binary models with non-binary datasets", "text": "Comparison of ML model computational performance by parameter optimisation setting: each model (i.e. GCN and XGBoost) is run with different parameter optimisation settings (\u2018ml_depth\u2019) to show the difference in increase of runtime between respectively a GNN and a boosting tree at maximum settings\n\nQuantitative performance metrics in Table 4 demonstrate GCN\u2019s superior discrimination ability across all model configurations, while Table 5 shows average runtimes and CPU usage for the full pipeline on the smaller artificial network with parameter emb_l=1 parameter, therefore the only steps that were performed are the ones involving the ML step. Results show that for the smaller network there is no big difference between using the \u2018light\u2019 or the \u2018ultralight\u2019 parameter optimisation degrees of complexity, and this is also true when considering just the training and prediction times with the full network.Fig. 3Performance evaluation of GCN single-run predictions on full network: Left panel shows receiver operating characteristic (ROC) curve comparing GCN performance (red) against random baseline (dotted black), demonstrating excellent discrimination ability with AUROC=0.9236. Right panel shows precision-recall curve with GCN achieving AUPRC=0.7077 (red) substantially outperforming the random baseline at 0.0909 (dotted black), confirming strong predictive performance for gene-to-drug edge prediction on the HD network", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction of repurposed drugs in binary models with non-binary datasets", "text": "Performance evaluation of GCN single-run predictions on full network: Left panel shows receiver operating characteristic (ROC) curve comparing GCN performance (red) against random baseline (dotted black), demonstrating excellent discrimination ability with AUROC=0.9236. Right panel shows precision-recall curve with GCN achieving AUPRC=0.7077 (red) substantially outperforming the random baseline at 0.0909 (dotted black), confirming strong predictive performance for gene-to-drug edge prediction on the HD network", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction of repurposed drugs in binary models with non-binary datasets", "text": "Single-run predictions on the full network demonstrate substantial differences between models in both prediction volume and discrimination ability. Quantitative evaluation using an 80/20 train/test split \u2013with held-out edges from the original network\u2013 reveals that GCN achieves excellent raw discrimination (i.e. AUROC=0.9313 [0.9222, 0.9407], AUPRC=0.7280 [0.6979, 0.7577], F1=0.5922) while XGBoost shows limited discrimination ability (i.e. AUROC=0.3958 [0.3778, 0.4135], AUPRC=0.0706 [0.0649, 0.0769], F1=0.1667) (Fig.\u00a03). When probability filtering is applied (i.e. prob>0.65, cluster>0.8), GCN maintains strong performance (i.e. AUROC=0.9236 [0.9139, 0.9323], AUPRC=0.7077 [0.6752, 0.7405], F1=0.5590) with perfect Precision@10, while XGBoost performance shows mixed results (AUROC=0.4152 [0.3950, 0.4364], AUPRC=0.0749 [0.0685, 0.0819], F1=0.1667) (Fig.\u00a03). Compared to Reep\u2019s reported F1-score of 0.867, our models return lower values due to architectural differences: in fact compared to their version that only predicted repurposed drugs from a smaller network only with drugs already associated to the disease of interest, our model takes into consideration a much larger and more sparse network where predictions are hard to be quantified with F1 scores. Moreover, our dual-model approach with consensus-based ranking provides additional robustness through model complementarity as demonstrated in Section\u00a0\u201cPrediction iterations as a solution for seed variability in ML methods\u201d and in Appendix C.Fig. 4XGBoost single run prediction: Full network of associations to HD from Figure 1 showing newly predicted gene-to-drug edges in green via the XGBoost model as described in Section \u201cPrediction of repurposed drugs in binary models with non-binary datasets\u201dFig. 5GCN single run prediction: Full network of associations to HD from Figure 1 showing newly predicted gene-to-drug edges in green via the GCN model as described in Section \u201cPrediction of repurposed drugs in binary models with non-binary datasets\u201d", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction of repurposed drugs in binary models with non-binary datasets", "text": "XGBoost single run prediction: Full network of associations to HD from Figure 1 showing newly predicted gene-to-drug edges in green via the XGBoost model as described in Section \u201cPrediction of repurposed drugs in binary models with non-binary datasets\u201d\n\nGCN single run prediction: Full network of associations to HD from Figure 1 showing newly predicted gene-to-drug edges in green via the GCN model as described in Section \u201cPrediction of repurposed drugs in binary models with non-binary datasets\u201d\n\nIn terms of prediction volume, starting from a network with 251 positive gene-to-drug edges and approximately 5\u00a0M possible gene-to-drug edges, the two models exhibit markedly different behavior: XGBoost returns between 20 and 160 predicted edges respectively with 0.95 and 0.6 probability filter, while GCN returns approximately 120,000 (0.95 filter) to 500,000 (0.65 filter) edges. This indicates that GCN is substantially more lenient in positive prediction even at high probability thresholds.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction of repurposed drugs in binary models with non-binary datasets", "text": "To evaluate whether these architecturally distinct models converge on similar drug candidates, we performed convergence analysis using Jaccard similarity across top-K rankings. Under standard filtering thresholds (i.e. prob\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge 0.65$$\\end{document}, cluster\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge 0.8$$\\end{document}; more details in Section \u201cML training and prediction in enriched KG\u201d), single-run predictions showed minimal overlap due to XGBoost\u2019s limited prediction volume of 3 drugs compared to GCN with 298 drugs, with Jaccard similarity near zero for top-K rankings. However, this divergence appears to be an artifact of filtering rather than fundamental model disagreement (detailed analysis in Appendix C).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction of repurposed drugs in binary models with non-binary datasets", "text": "Individual runs with GCN return many predicted repurposed drugs (i.e. \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$> 1100$$\\end{document}) with little edges associated to them (i.e. \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\le 10$$\\end{document}) (Fig.\u00a05), while individual runs with XGBoost return fewer predicted repurposed drugs (i.e. 5\u201340) with many more edges associated to them (i.e. \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$> 200$$\\end{document}) (Fig.\u00a04). Among the latter, several are associated to genes involved in neurological diseases (e.g. RECQL gene involved in Alzheimer\u2019s disease to Koparin, SMN1 and SMN2 genes involved in spinal muscular atrophy to Methyl Robustone) according to validation via PubChem (National Center for Biotechnology Information, part of the USA National Institute of Health and National Library of Medicine). Other predicted repurposed drugs were only associated to unrelated genes.Fig. 6Prediction agreement between XGBoost and GCN on full network single runs: Venn diagrams showing overlap of drugs from unfiltered predicted gene-to-drug edges classified by both models. XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} and GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} represent drugs from edges predicted as positive (prob\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge $$\\end{document}0.5) by XGBoost and GCN respectively, while XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^-$$\\end{document} and GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^-$$\\end{document} represent drugs predicted from edges below these thresholds. The XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document}\n\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\cap $$\\end{document} GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} intersection (n=2002) indicates drugs with high confidence from both models, while XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document}\n\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\cap $$\\end{document} GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^-$$\\end{document} (n=4187) reveal model-specific predictions", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction of repurposed drugs in binary models with non-binary datasets", "text": "Prediction agreement between XGBoost and GCN on full network single runs: Venn diagrams showing overlap of drugs from unfiltered predicted gene-to-drug edges classified by both models. XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} and GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} represent drugs from edges predicted as positive (prob\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\ge $$\\end{document}0.5) by XGBoost and GCN respectively, while XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^-$$\\end{document} and GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^-$$\\end{document} represent drugs predicted from edges below these thresholds. The XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document}\n\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\cap $$\\end{document} GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} intersection (n=2002) indicates drugs with high confidence from both models, while XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document}\n\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\cap $$\\end{document} GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^-$$\\end{document} (n=4187) reveal model-specific predictions", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction of repurposed drugs in binary models with non-binary datasets", "text": "Convergence analysis across both single-run and iteration-based predictions (Section\u00a0\u201cPrediction iterations as a solution for seed variability in ML methods\u201d) identified several drugs predicted consistently by both models across different evaluation frameworks, suggesting robust cross-model agreement on high-priority candidates (Fig.\u00a06, detailed analysis in Appendix C).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction iterations as a solution for seed variability in ML methods", "text": "To assess prediction stability and enable consensus ranking across random seeds, we performed iteration-based evaluation on the validated subset network (Appendix A): 10 iterations for XGBoost (i.e. 7.8\u00a0h total, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\sim 47$$\\end{document} minutes per iteration) and 200 iterations for GCN (i.e. 5.5\u00a0h total, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\sim 35$$\\end{document} seconds per iteration). The occurrences of predicted gene-to-drug edges across iterations exhibit distinct distributional patterns (Figs.\u00a07 and 8): XGBoost predictions display a binomial distribution spanning from 0 to the maximum number of occurrences (Fig.\u00a09), while GCN predictions show no specific distribution with all predicted edges appearing in at least 182 out of 200 iterations (Fig.\u00a010), indicating substantially higher prediction stability.Fig. 7XGBoost 10-iteration run prediction: Subset network of associations to HD from Fig.\u00a03 showing newly predicted gene-to-drug edges in green via 10 iterations of the XGBoost model as described in Section \"Prediction iterations as a solution for seed variability in ML methods\" (detailed distribution in Fig.\u00a09)Fig. 8GCN 200-iteration run prediction: Subset network of associations to HD from Fig.\u00a03 showing newly predicted gene-to-drug edges in green via 200 iterations of the GCN model as described in Section \"Prediction iterations as a solution for seed variability in ML methods\" (detailed distribution in Fig.\u00a010)Fig. 9Distribution of occurrences for predicted edges in XGBoost 10-iteration run: predicted edges follow a binomial distribution, with few edges being predicted consistently in all 10 iterations; this distribution indicates substantial seed-dependent variability in XGBoost predictions, with most edges appearing in fewer iterations (see Appendix B for detailed stability analysis)Fig. 10Distribution of occurrences for predicted edges in GCN 200-iteration run: predicted edges do not follow a specific distribution with all edges at least being predicted in 182 off all 200 iterations (i.e. 91% minimum occurrence rate), and the majority of edges predicted in all 200 iterations. This highly skewed distribution (i.e. all edges being predicted in at least 182 runs) indicates substantially higher prediction stability for GCN compared to XGBoost (Fig.\u00a09) (see Appendix B for detailed stability analysis)", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction iterations as a solution for seed variability in ML methods", "text": "XGBoost 10-iteration run prediction: Subset network of associations to HD from Fig.\u00a03 showing newly predicted gene-to-drug edges in green via 10 iterations of the XGBoost model as described in Section \"Prediction iterations as a solution for seed variability in ML methods\" (detailed distribution in Fig.\u00a09)\n\nGCN 200-iteration run prediction: Subset network of associations to HD from Fig.\u00a03 showing newly predicted gene-to-drug edges in green via 200 iterations of the GCN model as described in Section \"Prediction iterations as a solution for seed variability in ML methods\" (detailed distribution in Fig.\u00a010)\n\nDistribution of occurrences for predicted edges in XGBoost 10-iteration run: predicted edges follow a binomial distribution, with few edges being predicted consistently in all 10 iterations; this distribution indicates substantial seed-dependent variability in XGBoost predictions, with most edges appearing in fewer iterations (see Appendix B for detailed stability analysis)", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction iterations as a solution for seed variability in ML methods", "text": "Distribution of occurrences for predicted edges in GCN 200-iteration run: predicted edges do not follow a specific distribution with all edges at least being predicted in 182 off all 200 iterations (i.e. 91% minimum occurrence rate), and the majority of edges predicted in all 200 iterations. This highly skewed distribution (i.e. all edges being predicted in at least 182 runs) indicates substantially higher prediction stability for GCN compared to XGBoost (Fig.\u00a09) (see Appendix B for detailed stability analysis)\n\nConsensus predictions from multiple iterations demonstrate dramatically different performance characteristics (Table 4). For GCN, consensus across 200 iterations achieves excellent discrimination (i.e. AUROC=0.9865 [0.9764, 0.9946], AUPRC=0.9131 [0.8546, 0.9610]) with outstanding top-K performance (i.e. Precision@10=1.0, Recall@100=0.9242), recovering 92.42% of test edges among the top 100 predictions. In contrast, XGBoost consensus across 10 iterations shows poor discrimination (i.e. AUROC=0.4556 [0.3719, 0.5367], AUPRC=0.1023 [0.0678, 0.1429]) with limited recall (i.e. Recall@100=0.1667), indicating that XGBoost suffers more from lower link prediction efficiency in 80/20 train/split conditions than GCN. We hypothise that the removal of that many positive gene-to-drug edges deeply affect the outcome of prediction, due to the sparce nature of biological networks\u2014especially in RDs.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction iterations as a solution for seed variability in ML methods", "text": "Figure\u00a011 presents a heatmap comparing prediction confidences for 20 selected drugs across different evaluation frameworks (single-run on full network, single-run on subset network, and iteration-based consensus on subset network for both XGBoost and GCN). The heatmap reveals systematic patterns in prediction confidence: drugs consistently predicted by both models across evaluation frameworks represent high-confidence candidates (e.g. Ergocristine, Fig.\u00a011), while drugs with framework-specific confidence patterns highlight the complementary nature of single-run and iteration-based approaches.Fig. 11Prediction confidence for 20 selected drugs across evaluation frameworks: Heatmap showing prediction probabilities for drugs identified through convergence analysis, evaluated across XGBoost and GCN using both the single-run approach on both full network and subset network, and the iteration-based consensus approach on the subset network. Colour represents prediction confidence (darker red = higher probability). Consistent high-confidence predictions across all six frameworks represent robust candidates, while framework-specific patterns reveal differences in model calibration and the stabilising effect of iteration-based consensus. Notably, Ergocristine has higher confidence in the GCN iteration consensus approach on the subset network than in the single run approach on the full network, in support of our thesis that iterations can lead to candidates that are overlooked over single runs. White cells indicate random predictions", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction iterations as a solution for seed variability in ML methods", "text": "Prediction confidence for 20 selected drugs across evaluation frameworks: Heatmap showing prediction probabilities for drugs identified through convergence analysis, evaluated across XGBoost and GCN using both the single-run approach on both full network and subset network, and the iteration-based consensus approach on the subset network. Colour represents prediction confidence (darker red = higher probability). Consistent high-confidence predictions across all six frameworks represent robust candidates, while framework-specific patterns reveal differences in model calibration and the stabilising effect of iteration-based consensus. Notably, Ergocristine has higher confidence in the GCN iteration consensus approach on the subset network than in the single run approach on the full network, in support of our thesis that iterations can lead to candidates that are overlooked over single runs. White cells indicate random predictions", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction iterations as a solution for seed variability in ML methods", "text": "The iteration-based consensus approach reveals an important pattern: some drugs without obvious disease-gene associations may still hold therapeutic relevance when evaluated through pharmacological mechanisms rather than network proximity alone. Ergocristine (CHEMBL1201247) exemplifies this pattern. While not being directly associated with HD-related genes in public databases, Ergocristine demonstrated higher and more stable prediction confidence in iteration-based consensus (i.e. 200 iterations on GCN) compared to single-run predictions on both full and subset networks (detailed probability distributions in Appendix C). Notably, Ergocristine emerged from the convergence analysis as one of the drugs predicted by both XGBoost and GCN across different evaluation frameworks: it appears in the overlap between models on both full network single runs and subset network iteration runs, demonstrating robust cross-model agreement. According to PubChem classification, Ergocristine functions as a Peripheral Nervous System Agent with mechanisms involving neurotransmitter modulation and motor pathway regulation\u2013pharmacologically relevant to HD given the disease\u2019s characteristic GABAergic medium spiny neuron degeneration and motor dysfunction [2]. While this higher prediction confidence may be coincidental, it also suggests that iteration-based consensus could distinguish genuine signal from stochastic noise\u2013and surface biologically plausible candidates that might be overlooked or underweighted in single-run predictions.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction iterations as a solution for seed variability in ML methods", "text": "Detailed iteration analysis\u2014including edge occurrence distributions, per-iteration variability, and filtering comparisons\u2013 is provided in Appendix B. Analysis of five hand-picked held-out edges involving key HD-associated genes (TUBB3, GSK3B, HTT, FASN) confirms this pattern: GCN recovers 4 out of 5 edges within its top predictions, while XGBoost recovers 1 out of 5. This demonstrates the substantial difference in recall capability between the models (detailed per-edge recovery analysis in Appendix B).Fig. 12Consensus prediction agreement between models on subset network: Venn diagram comparing drugs from unfiltered positive predictions from XGBoost (10 iterations) and all predictions from GCN (200 iterations) on the subset network. Only positive consensus predictions for XGBoost are stored in iterations. The overlap XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document}\n\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\cap $$\\end{document} GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} (n=591) represents 17.74% of XGBoost predicted drugs and 100% of GCN drugs from positive predictions, demonstrating moderate convergence on high-priority drug candidates despite different iteration counts and model architectures. Model-specific predictions (XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} only: n=2709) reflect differences in confidence calibration and prediction volume. Detailed top-K overlap analysis is provided in Appendix C (Fig.\u00a020)", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction iterations as a solution for seed variability in ML methods", "text": "Consensus prediction agreement between models on subset network: Venn diagram comparing drugs from unfiltered positive predictions from XGBoost (10 iterations) and all predictions from GCN (200 iterations) on the subset network. Only positive consensus predictions for XGBoost are stored in iterations. The overlap XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document}\n\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\cap $$\\end{document} GCN\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} (n=591) represents 17.74% of XGBoost predicted drugs and 100% of GCN drugs from positive predictions, demonstrating moderate convergence on high-priority drug candidates despite different iteration counts and model architectures. Model-specific predictions (XGB\\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$^+$$\\end{document} only: n=2709) reflect differences in confidence calibration and prediction volume. Detailed top-K overlap analysis is provided in Appendix C (Fig.\u00a020)", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction iterations as a solution for seed variability in ML methods", "text": "Convergence analysis using relaxed thresholds (Appendix C) confirms that filtering creates artificial divergence: when analysed without probability filters, consensus predictions from both models demonstrate strong agreement at top ranks\u2013100% overlap at top-10 (i.e. Jaccard=1.0), 95% at top-20, 94% at top-50, and 96% at top-100 (Fig.\u00a012). This indicates that despite substantial differences in discrimination ability and confidence calibration, both models fundamentally identify similar high-priority drug candidates for HD.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Methods", "text": "The aim of this work was to develop a structured and modular approach for drug repurposing pipelines that can be implemented in cloud computing environments\u2014enabling scalability\u2014and that can be modified by replacing or adding modules with different functionalities\u2014enabling reusability. Several steps also accept specific seeds for non-deterministic operations, or store the randomly generated ones if none are provided.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Methods", "text": "The entire pipeline has been divided into Python scripts (Table 6) and functional models (Fig.\u00a013).Fig. 13Pipeline modules: each block represents one of the functional modules from Table 6 that focuses on a specific set of operations to process data in the biological association network of interestTable 6Pipeline functions divided into modules: each module focuses on a specific set of operations to process data in the biological association network of interest, as described in its respective methods section; visual depictions in Fig.\u00a013ScriptModuleOperationsSectionMonarch.pyDatabase callsFetching biological associations up to the n-th degree of distance\u2014third by default\u2014given a disease ID as inputSection \u201cDatabase APIs for biological, gene-drug associations anddrug-drug similaritydgidb.pyDatabase callsFetching gene-to-drug edges given a gene ID as inputSection \u201cDatabase APIs for biological, gene-drug associations anddrug-drug similaritydrugsimilarity.pyDrug-to-drugGenerating drug-to-drug edges based on 2D feature similarity between drugsSection \u201cDatabase APIs for biological, gene-drug associations anddrug-drug similaritynegsamples.pyNegative samplesGenerating semantically valid negative samples by considering node centrality, gene-to-gene edges, and drug-to-drug edges above a certain similarity thresholdSection \u201cGeneration of semantically-valid negative samples based on centrality measures\u201dembeddings_{ml_method}.pyEmbeddingsConverting gene-to-drug edges into node embeddings according to requirements for the specific ML methodSection \u201cParallelisation in gene-drug embeddings\u201dnetworkmodel_{ml_method}.pyDrug predictionPredicts drugs based on association to genes and prediction probability via XGBoost or GCNSections \u201cPrediction of repurposed drugs in binary models withnon-binary datasets\u201d, \u201cPrediction iterations as a solution for seed variability in MLmethods\u201d", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Methods", "text": "Pipeline modules: each block represents one of the functional modules from Table 6 that focuses on a specific set of operations to process data in the biological association network of interest\n\nPipeline functions divided into modules: each module focuses on a specific set of operations to process data in the biological association network of interest, as described in its respective methods section; visual depictions in Fig.\u00a013", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Methods", "text": "We extended the work of Reep to accommodate for the new versions of the Monarch KG v3 FastAPI (https://api-v3.monarchinitiative.org/v3/api/) and the GraphQL 5.0 DGIdb API (https://dgidb.org/api/graphql) [31, 38]. Provided a disease URI, the pipeline fetches all biological associations to it up to the third degree of distance (d.o.d.). Monarch Initiative returns JSON files in the Biolink model format for each node query listing all nodes associated [53], and the pipeline then performs more calls based on the nodes returned. DGIdb queries are performed for all gene nodes from Monarch Initiative\u2019s calls, and it returns JSON files in the GraphQL format for each gene node listing all drug nodes associated [38]. The pipeline accepts Monarch Initiative\u2019s proprietary disease URIs (i.e. \u2018MONDO:\u2019) directly, while the more widely used \u2018OMIM:\u2019 IDs are not supported. Bianchi developed a custom JavaScript tool (https://github.com/NCMBianchi/OMIM-converter, [73]) that leverages Monarch Initiative\u2019s entity mapping endpoints to ensure accurate cross-referencing between identifier systems. All datasets and results presented in this study were retrieved or based on datasets via these APIs on October 7th and 10th, 2025 at the latest.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Methods", "text": "Moreover, the previous version of the tool by Reep only took into consideration drugs already in the network of associations to the disease for the newly predicted gene-to-drug edges [31]. We added a step that fetches all the drugs in DGIdb, and we handled them separately.\n\nThe generation of drug-to-drug edges was mostly untouched compared to Reep: provided \u2018chembl:\u2019 and \u2018wikidata:\u2019 drug nodes\u2019 URIs, pairwise similarity is computed by fetching the SMILES notation [31, 74] representing the molecule\u2019s structure for each drug, and converting them to RDkit 2-dimensional objects. Those are then used to build 4096-bit ECFP vectors representing the presence or absence of a given feature in the molecule, which are used to compute pair-wise similarity according to Tanimoto\u2019s coefficient [29, 75] \u2013maintaining the same approach used by Reep for molecular similarity assessment, although with a lower similarity threshold (i.e. 0.5 instead of 0.92 [31]) to accommodate for a higher number of weighted edges for negative samples generation (Section\u00a0\u201cGeneration of semantically-valid negative samples based on centrality measures\u201d) and GCN training (4.2).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Methods", "text": "For the purpose of testing prediction efficiency, runtimes and CPU strain on several runs within manageable time-frames (i.e. less than 10\u00a0h), we generated a smaller network of 1000 gene nodes, 600 drug nodes and 10% of all other nodes associated to the disease of interest and 4400 other drugs from DGIdb by computing the \u2018PageRank\u2019 centrality score for all the nodes, and selected a subset with the same distribution of centrality scores. More on this in Appendix A.\n\nDue to the lack of databases which store structured information on known lack of associations between gene and drugs, we generated semantically-valid negative samples (Section\u00a0\u201cAlgorithm for semantic generation of negative samples\u201d). Specifically, we aim to discriminate between valid and invalid negative samples among the missing gene-drug associations (i.e. nodes not connected by edges) in the KG, to ignore invalid negative samples during prediction.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Methods", "text": "By iterating through each positive sample (i.e. gene-to-drug edges fetched from DGIdb, labelled \u20181\u2019), a centrality score is computed for each gene and drug node (i.e. degree, centrality, eigenvector centrality or in-betweenness centrality depending on the user input). Whenever either of the nodes has a centrality score above a certain threshold, all the genes at one d.o.d. of that given gene and all the drugs at one d.o.d. of that drug above a certain similarity threshold are taken into consideration. Any combination of gene-to-drug edges from this pool is deemed semantically valid (i.e. labelled \u20180\u2019).\n\nML methods require information provided in vectors to perform a prediction: one node embedding dataset with all information for training with nodes and edges, and one for prediction only with nodes. Both mapping datasets contain all biological entity nodes and drug nodes associated to the disease of interest, while the prediction mapping dataset also contains all other drug nodes in DGIdb, to make predictions also outside of the associations to the disease of interest.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Methods", "text": "Node embeddings are generated using Node2Vec as implemented in the node2vec Python package [62], which creates dense vector representations through biased random walks on the network followed by skip-gram learning to predict node contexts [60]. In contrast to Reep, who employed RDF2Vec\u2014no longer supported\u2014with maximum walk depth of 4 and 10 walks per entity [31], our Node2Vec implementation provides more flexible control through parametrised biased random walks to the advantage of biological networks that are inherently sparse.\n\nIn our implementation, we used non-default parameters for embedding dimensions, walk length and number of walks. Embedding dimensions was set to 64 (i.e. default = 128) to avoid overfitting in biological networks, which are sparse compared to the social networks Node2Vec was developed for. Walk length instead was set to 30 (i.e. default = 80) due to the nodes in our KG never being more than 5 d.o.d. from each other\u2014far less than in social networks. Finally, the number of walks was set to 200 (i.e. default = 10) to ensure adequate sampling of the low proportion of positive gene-drug associations in our KG\u2014only 28,260 out of 13.8M possible pairs, as reported in the Results section (Table 2 and in Section \u201cParallelisation in gene-drug embeddings\u201d). The higher number of walks also improves embedding stability across iterative runs with different random seeds (more in Section \u201cPrediction iterations as a solution for seed variability in MLmethods\u201d).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Methods", "text": "To improve computational efficiency, the entire pipeline implements a caching mechanism\u2014triggered by the emb_l=1 parameter\u2013 that enables loading of previously computed embeddings when processing the same disease and date combination, avoiding redundant computation. The date parameter can be manually set using the set_date() utility function to enable reuse of existing embedding files. For computationally intensive tasks such as generating embeddings for all possible gene-drug combinations in the prediction matrix, parallel processing distributes the workload across multiple CPU cores\u2014configurable via the n_jobs parameter\u2014significantly reducing runtime compared to sequential processing.\n\nWhile the GCN model from the PyTorch-geometric library only requires the training node embedding map of all the gene-to-drug edges in the network of associations (i.e. positive samples labelled \u20181\u2019, and valid negative samples labelled \u20180\u2019) [47, 48], XGBoost also requires the training node embedding map for all the possible gene-to-drug edges not in the network (i.e. invalid negative samples, labelled \u2018-1\u2019) [59], which have to be computed by iterating through genes and drugs at this point of the pipeline. This step is extremely time-consuming, therefore we implemented a script that breaks the list of drugs into chunks, and each chunk is processed by a different CPU core concurrently. The maximum number of CPU cores to use can be set by the user.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Methods", "text": "Node embeddings are then fed to the appropriate model for training and subsequent prediction. Several parameters are made available to the user (Table 7) to customise the ML prediction step.Table 7Sample of the parameters for the run_networkmodelml_method() function from the \u2018Drug Prediction\u2019 module: \u2018run_jobs\u2019, \u2018run_depth\u2019 (Table 5) and \u2018batch_size\u2019 regulate the ML prediction step; \u2018prob_input\u2019 and \u2018clust_input\u2019 regulate the filtering of predicted edges according to confidence metrics, compared to Reep\u2019s single filtering threshold of 0.9 [31]; \u2018model_type\u2019 is a conditional parameter when using GCN and not XGBoostParameterDefaultDescriptionrun_jobsHalf of CPU coresNumber of CPU cores to use for training/predictionrun_depth\u2018ultralight\u2019Depth of parameter optimisation (\u2018ultralight\u2019, \u2018light\u2019, \u2018full\u2019)prob_input0.65Probability threshold for filtering predictionsclust_input0.8Cluster threshold for ranking drugsbatch_s2000Batch size for predictionmodel_type [PyG]\u2018gcn\u2019Type of Graph Neural Network model (\u2018gcn\u2019, \u2018gat\u2019, \u2018sage\u2019)\n\nSample of the parameters for the run_networkmodelml_method() function from the \u2018Drug Prediction\u2019 module: \u2018run_jobs\u2019, \u2018run_depth\u2019 (Table 5) and \u2018batch_size\u2019 regulate the ML prediction step; \u2018prob_input\u2019 and \u2018clust_input\u2019 regulate the filtering of predicted edges according to confidence metrics, compared to Reep\u2019s single filtering threshold of 0.9 [31]; \u2018model_type\u2019 is a conditional parameter when using GCN and not XGBoost", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Methods", "text": "The GCN model takes as training input the node embeddings that only consider existing edges with various labels (i.e. \u20181\u2019 for positive samples, \u20180\u2019 for valid negative samples), thus it only requires one run and it returns all edges that are predicted as positive. XGBoost instead mostly requires a binary mapping of edges that are deemed either positive or negative [59]. To take advantage of the negative samples information, the model runs twice: once where valid negative samples are considered as positive, and once where they are considered as negative. Results from both runs are pooled: any edge that appeared at list once is ranked, and the prediction probabilities is always averaged across runs with \u20180.5\u2019 (random) for runs where the edge does not appear.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Methods", "text": "For quantitative evaluation of model performance, we employed an 80/20 train/test split on gene-to-drug edges from DGIdb. Of the 69,907 total drugs in their database, only 11,603 (16.6%) could be successfully mapped to SMILES representations required for drug similarity calculations (detailed mappability analysis in Section\u00a0\u201cDatabase APIs for biological, gene-drug associations anddrug-drug similarity\u201d), constraining the prediction space to drugs with available structural information. In each evaluation run, 80% of known gene-drug associations were randomly selected for training \u2013including both positive and semantically-valid negative samples generated as described in section \u201cGeneration of semantically-valid negative samples based on centrality measures\u201d\u2014while the remaining 20% were held out for testing. This split ratio follows standard ML practice, providing sufficient training data while maintaining an adequate test set for robust performance estimation. Split selection was performed using a fixed random seed (i.e. seed=42) to ensure reproducibility across experiments. For iteration-based evaluations, the same train/test split was maintained across all iterations\u2013only the model\u2019s random seed was varied\u2013ensuring that performance metrics reflect model stability rather than data variability. Prediction efficiency is also estimated by verifying the presence of a smaller subset gene-to-drug edges (i.e. n=5) that were in the original network, but that were removed ahead of training: such edges were hand-picked based on their centrality score to be relevant enough in the network, yet not too relevant (Table 8).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Methods", "text": "XGBoost model training employs randomised hyperparameter search using scikit-learn\u2019s RandomizedSearchCV() with repeated stratified k-fold cross-validation. The pipeline offers three optimisation depth settings (run_depth parameter): ultralight, light, and full, which control both the hyperparameter search space and cross-validation complexity. The full setting explores a comprehensive parameter grid including min_child_weight (i.e. seven values between 2 and 30), gamma (i.e. nine values between 0 and 6), regularisation parameters (i.e. reg_alpha and reg_lambda, six values each between 0 and 10), subsample and colsample_bytree (i.e. uniform distributions respectively between 0.5 and 1.0, and between 0.2 and 1.0), max_depth (i.e. seven values between 4 and 16), n_estimators (i.e. seven values between 35 and 100), and learning_rate (i.e. uniform from 0 to 0.3), with 10-split 5-repeat cross-validation and 20 random iterations. The light and ultralight settings use reduced parameter grids and fewer cross-validation splits (i.e. 3/2 and 2/1 splits/repeats respectively) to balance optimisation quality with computational efficiency (Table 5). Due to the dual-run embedding XGBoost strategy described above, two separate models are trained with optimised hyperparameters, and predictions from each of the two runs are subsequently pooled.Fig. 14Degree distribution comparison between full and subset networks: Left panel shows degree distribution density plots for full network (blue) and subset network (orange), demonstrating that the subset preserves the characteristic peak near zero degree typical of biological networks\u2014 even though with higher counts due to the smaller network mostly having 1 d.o.d. associations (Appendix A). Right panel presents degree distribution on log-log scale, confirming that both networks exhibit scale-free topology with similar power-law decay patterns. The subset network maintains representative coverage of high-degree hub nodes while reducing overall network size by approximately 95%, enabling computationally feasible iteration-based evaluations. Peaks in both curves represent highly interconnected gene hubs. Complete statistical validation including betweenness centrality preservation (KS-test p>0.05), and entity type proportions are provided in Appendix A", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Methods", "text": "Degree distribution comparison between full and subset networks: Left panel shows degree distribution density plots for full network (blue) and subset network (orange), demonstrating that the subset preserves the characteristic peak near zero degree typical of biological networks\u2014 even though with higher counts due to the smaller network mostly having 1 d.o.d. associations (Appendix A). Right panel presents degree distribution on log-log scale, confirming that both networks exhibit scale-free topology with similar power-law decay patterns. The subset network maintains representative coverage of high-degree hub nodes while reducing overall network size by approximately 95%, enabling computationally feasible iteration-based evaluations. Peaks in both curves represent highly interconnected gene hubs. Complete statistical validation including betweenness centrality preservation (KS-test p>0.05), and entity type proportions are provided in Appendix A", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Methods", "text": "To enable comprehensive evaluation through multiple iterations while maintaining manageable computational requirements, we generated a representative subset network (i.e. 1000 genes, 600 drugs) using PageRank centrality-based stratified sampling that preserves the topological characteristics of the full network (i.e. p>0.05 for betweenness centrality distributions; see Appendix A for complete methodology and statistical validation; Fig.\u00a014), while node type proportions were intentionally enriched for genes and drugs to maximise prediction space. Single-run evaluations and ablation studies were performed on the full network, while iteration-based consensus predictions were evaluated on this validated subset network due to computational constraints (see Appendix A for detailed justification).Table 8Gene-to-drug edges removed for prediction efficiency check: curated subset of edges involving key gene nodes in the network of associations to HD; removed according to section \u201cPrediction of repurposed drugs in binary models with non-binary datasets\u201d; validated in section \u201cPrediction iterations as a solution for seed variability in ML methods\u201dGene IDGene nameDrug IDDrug nameScoreHGNC:20,772TUBB3CHEMBL1173445LARGAZOLE0.0434HGNC:4617GSK3BCHEMBL546797CHEMBL5467970.0355HGNC:4851HTTCHEMBL581880CHEMBL5818800.0576HGNC:4851HTTCHEMBL472437KAMEBANIN0.0432HGNC:3594FASNCHEMBL210208PYROGALLOL RED0.0773", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Methods", "text": "Gene-to-drug edges removed for prediction efficiency check: curated subset of edges involving key gene nodes in the network of associations to HD; removed according to section \u201cPrediction of repurposed drugs in binary models with non-binary datasets\u201d; validated in section \u201cPrediction iterations as a solution for seed variability in ML methods\u201d\n\nSince single runs are influenced by the inherent variability of prediction methods represented by the input seed, we implemented an iteration-based consensus approach that executes multiple runs with different random seeds. For computational feasibility, we performed 10 iterations for XGBoost and 200 iterations for GCN on the validated subset network (Appendix A), thus having shorter runtimes than on the full network (i.e. \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\sim 47$$\\end{document} minutes for XGBoost vs. \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\sim 35$$\\end{document} seconds for GCN).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Methods", "text": "Results from all iterations are collected in a mapping dictionary tracking prediction probabilities across runs. Gene-to-drug edges are then ranked using a consensus strategy based on two criteria: occurrence count across iterations, and average prediction probability. By default, only the top 5% of predictions at a maximum of 100 edges (ml_iter_max_edges=100) from each prediction iteration are labelled as positive and returned in the final results. This consensus approach reduces sensitivity to random seed variation and improves prediction stability (detailed iteration analysis in Appendix B).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Pipeline upgrades", "text": "We extended the work of Reep to accommodate for the new versions of the Monarch KG v3 FastAPI (https://api-v3.monarchinitiative.org/v3/api/) and the GraphQL 5.0 DGIdb API (https://dgidb.org/api/graphql) [31, 38]. Provided a disease URI, the pipeline fetches all biological associations to it up to the third degree of distance (d.o.d.). Monarch Initiative returns JSON files in the Biolink model format for each node query listing all nodes associated [53], and the pipeline then performs more calls based on the nodes returned. DGIdb queries are performed for all gene nodes from Monarch Initiative\u2019s calls, and it returns JSON files in the GraphQL format for each gene node listing all drug nodes associated [38]. The pipeline accepts Monarch Initiative\u2019s proprietary disease URIs (i.e. \u2018MONDO:\u2019) directly, while the more widely used \u2018OMIM:\u2019 IDs are not supported. Bianchi developed a custom JavaScript tool (https://github.com/NCMBianchi/OMIM-converter, [73]) that leverages Monarch Initiative\u2019s entity mapping endpoints to ensure accurate cross-referencing between identifier systems. All datasets and results presented in this study were retrieved or based on datasets via these APIs on October 7th and 10th, 2025 at the latest.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Pipeline upgrades", "text": "Moreover, the previous version of the tool by Reep only took into consideration drugs already in the network of associations to the disease for the newly predicted gene-to-drug edges [31]. We added a step that fetches all the drugs in DGIdb, and we handled them separately.\n\nThe generation of drug-to-drug edges was mostly untouched compared to Reep: provided \u2018chembl:\u2019 and \u2018wikidata:\u2019 drug nodes\u2019 URIs, pairwise similarity is computed by fetching the SMILES notation [31, 74] representing the molecule\u2019s structure for each drug, and converting them to RDkit 2-dimensional objects. Those are then used to build 4096-bit ECFP vectors representing the presence or absence of a given feature in the molecule, which are used to compute pair-wise similarity according to Tanimoto\u2019s coefficient [29, 75] \u2013maintaining the same approach used by Reep for molecular similarity assessment, although with a lower similarity threshold (i.e. 0.5 instead of 0.92 [31]) to accommodate for a higher number of weighted edges for negative samples generation (Section\u00a0\u201cGeneration of semantically-valid negative samples based on centrality measures\u201d) and GCN training (4.2).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Pipeline upgrades", "text": "For the purpose of testing prediction efficiency, runtimes and CPU strain on several runs within manageable time-frames (i.e. less than 10\u00a0h), we generated a smaller network of 1000 gene nodes, 600 drug nodes and 10% of all other nodes associated to the disease of interest and 4400 other drugs from DGIdb by computing the \u2018PageRank\u2019 centrality score for all the nodes, and selected a subset with the same distribution of centrality scores. More on this in Appendix A.\n\nDue to the lack of databases which store structured information on known lack of associations between gene and drugs, we generated semantically-valid negative samples (Section\u00a0\u201cAlgorithm for semantic generation of negative samples\u201d). Specifically, we aim to discriminate between valid and invalid negative samples among the missing gene-drug associations (i.e. nodes not connected by edges) in the KG, to ignore invalid negative samples during prediction.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Pipeline upgrades", "text": "By iterating through each positive sample (i.e. gene-to-drug edges fetched from DGIdb, labelled \u20181\u2019), a centrality score is computed for each gene and drug node (i.e. degree, centrality, eigenvector centrality or in-betweenness centrality depending on the user input). Whenever either of the nodes has a centrality score above a certain threshold, all the genes at one d.o.d. of that given gene and all the drugs at one d.o.d. of that drug above a certain similarity threshold are taken into consideration. Any combination of gene-to-drug edges from this pool is deemed semantically valid (i.e. labelled \u20180\u2019).\n\nML methods require information provided in vectors to perform a prediction: one node embedding dataset with all information for training with nodes and edges, and one for prediction only with nodes. Both mapping datasets contain all biological entity nodes and drug nodes associated to the disease of interest, while the prediction mapping dataset also contains all other drug nodes in DGIdb, to make predictions also outside of the associations to the disease of interest.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Pipeline upgrades", "text": "Node embeddings are generated using Node2Vec as implemented in the node2vec Python package [62], which creates dense vector representations through biased random walks on the network followed by skip-gram learning to predict node contexts [60]. In contrast to Reep, who employed RDF2Vec\u2014no longer supported\u2014with maximum walk depth of 4 and 10 walks per entity [31], our Node2Vec implementation provides more flexible control through parametrised biased random walks to the advantage of biological networks that are inherently sparse.\n\nIn our implementation, we used non-default parameters for embedding dimensions, walk length and number of walks. Embedding dimensions was set to 64 (i.e. default = 128) to avoid overfitting in biological networks, which are sparse compared to the social networks Node2Vec was developed for. Walk length instead was set to 30 (i.e. default = 80) due to the nodes in our KG never being more than 5 d.o.d. from each other\u2014far less than in social networks. Finally, the number of walks was set to 200 (i.e. default = 10) to ensure adequate sampling of the low proportion of positive gene-drug associations in our KG\u2014only 28,260 out of 13.8M possible pairs, as reported in the Results section (Table 2 and in Section \u201cParallelisation in gene-drug embeddings\u201d). The higher number of walks also improves embedding stability across iterative runs with different random seeds (more in Section \u201cPrediction iterations as a solution for seed variability in MLmethods\u201d).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Pipeline upgrades", "text": "To improve computational efficiency, the entire pipeline implements a caching mechanism\u2014triggered by the emb_l=1 parameter\u2013 that enables loading of previously computed embeddings when processing the same disease and date combination, avoiding redundant computation. The date parameter can be manually set using the set_date() utility function to enable reuse of existing embedding files. For computationally intensive tasks such as generating embeddings for all possible gene-drug combinations in the prediction matrix, parallel processing distributes the workload across multiple CPU cores\u2014configurable via the n_jobs parameter\u2014significantly reducing runtime compared to sequential processing.\n\nWhile the GCN model from the PyTorch-geometric library only requires the training node embedding map of all the gene-to-drug edges in the network of associations (i.e. positive samples labelled \u20181\u2019, and valid negative samples labelled \u20180\u2019) [47, 48], XGBoost also requires the training node embedding map for all the possible gene-to-drug edges not in the network (i.e. invalid negative samples, labelled \u2018-1\u2019) [59], which have to be computed by iterating through genes and drugs at this point of the pipeline. This step is extremely time-consuming, therefore we implemented a script that breaks the list of drugs into chunks, and each chunk is processed by a different CPU core concurrently. The maximum number of CPU cores to use can be set by the user.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Database APIs for biological, gene-drug associations and drug-drug similarity", "text": "We extended the work of Reep to accommodate for the new versions of the Monarch KG v3 FastAPI (https://api-v3.monarchinitiative.org/v3/api/) and the GraphQL 5.0 DGIdb API (https://dgidb.org/api/graphql) [31, 38]. Provided a disease URI, the pipeline fetches all biological associations to it up to the third degree of distance (d.o.d.). Monarch Initiative returns JSON files in the Biolink model format for each node query listing all nodes associated [53], and the pipeline then performs more calls based on the nodes returned. DGIdb queries are performed for all gene nodes from Monarch Initiative\u2019s calls, and it returns JSON files in the GraphQL format for each gene node listing all drug nodes associated [38]. The pipeline accepts Monarch Initiative\u2019s proprietary disease URIs (i.e. \u2018MONDO:\u2019) directly, while the more widely used \u2018OMIM:\u2019 IDs are not supported. Bianchi developed a custom JavaScript tool (https://github.com/NCMBianchi/OMIM-converter, [73]) that leverages Monarch Initiative\u2019s entity mapping endpoints to ensure accurate cross-referencing between identifier systems. All datasets and results presented in this study were retrieved or based on datasets via these APIs on October 7th and 10th, 2025 at the latest.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Database APIs for biological, gene-drug associations and drug-drug similarity", "text": "Moreover, the previous version of the tool by Reep only took into consideration drugs already in the network of associations to the disease for the newly predicted gene-to-drug edges [31]. We added a step that fetches all the drugs in DGIdb, and we handled them separately.\n\nThe generation of drug-to-drug edges was mostly untouched compared to Reep: provided \u2018chembl:\u2019 and \u2018wikidata:\u2019 drug nodes\u2019 URIs, pairwise similarity is computed by fetching the SMILES notation [31, 74] representing the molecule\u2019s structure for each drug, and converting them to RDkit 2-dimensional objects. Those are then used to build 4096-bit ECFP vectors representing the presence or absence of a given feature in the molecule, which are used to compute pair-wise similarity according to Tanimoto\u2019s coefficient [29, 75] \u2013maintaining the same approach used by Reep for molecular similarity assessment, although with a lower similarity threshold (i.e. 0.5 instead of 0.92 [31]) to accommodate for a higher number of weighted edges for negative samples generation (Section\u00a0\u201cGeneration of semantically-valid negative samples based on centrality measures\u201d) and GCN training (4.2).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Database APIs for biological, gene-drug associations and drug-drug similarity", "text": "For the purpose of testing prediction efficiency, runtimes and CPU strain on several runs within manageable time-frames (i.e. less than 10\u00a0h), we generated a smaller network of 1000 gene nodes, 600 drug nodes and 10% of all other nodes associated to the disease of interest and 4400 other drugs from DGIdb by computing the \u2018PageRank\u2019 centrality score for all the nodes, and selected a subset with the same distribution of centrality scores. More on this in Appendix A.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Generation of semantically-valid negative samples based on centrality measures", "text": "Due to the lack of databases which store structured information on known lack of associations between gene and drugs, we generated semantically-valid negative samples (Section\u00a0\u201cAlgorithm for semantic generation of negative samples\u201d). Specifically, we aim to discriminate between valid and invalid negative samples among the missing gene-drug associations (i.e. nodes not connected by edges) in the KG, to ignore invalid negative samples during prediction.\n\nBy iterating through each positive sample (i.e. gene-to-drug edges fetched from DGIdb, labelled \u20181\u2019), a centrality score is computed for each gene and drug node (i.e. degree, centrality, eigenvector centrality or in-betweenness centrality depending on the user input). Whenever either of the nodes has a centrality score above a certain threshold, all the genes at one d.o.d. of that given gene and all the drugs at one d.o.d. of that drug above a certain similarity threshold are taken into consideration. Any combination of gene-to-drug edges from this pool is deemed semantically valid (i.e. labelled \u20180\u2019).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Parallelisation in gene-drug embeddings", "text": "ML methods require information provided in vectors to perform a prediction: one node embedding dataset with all information for training with nodes and edges, and one for prediction only with nodes. Both mapping datasets contain all biological entity nodes and drug nodes associated to the disease of interest, while the prediction mapping dataset also contains all other drug nodes in DGIdb, to make predictions also outside of the associations to the disease of interest.\n\nNode embeddings are generated using Node2Vec as implemented in the node2vec Python package [62], which creates dense vector representations through biased random walks on the network followed by skip-gram learning to predict node contexts [60]. In contrast to Reep, who employed RDF2Vec\u2014no longer supported\u2014with maximum walk depth of 4 and 10 walks per entity [31], our Node2Vec implementation provides more flexible control through parametrised biased random walks to the advantage of biological networks that are inherently sparse.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Parallelisation in gene-drug embeddings", "text": "In our implementation, we used non-default parameters for embedding dimensions, walk length and number of walks. Embedding dimensions was set to 64 (i.e. default = 128) to avoid overfitting in biological networks, which are sparse compared to the social networks Node2Vec was developed for. Walk length instead was set to 30 (i.e. default = 80) due to the nodes in our KG never being more than 5 d.o.d. from each other\u2014far less than in social networks. Finally, the number of walks was set to 200 (i.e. default = 10) to ensure adequate sampling of the low proportion of positive gene-drug associations in our KG\u2014only 28,260 out of 13.8M possible pairs, as reported in the Results section (Table 2 and in Section \u201cParallelisation in gene-drug embeddings\u201d). The higher number of walks also improves embedding stability across iterative runs with different random seeds (more in Section \u201cPrediction iterations as a solution for seed variability in MLmethods\u201d).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Parallelisation in gene-drug embeddings", "text": "To improve computational efficiency, the entire pipeline implements a caching mechanism\u2014triggered by the emb_l=1 parameter\u2013 that enables loading of previously computed embeddings when processing the same disease and date combination, avoiding redundant computation. The date parameter can be manually set using the set_date() utility function to enable reuse of existing embedding files. For computationally intensive tasks such as generating embeddings for all possible gene-drug combinations in the prediction matrix, parallel processing distributes the workload across multiple CPU cores\u2014configurable via the n_jobs parameter\u2014significantly reducing runtime compared to sequential processing.\n\nWhile the GCN model from the PyTorch-geometric library only requires the training node embedding map of all the gene-to-drug edges in the network of associations (i.e. positive samples labelled \u20181\u2019, and valid negative samples labelled \u20180\u2019) [47, 48], XGBoost also requires the training node embedding map for all the possible gene-to-drug edges not in the network (i.e. invalid negative samples, labelled \u2018-1\u2019) [59], which have to be computed by iterating through genes and drugs at this point of the pipeline. This step is extremely time-consuming, therefore we implemented a script that breaks the list of drugs into chunks, and each chunk is processed by a different CPU core concurrently. The maximum number of CPU cores to use can be set by the user.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "Node embeddings are then fed to the appropriate model for training and subsequent prediction. Several parameters are made available to the user (Table 7) to customise the ML prediction step.Table 7Sample of the parameters for the run_networkmodelml_method() function from the \u2018Drug Prediction\u2019 module: \u2018run_jobs\u2019, \u2018run_depth\u2019 (Table 5) and \u2018batch_size\u2019 regulate the ML prediction step; \u2018prob_input\u2019 and \u2018clust_input\u2019 regulate the filtering of predicted edges according to confidence metrics, compared to Reep\u2019s single filtering threshold of 0.9 [31]; \u2018model_type\u2019 is a conditional parameter when using GCN and not XGBoostParameterDefaultDescriptionrun_jobsHalf of CPU coresNumber of CPU cores to use for training/predictionrun_depth\u2018ultralight\u2019Depth of parameter optimisation (\u2018ultralight\u2019, \u2018light\u2019, \u2018full\u2019)prob_input0.65Probability threshold for filtering predictionsclust_input0.8Cluster threshold for ranking drugsbatch_s2000Batch size for predictionmodel_type [PyG]\u2018gcn\u2019Type of Graph Neural Network model (\u2018gcn\u2019, \u2018gat\u2019, \u2018sage\u2019)\n\nSample of the parameters for the run_networkmodelml_method() function from the \u2018Drug Prediction\u2019 module: \u2018run_jobs\u2019, \u2018run_depth\u2019 (Table 5) and \u2018batch_size\u2019 regulate the ML prediction step; \u2018prob_input\u2019 and \u2018clust_input\u2019 regulate the filtering of predicted edges according to confidence metrics, compared to Reep\u2019s single filtering threshold of 0.9 [31]; \u2018model_type\u2019 is a conditional parameter when using GCN and not XGBoost", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "The GCN model takes as training input the node embeddings that only consider existing edges with various labels (i.e. \u20181\u2019 for positive samples, \u20180\u2019 for valid negative samples), thus it only requires one run and it returns all edges that are predicted as positive. XGBoost instead mostly requires a binary mapping of edges that are deemed either positive or negative [59]. To take advantage of the negative samples information, the model runs twice: once where valid negative samples are considered as positive, and once where they are considered as negative. Results from both runs are pooled: any edge that appeared at list once is ranked, and the prediction probabilities is always averaged across runs with \u20180.5\u2019 (random) for runs where the edge does not appear.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "For quantitative evaluation of model performance, we employed an 80/20 train/test split on gene-to-drug edges from DGIdb. Of the 69,907 total drugs in their database, only 11,603 (16.6%) could be successfully mapped to SMILES representations required for drug similarity calculations (detailed mappability analysis in Section\u00a0\u201cDatabase APIs for biological, gene-drug associations anddrug-drug similarity\u201d), constraining the prediction space to drugs with available structural information. In each evaluation run, 80% of known gene-drug associations were randomly selected for training \u2013including both positive and semantically-valid negative samples generated as described in section \u201cGeneration of semantically-valid negative samples based on centrality measures\u201d\u2014while the remaining 20% were held out for testing. This split ratio follows standard ML practice, providing sufficient training data while maintaining an adequate test set for robust performance estimation. Split selection was performed using a fixed random seed (i.e. seed=42) to ensure reproducibility across experiments. For iteration-based evaluations, the same train/test split was maintained across all iterations\u2013only the model\u2019s random seed was varied\u2013ensuring that performance metrics reflect model stability rather than data variability. Prediction efficiency is also estimated by verifying the presence of a smaller subset gene-to-drug edges (i.e. n=5) that were in the original network, but that were removed ahead of training: such edges were hand-picked based on their centrality score to be relevant enough in the network, yet not too relevant (Table 8).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "XGBoost model training employs randomised hyperparameter search using scikit-learn\u2019s RandomizedSearchCV() with repeated stratified k-fold cross-validation. The pipeline offers three optimisation depth settings (run_depth parameter): ultralight, light, and full, which control both the hyperparameter search space and cross-validation complexity. The full setting explores a comprehensive parameter grid including min_child_weight (i.e. seven values between 2 and 30), gamma (i.e. nine values between 0 and 6), regularisation parameters (i.e. reg_alpha and reg_lambda, six values each between 0 and 10), subsample and colsample_bytree (i.e. uniform distributions respectively between 0.5 and 1.0, and between 0.2 and 1.0), max_depth (i.e. seven values between 4 and 16), n_estimators (i.e. seven values between 35 and 100), and learning_rate (i.e. uniform from 0 to 0.3), with 10-split 5-repeat cross-validation and 20 random iterations. The light and ultralight settings use reduced parameter grids and fewer cross-validation splits (i.e. 3/2 and 2/1 splits/repeats respectively) to balance optimisation quality with computational efficiency (Table 5). Due to the dual-run embedding XGBoost strategy described above, two separate models are trained with optimised hyperparameters, and predictions from each of the two runs are subsequently pooled.Fig. 14Degree distribution comparison between full and subset networks: Left panel shows degree distribution density plots for full network (blue) and subset network (orange), demonstrating that the subset preserves the characteristic peak near zero degree typical of biological networks\u2014 even though with higher counts due to the smaller network mostly having 1 d.o.d. associations (Appendix A). Right panel presents degree distribution on log-log scale, confirming that both networks exhibit scale-free topology with similar power-law decay patterns. The subset network maintains representative coverage of high-degree hub nodes while reducing overall network size by approximately 95%, enabling computationally feasible iteration-based evaluations. Peaks in both curves represent highly interconnected gene hubs. Complete statistical validation including betweenness centrality preservation (KS-test p>0.05), and entity type proportions are provided in Appendix A", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "Degree distribution comparison between full and subset networks: Left panel shows degree distribution density plots for full network (blue) and subset network (orange), demonstrating that the subset preserves the characteristic peak near zero degree typical of biological networks\u2014 even though with higher counts due to the smaller network mostly having 1 d.o.d. associations (Appendix A). Right panel presents degree distribution on log-log scale, confirming that both networks exhibit scale-free topology with similar power-law decay patterns. The subset network maintains representative coverage of high-degree hub nodes while reducing overall network size by approximately 95%, enabling computationally feasible iteration-based evaluations. Peaks in both curves represent highly interconnected gene hubs. Complete statistical validation including betweenness centrality preservation (KS-test p>0.05), and entity type proportions are provided in Appendix A", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "To enable comprehensive evaluation through multiple iterations while maintaining manageable computational requirements, we generated a representative subset network (i.e. 1000 genes, 600 drugs) using PageRank centrality-based stratified sampling that preserves the topological characteristics of the full network (i.e. p>0.05 for betweenness centrality distributions; see Appendix A for complete methodology and statistical validation; Fig.\u00a014), while node type proportions were intentionally enriched for genes and drugs to maximise prediction space. Single-run evaluations and ablation studies were performed on the full network, while iteration-based consensus predictions were evaluated on this validated subset network due to computational constraints (see Appendix A for detailed justification).Table 8Gene-to-drug edges removed for prediction efficiency check: curated subset of edges involving key gene nodes in the network of associations to HD; removed according to section \u201cPrediction of repurposed drugs in binary models with non-binary datasets\u201d; validated in section \u201cPrediction iterations as a solution for seed variability in ML methods\u201dGene IDGene nameDrug IDDrug nameScoreHGNC:20,772TUBB3CHEMBL1173445LARGAZOLE0.0434HGNC:4617GSK3BCHEMBL546797CHEMBL5467970.0355HGNC:4851HTTCHEMBL581880CHEMBL5818800.0576HGNC:4851HTTCHEMBL472437KAMEBANIN0.0432HGNC:3594FASNCHEMBL210208PYROGALLOL RED0.0773", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "Gene-to-drug edges removed for prediction efficiency check: curated subset of edges involving key gene nodes in the network of associations to HD; removed according to section \u201cPrediction of repurposed drugs in binary models with non-binary datasets\u201d; validated in section \u201cPrediction iterations as a solution for seed variability in ML methods\u201d\n\nSince single runs are influenced by the inherent variability of prediction methods represented by the input seed, we implemented an iteration-based consensus approach that executes multiple runs with different random seeds. For computational feasibility, we performed 10 iterations for XGBoost and 200 iterations for GCN on the validated subset network (Appendix A), thus having shorter runtimes than on the full network (i.e. \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\sim 47$$\\end{document} minutes for XGBoost vs. \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\sim 35$$\\end{document} seconds for GCN).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "ML training and prediction in enriched KG", "text": "Results from all iterations are collected in a mapping dictionary tracking prediction probabilities across runs. Gene-to-drug edges are then ranked using a consensus strategy based on two criteria: occurrence count across iterations, and average prediction probability. By default, only the top 5% of predictions at a maximum of 100 edges (ml_iter_max_edges=100) from each prediction iteration are labelled as positive and returned in the final results. This consensus approach reduces sensitivity to random seed variation and improves prediction stability (detailed iteration analysis in Appendix B).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction of repurposed drugs in binary models with non-binary datasets", "text": "Node embeddings are then fed to the appropriate model for training and subsequent prediction. Several parameters are made available to the user (Table 7) to customise the ML prediction step.Table 7Sample of the parameters for the run_networkmodelml_method() function from the \u2018Drug Prediction\u2019 module: \u2018run_jobs\u2019, \u2018run_depth\u2019 (Table 5) and \u2018batch_size\u2019 regulate the ML prediction step; \u2018prob_input\u2019 and \u2018clust_input\u2019 regulate the filtering of predicted edges according to confidence metrics, compared to Reep\u2019s single filtering threshold of 0.9 [31]; \u2018model_type\u2019 is a conditional parameter when using GCN and not XGBoostParameterDefaultDescriptionrun_jobsHalf of CPU coresNumber of CPU cores to use for training/predictionrun_depth\u2018ultralight\u2019Depth of parameter optimisation (\u2018ultralight\u2019, \u2018light\u2019, \u2018full\u2019)prob_input0.65Probability threshold for filtering predictionsclust_input0.8Cluster threshold for ranking drugsbatch_s2000Batch size for predictionmodel_type [PyG]\u2018gcn\u2019Type of Graph Neural Network model (\u2018gcn\u2019, \u2018gat\u2019, \u2018sage\u2019)\n\nSample of the parameters for the run_networkmodelml_method() function from the \u2018Drug Prediction\u2019 module: \u2018run_jobs\u2019, \u2018run_depth\u2019 (Table 5) and \u2018batch_size\u2019 regulate the ML prediction step; \u2018prob_input\u2019 and \u2018clust_input\u2019 regulate the filtering of predicted edges according to confidence metrics, compared to Reep\u2019s single filtering threshold of 0.9 [31]; \u2018model_type\u2019 is a conditional parameter when using GCN and not XGBoost", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction of repurposed drugs in binary models with non-binary datasets", "text": "The GCN model takes as training input the node embeddings that only consider existing edges with various labels (i.e. \u20181\u2019 for positive samples, \u20180\u2019 for valid negative samples), thus it only requires one run and it returns all edges that are predicted as positive. XGBoost instead mostly requires a binary mapping of edges that are deemed either positive or negative [59]. To take advantage of the negative samples information, the model runs twice: once where valid negative samples are considered as positive, and once where they are considered as negative. Results from both runs are pooled: any edge that appeared at list once is ranked, and the prediction probabilities is always averaged across runs with \u20180.5\u2019 (random) for runs where the edge does not appear.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction of repurposed drugs in binary models with non-binary datasets", "text": "For quantitative evaluation of model performance, we employed an 80/20 train/test split on gene-to-drug edges from DGIdb. Of the 69,907 total drugs in their database, only 11,603 (16.6%) could be successfully mapped to SMILES representations required for drug similarity calculations (detailed mappability analysis in Section\u00a0\u201cDatabase APIs for biological, gene-drug associations anddrug-drug similarity\u201d), constraining the prediction space to drugs with available structural information. In each evaluation run, 80% of known gene-drug associations were randomly selected for training \u2013including both positive and semantically-valid negative samples generated as described in section \u201cGeneration of semantically-valid negative samples based on centrality measures\u201d\u2014while the remaining 20% were held out for testing. This split ratio follows standard ML practice, providing sufficient training data while maintaining an adequate test set for robust performance estimation. Split selection was performed using a fixed random seed (i.e. seed=42) to ensure reproducibility across experiments. For iteration-based evaluations, the same train/test split was maintained across all iterations\u2013only the model\u2019s random seed was varied\u2013ensuring that performance metrics reflect model stability rather than data variability. Prediction efficiency is also estimated by verifying the presence of a smaller subset gene-to-drug edges (i.e. n=5) that were in the original network, but that were removed ahead of training: such edges were hand-picked based on their centrality score to be relevant enough in the network, yet not too relevant (Table 8).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction of repurposed drugs in binary models with non-binary datasets", "text": "XGBoost model training employs randomised hyperparameter search using scikit-learn\u2019s RandomizedSearchCV() with repeated stratified k-fold cross-validation. The pipeline offers three optimisation depth settings (run_depth parameter): ultralight, light, and full, which control both the hyperparameter search space and cross-validation complexity. The full setting explores a comprehensive parameter grid including min_child_weight (i.e. seven values between 2 and 30), gamma (i.e. nine values between 0 and 6), regularisation parameters (i.e. reg_alpha and reg_lambda, six values each between 0 and 10), subsample and colsample_bytree (i.e. uniform distributions respectively between 0.5 and 1.0, and between 0.2 and 1.0), max_depth (i.e. seven values between 4 and 16), n_estimators (i.e. seven values between 35 and 100), and learning_rate (i.e. uniform from 0 to 0.3), with 10-split 5-repeat cross-validation and 20 random iterations. The light and ultralight settings use reduced parameter grids and fewer cross-validation splits (i.e. 3/2 and 2/1 splits/repeats respectively) to balance optimisation quality with computational efficiency (Table 5). Due to the dual-run embedding XGBoost strategy described above, two separate models are trained with optimised hyperparameters, and predictions from each of the two runs are subsequently pooled.Fig. 14Degree distribution comparison between full and subset networks: Left panel shows degree distribution density plots for full network (blue) and subset network (orange), demonstrating that the subset preserves the characteristic peak near zero degree typical of biological networks\u2014 even though with higher counts due to the smaller network mostly having 1 d.o.d. associations (Appendix A). Right panel presents degree distribution on log-log scale, confirming that both networks exhibit scale-free topology with similar power-law decay patterns. The subset network maintains representative coverage of high-degree hub nodes while reducing overall network size by approximately 95%, enabling computationally feasible iteration-based evaluations. Peaks in both curves represent highly interconnected gene hubs. Complete statistical validation including betweenness centrality preservation (KS-test p>0.05), and entity type proportions are provided in Appendix A", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction of repurposed drugs in binary models with non-binary datasets", "text": "Degree distribution comparison between full and subset networks: Left panel shows degree distribution density plots for full network (blue) and subset network (orange), demonstrating that the subset preserves the characteristic peak near zero degree typical of biological networks\u2014 even though with higher counts due to the smaller network mostly having 1 d.o.d. associations (Appendix A). Right panel presents degree distribution on log-log scale, confirming that both networks exhibit scale-free topology with similar power-law decay patterns. The subset network maintains representative coverage of high-degree hub nodes while reducing overall network size by approximately 95%, enabling computationally feasible iteration-based evaluations. Peaks in both curves represent highly interconnected gene hubs. Complete statistical validation including betweenness centrality preservation (KS-test p>0.05), and entity type proportions are provided in Appendix A", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction of repurposed drugs in binary models with non-binary datasets", "text": "To enable comprehensive evaluation through multiple iterations while maintaining manageable computational requirements, we generated a representative subset network (i.e. 1000 genes, 600 drugs) using PageRank centrality-based stratified sampling that preserves the topological characteristics of the full network (i.e. p>0.05 for betweenness centrality distributions; see Appendix A for complete methodology and statistical validation; Fig.\u00a014), while node type proportions were intentionally enriched for genes and drugs to maximise prediction space. Single-run evaluations and ablation studies were performed on the full network, while iteration-based consensus predictions were evaluated on this validated subset network due to computational constraints (see Appendix A for detailed justification).Table 8Gene-to-drug edges removed for prediction efficiency check: curated subset of edges involving key gene nodes in the network of associations to HD; removed according to section \u201cPrediction of repurposed drugs in binary models with non-binary datasets\u201d; validated in section \u201cPrediction iterations as a solution for seed variability in ML methods\u201dGene IDGene nameDrug IDDrug nameScoreHGNC:20,772TUBB3CHEMBL1173445LARGAZOLE0.0434HGNC:4617GSK3BCHEMBL546797CHEMBL5467970.0355HGNC:4851HTTCHEMBL581880CHEMBL5818800.0576HGNC:4851HTTCHEMBL472437KAMEBANIN0.0432HGNC:3594FASNCHEMBL210208PYROGALLOL RED0.0773", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction of repurposed drugs in binary models with non-binary datasets", "text": "Gene-to-drug edges removed for prediction efficiency check: curated subset of edges involving key gene nodes in the network of associations to HD; removed according to section \u201cPrediction of repurposed drugs in binary models with non-binary datasets\u201d; validated in section \u201cPrediction iterations as a solution for seed variability in ML methods\u201d", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction iterations as a solution for seed variability in ML methods", "text": "Since single runs are influenced by the inherent variability of prediction methods represented by the input seed, we implemented an iteration-based consensus approach that executes multiple runs with different random seeds. For computational feasibility, we performed 10 iterations for XGBoost and 200 iterations for GCN on the validated subset network (Appendix A), thus having shorter runtimes than on the full network (i.e. \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\sim 47$$\\end{document} minutes for XGBoost vs. \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\sim 35$$\\end{document} seconds for GCN).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Prediction iterations as a solution for seed variability in ML methods", "text": "Results from all iterations are collected in a mapping dictionary tracking prediction probabilities across runs. Gene-to-drug edges are then ranked using a consensus strategy based on two criteria: occurrence count across iterations, and average prediction probability. By default, only the top 5% of predictions at a maximum of 100 edges (ml_iter_max_edges=100) from each prediction iteration are labelled as positive and returned in the final results. This consensus approach reduces sensitivity to random seed variation and improves prediction stability (detailed iteration analysis in Appendix B).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Discussion", "text": "Our pipeline differs from current approaches mostly via the real-time integration of data from multiple heterogeneous databases (i.e. Monarch Initiative, DGIdb, ChEMBL, Wikidata), and via the semantic generation of negative samples using combined network centrality and molecular similarity metrics. By not relying on pre-computed predictions and by providing a semantic-based solution to most databases not providing negative samples (i.e. known lack of associations between biological entities), we addresses the critical limitations pointed out by Tanoli et al. on the current state of drug repurposing tools that mostly require curated large-scale data [10] (Section\u00a0\u201cClinical approaches to drug repurposing\u201d).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Discussion", "text": "While Tanoli et al. emphasise the importance of benchmarking and standard performance metrics [10]\u2014which we incorporate in our validation strategy (Section\u00a0\u201cML training and prediction in enriched KG\u201d)\u2014and most catalogued resources focus either on structure-based or knowledge-based approaches in isolation, our framework synthesises both paradigms within a unified semantic network. Furthermore, Tanoli et al. note that the majority of surveyed resources (73% of top-ranked resources) are disease-agnostic, but lack targeted tools for specific stages of the repurposing pipeline for both target-based and phenotype-based approaches [10]. Our contribution directly addresses this gap by developing a disease-specific pipeline that can be adapted to any rare or non-rare disease (RD) for which gene-disease and drug-target associations are available in public databases, while implementing resource-efficient parallelisation strategies to overcome the computational bottlenecks that plague many semi-automated approaches.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Discussion", "text": "As demonstrated by the results in drug repurposing predictions (section \u201cPrediction iterations as a solution for seed variability in ML methods\u201d) and in hardware performance (sections \u201cPrediction of repurposed drugs in binary models withnon-binary datasets\u201d and \u201cPrediction iterations as a solution for seed variability in ML methods\u201d), we successfully developed a fully functioning semi-automated pipeline based on KG and semantic associations. The pipeline implements best practices for research software and data management to improve the reproducibility of results and to adhere to the FAIR software principles:Documentation and discoverability: all development and testing steps involving the full network and smaller test networks\u2014single or multiple iterations\u2014are fully documented and annotated; data originates from publicly accessible databases via reproducible API calls with documented versions and parameters;Accessibility and openness: scripts are available on Github and Gitlab in widely used file formats (e.g..py,.ipynb); data from public databases have been uploaded on Zenodo to provide reference datasets for validation even if they become unavailable from the original source;Modularity and interoperability: pipeline functionalities are divided into distinct modules that they can be independently replaced or scaled up; input and output data structures for each module are well documented; specific modules can be readily modified to accommodate changes in integrated tools (e.g. Monarch Initiative\u2019s V5 API);Reproducibility and reusability: intermediate datasets are preserved at each processing step (i.e. initial, intermediate and final) in several data structures and formats (e.g. DataFrames in.csv, embeddings in.pkl) so that they can also be transformed for other research purposes that have no connection to this drug repurposing pipeline; they also have been uploaded on Zenodo.These design principles align with established guidelines for research data management (FAIR) [21] and research software (FAIR4RS) [26], facilitating validation, extension, and adaptation of the pipeline to other RDs.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Discussion", "text": "Documentation and discoverability: all development and testing steps involving the full network and smaller test networks\u2014single or multiple iterations\u2014are fully documented and annotated; data originates from publicly accessible databases via reproducible API calls with documented versions and parameters;\n\nAccessibility and openness: scripts are available on Github and Gitlab in widely used file formats (e.g..py,.ipynb); data from public databases have been uploaded on Zenodo to provide reference datasets for validation even if they become unavailable from the original source;\n\nModularity and interoperability: pipeline functionalities are divided into distinct modules that they can be independently replaced or scaled up; input and output data structures for each module are well documented; specific modules can be readily modified to accommodate changes in integrated tools (e.g. Monarch Initiative\u2019s V5 API);\n\nReproducibility and reusability: intermediate datasets are preserved at each processing step (i.e. initial, intermediate and final) in several data structures and formats (e.g. DataFrames in.csv, embeddings in.pkl) so that they can also be transformed for other research purposes that have no connection to this drug repurposing pipeline; they also have been uploaded on Zenodo.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Discussion", "text": "In fact, between the previous version of this pipeline developed by Reep and this work the \u2018Gene-to-drug Embeddings\u2019 and \u2018Drug Prediction\u2019 modules\u2014initially only working with XGBoost [31]\u2014have been juxtaposed to those for GCN (Graph Convolution Networks). And the network of biological associations to HD developed by the \u2018Database Calls\u2019 and \u2018Drug-to-Drug\u2019 modules could be used elsewhere.\n\nOur pipeline addresses critical gaps in the current drug repurposing landscape by integrating real-time heterogeneous data from multiple public databases, implementing semantically-informed negative sampling strategies, and providing computationally efficient approaches that enable robust consensus predictions. These advances build upon and extend existing methodologies while maintaining generalisability to RDs where data sparsity remains a fundamental challenge.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Discussion", "text": "This work builds upon and substantially extends the automated drug repurposing workflow initially developed by Reep [31]. We reorganised the pipeline into a core modular architecture, while preserving the same underlying functionality. Methodologically, we update to recent versions of the same APIs used by Reep and we switched from RDF2vec to Node2Vec\u2014as reported in Section\u00a0\u201cPipeline upgrades\u201d). As a result the biological knowledge was expanded compared to 2022 data, and Node2Vec\u2019s enhanced flexibility better accommodates the scale-free and hierarchical characteristics typical of heterogeneous biomedical networks. Perhaps most significantly, we introduce a dual-model architecture implementing both XGBoost\u2014as in Reep\u2014and GCN via PyG to capture complementary aspects of drug-target relationships: XGBoost excels at learning complex feature interactions from concatenated embeddings [46], while GCN leverages graph structure directly through message-passing mechanisms (i.e. a GNN approach) [47, 48]. Paired with an iteration-based consensus approach that executes multiple runs with different random seeds and then aggregates predictions weighted by occurrence frequency and average probability, we reduced the stochastic variability inherent in non-deterministic model training.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Discussion", "text": "Our evaluation strategy extends beyond the F1-score metric employed by Reep to incorporate several others (e.g. AUROC, Recall@K) (Section\u00a0\u201cML training and prediction in enriched KG\u201d). Additionally, we introduced a clustering parameter to further tweak the filtering of repurposed drugs based on how interconnected they are to genes in the network, and as a result we achieve predictions that are consistent with those validated by Reep\u2014suggesting that this approach identifies genuinely promising gene-drug associations.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Discussion", "text": "We addressed specific challenges in RD drug repurposing reported in the literature [10] via several architectural choices described throughout this paper. For example, where most approaches leverage biological knowledge that is fixed in time and specific to a single disease, our heterogeneous KG fetches data from multiple public APIs (e.g. Monarch Initiative, DGIdb) for biological associations in real time\u2014thus positioning it as a hybrid approach that can be run at any point to update results for any disease of interest. Our suggested dual-model strategy with consensus-based ranking further distinguishes our research from single-model implementations [29, 31], as our convergence analysis demonstrates that despite different confidence calibration characteristics, both GCN and XGBoost fundamentally identify similar high-priority drug candidates when filtering constraints are relaxed, achieving 100% overlap at top-10 (Jaccard=1.0) and maintaining 96% agreement at top-100 candidates (Appendix C).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Discussion", "text": "To enable these iteration-based convergence analyses, we evaluated the model on a validated subset network of 1000 genes and 600 drugs (Appendix A) generated via PageRank centrality-based stratified sampling, which preserves the betweenness centrality distributions of the full network (i.e. p>0.05). This was necessary as running the same iteration-based model (i.e. 200 GCN iterations) on the whole network would require much more than the \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\sim 5$$\\end{document} hours necessary on the subset network, making extensive iterative prediction computationally prohibitive with the resources at our disposal. We therefore validated the iterative framework through comprehensive analysis of prediction stability, edge occurrence distributions, and model convergence patterns on this representative subset (Appendices B and C). While these analyses address secondary objectives beyond the pipeline\u2019s core functionality, they demonstrate that given sufficient computational resources \u2013such as those described in Section\u00a0\u201cNext steps\u201d\u2014the dual-model consensus approach could be scaled to the full network, potentially combining predictions from both XGBoost and GCN iterations to further enhance candidate prioritization.Fig. 15Disease-to-drug associations in RepurposeDrugs: Association matrix showing predicted drugs according to the RepurposeDrugs tool as of November 3rd 2025 (courtesy of https://repurposedrugs.aittokallio.group). Highligthed in red the disease of interest (i.e. Huntington\u2019s Disease) and in blue drugs that could be mapped to identifiers compatible with our pipeline\u2019s drug similarity network requirements (i.e. DGIdb identifiers with SMILES availability). The image has been re-oriented and highlighted compared to the original PNG export for better viewingFig. 16Prediction confidence for 37 RepurposeDrugs predictions for Huntington\u2019s disease against our models: Heatmap showing prediction probabilities in our model predictions for drugs identified by RepurposeDrugs as candidates against Huntington\u2019s disease, evaluated across our 6 frameworks as in Fig.\u00a011. Colours represent prediction confidence (darker red = higher probability). Of the 37 drugs predicted by RepurposeDrugs for HD, only 4 could be mapped to identifiers compatible with our pipeline\u2019s drug similarity network requirements (i.e. DGIdb identifiers with SMILES availability), all of which were independently identified by at least one of our modelling approaches. This limited overlap primarily reflects differences in database coverage, compound representation schemes, and structural information availability rather than fundamental methodological disagreement. The comparison highlights complementary strengths: RepurposeDrugs leverages clinical trial outcomes for approval likelihood prediction [76]", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Discussion", "text": "Disease-to-drug associations in RepurposeDrugs: Association matrix showing predicted drugs according to the RepurposeDrugs tool as of November 3rd 2025 (courtesy of https://repurposedrugs.aittokallio.group). Highligthed in red the disease of interest (i.e. Huntington\u2019s Disease) and in blue drugs that could be mapped to identifiers compatible with our pipeline\u2019s drug similarity network requirements (i.e. DGIdb identifiers with SMILES availability). The image has been re-oriented and highlighted compared to the original PNG export for better viewing", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Discussion", "text": "Prediction confidence for 37 RepurposeDrugs predictions for Huntington\u2019s disease against our models: Heatmap showing prediction probabilities in our model predictions for drugs identified by RepurposeDrugs as candidates against Huntington\u2019s disease, evaluated across our 6 frameworks as in Fig.\u00a011. Colours represent prediction confidence (darker red = higher probability). Of the 37 drugs predicted by RepurposeDrugs for HD, only 4 could be mapped to identifiers compatible with our pipeline\u2019s drug similarity network requirements (i.e. DGIdb identifiers with SMILES availability), all of which were independently identified by at least one of our modelling approaches. This limited overlap primarily reflects differences in database coverage, compound representation schemes, and structural information availability rather than fundamental methodological disagreement. The comparison highlights complementary strengths: RepurposeDrugs leverages clinical trial outcomes for approval likelihood prediction [76]", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Discussion", "text": "To contextualise our predictions within the broader drug repurposing landscape, we compared our consensus predictions against RepurposeDrugs [76], a comprehensive web-portal that combines curated drug-indication associations with ML predictions across 4314 drugs and 1756 diseases. Figure\u00a015 illustrates predicted drugs in RepurposeDrugs, and Fig.\u00a016 illustrates a heatmap showing drug candidates against HD predicted by Ianevski et al. and their prediction probability in our own results. Of the 37 drugs predicted by RepurposeDrugs for HD, only 4 could be mapped to identifiers for STRING conversion (i.e. 5 DGIdb identifiers out of 64) -thus being present as enriched drug-to-drug feature similarity edges in our approach\u2019s KG (Section\u00a0\u201cDatabase APIs for biological, gene-drug associations anddrug-drug similarity\u201d. Notably, all 4 mappable drugs were also in the subset network (Section\u00a0\u201cPrediction iterations as a solution for seed variability in ML methods\u201d, Appendix A) and they were independently identified by at least one of our modelling approaches (Fig.\u00a015). These findings suggest convergence on biologically plausible targets despite different methodological approaches and limited overlap due to database heterogeneity.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Discussion", "text": "This comparison highlights methodological complementarity: where RepurposeDrugs leverages large-scale drug-indication associations with clinical trial outcomes to predict approval likelihood [76], our approach constructs disease-specific heterogeneous KGs that integrate phenotypic relationships and molecular interactions through real-time API access. This distinction is significant given that only 39 out of 100 drug repurposing databases surveyed by Tanoli et al. implement APIs for programmatic access [10], whereas our pipeline ensures reproducibility and enables automatic updates as knowledge evolves.\n\nOur evaluation framework faces several important limitations that should be considered when interpreting the results. Computing time remains a challenge despite our optimisation efforts. XGBoost and GCN models could not be run for more than 10 and 200 iterations respectively due to runtimes on the subset network of about 5\u20138\u00a0h (Appendix B)\u2014which lead to high CPU temperatures in a non-dedicated environment. This computational constraint limited our ability to perform more extensive iterations on the full network, necessitating the use of a validated subset for iteration-based analyses as described in Section\u00a0\u201cComparison to current state-of-the-art drug repurposingmethods\u201d.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Discussion", "text": "A fundamental limitation of our negative sampling approach is the potential presence of unknown true positives within the negative sample set. While we employed centrality and similarity-based filtering to generate semantically-valid negatives, some gene-drug pairs currently lacking documented associations may represent genuine\u2014yet undiscovered\u2014therapeutic relationships. This \"unknown positive\" problem is inherent to drug repurposing research and means our model performance metrics such as AUROC and AUPRC may underestimate true predictive capability, as the model may be correctly identifying valid associations that are currently labelled as negatives in our evaluation framework.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Discussion", "text": "A technical limitation of our drug similarity analysis is that only 16.6% of DGIdb drugs (i.e. 11,603 out of 69,907) could be successfully mapped to SMILES representations for molecular similarity calculations\u2014as reported in Section\u00a0\u201cDatabase APIs for biological, gene-drug associations anddrug-drug similarity\u201d. This occurs because many drug entries in DGIdb lack ChEMBL or Wikidata identifiers, or these identifiers cannot be resolved to chemical structures. As a result, the drug-drug similarity network used in negative sample generation is sparser than ideal. This sparsity explains why the similarity-only negative sampling strategy showed limited standalone effectiveness\u2014with only 219 similarity-based negative samples in the ablation study compared to 2000 centrality-based negatives (Section\u00a0\u201cGeneration of semantically-valid negative samples based oncentrality measures\u201d, Appendix D). Nonetheless, the combined centrality and similarity approach still achieved the best overall performance, thus these limitations do not affect the primary prediction task\u2014as drug similarity is used only for negative sample refinement rather than for defining the prediction space itself.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Discussion", "text": "Model-wise, GCN returns too many edges\u2014therefore filtering is still too dependent on prediction probability alone since more than 100,000 edges occur in every run. Yet, the fact that no edge occurs for less than 90% of the times is an indicator that the prediction is consistent. XGBoost on the other hand provides a better occurrences distribution where the number of edges that are returned in all iterations is low enough to provide a meaningful result. Moreover, the fact that most of the edges only return once proves that the iterations are necessary to overcome the inconsistency issues in this model. These differences in prediction behaviour between the two models highlight the value of the dual-model consensus approach, though they also indicate that optimal filtering strategies may need to be model-specific rather than universal.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Discussion", "text": "Our validation approach relies primarily on literature-based evidence through PubChem lookups and comparison with previously validated predictions [31, 76]. While we demonstrate computational consistency and convergence between architecturally distinct models, we have not performed systematic cross-referencing with clinical trial databases or comprehensive literature mining beyond individual candidate verification. This literature-based validation provides supporting evidence for predicted associations\u2014but cannot substitute for experimental validation. Ultimately, the predicted gene-drug associations require experimental validation through in vitro or in vivo studies to assess therapeutic efficacy and safety profiles in HD contexts. Our pipeline serves to prioritise candidates for further investigation, but wet-lab validation, preclinical studies, and eventually clinical trials remain essential next steps to translate computational predictions into therapeutic interventions for RD patients.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Discussion", "text": "Future work should include more extensive iteration-based validation with diverse held-out edge sets to comprehensively assess prediction stability and generalisation across different network configurations. The computational infrastructure improvements described in Section\u00a0\u201cNext steps\u201d would enable such expanded validation efforts on the full network, potentially addressing several of the limitations identified here through increased iteration counts and more comprehensive baseline comparisons.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Discussion", "text": "The improvements we have made\u2014compared to the source pipeline by Reep\u2014lead to having even more coherent predictions for drugs related to neurological diseases\u2014as mentioned in Section\u00a0\u201cML training and prediction in enriched KG\u201d [31]: top results for each model run consistently show the same gene hubs that are associated with phenotypes more than once in filtered predicted gene-to-drug edges. This is also true for drugs predicted in the full network runs that are unlikely candidates against HD based on ChEMBL validation. As an example, Derrusnin is associated to the CYP3A4 gene involved in Vitamin D-dependent rickets, yet also Methyl Robustone\u2014a drug that is instead indeed validated in section \u201cPrediction iterations as a solution for seed variability in ML methods\u201d as a viable candidate drug against HD\u2014is associated to CYP3A4. This suggests that the network of association itself is rich and predictions are consistent, yet there is still room for improvement in how results are filtered based on the functional roles of genes within the network.\n\nOur contributions are in the generation of semantically-valid negative samples, and in the proposed iteration-consensus approach to mitigate the variability of non-deterministic ML approaches.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Discussion", "text": "Future work should include more extensive iteration-based validation with diverse held-out edge sets to comprehensively assess prediction stability and generalisation across different network configurations \u2013as suggested in Appendix C. Systematic investigation of how many predicted drugs exhibit patterns of improved confidence through iteration-based validation\u2014as exemplified by Ergocristine (Section\u00a0\u201cPrediction iterations as a solution for seed variability in ML methods\u201d)\u2014may reveal therapeutically relevant compounds that operate through mechanisms not fully captured by gene-disease association networks. More comprehensive experimental validation through clinical trial database cross-referencing, pathway-level analyses, and drug-target interaction (DTI) integration [77\u201381] represents important directions for enhancing biological validation, while systematic comparison against established prediction platforms like RepurposeDrugs offers a pragmatic intermediate validation strategy that balances computational reproducibility with biological credibility. Ultimately, wet-lab experimental validation remains the essential next step to confirm therapeutic potential of computationally predicted gene-drug associations.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Discussion", "text": "There is the option to scale up this drug repurposing pipeline by taking advantage of a multi-Docker structure: one main container handles the launch of pipeline runs, one operational container runs the actual steps of the pipeline, while all other pipeline modules (e.g. \u2018Database calls\u2019 for the steps involving Monarch.py and dgidb.py) are handled by their own containers. The advantage of this structure is that each module can have multiple instance, thus allowing for concurrent operations. Moreover, a Docker container is fully isolated and easy to build thanks to the necessary instruction files. The user only has to install the Docker desktop client and launch a single command without additional coding skills. Finally, such a setup can be installed on a dedicated device with optimised CPU or GPU for longer run times.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Discussion", "text": "Once implemented fully, the computational infrastructure improvements described above would enable expanded validation efforts on the full network, potentially addressing several of the limitations identified in Section\u00a0\u201cLimitations\u201d through increased iteration counts and more comprehensive baseline comparisons. Pipeline runs could be performed with iterative and parallelised prediction steps on the entire network\u2014and not just on a subset of it as in Sections\u00a0\u201cML training and prediction in enriched KG\u201d and \u201cML training and prediction in enriched KG\u201d. Results could be validated in PubChem. Finally, home-keeping genes could be filtered out for prediction, if they could be identified via an ontology.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Achievements", "text": "As demonstrated by the results in drug repurposing predictions (section \u201cPrediction iterations as a solution for seed variability in ML methods\u201d) and in hardware performance (sections \u201cPrediction of repurposed drugs in binary models withnon-binary datasets\u201d and \u201cPrediction iterations as a solution for seed variability in ML methods\u201d), we successfully developed a fully functioning semi-automated pipeline based on KG and semantic associations. The pipeline implements best practices for research software and data management to improve the reproducibility of results and to adhere to the FAIR software principles:Documentation and discoverability: all development and testing steps involving the full network and smaller test networks\u2014single or multiple iterations\u2014are fully documented and annotated; data originates from publicly accessible databases via reproducible API calls with documented versions and parameters;Accessibility and openness: scripts are available on Github and Gitlab in widely used file formats (e.g..py,.ipynb); data from public databases have been uploaded on Zenodo to provide reference datasets for validation even if they become unavailable from the original source;Modularity and interoperability: pipeline functionalities are divided into distinct modules that they can be independently replaced or scaled up; input and output data structures for each module are well documented; specific modules can be readily modified to accommodate changes in integrated tools (e.g. Monarch Initiative\u2019s V5 API);Reproducibility and reusability: intermediate datasets are preserved at each processing step (i.e. initial, intermediate and final) in several data structures and formats (e.g. DataFrames in.csv, embeddings in.pkl) so that they can also be transformed for other research purposes that have no connection to this drug repurposing pipeline; they also have been uploaded on Zenodo.These design principles align with established guidelines for research data management (FAIR) [21] and research software (FAIR4RS) [26], facilitating validation, extension, and adaptation of the pipeline to other RDs.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Achievements", "text": "Documentation and discoverability: all development and testing steps involving the full network and smaller test networks\u2014single or multiple iterations\u2014are fully documented and annotated; data originates from publicly accessible databases via reproducible API calls with documented versions and parameters;\n\nAccessibility and openness: scripts are available on Github and Gitlab in widely used file formats (e.g..py,.ipynb); data from public databases have been uploaded on Zenodo to provide reference datasets for validation even if they become unavailable from the original source;\n\nModularity and interoperability: pipeline functionalities are divided into distinct modules that they can be independently replaced or scaled up; input and output data structures for each module are well documented; specific modules can be readily modified to accommodate changes in integrated tools (e.g. Monarch Initiative\u2019s V5 API);\n\nReproducibility and reusability: intermediate datasets are preserved at each processing step (i.e. initial, intermediate and final) in several data structures and formats (e.g. DataFrames in.csv, embeddings in.pkl) so that they can also be transformed for other research purposes that have no connection to this drug repurposing pipeline; they also have been uploaded on Zenodo.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Achievements", "text": "In fact, between the previous version of this pipeline developed by Reep and this work the \u2018Gene-to-drug Embeddings\u2019 and \u2018Drug Prediction\u2019 modules\u2014initially only working with XGBoost [31]\u2014have been juxtaposed to those for GCN (Graph Convolution Networks). And the network of biological associations to HD developed by the \u2018Database Calls\u2019 and \u2018Drug-to-Drug\u2019 modules could be used elsewhere.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Comparison to current state-of-the-art drug repurposing methods", "text": "Our pipeline addresses critical gaps in the current drug repurposing landscape by integrating real-time heterogeneous data from multiple public databases, implementing semantically-informed negative sampling strategies, and providing computationally efficient approaches that enable robust consensus predictions. These advances build upon and extend existing methodologies while maintaining generalisability to RDs where data sparsity remains a fundamental challenge.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Comparison to current state-of-the-art drug repurposing methods", "text": "This work builds upon and substantially extends the automated drug repurposing workflow initially developed by Reep [31]. We reorganised the pipeline into a core modular architecture, while preserving the same underlying functionality. Methodologically, we update to recent versions of the same APIs used by Reep and we switched from RDF2vec to Node2Vec\u2014as reported in Section\u00a0\u201cPipeline upgrades\u201d). As a result the biological knowledge was expanded compared to 2022 data, and Node2Vec\u2019s enhanced flexibility better accommodates the scale-free and hierarchical characteristics typical of heterogeneous biomedical networks. Perhaps most significantly, we introduce a dual-model architecture implementing both XGBoost\u2014as in Reep\u2014and GCN via PyG to capture complementary aspects of drug-target relationships: XGBoost excels at learning complex feature interactions from concatenated embeddings [46], while GCN leverages graph structure directly through message-passing mechanisms (i.e. a GNN approach) [47, 48]. Paired with an iteration-based consensus approach that executes multiple runs with different random seeds and then aggregates predictions weighted by occurrence frequency and average probability, we reduced the stochastic variability inherent in non-deterministic model training.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Comparison to current state-of-the-art drug repurposing methods", "text": "Our evaluation strategy extends beyond the F1-score metric employed by Reep to incorporate several others (e.g. AUROC, Recall@K) (Section\u00a0\u201cML training and prediction in enriched KG\u201d). Additionally, we introduced a clustering parameter to further tweak the filtering of repurposed drugs based on how interconnected they are to genes in the network, and as a result we achieve predictions that are consistent with those validated by Reep\u2014suggesting that this approach identifies genuinely promising gene-drug associations.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Comparison to current state-of-the-art drug repurposing methods", "text": "We addressed specific challenges in RD drug repurposing reported in the literature [10] via several architectural choices described throughout this paper. For example, where most approaches leverage biological knowledge that is fixed in time and specific to a single disease, our heterogeneous KG fetches data from multiple public APIs (e.g. Monarch Initiative, DGIdb) for biological associations in real time\u2014thus positioning it as a hybrid approach that can be run at any point to update results for any disease of interest. Our suggested dual-model strategy with consensus-based ranking further distinguishes our research from single-model implementations [29, 31], as our convergence analysis demonstrates that despite different confidence calibration characteristics, both GCN and XGBoost fundamentally identify similar high-priority drug candidates when filtering constraints are relaxed, achieving 100% overlap at top-10 (Jaccard=1.0) and maintaining 96% agreement at top-100 candidates (Appendix C).", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Comparison to current state-of-the-art drug repurposing methods", "text": "To enable these iteration-based convergence analyses, we evaluated the model on a validated subset network of 1000 genes and 600 drugs (Appendix A) generated via PageRank centrality-based stratified sampling, which preserves the betweenness centrality distributions of the full network (i.e. p>0.05). This was necessary as running the same iteration-based model (i.e. 200 GCN iterations) on the whole network would require much more than the \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\sim 5$$\\end{document} hours necessary on the subset network, making extensive iterative prediction computationally prohibitive with the resources at our disposal. We therefore validated the iterative framework through comprehensive analysis of prediction stability, edge occurrence distributions, and model convergence patterns on this representative subset (Appendices B and C). While these analyses address secondary objectives beyond the pipeline\u2019s core functionality, they demonstrate that given sufficient computational resources \u2013such as those described in Section\u00a0\u201cNext steps\u201d\u2014the dual-model consensus approach could be scaled to the full network, potentially combining predictions from both XGBoost and GCN iterations to further enhance candidate prioritization.Fig. 15Disease-to-drug associations in RepurposeDrugs: Association matrix showing predicted drugs according to the RepurposeDrugs tool as of November 3rd 2025 (courtesy of https://repurposedrugs.aittokallio.group). Highligthed in red the disease of interest (i.e. Huntington\u2019s Disease) and in blue drugs that could be mapped to identifiers compatible with our pipeline\u2019s drug similarity network requirements (i.e. DGIdb identifiers with SMILES availability). The image has been re-oriented and highlighted compared to the original PNG export for better viewingFig. 16Prediction confidence for 37 RepurposeDrugs predictions for Huntington\u2019s disease against our models: Heatmap showing prediction probabilities in our model predictions for drugs identified by RepurposeDrugs as candidates against Huntington\u2019s disease, evaluated across our 6 frameworks as in Fig.\u00a011. Colours represent prediction confidence (darker red = higher probability). Of the 37 drugs predicted by RepurposeDrugs for HD, only 4 could be mapped to identifiers compatible with our pipeline\u2019s drug similarity network requirements (i.e. DGIdb identifiers with SMILES availability), all of which were independently identified by at least one of our modelling approaches. This limited overlap primarily reflects differences in database coverage, compound representation schemes, and structural information availability rather than fundamental methodological disagreement. The comparison highlights complementary strengths: RepurposeDrugs leverages clinical trial outcomes for approval likelihood prediction [76]", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Comparison to current state-of-the-art drug repurposing methods", "text": "Disease-to-drug associations in RepurposeDrugs: Association matrix showing predicted drugs according to the RepurposeDrugs tool as of November 3rd 2025 (courtesy of https://repurposedrugs.aittokallio.group). Highligthed in red the disease of interest (i.e. Huntington\u2019s Disease) and in blue drugs that could be mapped to identifiers compatible with our pipeline\u2019s drug similarity network requirements (i.e. DGIdb identifiers with SMILES availability). The image has been re-oriented and highlighted compared to the original PNG export for better viewing", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Comparison to current state-of-the-art drug repurposing methods", "text": "Prediction confidence for 37 RepurposeDrugs predictions for Huntington\u2019s disease against our models: Heatmap showing prediction probabilities in our model predictions for drugs identified by RepurposeDrugs as candidates against Huntington\u2019s disease, evaluated across our 6 frameworks as in Fig.\u00a011. Colours represent prediction confidence (darker red = higher probability). Of the 37 drugs predicted by RepurposeDrugs for HD, only 4 could be mapped to identifiers compatible with our pipeline\u2019s drug similarity network requirements (i.e. DGIdb identifiers with SMILES availability), all of which were independently identified by at least one of our modelling approaches. This limited overlap primarily reflects differences in database coverage, compound representation schemes, and structural information availability rather than fundamental methodological disagreement. The comparison highlights complementary strengths: RepurposeDrugs leverages clinical trial outcomes for approval likelihood prediction [76]", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Comparison to current state-of-the-art drug repurposing methods", "text": "To contextualise our predictions within the broader drug repurposing landscape, we compared our consensus predictions against RepurposeDrugs [76], a comprehensive web-portal that combines curated drug-indication associations with ML predictions across 4314 drugs and 1756 diseases. Figure\u00a015 illustrates predicted drugs in RepurposeDrugs, and Fig.\u00a016 illustrates a heatmap showing drug candidates against HD predicted by Ianevski et al. and their prediction probability in our own results. Of the 37 drugs predicted by RepurposeDrugs for HD, only 4 could be mapped to identifiers for STRING conversion (i.e. 5 DGIdb identifiers out of 64) -thus being present as enriched drug-to-drug feature similarity edges in our approach\u2019s KG (Section\u00a0\u201cDatabase APIs for biological, gene-drug associations anddrug-drug similarity\u201d. Notably, all 4 mappable drugs were also in the subset network (Section\u00a0\u201cPrediction iterations as a solution for seed variability in ML methods\u201d, Appendix A) and they were independently identified by at least one of our modelling approaches (Fig.\u00a015). These findings suggest convergence on biologically plausible targets despite different methodological approaches and limited overlap due to database heterogeneity.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Comparison to current state-of-the-art drug repurposing methods", "text": "This comparison highlights methodological complementarity: where RepurposeDrugs leverages large-scale drug-indication associations with clinical trial outcomes to predict approval likelihood [76], our approach constructs disease-specific heterogeneous KGs that integrate phenotypic relationships and molecular interactions through real-time API access. This distinction is significant given that only 39 out of 100 drug repurposing databases surveyed by Tanoli et al. implement APIs for programmatic access [10], whereas our pipeline ensures reproducibility and enables automatic updates as knowledge evolves.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Limitations", "text": "Our evaluation framework faces several important limitations that should be considered when interpreting the results. Computing time remains a challenge despite our optimisation efforts. XGBoost and GCN models could not be run for more than 10 and 200 iterations respectively due to runtimes on the subset network of about 5\u20138\u00a0h (Appendix B)\u2014which lead to high CPU temperatures in a non-dedicated environment. This computational constraint limited our ability to perform more extensive iterations on the full network, necessitating the use of a validated subset for iteration-based analyses as described in Section\u00a0\u201cComparison to current state-of-the-art drug repurposingmethods\u201d.\n\nA fundamental limitation of our negative sampling approach is the potential presence of unknown true positives within the negative sample set. While we employed centrality and similarity-based filtering to generate semantically-valid negatives, some gene-drug pairs currently lacking documented associations may represent genuine\u2014yet undiscovered\u2014therapeutic relationships. This \"unknown positive\" problem is inherent to drug repurposing research and means our model performance metrics such as AUROC and AUPRC may underestimate true predictive capability, as the model may be correctly identifying valid associations that are currently labelled as negatives in our evaluation framework.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Limitations", "text": "A technical limitation of our drug similarity analysis is that only 16.6% of DGIdb drugs (i.e. 11,603 out of 69,907) could be successfully mapped to SMILES representations for molecular similarity calculations\u2014as reported in Section\u00a0\u201cDatabase APIs for biological, gene-drug associations anddrug-drug similarity\u201d. This occurs because many drug entries in DGIdb lack ChEMBL or Wikidata identifiers, or these identifiers cannot be resolved to chemical structures. As a result, the drug-drug similarity network used in negative sample generation is sparser than ideal. This sparsity explains why the similarity-only negative sampling strategy showed limited standalone effectiveness\u2014with only 219 similarity-based negative samples in the ablation study compared to 2000 centrality-based negatives (Section\u00a0\u201cGeneration of semantically-valid negative samples based oncentrality measures\u201d, Appendix D). Nonetheless, the combined centrality and similarity approach still achieved the best overall performance, thus these limitations do not affect the primary prediction task\u2014as drug similarity is used only for negative sample refinement rather than for defining the prediction space itself.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Limitations", "text": "Model-wise, GCN returns too many edges\u2014therefore filtering is still too dependent on prediction probability alone since more than 100,000 edges occur in every run. Yet, the fact that no edge occurs for less than 90% of the times is an indicator that the prediction is consistent. XGBoost on the other hand provides a better occurrences distribution where the number of edges that are returned in all iterations is low enough to provide a meaningful result. Moreover, the fact that most of the edges only return once proves that the iterations are necessary to overcome the inconsistency issues in this model. These differences in prediction behaviour between the two models highlight the value of the dual-model consensus approach, though they also indicate that optimal filtering strategies may need to be model-specific rather than universal.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Limitations", "text": "Our validation approach relies primarily on literature-based evidence through PubChem lookups and comparison with previously validated predictions [31, 76]. While we demonstrate computational consistency and convergence between architecturally distinct models, we have not performed systematic cross-referencing with clinical trial databases or comprehensive literature mining beyond individual candidate verification. This literature-based validation provides supporting evidence for predicted associations\u2014but cannot substitute for experimental validation. Ultimately, the predicted gene-drug associations require experimental validation through in vitro or in vivo studies to assess therapeutic efficacy and safety profiles in HD contexts. Our pipeline serves to prioritise candidates for further investigation, but wet-lab validation, preclinical studies, and eventually clinical trials remain essential next steps to translate computational predictions into therapeutic interventions for RD patients.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Limitations", "text": "Future work should include more extensive iteration-based validation with diverse held-out edge sets to comprehensively assess prediction stability and generalisation across different network configurations. The computational infrastructure improvements described in Section\u00a0\u201cNext steps\u201d would enable such expanded validation efforts on the full network, potentially addressing several of the limitations identified here through increased iteration counts and more comprehensive baseline comparisons.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Contributions to drug repurposing", "text": "The improvements we have made\u2014compared to the source pipeline by Reep\u2014lead to having even more coherent predictions for drugs related to neurological diseases\u2014as mentioned in Section\u00a0\u201cML training and prediction in enriched KG\u201d [31]: top results for each model run consistently show the same gene hubs that are associated with phenotypes more than once in filtered predicted gene-to-drug edges. This is also true for drugs predicted in the full network runs that are unlikely candidates against HD based on ChEMBL validation. As an example, Derrusnin is associated to the CYP3A4 gene involved in Vitamin D-dependent rickets, yet also Methyl Robustone\u2014a drug that is instead indeed validated in section \u201cPrediction iterations as a solution for seed variability in ML methods\u201d as a viable candidate drug against HD\u2014is associated to CYP3A4. This suggests that the network of association itself is rich and predictions are consistent, yet there is still room for improvement in how results are filtered based on the functional roles of genes within the network.\n\nOur contributions are in the generation of semantically-valid negative samples, and in the proposed iteration-consensus approach to mitigate the variability of non-deterministic ML approaches.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Next steps", "text": "Future work should include more extensive iteration-based validation with diverse held-out edge sets to comprehensively assess prediction stability and generalisation across different network configurations \u2013as suggested in Appendix C. Systematic investigation of how many predicted drugs exhibit patterns of improved confidence through iteration-based validation\u2014as exemplified by Ergocristine (Section\u00a0\u201cPrediction iterations as a solution for seed variability in ML methods\u201d)\u2014may reveal therapeutically relevant compounds that operate through mechanisms not fully captured by gene-disease association networks. More comprehensive experimental validation through clinical trial database cross-referencing, pathway-level analyses, and drug-target interaction (DTI) integration [77\u201381] represents important directions for enhancing biological validation, while systematic comparison against established prediction platforms like RepurposeDrugs offers a pragmatic intermediate validation strategy that balances computational reproducibility with biological credibility. Ultimately, wet-lab experimental validation remains the essential next step to confirm therapeutic potential of computationally predicted gene-drug associations.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Next steps", "text": "There is the option to scale up this drug repurposing pipeline by taking advantage of a multi-Docker structure: one main container handles the launch of pipeline runs, one operational container runs the actual steps of the pipeline, while all other pipeline modules (e.g. \u2018Database calls\u2019 for the steps involving Monarch.py and dgidb.py) are handled by their own containers. The advantage of this structure is that each module can have multiple instance, thus allowing for concurrent operations. Moreover, a Docker container is fully isolated and easy to build thanks to the necessary instruction files. The user only has to install the Docker desktop client and launch a single command without additional coding skills. Finally, such a setup can be installed on a dedicated device with optimised CPU or GPU for longer run times.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Next steps", "text": "Once implemented fully, the computational infrastructure improvements described above would enable expanded validation efforts on the full network, potentially addressing several of the limitations identified in Section\u00a0\u201cLimitations\u201d through increased iteration counts and more comprehensive baseline comparisons. Pipeline runs could be performed with iterative and parallelised prediction steps on the entire network\u2014and not just on a subset of it as in Sections\u00a0\u201cML training and prediction in enriched KG\u201d and \u201cML training and prediction in enriched KG\u201d. Results could be validated in PubChem. Finally, home-keeping genes could be filtered out for prediction, if they could be identified via an ontology.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Conclusion", "text": "The combination of generating semantically valid negative samples and iterating through predictions with different seeds proved to be beneficial in developing a structured framework for drug repurposing.\n\nThe use of individual modules for specific operations allows for our project\u2019s adherence to FAIR and FAIR4RS principles, and the introduction of iterations loops as a solution to the inherent variability of results from the ML steps by selecting the most recurring predicted gene-to-drug edges in the drug repurposing pipeline. Specifically, iterations of runs using XGBoost select the few edges that return several times across the very selective -yet highly variable- individual runs; while iterations of runs using GCN allow to identify the most meaningful predicted edges among the very many returned\u2014rather consistently\u2014across the individual runs. Combined, the results from iteration-consensus with the two models lead to predicted drugs that are recurring with higher prediction probability and stability across iterations in both XGBoost and GCN. Moreover, both approaches are optimised via process parallelisation that leads to shorter runs compared to runs in series without any parallelisation.", "title": "" }, { "pmid": 41639620, "pmcid": "PMC12964631", "section": "Conclusion", "text": "The end result is a highly customisable pipeline that that bridges the gap between computational predictions and clinical possibilities for patients affected by RDs. Semantic biological networks and ML algorithms are leveraged within a modular framework to systematically identify overlooked therapeutic opportunities, bringing us closer to delivering effective treatments for HD and other underserved RDs.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Introduction", "text": "Huntington's disease (HD) is a neurodegenerative disorder caused by an abnormal expansion of CAG trinucleotide repeats in the HTT gene on chromosome 41. This mutation triggers a cascade of progressive neuropathological changes, ultimately leading to neuronal dysfunction and death, particularly in the striatum, although widespread involvement of cortical regions and white matter is also observed (Ross and Tabrizi, 2011; Rosas et al., 2008; Adanyeguh et al., 2021).\n\nCognitive decline is a core feature of HD, typically emerging years before the onset of motor symptoms (Snowden, 2017). This decline follows a predominantly fronto-striatal pattern, progressively affecting multiple cognitive domains and culminating in a dementia syndrome (Martinez-Horta et al., 2023). While features such as psychomotor slowing, attentional deficits, and frontal-dysexecutive dysfunction are prominent, the cognitive phenotype of HD is heterogeneous, both in terms of domain involvement and progression trajectory across individuals (Snowden, 2017; Martinez-Horta et al., 2023; Horta-Barba et al., 2023; Sierra et al., 2023).", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Introduction", "text": "Longitudinal studies have shown that tests like the Symbol Digit Modalities Test (SDMT) and Stroop Word Reading are sensitive to cognitive changes in HD (Tabrizi et al., 2012; Schobel et al., 2017). While useful as markers of disease progression, they provide limited information on patients' overall cognitive status (Martinez-Horta et al., 2020a; Mestre et al., 2018). Although not a replacement for full neuropsychological assessment, screening tools offer a practical, time-efficient way to detect cognitive impairment and they are particularly valuable in longitudinal studies and clinical trials, where consistent and reproducible measures are crucial for tracking change and comparing outcomes (Litvan et al., 2012; Marras et al., 2014; Reber et al., 2021).", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Introduction", "text": "Several screening instruments have been employed in HD (Begeti et al., 2013; Folstein et al., 1975; Group, 1996; Rosca and Simu, 2022; Rosen et al., 1984; Rosser and Hodges, 1994; Stout et al., 2014). However, a systematic review of these tools concluded that none currently fulfil the criteria for general recommendation in HD (Mestre et al., 2018). In response to this gap, our group conducted a series of studies evaluating the Parkinson's Disease - Cognitive Rating Scale (PD-CRS) in the context of HD (Horta-Barba et al., 2023; Martinez-Horta et al., 2020a, 2020b). These investigations demonstrated the tool's strong psychometric properties, its ability to differentiate between cognitive states, and its sensitivity to longitudinal changes, suggesting its potential as a viable screening instrument for global cognitive assessment in HD (Horta-Barba et al., 2023; Martinez-Horta et al., 2020a).", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Introduction", "text": "Beyond psychometric validation, identifying the neuroanatomical correlates of cognitive screening tool performance is key to linking test scores with brain pathology. However, few studies have examined the structural correlates of such tools in HD, limiting efforts to biologically ground their clinical use. To address this, we investigate the neuroanatomical correlates of PD-CRS performance in individuals with HD. Using voxel-based morphometry (VBM), we examine associations between PD-CRS scores and regional gray matter volume (GMV) in a well-characterized cohort. We also assess plasma neurofilament light chain (NfL) levels in relation to cognition and brain atrophy. By combining imaging and fluid biomarkers, we aim to validate the PD-CRS as both a clinically useful screening tool and a marker of disease-related neurodegeneration.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Methods", "text": "Fifty-two symptomatic gene-expansion carriers (CAG >39) were recruited from the Movement Disorders Unit at Hospital de la Santa Creu i Sant Pau (Barcelona). Demographic and clinical data were recorded. Motor severity was assessed via the UHDRS Total Motor Score (UHDRS-TMS), and functional status via the Total Functional Capacity (TFC) and Functional Independence Scale (FIS)(Group, 1996). Neuropsychiatric symptoms were evaluated using the Problem Behaviors Assessment-Short Form (PBA-s), using the product of frequency and severity scores (Callaghan et al., 2015).\n\nParticipants were classified into HD-ISS stages 1\u20133 based on biological criteria (Tabrizi et al., 2022). Stage 0 individuals were excluded, as they show no detectable neuropathology. Stage 1 participants were included despite being premanifest, given biomarker evidence of early neurodegeneration. Cumulative CAG-related disease burden was estimated using the CAP score: age\u00a0\u00d7\u00a0(CAG - 33.66) (Warner et al., 2022).\n\nExclusion criteria included other neurological/psychiatric disorders, traumatic brain injury with loss of consciousness, uncontrolled medical illness, active substance abuse, or psychosis. Patients unable to understand or perform study tasks were also excluded.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Methods", "text": "Cognitive functioning was evaluated in all participants using the PD-CRS (Pagonabarraga et al., 2008). Previous work from our group has validated the PD-CRS as a sensitive tool for staging cognitive impairment in HD, distinguishing between normal cognition, mild cognitive impairment (MCI), and dementia. The PD-CRS is composed of nine subtests assessing immediate verbal memory, confrontation naming, sustained attention, verbal working memory, unprompted clock drawing, clock copy, delayed free recall, alternating verbal fluency, and action verbal fluency. The posterior\u2013cortical subscore (range 0\u201330) is derived from confrontation naming and the clock copy task, while the remaining subtests contribute to the frontal\u2013subcortical subscore (range 0\u2013104). The sum of all subtests yields the total score, which ranges from 0 to 134, with higher scores indicating better cognitive performance. The PD-CRS A total score of \u226482 optimally identifies MCI, typically associated with minimal functional impact, while a score of \u226464 indicates dementia-level impairment with significant loss of autonomy. These thresholds have been empirically linked to functional outcomes, supporting the PD-CRS as both practical and clinically relevant for monitoring cognitive decline in HD (Horta-Barba et al., 2023; Martinez-Horta et al., 2020a).", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Methods", "text": "High-resolution T1-weighted MRI images were obtained on a 3T Philips Achieva scanner using an MPRAGE sequence (TR/TE\u00a0=\u00a012.65/7.08\u00a0ms, flip angle\u00a0=\u00a08\u00b0, field of view\u00a0=\u00a023\u00a0cm, matrix\u00a0=\u00a0256\u00a0\u00d7\u00a0256, slice thickness\u00a0=\u00a01\u00a0mm). Standard VBM procedures were performed using the SPM12 software package. (http://www.fil.ion.ucl.ac.uk/spm). GMV tissue probability maps were generated from the T1-weighted scans and subsequently normalized to the Montreal Neurological Institute (MNI) space using the DARTEL algorithm. To account for inter-individual variability, the normalized GMV maps were smoothed with an isotropic Gaussian filter with a full-width at half-maximum (FWHM) of 8\u00a0mm in all dimensions.\n\nPeripheral blood samples were collected into EDTA tubes, centrifuged (2000\u00d7g, 10\u00a0min), and plasma was aliquoted and stored at \u221280\u00a0\u00b0C per international guidelines. Plasma NfL levels were measured using the Simoa\u00ae Human NF-light Advantage Kit (Quanterix) on the SR-X\u2122 system, following the manufacturer's protocol. Calibration standards, quality controls, and duplicate samples were run on a 96-well plate. After incubation with antibody-coated beads and biotinylated detection antibodies, SBG enzyme and RGP substrate were added. Fluorescent signals were detected and quantified using a four-parameter logistic curve. All samples were analyzed in duplicate; intra-assay CVs remained below 20\u00a0%.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Methods", "text": "Descriptive statistics were computed to summarize sociodemographic, clinical, cognitive, and behavioral characteristics. Continuous variables were expressed as means and standard deviations, while categorical variables were summarized as frequencies and percentages. Group comparisons across HD-ISS stages were performed using one-way analyses of variance (ANOVA) for normally distributed variables, followed by Tukey's HSD post hoc tests to identify specific between-group differences. For pairwise comparisons where homogeneity of variance was violated, Welch's t-tests were applied. Normality assumptions were evaluated via Shapiro-Wilk tests and visual inspection of histograms and Q-Q plots. Test statistics (F or t), degrees of freedom, p-values, and effect sizes (Cohen's d) were reported as appropriate.\n\nCognitive staging was based on PD-CRS total score thresholds, classifying participants as cognitively normal, mildly impaired (PD-CRS \u226482), or demented (PD-CRS \u226464)7,11. To assess relative performance across PD-CRS subdomains, raw scores were standardized into z-scores, which were calculated using the mean and standard deviation of the internal study cohort. This within-cohort normalization accounted for differing scoring ranges across subtests (e.g., action fluency max\u00a0=\u00a030 vs. clock drawing max\u00a0=\u00a010), thereby enabling direct item-level comparisons and clearer identification of the domains most sensitive to disease-related cognitive decline.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Methods", "text": "Multiple linear regression analyses were conducted to identify clinical and behavioral predictors of global cognitive performance controlling for potential confounders. Standardized beta coefficients (B), associated p-values, and R2 values were reported. To investigate structural correlates of cognitive function, GMV maps from VBM analyses were included in a general linear model (GLM). Two analytical frameworks were implemented according to the primary target of inference: (i) PD-CRS total scores were modelled as the dependent variable when assessing the contribution of motor severity, perseverative behaviours and demographic variables to cognitive performance; and (ii) regional GMV from VBM-derived clusters was modelled as the dependent variable when assessing neuroanatomical correlates of cognitive impairment. Accordingly, PD-CRS served as predictor or outcome depending on the objective of each model. Additionally, exploratory stage-stratified analyses of the VBM-derived clusters were conducted across HD-ISS stages 1\u20133. Whole-brain VBM analyses applied cluster-level family-wise error (FWE) correction (p\u00a0<\u00a00.05) using random field theory (RFT) in SPM12. Stage-stratified and ROI-based analyses were considered exploratory given the limited subgroup sample sizes, and therefore effect sizes and spatial consistency were emphasized over additional multiple-comparison correction. GMV values from significant clusters were extracted for follow-up regressions with relevant clinical measures and PD-CRS subdomains.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Methods", "text": "Finally, multiple regression models assessed the relationship between plasma NfL concentrations and both regional GMV and PD-CRS total scores. Age was included as a covariate to account for known age-related effects on NfL. Results were reported with corresponding B values, p-values, and R2 statistics. For visualization, VBM maps were thresholded at an uncorrected level of p\u00a0<\u00a00.0001 (cluster extent k\u00a0=\u00a020 voxels).\n\nThe study protocol was reviewed and approved by the institutional ethics committee. Written informed consent was obtained from all participants prior to their inclusion in the study. All procedures were carried out in accordance with the ethical standards of the 1964 Declaration of Helsinki and its later amendments. The data supporting the findings of this study are available from the corresponding author upon reasonable request.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Participants", "text": "Fifty-two symptomatic gene-expansion carriers (CAG >39) were recruited from the Movement Disorders Unit at Hospital de la Santa Creu i Sant Pau (Barcelona). Demographic and clinical data were recorded. Motor severity was assessed via the UHDRS Total Motor Score (UHDRS-TMS), and functional status via the Total Functional Capacity (TFC) and Functional Independence Scale (FIS)(Group, 1996). Neuropsychiatric symptoms were evaluated using the Problem Behaviors Assessment-Short Form (PBA-s), using the product of frequency and severity scores (Callaghan et al., 2015).\n\nParticipants were classified into HD-ISS stages 1\u20133 based on biological criteria (Tabrizi et al., 2022). Stage 0 individuals were excluded, as they show no detectable neuropathology. Stage 1 participants were included despite being premanifest, given biomarker evidence of early neurodegeneration. Cumulative CAG-related disease burden was estimated using the CAP score: age\u00a0\u00d7\u00a0(CAG - 33.66) (Warner et al., 2022).\n\nExclusion criteria included other neurological/psychiatric disorders, traumatic brain injury with loss of consciousness, uncontrolled medical illness, active substance abuse, or psychosis. Patients unable to understand or perform study tasks were also excluded.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Cognitive assessment", "text": "Cognitive functioning was evaluated in all participants using the PD-CRS (Pagonabarraga et al., 2008). Previous work from our group has validated the PD-CRS as a sensitive tool for staging cognitive impairment in HD, distinguishing between normal cognition, mild cognitive impairment (MCI), and dementia. The PD-CRS is composed of nine subtests assessing immediate verbal memory, confrontation naming, sustained attention, verbal working memory, unprompted clock drawing, clock copy, delayed free recall, alternating verbal fluency, and action verbal fluency. The posterior\u2013cortical subscore (range 0\u201330) is derived from confrontation naming and the clock copy task, while the remaining subtests contribute to the frontal\u2013subcortical subscore (range 0\u2013104). The sum of all subtests yields the total score, which ranges from 0 to 134, with higher scores indicating better cognitive performance. The PD-CRS A total score of \u226482 optimally identifies MCI, typically associated with minimal functional impact, while a score of \u226464 indicates dementia-level impairment with significant loss of autonomy. These thresholds have been empirically linked to functional outcomes, supporting the PD-CRS as both practical and clinically relevant for monitoring cognitive decline in HD (Horta-Barba et al., 2023; Martinez-Horta et al., 2020a).", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Neuroimaging acquisition and preprocessing", "text": "High-resolution T1-weighted MRI images were obtained on a 3T Philips Achieva scanner using an MPRAGE sequence (TR/TE\u00a0=\u00a012.65/7.08\u00a0ms, flip angle\u00a0=\u00a08\u00b0, field of view\u00a0=\u00a023\u00a0cm, matrix\u00a0=\u00a0256\u00a0\u00d7\u00a0256, slice thickness\u00a0=\u00a01\u00a0mm). Standard VBM procedures were performed using the SPM12 software package. (http://www.fil.ion.ucl.ac.uk/spm). GMV tissue probability maps were generated from the T1-weighted scans and subsequently normalized to the Montreal Neurological Institute (MNI) space using the DARTEL algorithm. To account for inter-individual variability, the normalized GMV maps were smoothed with an isotropic Gaussian filter with a full-width at half-maximum (FWHM) of 8\u00a0mm in all dimensions.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Biosample collection and processing", "text": "Peripheral blood samples were collected into EDTA tubes, centrifuged (2000\u00d7g, 10\u00a0min), and plasma was aliquoted and stored at \u221280\u00a0\u00b0C per international guidelines. Plasma NfL levels were measured using the Simoa\u00ae Human NF-light Advantage Kit (Quanterix) on the SR-X\u2122 system, following the manufacturer's protocol. Calibration standards, quality controls, and duplicate samples were run on a 96-well plate. After incubation with antibody-coated beads and biotinylated detection antibodies, SBG enzyme and RGP substrate were added. Fluorescent signals were detected and quantified using a four-parameter logistic curve. All samples were analyzed in duplicate; intra-assay CVs remained below 20\u00a0%.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Statistical analysis", "text": "Descriptive statistics were computed to summarize sociodemographic, clinical, cognitive, and behavioral characteristics. Continuous variables were expressed as means and standard deviations, while categorical variables were summarized as frequencies and percentages. Group comparisons across HD-ISS stages were performed using one-way analyses of variance (ANOVA) for normally distributed variables, followed by Tukey's HSD post hoc tests to identify specific between-group differences. For pairwise comparisons where homogeneity of variance was violated, Welch's t-tests were applied. Normality assumptions were evaluated via Shapiro-Wilk tests and visual inspection of histograms and Q-Q plots. Test statistics (F or t), degrees of freedom, p-values, and effect sizes (Cohen's d) were reported as appropriate.\n\nCognitive staging was based on PD-CRS total score thresholds, classifying participants as cognitively normal, mildly impaired (PD-CRS \u226482), or demented (PD-CRS \u226464)7,11. To assess relative performance across PD-CRS subdomains, raw scores were standardized into z-scores, which were calculated using the mean and standard deviation of the internal study cohort. This within-cohort normalization accounted for differing scoring ranges across subtests (e.g., action fluency max\u00a0=\u00a030 vs. clock drawing max\u00a0=\u00a010), thereby enabling direct item-level comparisons and clearer identification of the domains most sensitive to disease-related cognitive decline.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Statistical analysis", "text": "Multiple linear regression analyses were conducted to identify clinical and behavioral predictors of global cognitive performance controlling for potential confounders. Standardized beta coefficients (B), associated p-values, and R2 values were reported. To investigate structural correlates of cognitive function, GMV maps from VBM analyses were included in a general linear model (GLM). Two analytical frameworks were implemented according to the primary target of inference: (i) PD-CRS total scores were modelled as the dependent variable when assessing the contribution of motor severity, perseverative behaviours and demographic variables to cognitive performance; and (ii) regional GMV from VBM-derived clusters was modelled as the dependent variable when assessing neuroanatomical correlates of cognitive impairment. Accordingly, PD-CRS served as predictor or outcome depending on the objective of each model. Additionally, exploratory stage-stratified analyses of the VBM-derived clusters were conducted across HD-ISS stages 1\u20133. Whole-brain VBM analyses applied cluster-level family-wise error (FWE) correction (p\u00a0<\u00a00.05) using random field theory (RFT) in SPM12. Stage-stratified and ROI-based analyses were considered exploratory given the limited subgroup sample sizes, and therefore effect sizes and spatial consistency were emphasized over additional multiple-comparison correction. GMV values from significant clusters were extracted for follow-up regressions with relevant clinical measures and PD-CRS subdomains.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Statistical analysis", "text": "Finally, multiple regression models assessed the relationship between plasma NfL concentrations and both regional GMV and PD-CRS total scores. Age was included as a covariate to account for known age-related effects on NfL. Results were reported with corresponding B values, p-values, and R2 statistics. For visualization, VBM maps were thresholded at an uncorrected level of p\u00a0<\u00a00.0001 (cluster extent k\u00a0=\u00a020 voxels).\n\nThe study protocol was reviewed and approved by the institutional ethics committee. Written informed consent was obtained from all participants prior to their inclusion in the study. All procedures were carried out in accordance with the ethical standards of the 1964 Declaration of Helsinki and its later amendments. The data supporting the findings of this study are available from the corresponding author upon reasonable request.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Results", "text": "The total sample comprised 52 symptomatic gene-expansion carriers (53.8\u00a0% female) with an average of 43.9 (3.5) CAG repeats and a mean CAP score of 500.6 (89.8). Motor and functional assessments revealed a mean UHDRS-TMS score of 36.8 (21.4), a TFC score of 10.3 (2.6), a FAS score of 21.0 (3.8), and a composite Unified Huntington's Disease Rating Scale score (cUHDRS) score of 9.3 (4.8).", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Results", "text": "PD-CRS total score was 79.7 (19.1), with subcortical and cortical subscores averaging 52.7 (16.2) and 27.0 (3.3), respectively. Based on established cut-off scores, 30 participants (57.7\u00a0%) scored \u226482, indicating cognitive impairment consistent with mild cognitive impairment (MCI), while 10 participants (19.2\u00a0%) scored \u226464, consistent with dementia-level impairment. Among participants in HD-ISS stage 1, all 10 (100\u00a0%) exhibited normal cognitive status. In stage 2, 4 individuals (44.4\u00a0%) met criteria for MCI and 1 (11.1\u00a0%) for dementia. In contrast, in stage 3, 16 participants (48.5\u00a0%) fell within the MCI range and 8 (24.2\u00a0%) met the threshold for dementia. Table 1 summarizes the main clinical and sociodemographic characteristics of the sample (See Supplementary Table 1 for all clinical and sociodemographic data).Table 1Clinical and sociodemographic data of the sample.Table 1Mean (SD) & RangeVariableStage 1 Mean (SD)Stage 1 RangeStage 2 Mean (SD)Stage 2 RangeStage 3 Mean (SD)Stage 3 RangeGender(5/5)\u2013(6/3)\u2013(14/19)\u2013Age (years)45.2 (8.6)35\u20136346.1 (7.7)32\u20136053.3 (10.8)26\u201372Education (years)15.9 (3.7)11\u20132214.3 (2.2)11\u20131813.3 (3.8)7\u201322CAG42.3 (2.3)39\u20134643.4 (1.9)41\u20134744.1 (3.8)39\u201356CAP score378.5 (62.3)310\u2013470442.3 (60.9)333\u2013526521.5 (92.4)373\u2013666cUHDRS17.6 (1.4)15\u20131913.5 (1.8)11\u2013177.5 (4.3)0\u201316UHDRS-TMS2.4 (2.3)0\u2013713.2 (8.7)2\u20132844.7 (19.4)5\u201380TFC13.0 (0.0)13\u20131312.8 (0.4)12\u2013139.2 (2.5)4\u201313FIS25.0 (0.0)25\u20132524.8 (0.7)24\u20132619.6 (3.5)15\u201329PD-CRS Total111.6 (9.9)99\u201312880.6 (15.2)54\u201310173.7 (17.9)35\u2013117\u00a0Cortical score28.7 (1.3)27\u20133027.6 (3.4)20\u20133026.0 (3.7)15\u201330\u00a0Subcortical score82.9 (10.0)70\u20139853.0 (13.1)34\u20137147.7 (15.6)20\u201388CAG: CAG repeat length; CAP score: CAG x AGE product; UHDRS-TMS: Unified Huntington's Disease Rating Scale Total Motor Score; cUHDRS: Composite UHDRS; TFC: Total Functional Capacity; FIS: Functional Independence Score; PD-CRS: Parkinson's Disease \u2013 Cognitive Rating Scale.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Results", "text": "CAG: CAG repeat length; CAP score: CAG x AGE product; UHDRS-TMS: Unified Huntington's Disease Rating Scale Total Motor Score; cUHDRS: Composite UHDRS; TFC: Total Functional Capacity; FIS: Functional Independence Score; PD-CRS: Parkinson's Disease \u2013 Cognitive Rating Scale.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Results", "text": "Using PD-CRS z-scores to allow direct comparison across items, the subtests with the lowest standardized scores were alternating verbal fluency (z\u00a0=\u00a0\u22120.63), working memory (z\u00a0=\u00a0\u22120.45), and delayed recall (z\u00a0=\u00a0\u22120.39). In contrast, the highest standardized performances were observed in naming (z\u00a0=\u00a00.62), clock copy (z\u00a0=\u00a00.45), and action verbal fluency (z\u00a0=\u00a00.31). Fig. 1 depicts these findings. Given the recognized cognitive heterogeneity in HD, we also performed an exploratory analysis to assess whether distinct phenotypic patterns could be identified within our cohort. To this end, we calculated within\u2013HD-ISS\u2013stage z-scores for PD-CRS subcortical and cortical subscores and screened for participants whose normalized performance deviated significantly (\u22651.5 SD below the stage mean), which would suggest a subcortical-predominant, cortical-predominant, or mixed phenotype. This analysis revealed no extreme phenotypes in stages 1 and 2, and only two cases in stage 3 (\u22486\u00a0%), one with a cortical-predominant and one with a mixed profile. Overall, while variability was greater in subcortical than in cortical subscores, the absence of consistent extreme profiles across stages argues against the existence of clearly differentiated phenotypes in this dataset. For this reason, no further phenotype-specific analyses were conducted.Fig. 1Distribution of the PD-CRS scores across groups.Fig. 1", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Results", "text": "To identify clinical and behavioral predictors of global cognition, we conducted a multiple linear regression with PD-CRS total score as the dependent variable. Predictors included total scores from the PBAs, age, CAG repeat length, education, CAP score, and UHDRS-TMS. The model was significant, explaining a large proportion of variance in cognition (R2\u00a0=\u00a00.862; F(16,35)\u00a0=\u00a013.65; p\u00a0<\u00a00.001). Perseverative behavior was the strongest negative correlate of PD-CRS scores (B\u00a0=\u00a0\u22124.26, p\u00a0<\u00a00.001), followed by motor symptom severity (B\u00a0=\u00a0\u22120.35, p\u00a0=\u00a00.006). Education showed a positive association (B\u00a0=\u00a01.73, p\u00a0=\u00a00.001). Other PBAs domains, age, CAG repeat length, and CAP score had no significant effects.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Results", "text": "PD-CRS scores declined significantly with disease progression [F(2,49)\u00a0=\u00a04.29; p\u00a0=\u00a00.019]. Stage 3 showed lower total scores than stage 1 [t(41)\u00a0=\u00a03.08; p\u00a0=\u00a00.004; d\u00a0=\u00a01.30], and stage 2 was also lower than stage 1 [t(17)\u00a0=\u00a04.84; p\u00a0<\u00a00.001; d\u00a0=\u00a02.30]. Subcortical scores followed a similar pattern [F(2,49)\u00a0=\u00a03.84; p\u00a0=\u00a00.027], with significant reductions in stage 3 and 2 vs stage 1. Cortical scores declined across stages [F(2,49)\u00a0=\u00a03.50; p\u00a0=\u00a00.037], with a significant drop in stage 3 vs stage 1 [t(41)\u00a0=\u00a02.50; p\u00a0=\u00a00.017; d\u00a0=\u00a01.06], but no difference between stages 1 and 2. PD-CRS-based cognitive staging confirmed this gradient. All participants in stage 1 (100\u00a0%) were cognitively normal. In stage 2, 4 (44.4\u00a0%) met criteria for MCI (PD-CRS\u226482) and 1 (11.1\u00a0%) for dementia (PD-CRS\u226464). In stage 3, 16 (48.5\u00a0%) met the MCI threshold and 8 (24.2\u00a0%) met dementia criteria.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Results", "text": "As reported in Fig. 2 and Table 2, Two clusters survived whole-brain family-wise error (FWE) correction at the cluster level. The largest cluster (24,263 voxels; p\u00a0<\u00a00.001, FWE-corrected) encompassed a broad bilateral frontostriatal and limbic-paralimbic network. On the left hemisphere, significant involvement was observed in the anterior and dorsolateral putamen, head of the caudate nucleus, ventral striatum, and both the external and internal segments of the globus pallidus. Extending anteriorly, the cluster reached the subcallosal cortex and orbitofrontal areas, while laterally it engaged the anterior insula to the anterior putamen. Importantly, this cluster also extended into homologous regions of the right hemisphere. Right-lateralized effects were observed in the ventral striatum, pallidum, and putamen, as well as in paralimbic areas including the anterior insula. The second cluster (9565 voxels; p\u00a0<\u00a00.001, FWE-corrected) was located predominantly in the right posterior parietal and occipital cortices. The more significant effects were found in the superior parietal lobule, angular gyrus, precuneus, and lateral occipital regions, including the parieto-occipital junction and cuneus.Fig. 2Neuroimaging resultsSection A displays the association between GMV extracted from the VBM-derived clusters and serum NfL concentrations. Each dot represents an individual participant. Regression lines indicate the direction and strength of the association, with shaded areas showing the 95\u00a0% confidence interval. Higher NfL levels reflect increased neuroaxonal damage and are associated with lower GMV in these regions.Section B illustrates the spatial distribution of GMV reductions associated with poorer PD-CRS total performance. Statistical parametric maps are shown on the MNI template in neurological convention, with warmer colors indicating stronger negative associations (cluster-level FWE-corrected p\u00a0<\u00a00.05).Section C depicts representative associations between GMV in selected regions and performance on individual PD-CRS subtests, highlighting domain-specific vulnerability. Higher scores indicate better cognitive performance. Axes, slopes and confidence intervals are reported to facilitate interpretation of effect sizes.Fig. 2Table 2GMV correlates of the PD-CRS.Table 2Anatomical LabelMNI (X,Y,Z)T valueHemisphereClusterPutamen anterior\u221230, \u221215, \u221254.78Left1Occipital cortex/Lingual gyrus35, \u221275, 54.32Right2Angular gyrus/Inferior parietal lobule\u221233, \u221263, 244.25Left2Cuneus/Occipital pole20, \u221290, 314.21Right2Superior temporal gyrus/TPJ\u221254, \u221230, 314.12Left1Precuneus/Posterior cingulate cortex10, \u221248, 423.98Right2Angular gyrus/Inferior parietal\u221231, \u221264, 203.92Left2Middle occipital gyrus51, \u221275, 313.90Right2Putamen\u221227, \u221224, \u221293.90Left1Lateral occipital cortex45, \u221269, 293.85Right2Ventral pallidum/Sub-lenticular region\u221236, \u221235, \u2212113.85Left1Superior parietal lobule21, \u221260, 503.81Right2Globus pallidus interna\u221224, \u22126, \u2212123.78Left1Angular gyrus/Posterior STG36, \u221265, 203.73Right2", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Results", "text": "Section A displays the association between GMV extracted from the VBM-derived clusters and serum NfL concentrations. Each dot represents an individual participant. Regression lines indicate the direction and strength of the association, with shaded areas showing the 95\u00a0% confidence interval. Higher NfL levels reflect increased neuroaxonal damage and are associated with lower GMV in these regions.\n\nSection B illustrates the spatial distribution of GMV reductions associated with poorer PD-CRS total performance. Statistical parametric maps are shown on the MNI template in neurological convention, with warmer colors indicating stronger negative associations (cluster-level FWE-corrected p\u00a0<\u00a00.05).\n\nSection C depicts representative associations between GMV in selected regions and performance on individual PD-CRS subtests, highlighting domain-specific vulnerability. Higher scores indicate better cognitive performance. Axes, slopes and confidence intervals are reported to facilitate interpretation of effect sizes.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Results", "text": "To assess whether the relationship between GMV and cognition was independent of motor and behavioral symptoms, we ran a multiple linear regression across the full sample. The dependent variable was mean GMV across significant VBM-derived ROIs; predictors included UHDRS-TMS and PBA perseveration scores, previously identified as strong cognitive correlates. The model was significant, explaining a moderate proportion of GMV variance (R2\u00a0=\u00a00.326, adjusted R2\u00a0=\u00a00.298; F(2,49)\u00a0=\u00a011.83; p\u00a0<\u00a00.001). Both predictors contributed: greater motor severity (B\u00a0=\u00a0\u22120.0104, p\u00a0=\u00a00.002) and higher perseveration (B\u00a0=\u00a0\u22120.0323, p\u00a0=\u00a00.012) were linked to reduced GMV. However, when PD-CRS total scores were added, cognition emerged as the strongest predictor (B\u00a0=\u00a00.0038, p\u00a0<\u00a00.001), indicating a unique association with brain structure beyond motor and behavioral effects.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Results", "text": "To investigate whether the structural\u2013cognitive associations varied across disease progression, we repeated the analyses using regional GMV extracted from the VBM-derived fronto-striatal and posterior clusters. Significant effects of HD-ISS stage were observed for both clusters (fronto-striatal: F(2,45)\u00a0=\u00a011.37, p\u00a0=\u00a00.0001; posterior: F(2,45)\u00a0=\u00a014.00, p\u00a0=\u00a00.00002), with substantially reduced volumes in stage 3 compared to stages 1\u20132. Within-stage regressions, using as predictors the PD-CRS total score, the UHDRS-TMS and the PBAs perseveration scores, the analysis confirmed different contribution patterns across stages. In stage 1, all predictors significantly accounted for GMV variance in the fronto-striatal cluster (R2\u00a0=\u00a00.92, p\u00a0=\u00a00.004\u20130.005), and motor and perseverative symptoms also contributed in the posterior cluster (R2\u00a0=\u00a00.89, p\u00a0=\u00a00.002\u20130.019). In stage 2, no significant associations emerged (R2\u00a0\u2264\u00a00.21, all p\u00a0>\u00a00.62). In stage 3, the PD-CRS was the most consistent predictor of GMV in both clusters (R2\u00a0=\u00a00.29\u20130.37; p\u00a0\u2264\u00a00.08).", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Results", "text": "Significant relationships (p\u00a0<\u00a00.05) were found for all subdomains, with particularly strong effects for immediate memory, attention, and verbal fluency. Higher GMV consistently predicted better cognitive performance. Immediate memory was positively associated with GMV in the left striatum (B\u00a0=\u00a02.77, p\u00a0<\u00a00.001, R2\u00a0=\u00a00.29), posterior cortex (B\u00a0=\u00a00.62, p\u00a0<\u00a00.001, R2\u00a0=\u00a00.28), calcarine cortex (B\u00a0=\u00a02.01, p\u00a0=\u00a00.001, R2\u00a0=\u00a00.28), left midtemporal (B\u00a0=\u00a02.48, p\u00a0=\u00a00.003, R2\u00a0=\u00a00.23), and left frontal-striatal regions (B\u00a0=\u00a01.47, p\u00a0=\u00a00.018, R2\u00a0=\u00a00.17). Attention scores correlated strongly with GMV in the posterior cortex (B\u00a0=\u00a01.07, R2\u00a0=\u00a00.47), left striatum (B\u00a0=\u00a04.65, R2\u00a0=\u00a00.46), left midtemporal (B\u00a0=\u00a04.54, R2\u00a0=\u00a00.40), calcarine cortex (B\u00a0=\u00a03.21, R2\u00a0=\u00a00.40), and TPJ (B\u00a0=\u00a014.06, all p\u00a0<\u00a00.001, R2\u00a0=\u00a00.38). Copy of a clock performance showed significant associations with GMV in the left striatum (B\u00a0=\u00a01.26, p\u00a0=\u00a00.001), calcarine, TPJ, posterior cortex, and bilateral midtemporal/striatal regions (all p\u00a0<\u00a00.02). Unprompted drawing of a clock was related to GMV in the left striatum (B\u00a0=\u00a01.37, p\u00a0=\u00a00.004), right striatum, calcarine, TPJ, and midtemporal cortex. Working memory was linked to GMV in the right striatum (B\u00a0=\u00a06.08, p\u00a0<\u00a00.001), as well as the posterior cortex, left striatum, and left midtemporal regions. Verbal fluency (alternating and action) showed widespread associations across nearly all ROIs. Alternating fluency was linked to right striatum, left striatum, posterior cortex, calcarine, TPJ, and midtemporal cortex (all p\u00a0<\u00a00.005). Action fluency was most strongly related to right striatum (B\u00a0=\u00a028.31, p\u00a0<\u00a00.001), with additional associations in posterior cortex, calcarine, frontal-striatal, and TPJ regions. Recognition memory correlated with GMV in right striatum (B\u00a0=\u00a09.53, p\u00a0<\u00a00.001), left striatum, posterior cortex, and calcarine cortex. Naming was moderately associated with posterior cortex (B\u00a0=\u00a00.35, p\u00a0=\u00a00.033), bilateral striatum, and calcarine cortex (all p\u00a0<\u00a00.04). Across all models, age was not a significant covariate.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Results", "text": "Lower cognitive performance was significantly associated with higher NfL levels (B\u00a0=\u00a0\u22120.153, p\u00a0=\u00a00.048; R2\u00a0=\u00a00.076). This association remained significant after adjusting for age (B\u00a0=\u00a0\u22120.165, p\u00a0=\u00a00.032), while age alone was not a significant predictor (B\u00a0=\u00a0\u22120.255, p\u00a0=\u00a00.107). The age-adjusted model explained 12.4\u00a0% of the variance in NfL (R2\u00a0=\u00a00.124).\n\nResults consistently showed inverse associations between GMV and NfL, indicating that greater atrophy is linked to higher neurodegeneration. Strongest effects were observed in the left frontal-striatal region (B\u00a0=\u00a0\u221211.38, p\u00a0<\u00a00.001) and posterior cortex (B\u00a0=\u00a0\u22122.77, p\u00a0=\u00a00.004), both surviving age correction. Age remained significant in most models but with smaller effects than GMV. Regression coefficients (B) and p-values are summarized in Table 3.Table 3Association between plasma neurofilament and GMV.Table 3ROIsB (GMV)p (GMV)B (Age)p (Age)R (Ross and Tabrizi, 2011)Calcarine\u22128.500.007\u22120.380.0170.204Left Frontal-striatal\u221211.380.000\u22120.360.0130.301Left mid temporal\u22129.660.032\u22120.330.0400.153Left Striatum\u221211.370.007\u22120.350.0260.200Parietal-occipital\u22122.770.004\u22120.360.0190.222Right mid temporal\u221213.160.182\u22120.350.0410.098Right striatum\u221242.480.000\u22120.360.0130.325TPJ\u221232.920.021\u22120.370.0220.167", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Sociodemographic and clinical characteristics", "text": "The total sample comprised 52 symptomatic gene-expansion carriers (53.8\u00a0% female) with an average of 43.9 (3.5) CAG repeats and a mean CAP score of 500.6 (89.8). Motor and functional assessments revealed a mean UHDRS-TMS score of 36.8 (21.4), a TFC score of 10.3 (2.6), a FAS score of 21.0 (3.8), and a composite Unified Huntington's Disease Rating Scale score (cUHDRS) score of 9.3 (4.8).", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Sociodemographic and clinical characteristics", "text": "PD-CRS total score was 79.7 (19.1), with subcortical and cortical subscores averaging 52.7 (16.2) and 27.0 (3.3), respectively. Based on established cut-off scores, 30 participants (57.7\u00a0%) scored \u226482, indicating cognitive impairment consistent with mild cognitive impairment (MCI), while 10 participants (19.2\u00a0%) scored \u226464, consistent with dementia-level impairment. Among participants in HD-ISS stage 1, all 10 (100\u00a0%) exhibited normal cognitive status. In stage 2, 4 individuals (44.4\u00a0%) met criteria for MCI and 1 (11.1\u00a0%) for dementia. In contrast, in stage 3, 16 participants (48.5\u00a0%) fell within the MCI range and 8 (24.2\u00a0%) met the threshold for dementia. Table 1 summarizes the main clinical and sociodemographic characteristics of the sample (See Supplementary Table 1 for all clinical and sociodemographic data).Table 1Clinical and sociodemographic data of the sample.Table 1Mean (SD) & RangeVariableStage 1 Mean (SD)Stage 1 RangeStage 2 Mean (SD)Stage 2 RangeStage 3 Mean (SD)Stage 3 RangeGender(5/5)\u2013(6/3)\u2013(14/19)\u2013Age (years)45.2 (8.6)35\u20136346.1 (7.7)32\u20136053.3 (10.8)26\u201372Education (years)15.9 (3.7)11\u20132214.3 (2.2)11\u20131813.3 (3.8)7\u201322CAG42.3 (2.3)39\u20134643.4 (1.9)41\u20134744.1 (3.8)39\u201356CAP score378.5 (62.3)310\u2013470442.3 (60.9)333\u2013526521.5 (92.4)373\u2013666cUHDRS17.6 (1.4)15\u20131913.5 (1.8)11\u2013177.5 (4.3)0\u201316UHDRS-TMS2.4 (2.3)0\u2013713.2 (8.7)2\u20132844.7 (19.4)5\u201380TFC13.0 (0.0)13\u20131312.8 (0.4)12\u2013139.2 (2.5)4\u201313FIS25.0 (0.0)25\u20132524.8 (0.7)24\u20132619.6 (3.5)15\u201329PD-CRS Total111.6 (9.9)99\u201312880.6 (15.2)54\u201310173.7 (17.9)35\u2013117\u00a0Cortical score28.7 (1.3)27\u20133027.6 (3.4)20\u20133026.0 (3.7)15\u201330\u00a0Subcortical score82.9 (10.0)70\u20139853.0 (13.1)34\u20137147.7 (15.6)20\u201388CAG: CAG repeat length; CAP score: CAG x AGE product; UHDRS-TMS: Unified Huntington's Disease Rating Scale Total Motor Score; cUHDRS: Composite UHDRS; TFC: Total Functional Capacity; FIS: Functional Independence Score; PD-CRS: Parkinson's Disease \u2013 Cognitive Rating Scale.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Sociodemographic and clinical characteristics", "text": "CAG: CAG repeat length; CAP score: CAG x AGE product; UHDRS-TMS: Unified Huntington's Disease Rating Scale Total Motor Score; cUHDRS: Composite UHDRS; TFC: Total Functional Capacity; FIS: Functional Independence Score; PD-CRS: Parkinson's Disease \u2013 Cognitive Rating Scale.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Sociodemographic and clinical characteristics", "text": "Using PD-CRS z-scores to allow direct comparison across items, the subtests with the lowest standardized scores were alternating verbal fluency (z\u00a0=\u00a0\u22120.63), working memory (z\u00a0=\u00a0\u22120.45), and delayed recall (z\u00a0=\u00a0\u22120.39). In contrast, the highest standardized performances were observed in naming (z\u00a0=\u00a00.62), clock copy (z\u00a0=\u00a00.45), and action verbal fluency (z\u00a0=\u00a00.31). Fig. 1 depicts these findings. Given the recognized cognitive heterogeneity in HD, we also performed an exploratory analysis to assess whether distinct phenotypic patterns could be identified within our cohort. To this end, we calculated within\u2013HD-ISS\u2013stage z-scores for PD-CRS subcortical and cortical subscores and screened for participants whose normalized performance deviated significantly (\u22651.5 SD below the stage mean), which would suggest a subcortical-predominant, cortical-predominant, or mixed phenotype. This analysis revealed no extreme phenotypes in stages 1 and 2, and only two cases in stage 3 (\u22486\u00a0%), one with a cortical-predominant and one with a mixed profile. Overall, while variability was greater in subcortical than in cortical subscores, the absence of consistent extreme profiles across stages argues against the existence of clearly differentiated phenotypes in this dataset. For this reason, no further phenotype-specific analyses were conducted.Fig. 1Distribution of the PD-CRS scores across groups.Fig. 1", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Sociodemographic and clinical characteristics", "text": "To identify clinical and behavioral predictors of global cognition, we conducted a multiple linear regression with PD-CRS total score as the dependent variable. Predictors included total scores from the PBAs, age, CAG repeat length, education, CAP score, and UHDRS-TMS. The model was significant, explaining a large proportion of variance in cognition (R2\u00a0=\u00a00.862; F(16,35)\u00a0=\u00a013.65; p\u00a0<\u00a00.001). Perseverative behavior was the strongest negative correlate of PD-CRS scores (B\u00a0=\u00a0\u22124.26, p\u00a0<\u00a00.001), followed by motor symptom severity (B\u00a0=\u00a0\u22120.35, p\u00a0=\u00a00.006). Education showed a positive association (B\u00a0=\u00a01.73, p\u00a0=\u00a00.001). Other PBAs domains, age, CAG repeat length, and CAP score had no significant effects.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Sociodemographic and clinical characteristics", "text": "PD-CRS scores declined significantly with disease progression [F(2,49)\u00a0=\u00a04.29; p\u00a0=\u00a00.019]. Stage 3 showed lower total scores than stage 1 [t(41)\u00a0=\u00a03.08; p\u00a0=\u00a00.004; d\u00a0=\u00a01.30], and stage 2 was also lower than stage 1 [t(17)\u00a0=\u00a04.84; p\u00a0<\u00a00.001; d\u00a0=\u00a02.30]. Subcortical scores followed a similar pattern [F(2,49)\u00a0=\u00a03.84; p\u00a0=\u00a00.027], with significant reductions in stage 3 and 2 vs stage 1. Cortical scores declined across stages [F(2,49)\u00a0=\u00a03.50; p\u00a0=\u00a00.037], with a significant drop in stage 3 vs stage 1 [t(41)\u00a0=\u00a02.50; p\u00a0=\u00a00.017; d\u00a0=\u00a01.06], but no difference between stages 1 and 2. PD-CRS-based cognitive staging confirmed this gradient. All participants in stage 1 (100\u00a0%) were cognitively normal. In stage 2, 4 (44.4\u00a0%) met criteria for MCI (PD-CRS\u226482) and 1 (11.1\u00a0%) for dementia (PD-CRS\u226464). In stage 3, 16 (48.5\u00a0%) met the MCI threshold and 8 (24.2\u00a0%) met dementia criteria.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Neuroanatomical correlates of PD-CRS", "text": "As reported in Fig. 2 and Table 2, Two clusters survived whole-brain family-wise error (FWE) correction at the cluster level. The largest cluster (24,263 voxels; p\u00a0<\u00a00.001, FWE-corrected) encompassed a broad bilateral frontostriatal and limbic-paralimbic network. On the left hemisphere, significant involvement was observed in the anterior and dorsolateral putamen, head of the caudate nucleus, ventral striatum, and both the external and internal segments of the globus pallidus. Extending anteriorly, the cluster reached the subcallosal cortex and orbitofrontal areas, while laterally it engaged the anterior insula to the anterior putamen. Importantly, this cluster also extended into homologous regions of the right hemisphere. Right-lateralized effects were observed in the ventral striatum, pallidum, and putamen, as well as in paralimbic areas including the anterior insula. The second cluster (9565 voxels; p\u00a0<\u00a00.001, FWE-corrected) was located predominantly in the right posterior parietal and occipital cortices. The more significant effects were found in the superior parietal lobule, angular gyrus, precuneus, and lateral occipital regions, including the parieto-occipital junction and cuneus.Fig. 2Neuroimaging resultsSection A displays the association between GMV extracted from the VBM-derived clusters and serum NfL concentrations. Each dot represents an individual participant. Regression lines indicate the direction and strength of the association, with shaded areas showing the 95\u00a0% confidence interval. Higher NfL levels reflect increased neuroaxonal damage and are associated with lower GMV in these regions.Section B illustrates the spatial distribution of GMV reductions associated with poorer PD-CRS total performance. Statistical parametric maps are shown on the MNI template in neurological convention, with warmer colors indicating stronger negative associations (cluster-level FWE-corrected p\u00a0<\u00a00.05).Section C depicts representative associations between GMV in selected regions and performance on individual PD-CRS subtests, highlighting domain-specific vulnerability. Higher scores indicate better cognitive performance. Axes, slopes and confidence intervals are reported to facilitate interpretation of effect sizes.Fig. 2Table 2GMV correlates of the PD-CRS.Table 2Anatomical LabelMNI (X,Y,Z)T valueHemisphereClusterPutamen anterior\u221230, \u221215, \u221254.78Left1Occipital cortex/Lingual gyrus35, \u221275, 54.32Right2Angular gyrus/Inferior parietal lobule\u221233, \u221263, 244.25Left2Cuneus/Occipital pole20, \u221290, 314.21Right2Superior temporal gyrus/TPJ\u221254, \u221230, 314.12Left1Precuneus/Posterior cingulate cortex10, \u221248, 423.98Right2Angular gyrus/Inferior parietal\u221231, \u221264, 203.92Left2Middle occipital gyrus51, \u221275, 313.90Right2Putamen\u221227, \u221224, \u221293.90Left1Lateral occipital cortex45, \u221269, 293.85Right2Ventral pallidum/Sub-lenticular region\u221236, \u221235, \u2212113.85Left1Superior parietal lobule21, \u221260, 503.81Right2Globus pallidus interna\u221224, \u22126, \u2212123.78Left1Angular gyrus/Posterior STG36, \u221265, 203.73Right2", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Neuroanatomical correlates of PD-CRS", "text": "Section A displays the association between GMV extracted from the VBM-derived clusters and serum NfL concentrations. Each dot represents an individual participant. Regression lines indicate the direction and strength of the association, with shaded areas showing the 95\u00a0% confidence interval. Higher NfL levels reflect increased neuroaxonal damage and are associated with lower GMV in these regions.\n\nSection B illustrates the spatial distribution of GMV reductions associated with poorer PD-CRS total performance. Statistical parametric maps are shown on the MNI template in neurological convention, with warmer colors indicating stronger negative associations (cluster-level FWE-corrected p\u00a0<\u00a00.05).\n\nSection C depicts representative associations between GMV in selected regions and performance on individual PD-CRS subtests, highlighting domain-specific vulnerability. Higher scores indicate better cognitive performance. Axes, slopes and confidence intervals are reported to facilitate interpretation of effect sizes.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Neuroanatomical correlates of PD-CRS", "text": "To assess whether the relationship between GMV and cognition was independent of motor and behavioral symptoms, we ran a multiple linear regression across the full sample. The dependent variable was mean GMV across significant VBM-derived ROIs; predictors included UHDRS-TMS and PBA perseveration scores, previously identified as strong cognitive correlates. The model was significant, explaining a moderate proportion of GMV variance (R2\u00a0=\u00a00.326, adjusted R2\u00a0=\u00a00.298; F(2,49)\u00a0=\u00a011.83; p\u00a0<\u00a00.001). Both predictors contributed: greater motor severity (B\u00a0=\u00a0\u22120.0104, p\u00a0=\u00a00.002) and higher perseveration (B\u00a0=\u00a0\u22120.0323, p\u00a0=\u00a00.012) were linked to reduced GMV. However, when PD-CRS total scores were added, cognition emerged as the strongest predictor (B\u00a0=\u00a00.0038, p\u00a0<\u00a00.001), indicating a unique association with brain structure beyond motor and behavioral effects.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Neuroanatomical correlates of PD-CRS", "text": "To investigate whether the structural\u2013cognitive associations varied across disease progression, we repeated the analyses using regional GMV extracted from the VBM-derived fronto-striatal and posterior clusters. Significant effects of HD-ISS stage were observed for both clusters (fronto-striatal: F(2,45)\u00a0=\u00a011.37, p\u00a0=\u00a00.0001; posterior: F(2,45)\u00a0=\u00a014.00, p\u00a0=\u00a00.00002), with substantially reduced volumes in stage 3 compared to stages 1\u20132. Within-stage regressions, using as predictors the PD-CRS total score, the UHDRS-TMS and the PBAs perseveration scores, the analysis confirmed different contribution patterns across stages. In stage 1, all predictors significantly accounted for GMV variance in the fronto-striatal cluster (R2\u00a0=\u00a00.92, p\u00a0=\u00a00.004\u20130.005), and motor and perseverative symptoms also contributed in the posterior cluster (R2\u00a0=\u00a00.89, p\u00a0=\u00a00.002\u20130.019). In stage 2, no significant associations emerged (R2\u00a0\u2264\u00a00.21, all p\u00a0>\u00a00.62). In stage 3, the PD-CRS was the most consistent predictor of GMV in both clusters (R2\u00a0=\u00a00.29\u20130.37; p\u00a0\u2264\u00a00.08).", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Associations between regional gray matter volumes and PD-CRS subdomains", "text": "Significant relationships (p\u00a0<\u00a00.05) were found for all subdomains, with particularly strong effects for immediate memory, attention, and verbal fluency. Higher GMV consistently predicted better cognitive performance. Immediate memory was positively associated with GMV in the left striatum (B\u00a0=\u00a02.77, p\u00a0<\u00a00.001, R2\u00a0=\u00a00.29), posterior cortex (B\u00a0=\u00a00.62, p\u00a0<\u00a00.001, R2\u00a0=\u00a00.28), calcarine cortex (B\u00a0=\u00a02.01, p\u00a0=\u00a00.001, R2\u00a0=\u00a00.28), left midtemporal (B\u00a0=\u00a02.48, p\u00a0=\u00a00.003, R2\u00a0=\u00a00.23), and left frontal-striatal regions (B\u00a0=\u00a01.47, p\u00a0=\u00a00.018, R2\u00a0=\u00a00.17). Attention scores correlated strongly with GMV in the posterior cortex (B\u00a0=\u00a01.07, R2\u00a0=\u00a00.47), left striatum (B\u00a0=\u00a04.65, R2\u00a0=\u00a00.46), left midtemporal (B\u00a0=\u00a04.54, R2\u00a0=\u00a00.40), calcarine cortex (B\u00a0=\u00a03.21, R2\u00a0=\u00a00.40), and TPJ (B\u00a0=\u00a014.06, all p\u00a0<\u00a00.001, R2\u00a0=\u00a00.38). Copy of a clock performance showed significant associations with GMV in the left striatum (B\u00a0=\u00a01.26, p\u00a0=\u00a00.001), calcarine, TPJ, posterior cortex, and bilateral midtemporal/striatal regions (all p\u00a0<\u00a00.02). Unprompted drawing of a clock was related to GMV in the left striatum (B\u00a0=\u00a01.37, p\u00a0=\u00a00.004), right striatum, calcarine, TPJ, and midtemporal cortex. Working memory was linked to GMV in the right striatum (B\u00a0=\u00a06.08, p\u00a0<\u00a00.001), as well as the posterior cortex, left striatum, and left midtemporal regions. Verbal fluency (alternating and action) showed widespread associations across nearly all ROIs. Alternating fluency was linked to right striatum, left striatum, posterior cortex, calcarine, TPJ, and midtemporal cortex (all p\u00a0<\u00a00.005). Action fluency was most strongly related to right striatum (B\u00a0=\u00a028.31, p\u00a0<\u00a00.001), with additional associations in posterior cortex, calcarine, frontal-striatal, and TPJ regions. Recognition memory correlated with GMV in right striatum (B\u00a0=\u00a09.53, p\u00a0<\u00a00.001), left striatum, posterior cortex, and calcarine cortex. Naming was moderately associated with posterior cortex (B\u00a0=\u00a00.35, p\u00a0=\u00a00.033), bilateral striatum, and calcarine cortex (all p\u00a0<\u00a00.04). Across all models, age was not a significant covariate.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Association between plasma neurofilament levels and global cognitive function (PD-CRS)", "text": "Lower cognitive performance was significantly associated with higher NfL levels (B\u00a0=\u00a0\u22120.153, p\u00a0=\u00a00.048; R2\u00a0=\u00a00.076). This association remained significant after adjusting for age (B\u00a0=\u00a0\u22120.165, p\u00a0=\u00a00.032), while age alone was not a significant predictor (B\u00a0=\u00a0\u22120.255, p\u00a0=\u00a00.107). The age-adjusted model explained 12.4\u00a0% of the variance in NfL (R2\u00a0=\u00a00.124).", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Association between plasma neurofilament levels and regional gray matter volume", "text": "Results consistently showed inverse associations between GMV and NfL, indicating that greater atrophy is linked to higher neurodegeneration. Strongest effects were observed in the left frontal-striatal region (B\u00a0=\u00a0\u221211.38, p\u00a0<\u00a00.001) and posterior cortex (B\u00a0=\u00a0\u22122.77, p\u00a0=\u00a00.004), both surviving age correction. Age remained significant in most models but with smaller effects than GMV. Regression coefficients (B) and p-values are summarized in Table 3.Table 3Association between plasma neurofilament and GMV.Table 3ROIsB (GMV)p (GMV)B (Age)p (Age)R (Ross and Tabrizi, 2011)Calcarine\u22128.500.007\u22120.380.0170.204Left Frontal-striatal\u221211.380.000\u22120.360.0130.301Left mid temporal\u22129.660.032\u22120.330.0400.153Left Striatum\u221211.370.007\u22120.350.0260.200Parietal-occipital\u22122.770.004\u22120.360.0190.222Right mid temporal\u221213.160.182\u22120.350.0410.098Right striatum\u221242.480.000\u22120.360.0130.325TPJ\u221232.920.021\u22120.370.0220.167", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Discussion", "text": "The present study investigated the neuroanatomical correlates of global cognitive functioning in HD using the PD-CRS, a cognitive screening instrument originally validated in PD and recently explored in HD (Horta-Barba et al., 2023; Martinez-Horta et al., 2020a). Our findings support the PD-CRS as a sensitive and clinically relevant tool for detecting cognitive impairment in HD, and demonstrate its robust association with GMV across a distributed frontostriatal, limbic, and parietal-occipital network. However, further validation of the PD-CRS in independent HD cohorts will be essential to strengthen its generalizability.\n\nClinically, cognitive impairment was highly prevalent in our cohort, with 58\u00a0% of participants meeting criteria for MCI and 19\u00a0% for dementia, based on validated PD-CRS thresholds (Martinez-Horta et al., 2020a). These rates increased consistently across HD-ISS stages. Importantly, all participants in stage 1 displayed preserved cognition, supporting the construct validity of the PD-CRS as a staging tool even in early or premanifest disease. At the item level, performance varied by cognitive domain, with executive-attentional tasks, particularly sustained attention and alternating verbal fluency, emerging as the most sensitive to disease progression.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Discussion", "text": "VBM analyses identified two large clusters where GMV significantly correlated with global PD-CRS scores. The first encompassed a bilateral frontostriatal and paralimbic network (Alexander et al., 1986; Haber and Knutson, 2010; Menon and Uddin, 2010). These regions are among the earliest and more consistently affected structures in HD (Ross et al., 2014; Tabrizi et al., 2009, 2011). The involvement of orbitofrontal and insular cortices, as well as subcallosal and ventral striatal regions, underscores the progressive engagement of limbic-prefrontal circuits as the disease advances (Enzi et al., 2012; Martinez-Horta et al., 2018, 2021). These structures subserve a wide range of cognitive domains that are prominently impaired in HD and that are targeted by the PD-CRS. The caudate and putamen play a central role in action selection, procedural memory, and attentional gating, while the globus pallidus and ventral striatum contribute to motivation, decision-making, and cognitive flexibility (Haber and Knutson, 2010; Yin and Knowlton, 2006; Grahn et al., 2008). The anterior insula is involved in salience detection and the integration of cognitive and affective information, and the subcallosal area has been linked to mood and cognitive control (Menon and Uddin, 2010). Through its diverse subtests, the PD-CRS assesses many of these domains, such as attention, working memory, and verbal fluency, making it well suited to detect dysfunction across this network.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Discussion", "text": "The second cluster, centered in the right posterior parietal and occipital cortex, involved regions supporting visuospatial integration and cognitive flexibility (Kawasaki et al., 2008; Seghier, 2013; Cavanna and Trimble, 2006; Whitlock, 2017). Degeneration in posterior brain regions, particularly parietal and occipital cortices, was historically neglected in early descriptions of HD. However, accumulating evidence demonstrates that structural alterations in these regions emerge well before clinical diagnosis, with cortical thinning detectable even in prodromal and premanifest individuals (Rosas et al., 2008; Tabrizi et al., 2009; Martinez-Horta et al., 2024; Nopoulos et al., 2010). Our results align with this literature by showing that PD-CRS performance is significantly associated with GMV in parietal-occipital regions, suggesting that this scale captures early, spatially distributed cortical involvement.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Discussion", "text": "To better understand the factors influencing cognitive performance, we ran a multiple regression with PD-CRS total scores as the outcome. The model explained a large proportion of variance, identifying perseverative behavior, motor severity, and education as significant predictors. These findings highlight the combined association between cognitive reserve, motor decline, and psychiatric symptoms with cognition in HD, although with the present cross-sectional data we cannot attribute a specific directionality or infer a clear causal relationship beyond correlation. We then tested whether motor and behavioral symptoms could account for GMV reductions in PD-CRS-related regions. Without cognitive scores, UHDRS-TMS and PBA perseveration were significantly linked to GMV. However, when PD-CRS was added, it became the strongest predictor, suggesting that GMV loss reflects cognitive decline beyond motor or behavioral dysfunction. It should also be acknowledged that PBA-s scores were particularly low in our cohort, which may reflect cohort-specific or care-related factors that attenuate the expression of neuropsychiatric symptoms. In other regions or healthcare settings, these values might differ substantially, and thus the associations we observed are likely not fully generalizable and should be interpreted as dependent, at least in part, on the behavioral profile of the studied sample.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Discussion", "text": "Importantly, stage-stratified analyses revealed that the expression of these structure\u2013function relationships varies along disease progression. While multiple clinical dimensions (cognitive, motor and behavioural) were related to GMV in early disease, cognition became the principal correlate in advanced stages, in line with the dominant impact of neurodegeneration on daily functioning at later phases. The absence of robust associations in stage 2 likely reflects reduced variability once symptoms are clinically established, limiting sensitivity to detect differential contributions. Together, these results support the role of PD-CRS as a biologically meaningful index of neurodegeneration over the HD progression continuum.\n\nBeyond global cognition, we also examined how specific PD-CRS subdomains mapped onto regional gray matter volumes. Results demonstrated clear topographic organization: fronto-striatal regions were preferentially associated with verbal fluency and working memory, posterior cortices with visuospatial functions (e.g., clock copy), and temporo-parietal junctions with attentional processing. These domain-specific anatomical associations validate the internal structure of the PD-CRS and reinforce its relevance for tracking both focal and diffuse cognitive deterioration. Importantly, age was not a significant covariate in any of the subdomain analyses, further highlighting the specificity of these associations.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Discussion", "text": "At the biomarker level, plasma NfL levels were significantly associated with PD-CRS scores, even after adjusting for age, indicating that the PD-CRS reflects neurodegeneration beyond structural atrophy (Byrne et al., 2017; Parkin et al., 2024). NfL also showed inverse correlations with GMV in key regions, particularly the left frontal-striatal and posterior cortices, mirroring PD-CRS-related structural changes. These results support the PD-CRS as a meaningful cognitive marker of both brain pathology and fluid biomarkers of disease burden and build on and extend our previous works validating the PD-CRS in HD (Horta-Barba et al., 2023; Martinez-Horta et al., 2020a).", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Discussion", "text": "Consistent with prior findings, we show that both total PD-CRS scores and specific subdomains are significantly associated with gray matter volume in posterior parietal and occipital cortices (Martinez-Horta et al., 2020a, 2020b). These associations remained significant after controlling for age and motor severity, highlighting the PD-CRS specificity in capturing neural correlates of cognitive decline. Across independent samples and complementary methods, evidence converges to support the PD-CRS as both a clinically valid staging tool and a sensitive marker of widespread neurodegeneration in HD. Beyond these findings that reinforce the strength and utility of the PD-CRS, other factors further enhance its applicability. One notable advantage is its availability in multiple validated translations, including languages spoken in Western Europe, Asia, and Latin America, supporting its global use in diverse clinical and research contexts.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Discussion", "text": "Despite these strengths, several limitations should be acknowledged. First, while our sample spanned a wide range of disease stages, it remains cross-sectional in nature, precluding conclusions about the temporal progression of structural-cognitive relationships. Longitudinal studies will be essential to establish whether PD-CRS scores track within-subject changes in brain volume over time. Second, although the PD-CRS captures multiple cognitive domains, it remains a screening tool and does not replace comprehensive neuropsychological testing. In our stage 1 cohort, and while participants were classified as cognitively normal according to PD-CRS normative data, this does not preclude the possibility that they might perform differently from healthy controls in a direct comparison. Future studies with larger samples of stage 1 participants, appropriate control groups, and longitudinal designs will be needed to determine whether the PD-CRS can detect such subtle early differences and to further establish its sensitivity in premanifest and early disease phases. Moreover, the relatively small sample size in HD-ISS stage 2, combined with within-group heterogeneity, likely reduced statistical power and may have limited our ability to detect stage-specific associations and stage-wise analyses were exploratory and sample sizes, especially in stages 1\u20132, constrain statistical power; therefore, these results should be interpreted with caution pending replication in larger stratified cohorts.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Discussion", "text": "Overall, this study provides robust evidence that the PD-CRS is a neuroanatomically grounded, clinically sensitive tool for assessing cognition in HD, also capturing disease-related structural changes across a broad network of functionally relevant brain regions. These findings, together with previous work, highlight the value of further investigating the potential role of the PD-CRS for disease monitoring and as an outcome measure in clinical trials in Huntington's disease, and underscore the importance of anatomically informed cognitive assessment strategies in neurodegenerative disorders.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "CRediT authorship contribution statement", "text": "Saul Martinez-Horta: Writing \u2013 review & editing, Writing \u2013 original draft, Visualization, Validation, Resources, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Arnau Puig-Davi: Writing \u2013 original draft, Project administration, Methodology, Investigation, Formal analysis, Data curation. Frederic Sampedro: Writing \u2013 original draft, Methodology, Investigation, Formal analysis. Jes\u00fas P\u00e9rez-P\u00e9rez: Writing \u2013 review & editing, Writing \u2013 original draft, Project administration, Investigation, Funding acquisition, Formal analysis. Carla Franch-Mart\u00ed: Writing \u2013 original draft, Resources, Project administration, Methodology, Investigation. Gonzalo Olmedo-Saura: Writing \u2013 original draft, Project administration, Investigation, Formal analysis. Elisa Rivas-Asensio: Writing \u2013 original draft, Methodology, Investigation, Formal analysis. Anna Vazquez-Oliver: Writing \u2013 original draft, Software, Methodology, Investigation. Laura P\u00e9rez-Carasol: Writing \u2013 original draft, Validation, Supervision, Methodology, Investigation, Formal analysis. Andrea Horta-Barba: Writing \u2013 original draft, Project administration, Methodology, Investigation. Javier Pagonabarraga: Writing \u2013 review & editing, Writing \u2013 original draft, Resources, Project administration, Funding acquisition. Jaime Kulisevsky: Writing \u2013 review & editing, Writing \u2013 original draft, Validation, Supervision, Resources, Funding acquisition.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Ethical compliance statement", "text": "This study was approved by the Institutional Review Board of the Hospital de la Santa Creu i Sant Pau. Written informed consent was obtained from all patients prior to their participation and was properly documented in accordance with institutional policies. We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this work is consistent with those guidelines.", "title": "" }, { "pmid": 41492395, "pmcid": "PMC12765053", "section": "Declaration of competing interest", "text": "The authors declare the following financial interests/personal relationships which may be considered as potential competing interests:Saul Martinez-Horta reports financial support was provided by Carlos III Health Institute. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "1. Introduction", "text": "Huntington\u2019s disease (HD) is a fatal neurodegenerative disorder driven by a Cytosine-Adenine-Guanine (CAG) repeat expansion in the huntingtin (HTT) gene [1]. A hallmark of HD pathology is the selective vulnerability and progressive degeneration of striatal medium spiny neurons in the caudate and putamen [2]. Striatal atrophy begins well before the onset of diagnostic motor manifestations, providing a long prodromal window during which neurodegeneration can be tracked using structural magnetic resonance imaging (MRI) [3,4,5,6]. While MRI offers a macroscopic readout of tissue loss, fluid biomarkers are needed to interrogate the molecular processes associated with this degeneration and its modulation over time [7].", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "1. Introduction", "text": "Current biomarker research has largely emphasized cytoskeletal markers such as neurofilament light chain (NEFL), which increase in cerebrospinal fluid (CSF) and blood in association with neurodegeneration and track disease intensity and progression in HD. Although NEFL provides a robust gauge of aggregate axonal damage, it is less informative regarding upstream biological processes\u2014such as inflammation and immune dysregulation\u2014that may contribute to or modulate degeneration in HD [8,9]. Advances in high-sensitivity, high-multiplex proteomic technologies now enable improved characterization of low-abundance inflammatory and immune mediators in biofluids. In particular, the Next-Gen Ultra-Sensitive Immunoassay (NULISA) platform supports attomolar-sensitive, multiplexed quantification of large protein panels\u2014including cytokines and other immune-related targets\u2014offering a new window into the immune microenvironment associated with neurodegeneration [10].", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "1. Introduction", "text": "While previous research has highlighted broad immune activation and microglial involvement in HD, there is a critical need to evaluate specific immune-regulatory pathways to better understand phenotypic heterogeneity. One such candidate is TNFRSF8 (CD30), a member of the tumor necrosis factor (TNF) receptor superfamily known for its role in immune signaling and regulation. High-sensitivity CSF proteomics now allows us to quantify these low-abundance immune regulators to see whether they index distinct biological tracks of neurodegeneration. In the context of chronic neuroinflammation, we hypothesize that reduced levels of TNFRSF8 in CSF indicate a loss of putative immune-regulatory signaling. This decline would be consistent with an immune-associated signal emerging as striatal atrophy progresses, providing a distinct signal separable from aggregate cytoskeletal injury.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "1. Introduction", "text": "HD progression shows substantial inter-individual variability, and mounting evidence implicates immune dysregulation as one contributor to phenotypic heterogeneity and progression rates. To conceptualize this, we propose a \u201cTwo-Track\u201d (or dual-pathway) model of HD. In this framework, Track 1 represents the well-established cascade of structural and cytoskeletal disintegration, robustly captured by markers like NEFL. Track 2 represents a distinct, downstream immune-associated signal, indexed by low-abundance immune-regulatory proteins like TNFRSF8. The practical application of this dual-pathway model is highly relevant for future clinical trials: if these parallel tracks are statistically independent, it implies that interventions targeting structural neurodegeneration may not inherently correct immune dysfunction, necessitating multidimensional biomarker endpoints and combination therapies. In this study, we test whether immune-regulatory signaling measured in CSF explains variance in striatal atrophy that is statistically independent of neuroaxonal damage markers. To ensure that these profiles reflect HD-specific pathogenesis rather than generalized aging or non-specific neurodegeneration, we explicitly evaluate the anatomical specificity of this signal and decouple the effects of chronological age from the underlying genetic mutation. Furthermore, we interrogate the directionality of this neuro-immune axis to determine whether immune dysregulation acts as a primary driver of atrophy or as a secondary, downstream consequence of structural degeneration.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "2. Results", "text": "Volumetric screening confirmed significant, disease-specific atrophy in the striatum. The strongest correlation between gray matter volume and disease burden (CAG-Age-Product [CAP] score) was observed in the putamen (Spearman rho = \u22120.64, p < 0.001), followed by the caudate (Spearman rho = \u22120.44, p < 0.001). Preliminary analyses indicated no significant differences between male and female participants in normalized putamen volume (p = 0.103), CSF NEFL levels (p = 0.820), or CSF TNFRSF8 levels (p = 0.700). Measures of global brain volume showed weaker associations, validating the striatum as the primary locus of degeneration in this cohort (Table 1). These patterns are consistent with established longitudinal imaging findings in HD cohorts [4,5,7,11,12].\n\nTable 1 displays the 10 brain regions demonstrating the strongest inverse correlations with disease burden (CAP score). Associations with disease burden were evaluated using Pearson correlation coefficients (R) and their corresponding unadjusted significance levels (p-value (CAP)). To adjust for head size, raw regional volumes were divided by total intracranial volume, yielding the standardized values shown as Norm_Stats (SD). For enhanced interpretability of the most highly significant regions, the unadjusted p-values are also displayed on a negative logarithmic scale (\u2212log10 (p)).", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "2. Results", "text": "To confirm the regional specificity of this immune-structural axis and ensure that it did not merely reflect generalized, age-related neurodegeneration, we evaluated TNFRSF8 against structural control regions. Although TNFRSF8 was robustly correlated with striatal atrophy (both putamen and caudate), it showed no significant association with hippocampal or cerebellar volumes (Table S1). This regional divergence indicates that the TNFRSF8 signature is specifically tied to striatal degeneration rather than global structural decline.\n\nWe identified two distinct biological signals tracking with striatal atrophy. Screening results are provided in Table S2 (Spearman rho), whereas the region-of-interest associations below are reported as Pearson r.\n\nThe Damage Signal (NEFL): Levels of NEFL were strongly negatively correlated with putamen volume (Pearson r = \u22120.53, p < 0.001). As striatal tissue is lost, CSF NEFL concentrations rise, reflecting neuroaxonal injury and release of cytoskeletal debris, consistent with extensive prior work supporting NEFL as a biomarker of neurodegeneration in HD [13,14].", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "2. Results", "text": "The Immune Signal (TNFRSF8): In a divergent pattern, the immune regulator TNFRSF8 (CD30) showed a positive correlation with putamen volume (Pearson r = 0.36, p < 0.001). As the striatum degenerates, levels of this putative immune-regulatory signal increase. TNFRSF8/CD30 is a member of the tumor necrosis factor (TNF) receptor superfamily with immune-regulatory functions, supporting its interpretation as an immune-associated signal in this context [15,16].\n\nThe correlations differed in direction (Figure 1). Additionally, the difference between dependent correlations was statistically significant (Steiger\u2019s Z = \u22125.94, p < 0.001). Taken together, these patterns support opposing biological behaviors rather than a single underlying factor [17].\n\nTo determine the optimal baseline for our independence testing, we first performed prespecified disease-burden model comparisons on the complete-case sample (N = 88). The CAP-adjusted model showed the best performance (Adj R2 = 0.380; CV RMSE = 0.108), modestly outperforming the age-adjusted model (Adj R2 = 0.350; CV RMSE = 0.111) and the no-burden model (Adj R2 = 0.355; CV RMSE = 0.110). Based on these results (Table 2), the CAP-adjusted specification served as the base model for subsequent testing.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "2. Results", "text": "To determine whether the immune signal provides unique information beyond this baseline, we performed an independence test using multivariable regression (Figure 2). The base model with CAP score, sex, and NEFL explained 35.6% of the variance in putamen volume (R2 = 0.356). Adding TNFRSF8 increased explained variance to 40.9% (R2 = 0.409; \u0394R2 = 0.053), and it significantly improved model fit (p = 0.008). TNFRSF8 was statistically significant as an independent predictor (Beta = 0.24, p = 0.008) even after controlling for the robust effect of NEFL, cumulative genetic burden (CAP score), and sex. To ensure that this relationship was not confounded by general aging, we ran an additional model decoupling the CAP score into chronological age and CAG repeat length. TNFRSF8 maintained its strong independent association with putamen volume (p = 0.008) whereas chronological age did not (p = 0.055), confirming that this neuro-immune axis is driven by HD pathogenesis rather than the passage of time (Supplementary Table S3). To mitigate optimism from using the same cohort for screening and modeling, we estimated confidence intervals for Beta via bootstrap resampling (B = 1000). Internal validation confirmed the stability of the TNFRSF8 effect (Beta = 0.24, 95% Bootstrap CI: [0.06, 0.46]) [18].\n\nRepresentative individual-level examples are shown in Figure 3, illustrating two distinct biomarker signatures (high NEFL vs. low TNFRSF8) within the same putamen atrophy quartile.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "2. Results", "text": "To explore the directionality of the neuro-immune axis, we tested alternative causal models. Cross-sectional mediation of CAP effects on putamen volume via TNFRSF8 was not strongly supported (Sobel p = 0.106; Supplementary Table S4. Conversely, testing putamen volume as a mediator of CAP effects on TNFRSF8 levels revealed a significant indirect effect (Sobel Z = \u22122.56, p = 0.010; bootstrap 95% CI [\u22120.0030, \u22120.0005]). This directional asymmetry suggests that structural striatal atrophy mediates the relationship between genetic burden and subsequent immune dysregulation, rather than the reverse (Figure 4).", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "2.1. Validation of Striatal Atrophy", "text": "Volumetric screening confirmed significant, disease-specific atrophy in the striatum. The strongest correlation between gray matter volume and disease burden (CAG-Age-Product [CAP] score) was observed in the putamen (Spearman rho = \u22120.64, p < 0.001), followed by the caudate (Spearman rho = \u22120.44, p < 0.001). Preliminary analyses indicated no significant differences between male and female participants in normalized putamen volume (p = 0.103), CSF NEFL levels (p = 0.820), or CSF TNFRSF8 levels (p = 0.700). Measures of global brain volume showed weaker associations, validating the striatum as the primary locus of degeneration in this cohort (Table 1). These patterns are consistent with established longitudinal imaging findings in HD cohorts [4,5,7,11,12].\n\nTable 1 displays the 10 brain regions demonstrating the strongest inverse correlations with disease burden (CAP score). Associations with disease burden were evaluated using Pearson correlation coefficients (R) and their corresponding unadjusted significance levels (p-value (CAP)). To adjust for head size, raw regional volumes were divided by total intracranial volume, yielding the standardized values shown as Norm_Stats (SD). For enhanced interpretability of the most highly significant regions, the unadjusted p-values are also displayed on a negative logarithmic scale (\u2212log10 (p)).", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "2.1. Validation of Striatal Atrophy", "text": "To confirm the regional specificity of this immune-structural axis and ensure that it did not merely reflect generalized, age-related neurodegeneration, we evaluated TNFRSF8 against structural control regions. Although TNFRSF8 was robustly correlated with striatal atrophy (both putamen and caudate), it showed no significant association with hippocampal or cerebellar volumes (Table S1). This regional divergence indicates that the TNFRSF8 signature is specifically tied to striatal degeneration rather than global structural decline.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "2.2. Divergent Proteomic Signatures", "text": "We identified two distinct biological signals tracking with striatal atrophy. Screening results are provided in Table S2 (Spearman rho), whereas the region-of-interest associations below are reported as Pearson r.\n\nThe Damage Signal (NEFL): Levels of NEFL were strongly negatively correlated with putamen volume (Pearson r = \u22120.53, p < 0.001). As striatal tissue is lost, CSF NEFL concentrations rise, reflecting neuroaxonal injury and release of cytoskeletal debris, consistent with extensive prior work supporting NEFL as a biomarker of neurodegeneration in HD [13,14].\n\nThe Immune Signal (TNFRSF8): In a divergent pattern, the immune regulator TNFRSF8 (CD30) showed a positive correlation with putamen volume (Pearson r = 0.36, p < 0.001). As the striatum degenerates, levels of this putative immune-regulatory signal increase. TNFRSF8/CD30 is a member of the tumor necrosis factor (TNF) receptor superfamily with immune-regulatory functions, supporting its interpretation as an immune-associated signal in this context [15,16].\n\nThe correlations differed in direction (Figure 1). Additionally, the difference between dependent correlations was statistically significant (Steiger\u2019s Z = \u22125.94, p < 0.001). Taken together, these patterns support opposing biological behaviors rather than a single underlying factor [17].", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "2.3. Independent Contributions to Atrophy", "text": "To determine the optimal baseline for our independence testing, we first performed prespecified disease-burden model comparisons on the complete-case sample (N = 88). The CAP-adjusted model showed the best performance (Adj R2 = 0.380; CV RMSE = 0.108), modestly outperforming the age-adjusted model (Adj R2 = 0.350; CV RMSE = 0.111) and the no-burden model (Adj R2 = 0.355; CV RMSE = 0.110). Based on these results (Table 2), the CAP-adjusted specification served as the base model for subsequent testing.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "2.3. Independent Contributions to Atrophy", "text": "To determine whether the immune signal provides unique information beyond this baseline, we performed an independence test using multivariable regression (Figure 2). The base model with CAP score, sex, and NEFL explained 35.6% of the variance in putamen volume (R2 = 0.356). Adding TNFRSF8 increased explained variance to 40.9% (R2 = 0.409; \u0394R2 = 0.053), and it significantly improved model fit (p = 0.008). TNFRSF8 was statistically significant as an independent predictor (Beta = 0.24, p = 0.008) even after controlling for the robust effect of NEFL, cumulative genetic burden (CAP score), and sex. To ensure that this relationship was not confounded by general aging, we ran an additional model decoupling the CAP score into chronological age and CAG repeat length. TNFRSF8 maintained its strong independent association with putamen volume (p = 0.008) whereas chronological age did not (p = 0.055), confirming that this neuro-immune axis is driven by HD pathogenesis rather than the passage of time (Supplementary Table S3). To mitigate optimism from using the same cohort for screening and modeling, we estimated confidence intervals for Beta via bootstrap resampling (B = 1000). Internal validation confirmed the stability of the TNFRSF8 effect (Beta = 0.24, 95% Bootstrap CI: [0.06, 0.46]) [18].\n\nRepresentative individual-level examples are shown in Figure 3, illustrating two distinct biomarker signatures (high NEFL vs. low TNFRSF8) within the same putamen atrophy quartile.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "2.4. Mediation Analysis", "text": "To explore the directionality of the neuro-immune axis, we tested alternative causal models. Cross-sectional mediation of CAP effects on putamen volume via TNFRSF8 was not strongly supported (Sobel p = 0.106; Supplementary Table S4. Conversely, testing putamen volume as a mediator of CAP effects on TNFRSF8 levels revealed a significant indirect effect (Sobel Z = \u22122.56, p = 0.010; bootstrap 95% CI [\u22120.0030, \u22120.0005]). This directional asymmetry suggests that structural striatal atrophy mediates the relationship between genetic burden and subsequent immune dysregulation, rather than the reverse (Figure 4).", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "3. Discussion", "text": "Our findings propose a \u201cTwo-Track\u201d model of HD pathology. Track 1, represented by NEFL, is the familiar signal of structural disintegration\u2014a readout of neuroaxonal injury that is increasingly used in HD biomarker research [13,14]. Track 2, represented by TNFRSF8 (CD30), suggests a distinct, immune-associated signal that declines with advancing striatal degeneration. TNFRSF8/CD30 is a TNF receptor superfamily member with known roles in immune signaling and regulation, and its decline is a hallmark of immune dysregulation, consistent with an interpretation involving reduced immune-regulatory capacity as atrophy progresses [15,16].", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "3. Discussion", "text": "This immune interpretation is biologically plausible given a broader literature supporting immune alterations in HD, including immune activation detectable prior to clinical onset, neuroinflammatory pathways implicated in disease progression, and evidence of microglial activation in persons with HD before clinical motor diagnosis [8,19,20,21,22,23]. The positive correlation between TNFRSF8 (CD30) and putamen volume may initially appear counterintuitive. In many neurodegenerative contexts, immune markers rise concurrently with tissue damage as part of a reactive neuroinflammatory response. However, soluble CD30 is primarily shed by actively engaged and proliferating immune cells, particularly activated T-cells. The observation that CD30 levels decline as striatal atrophy advances suggests a longitudinal shift toward a distinct late-stage immune-associated signal. In the context of chronic neuroinflammation, prolonged exposure to disease pathology may lead to profound immune dysregulation, impairing the cells\u2019 ability to maintain active regulatory signaling and blunting the shedding of CD30 into the CSF. Alternatively, this decline may reflect the progressive loss of protective, CD30-expressing regulatory immune cells that normally help maintain neuro-glial homeostasis. Under either mechanism, lower CD30 levels do not indicate an absence of inflammation, but rather a maladaptive failure of active immune-regulatory networks as the disease worsens.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "3. Discussion", "text": "The clinical implication is that targeting the immune system in HD may require a different approach than nonspecific suppression of inflammation. Specifically, if the progressive decline in TNFRSF8/CD30 reflects a distinct, maladaptive immune-associated signal, future HD clinical trials might need to explore immune-modulating therapies designed to restore this specific signaling pathway. Further, utilizing CD30 as an independent biomarker could allow these trials to measure target engagement and immune restoration distinctly from structural cytoskeletal damage. Therefore, future research should evaluate whether preserving or restoring immune-regulatory signaling pathways (e.g., TNF/TNFR superfamily) yields clinical benefit, ideally using designs that can test temporal ordering and within-person coupling [24].\n\nCrucially, our modeling choices and mediation analyses reveal important insights into the fundamental nature of HD progression. The finding that CAP-adjusted models outperformed age-adjusted models reinforces the concept that these structural and immune alterations are driven by the specific toxic genetic burden of the disease\u2014integrating the CAG repeat expansion over time\u2014rather than chronological aging alone [25].", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "3. Discussion", "text": "Furthermore, our bidirectional mediation analyses are highly revealing. The data do not support a simple mechanistic chain where genetic burden (CAP) directly causes an immune-associated signal (TNFRSF8 decline), which in turn causes striatal atrophy. Instead, we found significant support for the reverse pathway: putamen volume formally mediates the relationship between CAP score and TNFRSF8 levels. This directional asymmetry provides crucial statistical backing for our hypothesis that TNFRSF8 represents a downstream, late-stage immune-associated signal. It suggests an \u2018atrophy-driven\u2019 model of neuro-immune collapse, where the primary genetic insult initiates structural cytoskeletal damage (Track 1), and the accumulating central nervous system burden subsequently drives the dysregulation of peripheral immune-regulatory networks (Track 2). This temporal hierarchy highlights the complexity of the neuro-immune axis in HD, implying that preventing primary structural damage may be a prerequisite for rescuing immune homeostasis.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "3. Discussion", "text": "Because these two tracks provide unique, non-overlapping information, relying solely on single markers like NEFL may obscure critical dimensions of a patient\u2019s disease state. A composite biomarker approach\u2014integrating both neuroaxonal (NEFL) and immune-regulatory (TNFRSF8) readouts\u2014could provide a much higher-resolution, multidimensional picture of HD pathology. Clinically, this dual-axis profiling could prove invaluable for next-generation trial designs, allowing for biomarker-guided patient stratification, prognostic enrichment, and the precise measurement of target engagement for combination therapies that address both neurodegeneration and neuroinflammation [13,14].", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "3. Discussion", "text": "Despite these findings, this study presents several limitations that may inform future research. First, cross-sectional mediation analysis may be biased in cases where the underlying causal process is dynamic or complex. Second, the comparatively small sample size of 88 persons with HD may limit the statistical power and generalizability of our findings. Third, the current study lacks a control group of gene-negative family members or healthy individuals. While our primary focus was evaluating continuous relationships between biofluid markers and structural atrophy within the HD continuum, the absence of a control cohort precludes direct comparisons to a healthy neuro-immune baseline. However, our supplementary regional specificity and chronological age-decoupling analyses strongly mitigate the concern that these findings merely reflect global aging or generalized neurodegeneration, instead confirming a highly localized, disease-specific striatal axis. Fourth, data regarding concomitant medications were based on self-reported clinical history. The lack of standardized, prospective medication logs or toxicology screening may limit the precision with which we could evaluate the influence of anti-inflammatory or other confounding substances on the CSF proteome.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "3. Discussion", "text": "To directly address these limitations and prospectively validate the disease-specificity observed in our internal structural controls, future work should extend this cross-sectional analysis using larger longitudinal datasets that include gene-negative controls and repeated CSF collection paired to structural MRI. Such designs would enable (i) mixed-effects trajectory models to quantify within-person change in TNFRSF8 and striatal volume, (ii) tests of temporal precedence using time-lagged associations, and (iii) longitudinal mediation or dynamic path models that are not identifiable from cross-sectional data [26,27,28]. Furthermore, expanding this framework into multi-pathway panels will be critical for capturing the full pathophysiological cascade of HD. Future investigations should integrate our findings with parallel markers such as neurofilament heavy chain (NEFH) to complement NEFL as an index of varied axonal pathology [29], myeloperoxidase (MPO) to capture the oxidative stress and innate immune dysregulation characteristic of HD [30,31], and ubiquitin C-terminal hydrolase L1 (UCHL1). As a core component of the ubiquitin-proteasome system, UCHL1 serves not only as a direct measure of neuronal somatic injury but also reflects the breakdown of proteostasis central to mutant huntingtin aggregation [32,33]. With larger controlled longitudinal samples, these expanded multi-marker approaches could be definitively evaluated for prognostic enrichment and sensitivity to therapeutic target engagement in interventional studies.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "3.1. Limitations", "text": "Despite these findings, this study presents several limitations that may inform future research. First, cross-sectional mediation analysis may be biased in cases where the underlying causal process is dynamic or complex. Second, the comparatively small sample size of 88 persons with HD may limit the statistical power and generalizability of our findings. Third, the current study lacks a control group of gene-negative family members or healthy individuals. While our primary focus was evaluating continuous relationships between biofluid markers and structural atrophy within the HD continuum, the absence of a control cohort precludes direct comparisons to a healthy neuro-immune baseline. However, our supplementary regional specificity and chronological age-decoupling analyses strongly mitigate the concern that these findings merely reflect global aging or generalized neurodegeneration, instead confirming a highly localized, disease-specific striatal axis. Fourth, data regarding concomitant medications were based on self-reported clinical history. The lack of standardized, prospective medication logs or toxicology screening may limit the precision with which we could evaluate the influence of anti-inflammatory or other confounding substances on the CSF proteome.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "3.2. Future Directions", "text": "To directly address these limitations and prospectively validate the disease-specificity observed in our internal structural controls, future work should extend this cross-sectional analysis using larger longitudinal datasets that include gene-negative controls and repeated CSF collection paired to structural MRI. Such designs would enable (i) mixed-effects trajectory models to quantify within-person change in TNFRSF8 and striatal volume, (ii) tests of temporal precedence using time-lagged associations, and (iii) longitudinal mediation or dynamic path models that are not identifiable from cross-sectional data [26,27,28]. Furthermore, expanding this framework into multi-pathway panels will be critical for capturing the full pathophysiological cascade of HD. Future investigations should integrate our findings with parallel markers such as neurofilament heavy chain (NEFH) to complement NEFL as an index of varied axonal pathology [29], myeloperoxidase (MPO) to capture the oxidative stress and innate immune dysregulation characteristic of HD [30,31], and ubiquitin C-terminal hydrolase L1 (UCHL1). As a core component of the ubiquitin-proteasome system, UCHL1 serves not only as a direct measure of neuronal somatic injury but also reflects the breakdown of proteostasis central to mutant huntingtin aggregation [32,33]. With larger controlled longitudinal samples, these expanded multi-marker approaches could be definitively evaluated for prognostic enrichment and sensitivity to therapeutic target engagement in interventional studies.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "4. Materials and Methods", "text": "We analyzed data from 88 people with HD (CAG repeats \u226536) spanning the stages before clinical motor diagnosis and the early stages after clinical motor diagnosis [34,35]. To minimize confounding effects on biofluid immune markers, patients taking potent immunomodulatory therapies were excluded from the study. Concomitant medications were screened for use of common anti-inflammatories, including aspirin, ibuprofen, and other non-steroidal anti-inflammatory drugs (NSAID). Demographics and clinical characteristics are summarized in Table 3. The cohort had a mean age of 39.0 years (SD = 11.8) and a mean CAP score of 336.5 (SD = 92.3) [25,36].", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "4. Materials and Methods", "text": "T1-weighted MRI scans were acquired on GE (GE HealthCare Technologies Inc. in Chicago, IL, USA) 3T scanners using harmonized protocols across sites and processed with FastSurfer (German Center for Neurodegenerative Diseases [DZNE], Bonn, Germany) to generate automated segmentations of subcortical structures [37]. Caudate and putamen volumes were normalized to intracranial volume (ICV) using the ratio method (Volume/ICV). To evaluate regional specificity, bilateral volumes of the hippocampus and cerebellum (including both cerebellar cortex and white matter) were also extracted and similarly normalized to ICV to serve as structural control regions. For the present analysis, which required paired CSF and MRI, we restricted the sample to participants with both structural MRI and CSF available from the University of Iowa and the University of Wisconsin\u2013Madison.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "4. Materials and Methods", "text": "CSF samples included in this analysis were collected at the University of Iowa and the University of Wisconsin\u2013Madison under standardized protocols, processed according to study procedures, and stored until assay. Samples were shipped to the BioSpecimen Exchange for Neurological Disorders (BioSEND) repository at Indiana University for centralized handling, where they were blinded and aliquoted before shipment to the University of Wisconsin Alzheimer\u2019s Disease Research Center (ADRC) Biomarker Lab for analysis [38]. Following completion of laboratory assays, results were unblinded, enabling linkage of CSF measures to imaging-derived outcomes for the present conjoined analyses.\n\nProteomic profiling was performed using the NULISAseq Central Nervous System (CNS) Disease Panel 120 and Inflammation Panel (NIF) 250 (Alamar Biosciences, Fremont, CA, USA), targeting a combined total of approximately 370 proteins [10,39,40]. Data were normalized using internal and inter-plate controls and log2-transformed to the NULISA Protein Quantification (NPQ) scale [10]. Values below the limit of detection (LOD) were retained (as NPQ values) consistent with manufacturer recommendations, and low-detectability analytes were filtered prior to screening, yielding 268 proteins for downstream analysis [10].", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "4. Materials and Methods", "text": "Sex was included a priori as a biological covariate in volumetric models. Preliminary group comparisons by sex were evaluated using Welch\u2019s t-tests. Other potential covariates (education, study site/scanner, race/ethnicity, and clinical severity measures such as total motor score) were not included in the primary model because they can be sparse/collinear in this sample and, for some measures, may be downstream of neurodegeneration. These were reserved for sensitivity analyses.\n\nScreening procedure: To identify candidate biomarkers of striatal neurodegeneration, we screened the 268 retained proteins by computing Spearman\u2019s rank correlations between protein abundance and normalized putamen volume. p-values were adjusted using the Benjamini\u2013Hochberg false discovery rate (FDR) procedure [41]. Significance was defined as q < 0.05, and analytes meeting a suggestive threshold (q < 0.10) were advanced for multivariable characterization. TNFRSF8 was pre-specified for follow-up given its immune-regulatory role and interpretability as a distinct immune-associated signal in the context of broader immune dysregulation; in this screen, it met the suggestive FDR threshold (q = 0.059).", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "4. Materials and Methods", "text": "Multivariable modeling: Analytes meeting the suggestive FDR threshold (q < 0.10) were entered into multivariable linear regression models. In addition, TNFRSF8 was included a priori as a pre-specified immune-regulatory target given its interpretability in the context of immune dysregulation. All continuous predictors (CAP score, NEFL, and candidate immune markers) were Z-score standardized prior to modeling to enable direct comparison of standardized regression coefficients (Beta). The primary model was specified as follows: Putamen_Volume ~ CAP + Sex + NEFL + TNFRSF8. While the screening step used Spearman correlations to reduce sensitivity to outliers, region-of-interest matrices and multivariable models used Pearson correlations and ordinary least squares (OLS) regression on standardized data to satisfy linear modeling assumptions.\n\nDisease-burden model comparison: We compared three prespecified models on a common complete-case sample (N = 88): (i) no disease-burden covariate, (ii) age-adjusted, and (iii) CAP-adjusted. Models were compared using adjusted R2, AIC/BIC, and k-fold cross-validated RMSE/MAE (k = 10 when available). CAP was retained as the primary disease-burden covariate because it provided superior fit and/or lower cross-validated error relative to age.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "4. Materials and Methods", "text": "Regional Specificity and Age-Decoupling Analyses: To ensure the identified neuro-immune associations reflected HD-specific pathogenesis rather than generalized aging or non-specific neurodegeneration, two pre-specified supplementary analyses were conducted. First, regional specificity was evaluated by calculating Pearson correlations between TNFRSF8 and the ICV-normalized volumes of structural control regions (hippocampus and cerebellum). Second, to mathematically decouple the effects of chronological aging from genetic burden, an additional multivariable OLS regression model was conducted. In this model, the composite CAP score was separated into its constituent variables\u2014chronological age and CAG repeat length\u2014to determine whether TNFRSF8 maintained a significant association with putamen volume independent of chronological age.\n\nModel diagnostics: Multicollinearity was assessed using the variance inflation factor (VIF) [42]; all predictors had VIF < 3 (max VIF = 1.80). Residuals were evaluated for normality (Shapiro\u2013Wilk) and homoscedasticity (Breusch\u2013Pagan) [43,44]. Influential observations were assessed using Cook\u2019s distance (threshold 4/N) [45].", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "4. Materials and Methods", "text": "Sensitivity analyses were performed to evaluate the potential influence of concomitant anti-inflammatory medications. No significant associations were observed between the use of aspirin or other NSAID and CSF TNFRSF8 levels. It is, however, noted that the number of patients taking these medications was minimal in the current cohort.\n\nValidation and limitations: To mitigate optimism from using the same cohort for screening and modeling, confidence intervals for regression coefficients were estimated via bootstrap resampling (B = 1000) [18]. Mediation analysis was conducted to assess statistical mediation, evaluating whether TNFRSF8 mediated the association between CAP score and putamen volume [46,47]. All results are presented as exploratory pending external replication.\n\nThe study employed a paper-as-code workflow, in which pre-specified analysis scripts automatically generate tables, figures, and corresponding manuscript text directly from the finalized statistical outputs. This approach ensures internal consistency, reproducibility, and traceability between the analysis code and the reported results.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "4. Materials and Methods", "text": "Any automated text generation was deterministic and rule-based, reflecting fixed templates populated by computed results, rather than probabilistic or generative modeling. No large language model or generative AI system was used to infer, summarize, or interpret findings beyond formatting and structured insertion of results into manuscript templates. All analytical decisions, statistical modeling choices, and scientific interpretations were defined a priori by the authors and verified manually prior to submission.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "4.1. Participants", "text": "We analyzed data from 88 people with HD (CAG repeats \u226536) spanning the stages before clinical motor diagnosis and the early stages after clinical motor diagnosis [34,35]. To minimize confounding effects on biofluid immune markers, patients taking potent immunomodulatory therapies were excluded from the study. Concomitant medications were screened for use of common anti-inflammatories, including aspirin, ibuprofen, and other non-steroidal anti-inflammatory drugs (NSAID). Demographics and clinical characteristics are summarized in Table 3. The cohort had a mean age of 39.0 years (SD = 11.8) and a mean CAP score of 336.5 (SD = 92.3) [25,36].", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "4.2. Neuroimaging", "text": "T1-weighted MRI scans were acquired on GE (GE HealthCare Technologies Inc. in Chicago, IL, USA) 3T scanners using harmonized protocols across sites and processed with FastSurfer (German Center for Neurodegenerative Diseases [DZNE], Bonn, Germany) to generate automated segmentations of subcortical structures [37]. Caudate and putamen volumes were normalized to intracranial volume (ICV) using the ratio method (Volume/ICV). To evaluate regional specificity, bilateral volumes of the hippocampus and cerebellum (including both cerebellar cortex and white matter) were also extracted and similarly normalized to ICV to serve as structural control regions. For the present analysis, which required paired CSF and MRI, we restricted the sample to participants with both structural MRI and CSF available from the University of Iowa and the University of Wisconsin\u2013Madison.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "4.3. CSF Collection and Handling", "text": "CSF samples included in this analysis were collected at the University of Iowa and the University of Wisconsin\u2013Madison under standardized protocols, processed according to study procedures, and stored until assay. Samples were shipped to the BioSpecimen Exchange for Neurological Disorders (BioSEND) repository at Indiana University for centralized handling, where they were blinded and aliquoted before shipment to the University of Wisconsin Alzheimer\u2019s Disease Research Center (ADRC) Biomarker Lab for analysis [38]. Following completion of laboratory assays, results were unblinded, enabling linkage of CSF measures to imaging-derived outcomes for the present conjoined analyses.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "4.4. Proteomics", "text": "Proteomic profiling was performed using the NULISAseq Central Nervous System (CNS) Disease Panel 120 and Inflammation Panel (NIF) 250 (Alamar Biosciences, Fremont, CA, USA), targeting a combined total of approximately 370 proteins [10,39,40]. Data were normalized using internal and inter-plate controls and log2-transformed to the NULISA Protein Quantification (NPQ) scale [10]. Values below the limit of detection (LOD) were retained (as NPQ values) consistent with manufacturer recommendations, and low-detectability analytes were filtered prior to screening, yielding 268 proteins for downstream analysis [10].", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "4.5. Statistical Analysis and Biomarker Screening", "text": "Sex was included a priori as a biological covariate in volumetric models. Preliminary group comparisons by sex were evaluated using Welch\u2019s t-tests. Other potential covariates (education, study site/scanner, race/ethnicity, and clinical severity measures such as total motor score) were not included in the primary model because they can be sparse/collinear in this sample and, for some measures, may be downstream of neurodegeneration. These were reserved for sensitivity analyses.\n\nScreening procedure: To identify candidate biomarkers of striatal neurodegeneration, we screened the 268 retained proteins by computing Spearman\u2019s rank correlations between protein abundance and normalized putamen volume. p-values were adjusted using the Benjamini\u2013Hochberg false discovery rate (FDR) procedure [41]. Significance was defined as q < 0.05, and analytes meeting a suggestive threshold (q < 0.10) were advanced for multivariable characterization. TNFRSF8 was pre-specified for follow-up given its immune-regulatory role and interpretability as a distinct immune-associated signal in the context of broader immune dysregulation; in this screen, it met the suggestive FDR threshold (q = 0.059).", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "4.5. Statistical Analysis and Biomarker Screening", "text": "Multivariable modeling: Analytes meeting the suggestive FDR threshold (q < 0.10) were entered into multivariable linear regression models. In addition, TNFRSF8 was included a priori as a pre-specified immune-regulatory target given its interpretability in the context of immune dysregulation. All continuous predictors (CAP score, NEFL, and candidate immune markers) were Z-score standardized prior to modeling to enable direct comparison of standardized regression coefficients (Beta). The primary model was specified as follows: Putamen_Volume ~ CAP + Sex + NEFL + TNFRSF8. While the screening step used Spearman correlations to reduce sensitivity to outliers, region-of-interest matrices and multivariable models used Pearson correlations and ordinary least squares (OLS) regression on standardized data to satisfy linear modeling assumptions.\n\nDisease-burden model comparison: We compared three prespecified models on a common complete-case sample (N = 88): (i) no disease-burden covariate, (ii) age-adjusted, and (iii) CAP-adjusted. Models were compared using adjusted R2, AIC/BIC, and k-fold cross-validated RMSE/MAE (k = 10 when available). CAP was retained as the primary disease-burden covariate because it provided superior fit and/or lower cross-validated error relative to age.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "4.5. Statistical Analysis and Biomarker Screening", "text": "Regional Specificity and Age-Decoupling Analyses: To ensure the identified neuro-immune associations reflected HD-specific pathogenesis rather than generalized aging or non-specific neurodegeneration, two pre-specified supplementary analyses were conducted. First, regional specificity was evaluated by calculating Pearson correlations between TNFRSF8 and the ICV-normalized volumes of structural control regions (hippocampus and cerebellum). Second, to mathematically decouple the effects of chronological aging from genetic burden, an additional multivariable OLS regression model was conducted. In this model, the composite CAP score was separated into its constituent variables\u2014chronological age and CAG repeat length\u2014to determine whether TNFRSF8 maintained a significant association with putamen volume independent of chronological age.\n\nModel diagnostics: Multicollinearity was assessed using the variance inflation factor (VIF) [42]; all predictors had VIF < 3 (max VIF = 1.80). Residuals were evaluated for normality (Shapiro\u2013Wilk) and homoscedasticity (Breusch\u2013Pagan) [43,44]. Influential observations were assessed using Cook\u2019s distance (threshold 4/N) [45].", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "4.5. Statistical Analysis and Biomarker Screening", "text": "Sensitivity analyses were performed to evaluate the potential influence of concomitant anti-inflammatory medications. No significant associations were observed between the use of aspirin or other NSAID and CSF TNFRSF8 levels. It is, however, noted that the number of patients taking these medications was minimal in the current cohort.\n\nValidation and limitations: To mitigate optimism from using the same cohort for screening and modeling, confidence intervals for regression coefficients were estimated via bootstrap resampling (B = 1000) [18]. Mediation analysis was conducted to assess statistical mediation, evaluating whether TNFRSF8 mediated the association between CAP score and putamen volume [46,47]. All results are presented as exploratory pending external replication.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "4.6. Generative Artificial Intelligence", "text": "The study employed a paper-as-code workflow, in which pre-specified analysis scripts automatically generate tables, figures, and corresponding manuscript text directly from the finalized statistical outputs. This approach ensures internal consistency, reproducibility, and traceability between the analysis code and the reported results.\n\nAny automated text generation was deterministic and rule-based, reflecting fixed templates populated by computed results, rather than probabilistic or generative modeling. No large language model or generative AI system was used to infer, summarize, or interpret findings beyond formatting and structured insertion of results into manuscript templates. All analytical decisions, statistical modeling choices, and scientific interpretations were defined a priori by the authors and verified manually prior to submission.", "title": "" }, { "pmid": 41828602, "pmcid": "PMC12985888", "section": "5. Conclusions", "text": "In summary, our study finds that altered immune signaling is a distinct pathological correlate in Huntington\u2019s disease, separable from general cytoskeletal damage and independent of chronological aging. Our findings support a \u201cTwo-Track\u201d model of HD pathology: Track 1, represented by NEFL, tracks structural disintegration while Track 2, represented by TNFRSF8 (CD30), reflects a parallel immune-associated signal that declines with advancing striatal degeneration. This separation highlights a specific, maladaptive pattern of immune dysregulation that occurs alongside physical atrophy. Ultimately, this study enhances our understanding of the multi-dimensional biological processes that drive HD progression. By extending this framework into larger, longitudinal cohorts, the field can further validate comprehensive multi-marker panels. Such approaches hold significant promise for prognostic enrichment, patient stratification, and the precise measurement of target engagement in future immune-modulating clinical trials for persons with HD.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Introduction", "text": "Neurodegenerative disorders (NDs) are a subset of neurological disorders characterized by progressive and irreversible neuronal loss, leading to cognitive, motor, or functional impairments (1). According to the Global Burden of Disease Study (GBD), NDs are the leading cause of disability-adjusted life years (DALYs) and the second leading cause of death worldwide. In 2016, over 276 million DALYs were attributed to NDs, with a significant proportion due to neurodegenerative diseases. These conditions not only reduce life expectancy but also impose an immense strain on healthcare systems, families, and caregivers (2).\n\nSome of the most prevalent NDs include Alzheimer's disease (AD), Parkinson's disease (PD), Huntington's disease (HD), and amyotrophic lateral sclerosis (ALS) (3, 4). The pathophysiology of these disorders is complex, often involving multifactorial genetic and environmental interactions that contribute to neuroinflammation, protein misfolding, mitochondrial dysfunction, and synaptic dysregulation (5).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Introduction", "text": "Despite extensive research, developing disease-modifying treatments for major NDs remains a formidable challenge. For AD, approved therapies (acetylcholinesterase inhibitors, memantine, and recent anti-amyloid antibodies) offer modest symptomatic benefit or, at best, a modest slowing of clinical decline, without halting the underlying pathological cascade (6, 7). PD management relies primarily on dopamine replacement (e.g., L-DOPA) and deep brain stimulation, which effectively manage motor symptoms but do not address the progressive neurodegeneration (8, 9). HD, a monogenic disorder, has no disease-modifying treatment; current care is purely supportive (10, 11).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Introduction", "text": "This critical therapeutic gap stems from several intrinsic obstacles. First, the complex, multifactorial pathogenesis of NDs involves diverse genetic susceptibilities, proteinopathies, and dysfunctional cellular pathways, making single-target pharmacological interventions often insufficient (12). Second, therapeutic delivery to the central nervous system (CNS) is profoundly constrained by the highly selective blood\u2013brain barrier (BBB). Rather than absolute exclusion, the BBB employs tight junctions, efflux transporters (e.g., P-glycoprotein), metabolic enzymes, and receptor-mediated trafficking to regulate xenobiotic entry based on molecular size, charge, lipophilicity, and carrier availability. Consequently, only ~2%\u22125% of small molecules and virtually no unmodified biologics reach therapeutically relevant brain concentrations without specialized delivery strategies (13, 14). Third, the limited intrinsic regenerative capacity of central neurons, coupled with late clinical detection and heterogeneous progression rates, significantly narrows the therapeutic window for disease-modifying interventions (15).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Introduction", "text": "Given this therapeutic stagnation, genome-editing technologies offer the unprecedented ability to intervene at the molecular origins of neurodegeneration (16). Many NDs possess well-validated genetic drivers that represent rational clustered regularly interspaced short palindromic repeats (CRISPR) targets, including huntingtin (HTT) CAG repeat expansions in HD, Presenilin-1 (PSEN1)/PSEN2 and amyloid precursor protein (APP) mutations in familial AD (fAD), and alpha-synuclein (SNCA) and leucine-rich repeat kinase 2 (LRRK2) in PD (17\u201319). Thus, the genetic tractability of major NDs provides a compelling mechanistic rationale for genome editing as a disease-modifying therapeutic strategy.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Introduction", "text": "The discovery of CRISPR-associated protein (Cas) technology has transformed biomedical research, offering precise and programmable tools to modify the genome and transcriptome. Classical Cas9 enables targeted gene knockout, correction, or insertion via guide RNA-directed cleavage (20). The CRISPR toolbox has rapidly expanded to encompass a range of complementary modalities. Catalytically inactive (dead) Cas 9 (dCas9) allows transcriptional repression or activation without altering the DNA sequence (CRISPR interference (CRISPRi/CRISPR activation (CRISPRa), base and prime editors facilitate single-nucleotide resolution corrections, and RNA-targeting Cas13 systems permit transient modulation of disease-associated transcripts (21\u201324). These diverse strategies are particularly advantageous in post-mitotic neurons, where permanent DNA breaks carry increased risk, and reversible or finely tuned modulation of gene expression or pathogenic transcripts may provide safer, more effective interventions. Neurons have intrinsically limited capacity for homology-directed repair (HDR), meaning that double-stranded breaks (DSBs) introduced by conventional nucleases can result in persistent genomic damage and cytotoxicity (25, 26).\n\nBy leveraging these diverse strategies, CRISPR technologies provide the ability not only to correct causative mutations but also to modulate pathogenic pathways, restore cellular homeostasis, and complement conventional symptomatic therapies.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Introduction", "text": "Despite these advances, significant hurdles remain, including efficient delivery across the BBB, off-target effects, immune activation, and ethical considerations. Addressing these barriers is essential to realizing the full therapeutic potential of genome editing in NDs (27).\n\nThis review aims to explore the potential applications of CRISPR\u2013Cas technology in the treatment of neurodegenerative disorders, examining its mechanisms and therapeutic possibilities across a range of conditions. Additionally, we will investigate the challenges in applying CRISPR\u2013Cas technology to neurodegenerative disorders, including delivery barriers, off-target effects, immune responses, and ethical concerns, while proposing potential strategies to overcome these obstacles. Finally, we will examine the status of CRISPR-based therapeutics in clinical trials, identify gaps, and propose future directions toward the successful application of CRISPR\u2013Cas in neurodegenerative diseases.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR\u2013Cas system", "text": "The CRISPR\u2013Cas system serves as an adaptive immune defense mechanism in bacteria and archaea, protecting against invading genetic elements such as bacteriophages and plasmids (28). This system operates through three main stages: adaptation, expression, and interference. During adaptation, foreign DNA fragments, known as protospacers, are incorporated into the CRISPR array as spacers. In the expression stage, the CRISPR array is transcribed into a precursor CRISPR RNA (pre-crRNA), which is then processed into mature crRNAs. Finally, in the interference stage, the mature crRNA directs Cas proteins, which function as endonucleases, to recognize the complementary sequences of the invading genetic element and create DSBs, thereby neutralizing the threat (29, 30).\n\nThe CRISPR\u2013Cas system is categorized into two major classes, Class 1 and Class 2, encompassing six types and 33 subtypes. Class 1 is characterized by multiprotein effector complexes and is prevalent in both bacteria and archaea, whereas Class 2, found exclusively in bacteria, relies on a single multifunctional effector protein (31). Each class is further divided into three types: Class 1 includes Types I, III, and IV, whereas Class 2 comprises Types II, V, and VI.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR\u2013Cas system", "text": "The Class 1 CRISPR\u2013Cas system, particularly Type I, employs multiple Cas proteins that interact with crRNA to form the CRISPR-associated complex for antiviral defense (Cascade). Cascade binds to foreign DNA, allowing crRNA to hybridize with the complementary strand, forming an R-loop, which is then cleaved by Cas3, an enzyme with helicase and nuclease activity (32). The Type I system is the most prevalent among prokaryotes and has further subtypes, such as Type I-A to I-F, each exhibiting distinct Cas protein compositions and mechanisms of interference (33). Similarly, Type III systems, which use Cas10 as the signature protein, exhibit RNA and DNA cleavage capabilities, often functioning through a co-transcriptional targeting mechanism that recognizes RNA and degrades DNA in a coordinated manner (34). Type IV systems, though less characterized, are generally found in plasmids rather than chromosomal CRISPR loci and exhibit a minimal set of Cas proteins, including Cas12a [formerly CRISPR-associated protein 1 (Csf1)]. Their exact function remains under investigation, though they appear to contribute to horizontal gene transfer regulation (35).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR\u2013Cas system", "text": "In contrast, Class 2 CRISPR\u2013Cas systems rely on single-effector proteins, such as Cas9 in Type II, Cas12 in Type V, and Cas13 in Type VI (36). Cas9, the most well-studied and widely utilized in genome editing, is guided by a single-guide RNA (sgRNA) that directs site-specific DNA cleavage at the protospacer (37). Type II CRISPR systems require a trans-activating CRISPR RNA (tracrRNA) to assist in crRNA maturation and target cleavage (38). Type V systems, which include Cas12 proteins, differ from Cas9 by generating staggered double-strand DNA breaks rather than blunt-ended cuts. Upon specific target DNA recognition, Cas12 undergoes conformational activation that triggers collateral, nonspecific degradation of surrounding single-stranded DNA (ssDNA) molecules. This property forms the mechanistic basis of CRISPR-based diagnostic platforms, where collateral ssDNA cleavage amplifies signal following accurate target detection (39). Type VI systems, featuring Cas13 effectors, uniquely target RNA instead of DNA. After binding to a complementary RNA substrate, Cas13 becomes catalytically activated and induces indiscriminate collateral cleavage of nearby RNA molecules, enabling programmable transcriptome editing as well as highly sensitive RNA diagnostics (40). The diversity of CRISPR\u2013Cas systems highlights their evolutionary innovation and functional specialization in prokaryotic adaptive immunity.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR\u2013Cas system", "text": "Despite their prevalence in nature and mechanistic diversity, Class 1 CRISPR\u2013Cas systems remain significantly underrepresented in genome-editing applications. Their effector function relies on large, multi-subunit complexes, which introduces major challenges for clinical translation, particularly regarding delivery into mammalian cells, vector packaging constraints, and protein complex assembly fidelity in heterologous systems (32, 41). Additionally, incomplete characterization of many Class 1 subtypes limits precise manipulation and safety assessments, further constraining their therapeutic development compared with Class 2 systems, where single-effector nucleases such as Cas9 and Cas12 are inherently more compatible with current delivery modalities and regulatory expectations (42).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR\u2013Cas system", "text": "The CRISPR\u2013Cas9 system has revolutionized genome editing due to its high precision and programmability. The breakthrough realization that CRISPR could be repurposed as a genome-editing tool came from the work of Doudna and Charpentier, who demonstrated that the Cas9 protein could be programmed with a synthetic sgRNA to target specific DNA sequences. Their findings laid the foundation for programmable genome modifications, allowing precise genetic alterations in diverse organisms (20, 43). This system consists of two primary components: the Cas9 endonuclease and a guide RNA (gRNA) (44). The Cas9 protein, originally derived from Streptococcus pyogenes (spCas9), is a multidomain endonuclease responsible for generating targeted DSBs in DNA, earning it the designation of a \u201cgenetic scissor.\u201d Cas9 comprises two distinct regions: the recognition lobe (REC) and the nuclease lobe (NUC). The NUC region contains the His-Asn-His (HNH) and RuvC nuclease domains, which are responsible for cleaving the complementary and non-complementary DNA strands, respectively, as well as the protospacer adjacent motif (PAM)-interacting domain, which is essential for target DNA recognition (45).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR\u2013Cas system", "text": "The CRISPR\u2013Cas9 genome editing process follows three key steps: target recognition, DNA cleavage, and repair. The gRNA comprises two essential components: a tracrRNA and a crRNA that guides the Cas9 enzyme to a specific target site within the genome (46). Target recognition occurs when the5\u2032 end of the crRNA hybridizes with a complementary sequence in the genome, while the adjacent PAM sequence facilitates Cas9 binding. Upon successful pairing, Cas9 induces a DSB at the target locus, activating the cellular DNA repair machinery (47). Following DSB formation, cells repair the damage through either non-homologous end joining (NHEJ) or HDR pathways. NHEJ is an error-prone mechanism that directly ligates DNA ends, often introducing insertions or deletions (indels) that can disrupt gene function (48). In contrast, HDR enables precise genome modifications by utilizing a homologous DNA template, facilitating accurate base substitutions or insertions (Figure 1) (49).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR\u2013Cas system", "text": "The schematic diagram represents the mechanism of CRISPR\u2013Cas genome editing. An sgRNA is necessary for CRISPR\u2013Cas genome editing because it guides the Cas endonuclease to a particular area of the genomic DNA, causing a double-strand break. Transgenic DNA can be produced by supplying donor DNA in trans, but without donor DNA, the host cell will repair the double-strand break, leading to an insertion or deletion that may disturb a gene's open reading frame. The figure was designed with BioRender (http://Biorender.com/). CRISPR, clustered regularly interspaced short palindromic repeats; Cas, CRISPR-associated protein; crRNA, CRISPR RNA; gRNA, guide RNA; DNA, deoxyribonucleic acid; PAM, protospacer adjacent motif; NHEJ, non-homologous end joining; HDR, homology-directed repair.\n\nSubstantial technological refinements have enhanced the specificity and versatility of Cas9. High-fidelity variants such as enhanced specificity Cas9 (eSpCas9) and Streptococcus pyogenes Cas9\u2013High-Fidelity 1 (SpCas9-HF1) have been engineered to minimize off-target cleavage, improving editing accuracy (50). Cas9 nickases (nCas9), which generate single-strand breaks instead of DSBs, further reduce genomic instability while enabling precise editing (51).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR\u2013Cas system", "text": "Base editors, such as cytosine base editors (CBEs) and adenine base editors (ABEs), enable targeted single-nucleotide modifications, reducing the risk of undesired genomic alterations (52, 53). Base editing is a kind of genome editing that can change a base pair at the target genomic locus directly and irreversibly without the need for donor templates, DSBs, or HDR procedures 41. Two parts make into DNA base editors: a single-stranded DNA modifying enzyme to target nucleotide alteration and a Cas enzyme for DNA binding. There are now two types of DNA base editors available: CBEs and ABEs. Therefore, four transition mutations (cytosine convent into thymine C \u2192 T, T \u2192 C, guanine into adenine G \u2192 A, and A \u2192 G) can be installed using base editing (23, 54). Cytosine is changed into uracil by the first base editor (CBE1) deaminating the exocyclic amine. Uracil's detection by cell replication machinery as a thymine, which resulting transition the C\u2013G to T\u2013A (55). Cas proteins are unrelated to gRNA's activity and rely on it for DNA binding. Therefore, by altering their amino acids or changing their structural makeup, one or both of its domains can be rendered inactive with no loss in their ability to bind. By applying this theory, researchers have created modified Cas endonuclease, resulting enable to disrupt the activation of Ruvc and HNH domains to create as nCas or a dCas. All Cas proteins, including dCas and nCas, are able to merge with various other molecules, such as enzymes, without affecting their ability to bind and cut independently (56).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR\u2013Cas system", "text": "Prime editing, which employs an engineered reverse transcriptase fused to Cas9 nickase, facilitates the precise insertion or deletion of genetic sequences without requiring donor templates, thereby increasing editing efficiency and accuracy (54). The prime editing method was developed by Anzalone in 2019, which decreases the number of inadvertent errors (54, 57). Researchers demonstrated that prime editing effectively installs all 12 potential base-to-base transformations without requiring donor templates of DNA or inducing DSBs in the target sequence, because this method doesn't depend on DSB 46. It consists of a prime editing guide RNA (pegRNA) and Cas9 (H840A) nickase; the pegRNA is longer than gRNA. The engineered reverse transcriptase (RT) results from fusing pegRNA and Cas9 (H840A) nickase. RT serves as the enzyme that uses the pegRNA as a template to synthesize the edited DNA sequence during prime editing. The prime editing has been designed and screened in human cells. There are prime editing systems that process by incorporating Cas9 (H840A) nickase and RT enzyme, and mutations are introduced when transversions occur (58\u201360). pegRNA includes tracrRNA and spacer, in addition to gRNA, gRNA connected to a specific gene of RNA sequence, which consists of a primer binding site (PBS), the PBS site contains a complementary sequence to the target region of the edited sequence of DNA. Once the target DNA is bound by the CRISPR\u2013Cas9 system, nCas9 cuts the opposing DNA strand to create a single-stranded break (SSB) in the DNA. Using Cas9 fused to RT, the SSB of the DNA sequence links to the PBS site in the sgRNA, acting as a primer to create a new DNA fragment. The targeted DNA will subsequently be replaced by this freshly created DNA (56).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR\u2013Cas system", "text": "RNA-targeting CRISPR systems (Cas13 and newer RNA effectors) allow for programmable changes to transcripts without changing DNA permanently. This is useful for reversible changes, antiviral strategies, and diagnostics. However, RNA editors have their own problems, such as collateral cleavage in some cases, activity that depends on the sequence, and the need for repeated dosing for therapeutic effects. Newer engineered variants, on the other hand, show better specificity and less collateral activity (61). Table 1 shows some strengths and challenges of contemporary CRISPR editing modalities.\n\nThe strengths and challenges of contemporary CRISPR editing modalities.\n\nCRISPR, clustered regularly interspaced short palindromic repeats; Cas, CRISPR-associated protein; DSB, double-strand break; HDR, homology-directed repair; PAM, protospacer adjacent motif; CBE, cytosine base editor; ABE, adenine base editor; RNA, ribonucleic acid.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR\u2013Cas system", "text": "Beyond Cas9, other CRISPR-associated proteins have significantly expanded the potential application. Cas12 (Type V) effectors differ from Cas9 by generating staggered double-strand breaks and exhibiting collateral ssDNA degradation, making them useful for genome editing and diagnostic applications such as DNA endonuclease-targeted CRISPR trans reporter (DETECTR) (62, 63). Cas14/Cas12f and engineered miniaturized variants such as miniaturized CRISPR-associated nuclease (CasMINI) provide compact alternatives suitable for delivery via adeno-associated viruses (AAVs), facilitating in vivo applications where vector size is limiting (64, 65). Cas13 (Type VI) targets RNA rather than DNA, and has been adapted into programmable RNA-editing systems such as RNA editing for programmable A to I replacement (REPAIR) and RNA editing for specific C to U exchange (RESCUE), which catalyze adenosine-to-inosine or cytosine-to-uracil conversions in transcripts, allowing transient, reversible editing without altering the genome (61, 66).\n\nEmerging CRISPR-associated transposase (CAST) systems integrate RNA-guided targeting with transposon machinery, enabling programmable, DSB-free DNA insertions that bypass host repair pathways, thus increasing precision and efficiency for certain genome engineering applications (67).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR\u2013Cas system", "text": "Finally, CRISPR has been harnessed for gene regulation without altering DNA sequences. CRISPRa employ dCas9 fused to transcriptional activators to upregulate gene expression, while CRISPRi use dCas9-repressor fusions to silence gene transcription, facilitating functional genomics studies and potential therapeutic gene modulation (68, 69). Collectively, these platforms demonstrate the increasing diversification of CRISPR-based technologies, spanning DNA and RNA editing, precision base modifications, delivery system optimization, and transcriptional regulation, thereby expanding their translational potential in next-generation therapeutics.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "An adaptive immune system of bacteria and archaea", "text": "The CRISPR\u2013Cas system serves as an adaptive immune defense mechanism in bacteria and archaea, protecting against invading genetic elements such as bacteriophages and plasmids (28). This system operates through three main stages: adaptation, expression, and interference. During adaptation, foreign DNA fragments, known as protospacers, are incorporated into the CRISPR array as spacers. In the expression stage, the CRISPR array is transcribed into a precursor CRISPR RNA (pre-crRNA), which is then processed into mature crRNAs. Finally, in the interference stage, the mature crRNA directs Cas proteins, which function as endonucleases, to recognize the complementary sequences of the invading genetic element and create DSBs, thereby neutralizing the threat (29, 30).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas systems and classes", "text": "The CRISPR\u2013Cas system is categorized into two major classes, Class 1 and Class 2, encompassing six types and 33 subtypes. Class 1 is characterized by multiprotein effector complexes and is prevalent in both bacteria and archaea, whereas Class 2, found exclusively in bacteria, relies on a single multifunctional effector protein (31). Each class is further divided into three types: Class 1 includes Types I, III, and IV, whereas Class 2 comprises Types II, V, and VI.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas systems and classes", "text": "The Class 1 CRISPR\u2013Cas system, particularly Type I, employs multiple Cas proteins that interact with crRNA to form the CRISPR-associated complex for antiviral defense (Cascade). Cascade binds to foreign DNA, allowing crRNA to hybridize with the complementary strand, forming an R-loop, which is then cleaved by Cas3, an enzyme with helicase and nuclease activity (32). The Type I system is the most prevalent among prokaryotes and has further subtypes, such as Type I-A to I-F, each exhibiting distinct Cas protein compositions and mechanisms of interference (33). Similarly, Type III systems, which use Cas10 as the signature protein, exhibit RNA and DNA cleavage capabilities, often functioning through a co-transcriptional targeting mechanism that recognizes RNA and degrades DNA in a coordinated manner (34). Type IV systems, though less characterized, are generally found in plasmids rather than chromosomal CRISPR loci and exhibit a minimal set of Cas proteins, including Cas12a [formerly CRISPR-associated protein 1 (Csf1)]. Their exact function remains under investigation, though they appear to contribute to horizontal gene transfer regulation (35).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas systems and classes", "text": "In contrast, Class 2 CRISPR\u2013Cas systems rely on single-effector proteins, such as Cas9 in Type II, Cas12 in Type V, and Cas13 in Type VI (36). Cas9, the most well-studied and widely utilized in genome editing, is guided by a single-guide RNA (sgRNA) that directs site-specific DNA cleavage at the protospacer (37). Type II CRISPR systems require a trans-activating CRISPR RNA (tracrRNA) to assist in crRNA maturation and target cleavage (38). Type V systems, which include Cas12 proteins, differ from Cas9 by generating staggered double-strand DNA breaks rather than blunt-ended cuts. Upon specific target DNA recognition, Cas12 undergoes conformational activation that triggers collateral, nonspecific degradation of surrounding single-stranded DNA (ssDNA) molecules. This property forms the mechanistic basis of CRISPR-based diagnostic platforms, where collateral ssDNA cleavage amplifies signal following accurate target detection (39). Type VI systems, featuring Cas13 effectors, uniquely target RNA instead of DNA. After binding to a complementary RNA substrate, Cas13 becomes catalytically activated and induces indiscriminate collateral cleavage of nearby RNA molecules, enabling programmable transcriptome editing as well as highly sensitive RNA diagnostics (40). The diversity of CRISPR\u2013Cas systems highlights their evolutionary innovation and functional specialization in prokaryotic adaptive immunity.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas systems and classes", "text": "Despite their prevalence in nature and mechanistic diversity, Class 1 CRISPR\u2013Cas systems remain significantly underrepresented in genome-editing applications. Their effector function relies on large, multi-subunit complexes, which introduces major challenges for clinical translation, particularly regarding delivery into mammalian cells, vector packaging constraints, and protein complex assembly fidelity in heterologous systems (32, 41). Additionally, incomplete characterization of many Class 1 subtypes limits precise manipulation and safety assessments, further constraining their therapeutic development compared with Class 2 systems, where single-effector nucleases such as Cas9 and Cas12 are inherently more compatible with current delivery modalities and regulatory expectations (42).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Genome editing tools: CRISPR-Cas9 and beyond", "text": "The CRISPR\u2013Cas9 system has revolutionized genome editing due to its high precision and programmability. The breakthrough realization that CRISPR could be repurposed as a genome-editing tool came from the work of Doudna and Charpentier, who demonstrated that the Cas9 protein could be programmed with a synthetic sgRNA to target specific DNA sequences. Their findings laid the foundation for programmable genome modifications, allowing precise genetic alterations in diverse organisms (20, 43). This system consists of two primary components: the Cas9 endonuclease and a guide RNA (gRNA) (44). The Cas9 protein, originally derived from Streptococcus pyogenes (spCas9), is a multidomain endonuclease responsible for generating targeted DSBs in DNA, earning it the designation of a \u201cgenetic scissor.\u201d Cas9 comprises two distinct regions: the recognition lobe (REC) and the nuclease lobe (NUC). The NUC region contains the His-Asn-His (HNH) and RuvC nuclease domains, which are responsible for cleaving the complementary and non-complementary DNA strands, respectively, as well as the protospacer adjacent motif (PAM)-interacting domain, which is essential for target DNA recognition (45).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Genome editing tools: CRISPR-Cas9 and beyond", "text": "The CRISPR\u2013Cas9 genome editing process follows three key steps: target recognition, DNA cleavage, and repair. The gRNA comprises two essential components: a tracrRNA and a crRNA that guides the Cas9 enzyme to a specific target site within the genome (46). Target recognition occurs when the5\u2032 end of the crRNA hybridizes with a complementary sequence in the genome, while the adjacent PAM sequence facilitates Cas9 binding. Upon successful pairing, Cas9 induces a DSB at the target locus, activating the cellular DNA repair machinery (47). Following DSB formation, cells repair the damage through either non-homologous end joining (NHEJ) or HDR pathways. NHEJ is an error-prone mechanism that directly ligates DNA ends, often introducing insertions or deletions (indels) that can disrupt gene function (48). In contrast, HDR enables precise genome modifications by utilizing a homologous DNA template, facilitating accurate base substitutions or insertions (Figure 1) (49).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Genome editing tools: CRISPR-Cas9 and beyond", "text": "The schematic diagram represents the mechanism of CRISPR\u2013Cas genome editing. An sgRNA is necessary for CRISPR\u2013Cas genome editing because it guides the Cas endonuclease to a particular area of the genomic DNA, causing a double-strand break. Transgenic DNA can be produced by supplying donor DNA in trans, but without donor DNA, the host cell will repair the double-strand break, leading to an insertion or deletion that may disturb a gene's open reading frame. The figure was designed with BioRender (http://Biorender.com/). CRISPR, clustered regularly interspaced short palindromic repeats; Cas, CRISPR-associated protein; crRNA, CRISPR RNA; gRNA, guide RNA; DNA, deoxyribonucleic acid; PAM, protospacer adjacent motif; NHEJ, non-homologous end joining; HDR, homology-directed repair.\n\nSubstantial technological refinements have enhanced the specificity and versatility of Cas9. High-fidelity variants such as enhanced specificity Cas9 (eSpCas9) and Streptococcus pyogenes Cas9\u2013High-Fidelity 1 (SpCas9-HF1) have been engineered to minimize off-target cleavage, improving editing accuracy (50). Cas9 nickases (nCas9), which generate single-strand breaks instead of DSBs, further reduce genomic instability while enabling precise editing (51).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Genome editing tools: CRISPR-Cas9 and beyond", "text": "Base editors, such as cytosine base editors (CBEs) and adenine base editors (ABEs), enable targeted single-nucleotide modifications, reducing the risk of undesired genomic alterations (52, 53). Base editing is a kind of genome editing that can change a base pair at the target genomic locus directly and irreversibly without the need for donor templates, DSBs, or HDR procedures 41. Two parts make into DNA base editors: a single-stranded DNA modifying enzyme to target nucleotide alteration and a Cas enzyme for DNA binding. There are now two types of DNA base editors available: CBEs and ABEs. Therefore, four transition mutations (cytosine convent into thymine C \u2192 T, T \u2192 C, guanine into adenine G \u2192 A, and A \u2192 G) can be installed using base editing (23, 54). Cytosine is changed into uracil by the first base editor (CBE1) deaminating the exocyclic amine. Uracil's detection by cell replication machinery as a thymine, which resulting transition the C\u2013G to T\u2013A (55). Cas proteins are unrelated to gRNA's activity and rely on it for DNA binding. Therefore, by altering their amino acids or changing their structural makeup, one or both of its domains can be rendered inactive with no loss in their ability to bind. By applying this theory, researchers have created modified Cas endonuclease, resulting enable to disrupt the activation of Ruvc and HNH domains to create as nCas or a dCas. All Cas proteins, including dCas and nCas, are able to merge with various other molecules, such as enzymes, without affecting their ability to bind and cut independently (56).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Genome editing tools: CRISPR-Cas9 and beyond", "text": "Prime editing, which employs an engineered reverse transcriptase fused to Cas9 nickase, facilitates the precise insertion or deletion of genetic sequences without requiring donor templates, thereby increasing editing efficiency and accuracy (54). The prime editing method was developed by Anzalone in 2019, which decreases the number of inadvertent errors (54, 57). Researchers demonstrated that prime editing effectively installs all 12 potential base-to-base transformations without requiring donor templates of DNA or inducing DSBs in the target sequence, because this method doesn't depend on DSB 46. It consists of a prime editing guide RNA (pegRNA) and Cas9 (H840A) nickase; the pegRNA is longer than gRNA. The engineered reverse transcriptase (RT) results from fusing pegRNA and Cas9 (H840A) nickase. RT serves as the enzyme that uses the pegRNA as a template to synthesize the edited DNA sequence during prime editing. The prime editing has been designed and screened in human cells. There are prime editing systems that process by incorporating Cas9 (H840A) nickase and RT enzyme, and mutations are introduced when transversions occur (58\u201360). pegRNA includes tracrRNA and spacer, in addition to gRNA, gRNA connected to a specific gene of RNA sequence, which consists of a primer binding site (PBS), the PBS site contains a complementary sequence to the target region of the edited sequence of DNA. Once the target DNA is bound by the CRISPR\u2013Cas9 system, nCas9 cuts the opposing DNA strand to create a single-stranded break (SSB) in the DNA. Using Cas9 fused to RT, the SSB of the DNA sequence links to the PBS site in the sgRNA, acting as a primer to create a new DNA fragment. The targeted DNA will subsequently be replaced by this freshly created DNA (56).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Genome editing tools: CRISPR-Cas9 and beyond", "text": "RNA-targeting CRISPR systems (Cas13 and newer RNA effectors) allow for programmable changes to transcripts without changing DNA permanently. This is useful for reversible changes, antiviral strategies, and diagnostics. However, RNA editors have their own problems, such as collateral cleavage in some cases, activity that depends on the sequence, and the need for repeated dosing for therapeutic effects. Newer engineered variants, on the other hand, show better specificity and less collateral activity (61). Table 1 shows some strengths and challenges of contemporary CRISPR editing modalities.\n\nThe strengths and challenges of contemporary CRISPR editing modalities.\n\nCRISPR, clustered regularly interspaced short palindromic repeats; Cas, CRISPR-associated protein; DSB, double-strand break; HDR, homology-directed repair; PAM, protospacer adjacent motif; CBE, cytosine base editor; ABE, adenine base editor; RNA, ribonucleic acid.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Genome editing tools: CRISPR-Cas9 and beyond", "text": "Beyond Cas9, other CRISPR-associated proteins have significantly expanded the potential application. Cas12 (Type V) effectors differ from Cas9 by generating staggered double-strand breaks and exhibiting collateral ssDNA degradation, making them useful for genome editing and diagnostic applications such as DNA endonuclease-targeted CRISPR trans reporter (DETECTR) (62, 63). Cas14/Cas12f and engineered miniaturized variants such as miniaturized CRISPR-associated nuclease (CasMINI) provide compact alternatives suitable for delivery via adeno-associated viruses (AAVs), facilitating in vivo applications where vector size is limiting (64, 65). Cas13 (Type VI) targets RNA rather than DNA, and has been adapted into programmable RNA-editing systems such as RNA editing for programmable A to I replacement (REPAIR) and RNA editing for specific C to U exchange (RESCUE), which catalyze adenosine-to-inosine or cytosine-to-uracil conversions in transcripts, allowing transient, reversible editing without altering the genome (61, 66).\n\nEmerging CRISPR-associated transposase (CAST) systems integrate RNA-guided targeting with transposon machinery, enabling programmable, DSB-free DNA insertions that bypass host repair pathways, thus increasing precision and efficiency for certain genome engineering applications (67).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Genome editing tools: CRISPR-Cas9 and beyond", "text": "Finally, CRISPR has been harnessed for gene regulation without altering DNA sequences. CRISPRa employ dCas9 fused to transcriptional activators to upregulate gene expression, while CRISPRi use dCas9-repressor fusions to silence gene transcription, facilitating functional genomics studies and potential therapeutic gene modulation (68, 69). Collectively, these platforms demonstrate the increasing diversification of CRISPR-based technologies, spanning DNA and RNA editing, precision base modifications, delivery system optimization, and transcriptional regulation, thereby expanding their translational potential in next-generation therapeutics.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "AD is a progressive ND that impairs cognitive functions, including memory, reasoning, and daily living tasks (70, 71). The disease is a major cause of dementia among the elderly. It is characterized by the presence of amyloid-beta (A\u03b2) plaques and neurofibrillary tangles in the brain, which impair neurons and contribute to the degeneration of brain tissue (72, 73). The disease initiates with slight memory challenges that progress to severe cognitive and physical impairments, ultimately resulting in complete dependency and premature death (74). The disease disseminates throughout diverse communities at a prominent global health scale due to its elevated prevalence rates. There are 5.5 million individuals in the United States with AD. In 2016, the global number of individuals with dementia reached 43.8 million, reflecting a 116% increase since 1990, according to recent studies (75).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "Development of AD largely driven by hereditary elements follows several pathophysiological paths. Many times, mutations in the APP, PSEN1, and PSEN2 genes link fAD to one another (76). The APP gene codes the amyloid precursor protein on chromosome 21, which \u03b2-secretase and \u03b3-secretase break to generate A\u03b2 peptides. Applying mutations in APP produces overly high synthesis of A\u03b2, particularly the A\u03b242 variant, which readily accumulates to generate amyloid plaques, a key hallmark of AD (77). Importantly, members of the \u03b3-secretase complex, PSEN1 and PSEN2, respectively encode PSEN1 and PSEN2 proteins (18, 78). Early-onset fAD is caused by mutations in these genes altering \u03b3-secretase activity, hence increasing the A\u03b242/A\u03b240 ratio and promoting A\u03b2 aggregation (79).\n\nApart from amyloid pathology, tau protein dysfunction also speeds the development of disease. Although microtubule-associated protein tau (MAPT) gene mutations are rare in AD, tau proteins in affected individuals hyperphosphorylated, therefore losing their ability to stabilize microtubules (80). Along with impairments in neuronal activity, this instability results in paired helical filaments and neurofibrillary tangles (81\u201383).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "Moreover, exacerbating AD development is neuroinflammation. Released by active microglia and astrocytes, reactive oxygen species and pro-inflammatory cytokines damage neurons. Variations in the triggering receptor expressed on myeloid cells 2 (TREM2) gene, which affects microglial activity, have also been associated to increased AD risk, therefore underlining the critical role immune responses play in the evolution of disorders (84\u201387).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "CRISPR\u2013Cas9 represents a formidable gene-editing technique for AD, facilitating the precise modification of pathogenic genes, such as rectifying or eliminating APP, \u03b2-site amyloid precursor protein cleaving enzyme 1 (BACE1), and PSEN1/2 mutations to diminish the production and aggregation of harmful A\u03b2 (88); the conversion of Apo-lipoprotein E \u03b54 allele (APOE4) to the protective APOE3 or APOE2 isoforms, thereby reducing A\u03b2 accumulation, tau pathology, and neuronal susceptibility; the modulation of neuroinflammation through the targeting of CD33, glia maturation factor (GMF), or other immune mediators to enhance microglial clearance of amyloid plaques and inhibit detrimental cytokine signaling (89); and the amplification of neuroprotective pathways by promoting non-amyloidogenic APP processing (\u03b1-cleavage) or bolstering microglial functions that maintain neuronal integrity (88). These attributes render CRISPR a versatile method for rectifying genetic disease etiologies, reducing the accumulation of deleterious proteins, mitigating inflammation, and enhancing cerebral resilience. This approach enables researchers to acquire accurate AD models in vivo and in vitro, enhancing the comprehension of disease mechanisms, and aiding in the identification of novel therapy alternatives (Figure 2).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "Diagrammatic illustration of the CRISPR\u2013Cas9 system in the study and treatment of Alzheimer's disease (AD) and shows how CRISPR\u2013Cas9 can be used to target genetic alterations and pathways linked to AD. The figure was designed with BioRender (http://Biorender.com/). CRISPR, clustered regularly interspaced short palindromic repeats; Cas9, CRISPR-associated protein 9; sgRNA, single-guide RNA; PAM, protospacer adjacent motif; HDR, homology-directed repair; MMEJ, microhomology-mediated end joining; DNA, deoxyribonucleic acid; A\u03b2 (amyloid beta), amyloid-\u03b2 peptide.\n\nCell models are extensively utilized in application on many neurodegenerative disorders, including AD, due to the absence of ethical dilemmas, the brevity of the experimental duration, and the cheap associated costs. Over the past few decades, numerous in vitro cellular models of AD have been developed.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "Researchers have effectively employed CRISPR\u2013Cas9 to eliminate the APP and BACE1 genes in neural cell types. This resulted in a significant reduction in the synthesis of A\u03b2, a primary therapeutic target for AD. The previous study demonstrated that the ablation of APP in brain cells via CRISPR\u2013Cas9 diminished both A\u03b2 expression and amyloid pathology. This indicates its potential as a gene therapy alternative for individuals with APP mutations (89, 90). Sun et al. (91) employed CRISPR\u2013Cas9 to modify the C-terminus of endogenous APP, markedly diminishing \u03b2-secretase cleavage and redirecting processing toward non-amyloidogenic \u03b1-secretase, thus decreasing A\u03b2 generation.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "Previous studies were conducted to reduce the A\u03b2 utilizing CRISPR technique the knockout of PSEN1 and PSEN2 genes (92\u201395). In addition, researchers effectively introduced this mutation into brain cells utilizing Cas9 nickase\u2013deaminase, significantly reducing A\u03b2 generation (96). Simultaneously, Song et al. (97) discovered that the suppression of KIBRA (Kidney and Brain expressed protein) in HT22 cells inhibited proliferation and induced death, whereas treatment with A\u03b21\u201342 oligomers was administered (96, 98). Prime-Base editing was used to treat AD via utilizing CRISPR to correct the gene defects, and the results shows reduction in A\u03b2 peptide accumulation. Several studies have been performed in vitro utilizing CRISPR\u2013Cas9 technology to modify and editing the APOE gene, as this method has the potential to convert high-risk APOE4 alleles into protective alleles such as APOE3 or APOE2 to reduce the AD risk (99).\n\nIn addition to constructing AD cell models, CRISPR\u2013Cas9 technology can also be utilized to develop AD animal models to improve and correct the cognitive function in AD. Serneels, T'Syen (100) Developed a novel model utilizing the CRISPR\u2013Cas9 technique to construct a humanized A sequence in the APP gene of both species (100). Another study utilized animal models in vivo to improve cognitive function and correct APP and BACE1 genes in AD. It demonstrated that the CRISPR system can suppress A\u03b2, having a significant effect on both pathology and cognitive function (101\u2013103).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "Furthermore, CRISPR-based gene editing techniques, particularly base editing and prime editing, exhibit the capacity to rectify harmful AD mutations without inducing double-strand DNA breaks. The genetic modification executed by base editors led to significant enhancements in tau pathology and cognitive recovery in the examined mice (58, 104, 105). APOE possesses various variations, including APOE2, APOE3, APOE3r, and APOE4. Komor et al. (59), the transfer of APOE3 to APOE4 in mouse astrocytes was achieved by changing Cincodon158 to T using the CRISPR\u2013Cas9 system, indicating that point mutations were corrected by the CRISPR\u2013Cas9 approach (59).\n\nPrime-Base editing serves as a genome-editing technique that offers extensive genetic modification possibilities via mutation repair without inducing DSBs to address AD. In vivo research focused on AD-related applications needs greater exploration (104\u2013106), such as a study by Park et al. (103) via utilizing Cas9 nanocomplexes with sgRNAs targeting the tyrosine hydroxylase and BACE1 genes were administered into mouse primary neural cells to evaluate the efficacy of Cas9 nanocomplexes. Furthermore, Takalo et al. (107) utilizing the CRISPR\u2013Cas9 gene editing technique and evaluating the preventive effects of the Plc2-P522R variant.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "Patient-derived induced pluripotent stem cells (iPSCs) have garnered growing interest due to their distinctive physiological characteristics. AD mutant genes, such as PSEN1, PSEN2, and APP, in iPSC-derived neurons provide exceptional platforms for elucidating disease mechanisms and formulating novel therapeutic strategies. Studies demonstrated that gene editing using CRISPR\u2013Cas9 can mitigate the impairments linked to AD by addressing synaptic dysfunction and altered signaling pathways (108\u2013111). Researchers have created gene-edited induced pluripotent stem cell models to investigate genetic mutations associated with AD, as well as knockout genes implicated in Alzheimer's pathogenesis, including ATP-binding cassette subfamily A member 7 (ABCA7) (112\u2013114). The integration of animal models and CRISPR-edited iPSCs enhances scientists' understanding of AD risk factors and therapeutic options (115). CRISPR-engineered iPSC and organoid models have elucidated critical mechanisms in AD, including lipid metabolism (APOE), endosomal dysfunction (APP/PSEN1), electrophysiological abnormalities (PSEN2), and neuroimmune interactions (APOE Christchurch). These human-relevant systems are highly effective for identifying disease pathways and evaluating treatment alternatives (116\u2013119). Stem cell models integrated with gene editing technologies facilitate the development of patient-specific therapies, paving the way for clinical applications.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "Stem cell-derived models, in conjunction with gene editing, exhibit potential to enhance the development of tailored AD therapies. Table 2 Illiterates applications of CRISPR\u2013Cas in AD. Utilizing CRISPR, scientists acquire enhanced insights into AD progression and develop successful therapeutic approaches.\n\nAD, Alzheimer's disease; CRISPR, clustered regularly interspaced short palindromic repeats; Cas9, CRISPR-associated protein 9; A\u03b2, amyloid-\u03b2 peptide; iPSC, induced pluripotent stem cell; KO, knockout; HDR, homology-directed repair; RNP, ribonucleoprotein complex; AAV, adeno-associated virus; BE, base editing/base editor; PE, prime editing; pegRNA, prime-editing guide RNA; APP, amyloid precursor protein; BACE1, \u03b2-site amyloid precursor protein cleaving enzyme 1; APOE, apolipoprotein E; TREM2, triggering receptor expressed on myeloid cells 2; MAPT, microtubule-associated protein tau gene; PSEN1/PSEN2, presenilin 1 and 2; SORL1, sortilin-related receptor 1; 2D, two-dimensional cell culture; 3D, three-dimensional culture.\n\nJia et al. (120) created a CRISPR-powered aptasensor that detects A\u03b2 biomarkers via fluorescence detection via CRISPR-Cas12a. There is currently little study on the use of CRISPR-Cas13 and CRISPR-Cas3 in AD diagnosis. Although its potential in diagnostic domains has not been thoroughly investigated, Ca3\u2032s primary DNA deletion function remains its most prominent feature.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "PD, the second prevalent neurodegenerative ailment, is marked by the degradation of dopaminergic neurons, which results in particular motor symptoms such as rigidity, bradykinesia, and tremor (121). After the loss of neuronal, the Lewy bodies appeared with cytoplasmic inclusions, which are mostly composed of aggregates of \u03b1-synuclein protein misfolded and perhaps diffusion in a prion-like shape among interconnected areas synoptically (122). PD can be classified into two forms: sporadic and familial. The familial form accounts for only 10%\u221215% of all PD cases. The hereditary versions are attributed to mutations in multiple genes, including SNCA, parkin (PRKN), PTEN-induced kinase 1 (PINK1), DJ-1/Parkinson's disease protein 7 (PARK7), and LRRK2. SNCA and LRRK2 mutations cause autosomal-dominant types of PD, while PRKN, PINK1, and DJ-1 mutations cause autosomal recessive variants (123). Although the origins of sporadic forms are currently unknown, certain genetic and environmental risk factors have been discovered. For instance, genetic polymorphisms in the LRRK2 and SNCA genes elevate the likelihood of developing sporadic PD (124).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "Between 1992 and 2021, global PD cases escalated from 450,000 to 1.34 million, with crude incidence rates increasing from 8.19 to 16.92 per 100,000 and age-standardized incidence rates rising from 11.54 to 15.63 per 100,000. Incidence rates increased significantly in persons aged 60 and beyond, reaching a zenith in those aged 85 and above. Forecasts indicate that by 2030, worldwide PD cases will total 1.93 million, with an age-standardized incidence rate of 27 per 100,000 (125).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "PD is a ND mostly attributed to genetic abnormalities and environmental stresses (126). Genetic alterations in critical genes such as SNCA, LRRK2, PARK2, PINK1, and Glucocerebrosidase 1 (GBA1) are central to the disorder. Mutations in SNCA result in the improper folding of SNCA protein, leading to aggregation into Lewy bodies, which are indicative of PD in the brain (127). Mutations in LRRK2 that enhance gene function result in increased kinase activity, which is detrimental to neurons. Mutations in PARK2 and PINK1 disrupt mitochondrial quality control, leading to mitochondrial malfunction, poor mitophagy, oxidative stress, and ultimately, the demise of dopaminergic neurons (128). Neurons exhibit increased toxicity when mutations in the LRRK2 gene lead to the enzyme acquiring excessive kinase activity. The PARK2 gene regulates parkin activity, which acts as an E3 ubiquitin ligase; however, its dysfunction results in the buildup of protein damage (128). Neuronal damage exacerbates when GBA1 mutations induce mitochondrial malfunction and elevate oxidative stress inside cells (129, 130). Genetic abnormalities alter protein degradation pathways, impair mitochondrial function, and interfere with oxidative stress responses, ultimately leading to PD symptoms through the degeneration of dopaminergic neurons in the substantia nigra (131). Impaired mitochondria result in heightened oxidative stress and diminished ATP synthesis, causing neuronal damage. The development of PD occurs due to mutations in PINK1 and PRKN that impair mitochondrial quality control mechanisms (131). PD neuroinflammation occurs when activated microglia secrete tumor necrosis factor-alpha (TNF-\u03b1), interleukin (IL)-1\u03b2, and reactive oxygen species, which concurrently harm dopaminergic neurons. The stimulation of microglial cells by misfolded SNCA protein initiates a persistent inflammatory response. Extracellular immune cells, in conjunction with activated astrocytes, facilitate the progression of PD (132).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "CRISPR\u2013Cas techniques have proven essential in elucidating PD pathophysiology by facilitating the precise alteration of genes associated with dopaminergic neuron susceptibility (133). CRISPR-mediated knock-in and knock-out mice of LRRK2, PINK1, and PARK have clarified abnormalities in mitochondrial quality control and dysregulation of autophagy (134, 135). CRISPR correction of LRRK2-G2019S in patient-derived iPSCs reinstates dopaminergic differentiation and mitochondrial function, underscoring its therapeutic significance. CRISPRi has been utilized to diminish SNCA overexpression, hence decreasing aggregation and related toxicity (136). Novel therapeutic approaches encompass AAV-mediated CRISPR\u2013Cas systems aimed at SNCA regulatory elements and base editors intended to rectify pathogenic point mutations; nonetheless, challenges related to delivery efficiency and off-target effects persist as significant obstacles (134).\n\nCRISPR\u2013Cas9 technologies, encompassing knockout, base editing, and gene correction, facilitate our comprehension of biological mechanisms and preserve phenotypes in models pertinent to PD. They possess significant potential for future gene therapy as they target genetic factors such as LRRK2, PINK1, and PARKIN, thereby aiding in the reduction of neuroinflammation and neuronal degeneration (137\u2013140).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "Recently, CRISPR\u2013Cas9 genome editing has been employed in a laboratory to rectify mutations responsible for PD and to elucidate their pathogenic mechanisms. For example, employing CRISPR\u2013Cas9 to eliminate the A53T-SNCA mutation resulted in a significant reduction in both SNCA mRNA expression and its protein levels (137, 141). Beligni et al. (142) demonstrated that CRISPR\u2013Cas9 could rectify the LRRK2 p.Gly2019Ser (G2019S) mutation in patient-derived cells. This indicates its potential as a therapeutic agent to restore normal kinase function (143). A separate work employed CRISPR\u2013Cas9 to entirely eliminate PINK1 in THP-1 cells, enhancing our understanding of PIN1\u2032s function in maintaining mitochondrial integrity and the progression of PD (142). A recent work applied by (142) addresses the role of PINK1 in PD by employing CRISPR\u2013Cas9 to generate PINK1 knock-out THP-1 cells. The PINK1 knockout exhibited diminished mitochondrial complex I activity alongside alterations in Tau protein expression patterns. In PINK1 knock-out macrophages, pro-inflammatory cytokines such as IL-6, IL-1\u03b2, and IL-23 were raised, while the lipid droplet count per cell decreased, indicating mitochondrial dysfunction and immune system pathological alterations (142).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "Tokuda et al. expanded upon these findings by employing HDR to correct LRRK2 G2019S and truncate the LRRK2 kinase domain in marmoset embryonic stem cells and iPSCs. This rendered non-human primate cellular models applicable (144). In pertinent research, elevating glial cell line-derived neurotrophic factor (GDNF) levels in MPTP-treated neuronal cells via CRISPR\u2013Cas9 was associated with reduced monoamine oxidase B (MAO-B) activity, diminished production of reactive oxygen species, and enhanced cell survival under neurotoxic stress (145). The in vitro applications of CRISPR\u2013Cas9, including gene deletion, point mutation correction, gene knockout, and gene upregulation, demonstrate its versatility and efficacy as a tool for investigating the biology of PD and developing gene-based therapies.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "Researchers have employed CRISPR\u2013Cas9 to develop large-animal models (in vivo) of PD by introducing deleterious SNCA mutations into existing brain systems (146). Zhu et al. (146) employed CRISPR\u2013Cas9 and SCNT to generate Guangxi Bama minipigs with the PD-associated E46K, H50Q, and G51D mutations in SNCA. Despite the presence of mutant SNCA in the early animals, they exhibited neither dopaminergic neurodegeneration nor Lewy body pathology at 3 months, highlighting the potential and limitations of this large-animal paradigm. Employing dCas9, CRISPRi effectively reduced SNCA mRNA and its protein levels in animal models (147). PINK1 and PARKIN constitute a pivotal signaling axis that significantly regulates the mitochondrial autophagy process in dopaminergic neurons. In PINK1 null Drosophila, the silencing of the ubiquitin c-terminal hydrolase L1 (UCHL1) gene using RNA interference can ameliorate the pathology associated with PD. The absence of UCH deubiquitination enhances mitochondrial autophagy by stimulating the expression of AMP-activated protein kinase (AMPK) and Unc-51\u2013like autophagy-activating kinase 1 (ULK1) (148).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "Researchers have developed the mitochondrial complex I (MCI)-Park model in mice using CRISPR Cas9 to generate dopamine neurons deficient in NADH dehydrogenase [ubiquinone] iron\u2013sulfur protein 2 (NDUSF2), which encodes mitochondrial complex I. Mice deficient in NDUSF2 exhibited neurodegenerative alterations (149). Point mutations in the vacuolar protein sorting 35 gene (VPS35) are linked to autosomal dominant late-onset PD (PARK17). The homozygous deletion of Vps35 using CRISPR Cas9 led to a survival disadvantage and a marked decrease in dopamine release in the caudate putamen of adult homozygous Vps35 mutant mice (150). Peroxidized phospholipids resulting from ferroptosis have been linked to the beginning of certain PD. Patatin-like phospholipase domain-containing protein 9 (PNPLA9), functioning as a hydrolytic enzyme, exhibits a preference for hydrolyzing peroxidized phospholipids. In mice, the ablation of PNPLA9 led to the advancement of Parkinsonian motor abnormalities and the accumulation of peroxidized phospholipids (151). This represents a promising method for gene repression in vivo PD models.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "Researchers have investigated the application of CRISPR techniques to activate genes. Narv\u00e1ez-P\u00e9rez et al. employed a pseudo-lentivirus to deliver a CRISPR/SAM system to rat astrocytes, enhancing the expression of tyrosine hydroxylase and resulting in dopamine production. When these altered astrocytes were introduced into a rat model exhibiting 6-hydroxydopamine (6-OHDA)-induced injury, they improved motor function (138). This demonstrated an innovative approach to treating PD. Increased CRISPR-Cas9 genome editing has produced mouse models that eliminate NDUSF2 (mitochondrial complex I) or knock out VPS35 and Pnpla9 (148\u2013152). This induces neurodegeneration and motor dysfunction akin to the manifestations of PD. Conversely, the ablation of PRKN, PINK1, and DJ-1 thrice in murine models did not induce neurodegeneration, indicating that some species exhibit greater resilience and that the genetic modeling of PD is complex (153).\n\niPSC models derived from individuals with PD are valuable for investigating genetic variations associated with the disorder, such as LRRK2, SNCA, PARKIN, and ubiquinol\u2013cytochrome c reductase core protein 1 (UQCRC1) (154\u2013156). Neural stem cells (NSCs) harboring the LRRK2 p.G2019S mutation are predisposed to proteasome stress, nuclear envelope dysfunction, clonal proliferation issues, and challenges in neural differentiation (157).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "Moreover, rectifying this mutation using knock-in techniques restores the phenotype to normal, indicating that nuclear morphology may serve as a biomarker for PD. LRRK2 p.G2019S disrupts mitochondrial mitophagy by impairing the Miro\u2013PINK1/PARKIN pathway, hence hindering cellular respiration and metabolism (158). Similarly, NSCs derived from individuals with PARK2 mutations differentiate into neurons exhibiting elevated oxidative stress, activated Nrf2 pathways, and abnormal mitochondrial morphology and dysregulated homeostasis (159, 160).\n\nResearch on dopaminergic SH-SY5Y cells with UQCRC1 mutations, observed in both familial and sporadic PD, indicates that axonal degradation occurs and mitochondrial respiratory chain function ceases. These iPSC-derived cellular models demonstrate how genetic risk factors influence pathways governing proteostasis, mitochondrial integrity, oxidative equilibrium, and cellular morphology (161). They demonstrate the utility of CRISPR\u2013Cas9-mediated gene repair or knockout in restoring cellular function, thereby enhancing the understanding of underlying mechanisms and identifying potential therapeutic targets in PD.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "CRISPR-modified PD organoids replicate dopaminergic neuron degeneration, oxidative stress, and \u03b1-synuclein pathology, offering a three-dimensional framework for longitudinal investigations. Three-dimensional human midbrain organoids generated from CRISPR-modified induced pluripotent stem cells augment the physiological relevance of these models. These organoids have a spatial organization akin to the human substantia nigra, with dopamine-producing neurons, astrocytes, and nascent microglial-like populations. CRISPR-induced PD mutations in organoids expedite SNCA fibrillization, impair neuromelanin production, cause lysosomal dysfunction, and lead to progressive neuronal loss, reflecting clinical illness with temporal advancement. These 3D systems enable longitudinal examination of synapse degeneration, neuroinflammation, and intercellular communication processes inadequately represented in 2D cultures (162).\n\nBiomedical applications illustrate the usefulness of CRISPR-Cas12 technology, which has shown promise in research for treating NDs, including PD. Healthcare research continues to develop and still under research, studies on the applicability of CRISPR-Cas12 for PD treatment, utilizing the existing CRISPR technology still limited.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "Researchers have discovered intriguing applications for CRISPR-Cas3\u2032s RNA-targeting DNA editing technologies in treating NDs, including PD. CRISPR systems based on DNA editing fail to target single-stranded RNA, but Cas13 can precisely control gene expression using RNA targets that do not alter genomic sequences (163). The application on PD and CRISPR-Cas13 still limited and shown in Table 3.\n\nPD, Parkinson's disease; CRISPR, clustered regularly interspaced short palindromic repeats; Cas9, CRISPR-associated protein 9; iPSC, induced pluripotent stem cell; KO, knockout; HDR, homology-directed repair; CRISPRi, CRISPR interference; dCas9, catalytically inactive Cas9; KRAB, Kr\u00fcppel-associated box transcriptional repression domain; ABE, adenine base editor; CBE, cytosine base editor; RNP, ribonucleoprotein complex; AAV, adeno-associated virus; HDAdV, helper-dependent adenoviral vector; ROS, reactive oxygen species; SNCA, \u03b1-synuclein gene; LRRK2, leucine-rich repeat kinase 2; GBA, glucocerebrosidase gene; PRKN (PARKIN), parkin gene; PINK1, PTEN-induced kinase 1; mtDNA, mitochondrial DNA.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "HD is a progressive ND, an autosomal dominant disease. The condition is attributed to an enlarged cytosine-guanine-adenine (CAG) repeat in the HTT gene located on chromosome 4 (4p16.3), resulting in a mutant huntingtin (mHTT) protein characterized by an elongated polyglutamine tract (164\u2013166). The deleterious gain-of-function characteristics of mHTT initiate a series of molecular disruptions that ultimately result in the specific destruction of medium spiny neurons within the striatum. At the molecular level, mHTT impairs transcriptional control by anomalously interacting with transcription factors, including CREB-binding protein (CBP), resulting in extensive transcriptional suppression and modified neuronal identity. Moreover, mHTT disrupts axonal transport by obstructing microtubule-associated motor proteins (dynein/dynactin), leading to impaired trafficking of vesicles, mitochondria, and neurotrophic signals such as brain-derived neurotrophic factor (BDNF) (167). Individuals with optimal health typically possess 16\u201320 CAG repeats, however, those afflicted with HD may have over 40 repetitions. Extended repeats are associated with earlier onset and more severe symptoms (164). The misfolded mHTT protein forms toxic aggregates that disrupt critical cellular activities such as transcription, energy metabolism, axonal transport, and apoptosis. This ultimately results in the demise of striatal and cortical neurons. In the clinic, HD manifests in middle age with issues related to movement, cognition, and mental health (168). The studies detected the mitochondrial dysfunction is a primary characteristic, marked by diminished respiratory chain activity, compromised calcium buffering, elevated oxidative stress, and increased vulnerability to apoptosis. mHTT also undermines autophagy and proteostasis by obstructing autophagosome cargo loading, impairing ubiquitin\u2013proteasome functionality, and facilitating the formation of insoluble aggregates. Synaptic dysfunction manifests early via altered NMDA receptor activation, excitotoxicity, and compromised corticostriatal connection. Recent data indicates that non-cell-autonomous mechanisms, including as impaired astrocytic glutamate absorption and microglial activation, contribute to the acceleration of neuronal death. These interrelated pathways collectively characterize HD as a progressive multisystem condition influenced by proteotoxicity, transcriptional dysregulation, compromised energy metabolism, and synaptic susceptibility (169, 170). A 2012 meta-analysis revealed significant regional variation in the prevalence of HD, with elevated rates in Europe, North America, and Australia (5.7 per 100,000) compared to diminished rates in Asian populations (0.4 per 100,000) (171).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "HD pathophysiology comes from the misfolding of mHTT into aggregates that impair cellular function, ultimately leading to neuronal cell death (172). The protein interferes with transcriptional regulators, leading to aberrant gene expression throughout the organism (173). Mitochondrial dysfunction, along with elevated oxidative stress, induces neuronal injury by diminishing cellular energy levels (174). Concurrent aberrant glutamate activity induces neuronal death pathways. The mHTT protein induces malfunction in the proteasome and autophagy processes, leading to the accumulation of toxic proteins (175). Neuronal damage increases due to chronic neuroinflammation caused by activated microglia and astrocytes in the nervous system (176). The prospective applications of CRISPR-Cas technology for the treatment of HD are significant due to its capacity for precise genetic material alteration. Gene therapy employing targeted CRISPR-Cas9 serves as a prospective treatment strategy for disease management by directly correcting genetic mutations at their origin (164, 177). Various studies have been demonstrated the potential of CRISR/Cas tool in HD application and theoretical aspects such as gene editing, selective targeting of mutant alleles, and excision of expanded cag repeats (178\u2013180). Shin et al. undertook a study with the goal of enhancing the specificity of alleles. They employed a customized approach using CRISPR\u2013Cas9 that selectively targeted specific alleles by modifying PAM-altering single nucleotide polymorphism (SNP). This approach focused on identifying CRISPR\u2013Cas9 sites that were unique to each patient and utilized a comprehensive understanding of the haplotype structure of the HTT gene. The objective was to deactivate the mutated HTT allele for a specific haplotype specifically (181).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "Researchers have employed CRISPR\u2013Cas9 genome editing to create isogenic cell models (in vitro) of HD by precisely modifying the CAG repeat region in the HTT gene. CRISPR has been used to allele-specific targeting and genetic correction, Allele-specific excision of the mutant HTT, Shin employed SNP-guided dual-CRISPR\u2013Cas9 targeting to excise approximately 44 kb from the mHTT allele in patient-derived cells. This eliminated mutant mRNA and protein while preserving the wild-type allele (181). In a separate investigation, Yang et al. showed that the application of CRISPR\u2013Cas9 in HD140Q-KI mice to remove HTT in a non-allele-specific way can successfully and permanently eradicate the harmful effects of polyglutamine expansion on neurons in the adult brain. The study population showed a significant reduction in reactive astrocytes and a notable improvement in motor impairment (182). Another study utilized CRISPR to CAG-repeat excision, such as a study introduced 150 CAG repeats into exon 1 of HEK293T cells to create clones resembling the condition, perhaps facilitating the investigation of disease mechanisms (183). In pertinent research, scholars employed Cas9 ribonucleoprotein complexes and single-stranded oligodeoxynucleotide donors to create HEK293T lines containing 41, 53, and 84 CAG repeats. This facilitated the examination of alterations in cellular phenotypes associated with varying repeat lengths and contributed to the development of novel treatments (184). Another approach employed Cas9 nickases and guide RNAs to excise CAG repeats, resulting in frameshifts that activated premature stop codons and inhibited mHTT production. This illustrates how accurate genome editing could reinstate the typical allele configuration (185). In transgenic BacHD mice, Monteys et al. (186) incorporated a modified human HD allele and targeted human mHTT exon 1. BacHD animals treated with rAAV.spCas9 and rAAV.sgHD1B/i3 showed decreased HTT mRNA levels in their right hemisphere but not in their uninfected left (186). Ekman et al. (187) revealed effectively inhibited the expression of the human mutant HTT gene in R6/2 mice that carried human mHTT exon 1 with roughly 115\u2013150 CAG repeats in 2019 using AAV-mediated CRISPR-Cas9. This study showed that R6/2 mice's lifespan and motor deficits were enhanced by inhibiting HTT expression. In conclusion, it has been shown in various animal models that HD may be effectively treated by employing CRISPR-Cas9 to eliminate mHTT gene expression (187).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "CRISPR-based gene-silencing techniques have been employed to reduce the expression of mHTT with the direct deletion of repeat expansions. CRISPRi exclusively inhibited mHTT transcription in fibroblasts derived from HD patients, without affecting wild-type HTT (188). Previous work employed CasRx (RNA-targeting Cas13) to downregulate HTT exon 1 in HEK293T cells, HD 140Q knock-in mice, and HD knock-in pigs. This decreased mHTT levels, diminished gliosis, and decelerated neurodegeneration (189). Ekman et al. (187) revealed effectively inhibited the expression of the human mHTT gene in R6/2 mice that carried human mHTT exon 1 with roughly 115\u2013150 CAG repeats in 2019 using AAV-mediated CRISPR-Cas9. This study showed that R6/2 mice's lifespan and motor deficits were enhanced by inhibiting HTT expression. It has been shown in various animal models that HD may be effectively treated by employing CRISPR-Cas9 to eliminate mHTT gene expression (187).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "Oura et al. (190) addressed the target range limits of WT-SpCas9, scientists engineered SpCas9-NG, which detects NGN PAMs. They used SpCas9-NG to precisely contract HD CAG repeat tracts in mouse embryonic stem cells. The procedure of rectifying mutations restored distinctive characteristics as well as health conditions in both neurons derived from modified stem cells and the animals produced by these cells. This study reveals that SpCas9-NG is a powerful way to treating enlarged CAG repeats and other disease variants that are difficult to target (190). Another study via Merienne et al. (191) found the development of the KamiCas9 self-inactivating CRISPR\u2013Cas9 system, which enables high-efficiency Cas9 expression during transitory intervals. Using their technique on adult mouse neuronal and glial cells, researchers were able to inactivate the HTT gene while concurrently improving HD disease markers. The KamiCas9 system made it possible to reduce unwanted edits when evaluating human iPSC-derived cells since it had less off-target effects than typical constitutive Cas9 systems. The study shows how self-inactivating CRISPR\u2013Cas9 technologies could become a feasible therapy option for NDs (191). Ultimately, allele-specific CRISPR\u2013Cas9 editing utilizing SNP-generated PAM sites facilitated the targeted silencing of mHTT by nonsense-mediated degradation in approximately 20% of European HD patients. This demonstrated exceptional accuracy and little off-target effects following GUIDE-Seq validation. This research demonstrates the versatility of CRISPR approaches in investigating the etiology of HD and formulating novel therapies. They can be utilized for genome editing, gene silencing, and allele-specific strategies (192).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "CRISPR-based techniques have been shown to be effective in vivo for HD by reducing levels of mHTT in the brain. Bunting and Donaldson demonstrated that employing antisense oligonucleotides to inhibit MSH3 activity reduced HTT CAG repeat expansions in HD mouse models. This indicates that gene-targeting methodologies may alter the trajectory of the disease (193). Similarly, administering CRISPR\u2013Cas9 to striatal neurons in HD140Q-knockin mice significantly reduced endogenous mHTT, inhibited aggregation formation, and ameliorated early symptoms of neurodegeneration and motor dysfunction, all without causing neuronal damage (182). In the prevalent R6/2 mouse model, CRISPR\u2013Cas9-induced deletion of HTT resulted in reduced protein inclusions, enhanced motor function, and an extended longevity. This indicates that targeted genome editing may be an effective approach for treating HD (103, 186, 187, 191).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "Enhanced editing precision has significantly improved safety and efficacy. Base editing was utilized in vivo to disrupt caspase-6 cleavage sites within the HTT gene. This resulted in mHTT isoforms that exhibit resistance to proteolytic processing. This approach effectively reduced the detrimental aggregation of huntingtin and safeguarded against neuronal loss in the striatum and brain of HD animal models (187). In vivo investigations, encompassing antisense, CRISPR\u2013Cas9 gene disruption, and base editing, demonstrate that reducing or altering mHTT expression can decelerate the advancement of neuropathological disorders and enhance functional outcomes. This indicates that CRISPR possesses the potential to serve as a disease-modifying instrument for HD (194).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "The integration of iPSCs with CRISPR\u2013Cas technology has advanced HD research by creating precise, isogenic models and revitalizing cells. Scientists developed homozygous HEK293T cell lines with varying amounts of CAG repeats in the HTT gene to facilitate mechanistic investigations and drug screening (184). Researchers have employed CRISPR\u2013Cas9 to rectify pathogenic CAG expansions in iPSCs, effectively reversing initial indicators of mitochondrial dysfunction in HD-derived neural stem cells, including respiratory issues, elevated reactive oxygen species, and calcium dysregulation (195\u2013198). The results indicate that the selective elimination of CAG repeats can restore mitochondrial function and maintain cellular equilibrium. CRISPR\u2013Cas9 editing facilitates allele-specific disruption, repeat excision, and gene repair in both in vitro and in vivo environments. Outstanding issues remain, including its efficacy, long-term safety, and performance on alternative genes. This gene-editing technique possesses significant potential to alter the trajectory of HD (199). Table 4 illustrates some HD applications and consequences.\n\nApplications of CRISPR\u2013Cas in Huntington's disease.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "HD, Huntington's disease; HTT, huntingtin gene; mHTT, mutant huntingtin; CRISPR, clustered regularly interspaced short palindromic repeats; Cas9, CRISPR-associated protein 9; iPSC, induced pluripotent stem cell; PAM, protospacer adjacent motif; SNP, single nucleotide polymorphism; PAS, PAM-altering SNP; KO, knockout; HDR, homology-directed repair; RNP, ribonucleoprotein complex; AAV, adeno-associated virus; PE, prime editing; pegRNA, prime-editing guide RNA; PE-CORE, prime-editing correction of repeat expansions; ABE, adenine base editor; CBE, cytosine base editor; CAG/CAA, trinucleotide codons encoding glutamine; 2D, two-dimensional cell culture; 3D, three-dimensional culture; ROS, reactive oxygen species.\n\nRecent advancements indicate that CRISPR-Cas13 systems may be effective in treating HD by targeting RNA transcripts associated with the disorder. Morelli and Wu (152) conducted a notable study that advanced an enhanced CRISPR-Cas13d system known as Cas13d-CAGEX. This treatment led to enduring enhancements in motor function, reduced neuronal degeneration, and sustained therapeutic effects with minimal adverse effects observed for up to 8 months post-injection (200).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "The efficient application of CRISPR-Cas13d demonstrates its considerable potential as an RNA-targeting tool for HD and possibly other dominant genetic disorders. Cas13d represents a safer alternative for prolonged gene regulation as it specifically inhibits pathogenic transcripts without altering genomic sequences. This differs from DNA-targeting techniques. Numerous studies employ CRISPR-Cas13 in Huntington's disease, although few utilize CRISPR-Cas12. No studies have utilized CRISPR-Cas3 in HD research, indicating that these methods remain underexplored for neurodegenerative therapeutics.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "ALS is a progressive ND characterized by the degeneration of both upper and lower motor neurons. This results in muscular weakening, atrophy, paralysis, and ultimately respiratory failure (201\u2013203). Epidemiological studies indicate that incidence significantly varies by geography, with rates between 0.26 and 23.46 cases per 100,000 person-years, whereas prevalence ranges from 1.57 to 11.80 instances per 100,000 individuals across various countries (203, 204). Approximately 75% of ALS cases occur sporadically, although around 25% are familial and associated with over 135 genetic loci (205\u2013208). A defining characteristic of ALS is the cytoplasmic mis localization and aggregation of TDP-43, which impairs RNA splicing, transport, and stability, resulting in extensive transcriptome dysregulation (202, 209). Mutant superoxide dismutase 1 (SOD1) and fused in sarcoma (FUS) intensify cellular stress through protein misfolding, oxidative injury, and impaired DNA damage repair. Intrinsic neuronal abnormalities are exacerbated by non-cell autonomous mechanisms, such as microglial activation, astrocytic glutamate toxicity due to diminished excitatory amino acid transporter 2 (EAAT2) expression, mitochondrial dysfunction, and compromised axonal transport. These mechanisms collectively begin a self-sustaining loop of neuroinflammation, energy depletion, and synaptic degeneration, ultimately resulting in motor neuron death (210, 211). The hexanucleotide (GGGGCC) expansion in C9orf72 is the predominant mutation among these. It induces toxic RNA foci and dipeptide repeat protein accumulation in around 40% of familial cases and 7% of sporadic cases (212\u2013214). TAR DNA-binding protein (TARDBP) mutations disrupt TDP-43 protein homeostasis (4% familial, 1% sporadic) (215, 216), while SOD1 mutations lead to oxidase dysregulation and protein aggregation (217, 218) in 12% of familial cases and 1%\u22122% of sporadic cases. Moreover, alterations in FUS, heterogeneous nuclear ribonucleoprotein A1 (HNRNPA1), profilin 1 (PFN1), and kinesin family member 5A (KIF5A) influence RNA metabolism, axonal transport, and cytoskeletal architecture, ultimately resulting in motor neuron degeneration in ALS (219\u2013223).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "The CRISPR\u2013Cas genome-editing technology offers a significant method for developing targeted ALS therapies through precise genetic modifications (224). The protentional applications of CRSPR/Cas in ALS disease was gene silencing and correction, disease modeling, gene therapy delivery (16, 224\u2013226) (Figure 3). In vivo CRISPR techniques have been employed to directly target pathogenic ALS mutations in animal models, yielding quantifiable functional improvements and molecular restoration. AAV-mediated CRISPR\u2013Cas9 targeting of mutant SOD1 decreased mutant SOD1 levels of proteins in the spinal cord and enhanced motor function and survival in transgenic mice, illustrating that somatic gene disruption can alleviate disease symptoms. Likewise, the excision of the pathogenic GGGGCC hexanucleotide repeat expansion (HRE) in C9orf72 by paired CRISPR\u2013Cas9 guides diminished RNA foci and dipeptide repeat proteins, therefore ameliorating molecular abnormalities in various transgenic mice models (227). Recently, RNA-targeting CRISPR systems (Cas13/CasRx) designed for high fidelity have been utilized in vivo to diminish harmful repetitive RNAs and subsequent dipeptide repeat proteins (DPR) toxicity, providing a non-DNA-cleaving therapeutic approach that may mitigate irreversible on-target genetic concerns. These investigations demonstrate the feasibility of in vivo CRISPR interventions to reverse fundamental ALS processes, while also emphasizing the problems of delivery, immunogenicity, and product purity (editing byproducts) that persist for clinical application (227).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "Schematic illustration shows the CRISPR-Cas9 uses in the investigation and treatment of Amyotrophic Lateral Sclerosis (ALS). (A) Traditional disease modeling involves obtaining diseased brain tissue biopsies to establish patient-specific primary cell lines. (B) CRISPR/Cas9-based disease modeling utilizes somatic reprogramming to create iPSCs, followed by gene editing (e.g., homologous recombination or deletion of G4C2 repeats) to investigate phenotypic changes in differentiated neurons. The figure was designed with BioRender (http://biorender.com/). CRISPR, clustered regularly interspaced short palindromic repeats; Cas9, CRISPR-associated protein 9; ALS, amyotrophic lateral sclerosis; iPSCs, induced pluripotent stem cells; qPCR, quantitative polymerase chain reaction ; G4C2, guanine\u2014cytosine hexanucleotide repeat; MiSeq, Illumina sequencing platform.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "In vitro CRISPR modification of human cells, particularly patient-derived fibroblasts and iPSC-derived motor neurons, has proved essential for elucidating ALS pathophysiology and confirming targets. The CRISPR\u2013Cas9-mediated removal or allelic modification of C9orf72 repetitions in patient-derived iPSC neurons diminishes RNA foci and DPR accumulation, while reinstating nucleocytoplasmic transport and cellular survival. CRISPR-induced knockouts of SOD1, FUS, or TARDBP in human cellular models replicate key mitochondrial, RNA metabolism, and proteostasis deficiencies, facilitating mechanistic comparisons among genotypes. The in vitro application of RNA-targeting Cas13 devices enables the direct reduction of hazardous repetitive RNAs in patient cells, facilitating swift assessments of guide efficacy and collateral activity prior to in vivo experimentation. Cell-based research is essential for determining dosage, assessing off-target effects, and developing biomarkers (228).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "CRISPR\u2013Cas9 technology has demonstrated its ability to rectify mutations associated with ALS in iPSC in vitro models derived from patient cells. For example, employing Cas9 along with a co-transfected donor template to rectify the SOD1 A272C mutation in fibroblast-derived iPSCs restored the wild-type sequence (229). Similarly, rectifying the SOD1 E100G mutation in iPSCs using homology-directed repair not only increased the quantity of motor neurons, the size of the soma, and the formation of neurites, but also diminished cell mortality post-differentiation (230). In C9orf72-associated ALS, employing CRISPR\u2013CasRx to diminish sense and antisense repeat transcripts in HEK293T cells and patient-derived iPSC neurons decreased RNA foci and dipeptide repeat proteins, thereby providing protection against glutamate-induced excitotoxicity. Furthermore, utilizing Cas9 HDR to accurately excise the deleterious HRE from C9orf72 maintained the gene's integrity, restored expression and methylation levels to baseline, reduced intron retention, and eliminated detrimental repeat artifacts in motor neurons derived from iPSCs. All of these in vitro experiments demonstrate that CRISPR-based editing is a versatile and promising approach to address the primary genetic causes of ALS (231). The generation of SOD1+/A272C iPSCs and FUS+/G1566A iPSCs from primary fibroblasts of two ALS individuals, one possessing a heterozygous A272C mutation in the SOD1 gene and the other a heterozygous G1566A mutation in the FUS gene. The CRISPR-Cas9 approach effectively rectified the SOD1+/A272C and FUS+/G1566A mutations in ALS iPSCs utilizing a donated plasmid or single-stranded oligodeoxynucleotide (ssODN) as the repair template, resulting in restored ALS iPSCs that exhibited normal pluripotency. Furthermore, SOD1+/A272C and repaired iPSCs were prompted to differentiate into motor neurons, resulting in the identification of 899 aberrant transcripts using RNA sequencing. This work may assist in identifying biomarkers and pathways for ALS and provides novel evidence for ALS therapy (229).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "Base editing and CRISPR\u2013Cas9 are two efficacious techniques for investigating and addressing ALS, particularly in vivo animals. An essential approach involves utilizing AAV9 to administer CRISPR components to transgenic ALS mice with the gain-of-function SOD1 G93A mutation. Initial endeavors to employ S. aureus Cas9 (SaCas9) and gRNAs for enhancing survival and muscle preservation yielded limited success; hence, alternative strategies were explored. Base editors, including cytidine or adenine deaminase linked to a nickase Cas, enable accurate single-nucleotide modifications without generating double-strand breaks an advantageous characteristic for rectifying or deactivating point mutations in familial ALS. A pivotal proof-of-concept study using adenine base editing to inactivate a mutant SOD1 allele in vivo, enhancing motor performance in an ALS mouse model. This demonstrated that base editing can effectively diminish hazardous protein levels and yield phenotypic advantages while circumventing several consequences associated with DSBs. Base editing is limited by editable sequence contexts and bystander edits, while the delivery of base-editor cargo to motor neurons presents a translational bottleneck; yet, these results demonstrate substantial potential for therapy for point-mutation ALS. Researchers employed a dual AAV9 split-intein methodology to circumvent the limitations of vector size, enabling the creation of specific nonsense mutations in SOD1 (227, 232, 233). We must exercise caution in simultaneous initiatives aimed at targeting FUS and TARDBP mutations to avoid the detrimental consequences of entirely eliminating a gene (234, 235). Despite these challenges, employing CRISPR to reduce mutant SOD1 in G93A-SOD1 mice significantly improved their motor functions, postponed illness onset by 37%, and extended their longevity by 25%. This indicates that gene editing possesses therapeutic potential (227).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "CRISPR\u2013Cas9 has been effectively employed to rectify faulty expansions in the C9orf72 gene, the most common genetic contributor to ALS. This pertains to SOD1-related ALS. Multiple groups have utilized paired guide RNAs around the hexanucleotide repeats to induce excision by NHEJ in iPSCs derived from patients. They have successfully eliminated up to 11% of the target DNA. Unanticipated deletions occurred; yet, C9orf72 expression remained constant (236\u2013238). To enhance fidelity, normal-length repeats were included by HDR techniques, restoring typical methylation and transcription profiles. Neurons derived from these modified iPSCs exhibited a reduced number of RNA foci, diminished dipeptide repeat protein aggregates, and enhanced stress resilience (231, 239). SOD1 participates in the detoxification of superoxide radicals, and mutations in SOD1 lead to enhanced harmful activity (209, 240). The G93A-SOD1 mouse model possesses roughly 25 tandem repeat copies of the human SOD1G93A transgene, demonstrating signs of amyotrophy and dyskinesia observed in ALS patients (241). CRISPR modification of iPSCs has emerged as a fundamental element in ALS research. By producing isogenic iPSC pairs (mutant vs. corrected control), researchers can associate genotype with phenotype while minimizing background interference. The differentiation into motor neurons and three-dimensional spinal/brain organoids facilitates the replication of TDP-43 mislocalization, RNA foci, DPR synthesis, mitochondrial dysfunction, axonal transport abnormalities, and non-cell-autonomous contributions from astrocytes and microglia found in ALS. These CRISPR-iPSC platforms are extensively utilized for mechanistic analysis, high-throughput drug screening, and preclinical assessment of gene-editing methodologies, including the evaluation of AAV/SNP-guided allele selectivity, base editors, and Cas13 guides. iPSC platforms are essential for precise off-target and transcriptome-wide RNA-editing evaluations before animal or clinical use (226).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "Recent advancements in CRISPR-based RNA-targeting methodologies suggest their potential utility in the treatment of ALS. Zhang et al. (2021) in their review article discussed an enhanced variant of CRISPR-Cas13d, termed CasRx, which can degrade both sense and antisense C9ORF72 repeat RNAs while preserving normal gene expression. Utilizing AAV vectors, CasRx was administered to HEK cells and patient-derived iPSC neurons in vitro, as well as to transgenic animal models in vivo. This resulted in significant reductions in RNA foci, DPRs, and glutamate-induced excitotoxicity (163). In another study, researchers employed Cas13 and Cas7-11 effectors to target ataxin-2 in models of TDP-43 proteinopathy. Direct injection of the Cas13 system into mice effectively reduced TDP-43 aggregation and the location of stress granules. This resulted in improved behavior, increased longevity, and reduced neurodegeneration in the mice. In comparison to previous Cas7-11 tools (242), Cas13 variants exhibiting enhanced fidelity demonstrated superior transcriptome specificity. The results indicate that RNA-targeting CRISPR systems possess significant potential as a therapeutic approach for ALS, as they can target critical disease characteristics. Table 5 illustrates CRISPR techniques in ALS.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Application of CRISPR\u2013Cas in neurodegenerative disorders", "text": "ALS, amyotrophic lateral sclerosis; CRISPR, clustered regularly interspaced short palindromic repeats; Cas, CRISPR-associated (e.g., Cas9, Cas13); SOD1\u2227G93A, SOD1 glycine-to-alanine substitution at codon 93 (G93A) model; iPSC, induced pluripotent stem cell; MN(s), motor neuron(s); HDR, homology-directed repair; KO, knock-out; KI, knock-in; KD, knockdown; CRISPRi, CRISPR interference; ABE, adenine base editor; CBE, cytosine base editor; DSB(s), double-strand break(s); AAV, adeno-associated virus (e.g., AAV9; AAV-PHP variants); RNP, ribonucleoprotein (Cas protein + guide RNA complex); ASO, antisense oligonucleotide; RNAi, RNA interference; DPR(s), dipeptide repeat protein(s); CasRx/Cas13d, Cas13d (also called CasRx) RNA-targeting CRISPR effector; PE, prime editing; pegRNA, prime-editing guide RNA; PBS, primer-binding site; PAM, protospacer adjacent motif; v-ATPase, vacuolar-type H+-ATPase; CNS, central nervous system; LNP(s), lipid nanoparticle(s).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR\u2013Cas in Alzheimer's disease", "text": "AD is a progressive ND that impairs cognitive functions, including memory, reasoning, and daily living tasks (70, 71). The disease is a major cause of dementia among the elderly. It is characterized by the presence of amyloid-beta (A\u03b2) plaques and neurofibrillary tangles in the brain, which impair neurons and contribute to the degeneration of brain tissue (72, 73). The disease initiates with slight memory challenges that progress to severe cognitive and physical impairments, ultimately resulting in complete dependency and premature death (74). The disease disseminates throughout diverse communities at a prominent global health scale due to its elevated prevalence rates. There are 5.5 million individuals in the United States with AD. In 2016, the global number of individuals with dementia reached 43.8 million, reflecting a 116% increase since 1990, according to recent studies (75).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR\u2013Cas in Alzheimer's disease", "text": "Development of AD largely driven by hereditary elements follows several pathophysiological paths. Many times, mutations in the APP, PSEN1, and PSEN2 genes link fAD to one another (76). The APP gene codes the amyloid precursor protein on chromosome 21, which \u03b2-secretase and \u03b3-secretase break to generate A\u03b2 peptides. Applying mutations in APP produces overly high synthesis of A\u03b2, particularly the A\u03b242 variant, which readily accumulates to generate amyloid plaques, a key hallmark of AD (77). Importantly, members of the \u03b3-secretase complex, PSEN1 and PSEN2, respectively encode PSEN1 and PSEN2 proteins (18, 78). Early-onset fAD is caused by mutations in these genes altering \u03b3-secretase activity, hence increasing the A\u03b242/A\u03b240 ratio and promoting A\u03b2 aggregation (79).\n\nApart from amyloid pathology, tau protein dysfunction also speeds the development of disease. Although microtubule-associated protein tau (MAPT) gene mutations are rare in AD, tau proteins in affected individuals hyperphosphorylated, therefore losing their ability to stabilize microtubules (80). Along with impairments in neuronal activity, this instability results in paired helical filaments and neurofibrillary tangles (81\u201383).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR\u2013Cas in Alzheimer's disease", "text": "Moreover, exacerbating AD development is neuroinflammation. Released by active microglia and astrocytes, reactive oxygen species and pro-inflammatory cytokines damage neurons. Variations in the triggering receptor expressed on myeloid cells 2 (TREM2) gene, which affects microglial activity, have also been associated to increased AD risk, therefore underlining the critical role immune responses play in the evolution of disorders (84\u201387).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR\u2013Cas in Alzheimer's disease", "text": "CRISPR\u2013Cas9 represents a formidable gene-editing technique for AD, facilitating the precise modification of pathogenic genes, such as rectifying or eliminating APP, \u03b2-site amyloid precursor protein cleaving enzyme 1 (BACE1), and PSEN1/2 mutations to diminish the production and aggregation of harmful A\u03b2 (88); the conversion of Apo-lipoprotein E \u03b54 allele (APOE4) to the protective APOE3 or APOE2 isoforms, thereby reducing A\u03b2 accumulation, tau pathology, and neuronal susceptibility; the modulation of neuroinflammation through the targeting of CD33, glia maturation factor (GMF), or other immune mediators to enhance microglial clearance of amyloid plaques and inhibit detrimental cytokine signaling (89); and the amplification of neuroprotective pathways by promoting non-amyloidogenic APP processing (\u03b1-cleavage) or bolstering microglial functions that maintain neuronal integrity (88). These attributes render CRISPR a versatile method for rectifying genetic disease etiologies, reducing the accumulation of deleterious proteins, mitigating inflammation, and enhancing cerebral resilience. This approach enables researchers to acquire accurate AD models in vivo and in vitro, enhancing the comprehension of disease mechanisms, and aiding in the identification of novel therapy alternatives (Figure 2).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR\u2013Cas in Alzheimer's disease", "text": "Diagrammatic illustration of the CRISPR\u2013Cas9 system in the study and treatment of Alzheimer's disease (AD) and shows how CRISPR\u2013Cas9 can be used to target genetic alterations and pathways linked to AD. The figure was designed with BioRender (http://Biorender.com/). CRISPR, clustered regularly interspaced short palindromic repeats; Cas9, CRISPR-associated protein 9; sgRNA, single-guide RNA; PAM, protospacer adjacent motif; HDR, homology-directed repair; MMEJ, microhomology-mediated end joining; DNA, deoxyribonucleic acid; A\u03b2 (amyloid beta), amyloid-\u03b2 peptide.\n\nCell models are extensively utilized in application on many neurodegenerative disorders, including AD, due to the absence of ethical dilemmas, the brevity of the experimental duration, and the cheap associated costs. Over the past few decades, numerous in vitro cellular models of AD have been developed.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR\u2013Cas in Alzheimer's disease", "text": "Researchers have effectively employed CRISPR\u2013Cas9 to eliminate the APP and BACE1 genes in neural cell types. This resulted in a significant reduction in the synthesis of A\u03b2, a primary therapeutic target for AD. The previous study demonstrated that the ablation of APP in brain cells via CRISPR\u2013Cas9 diminished both A\u03b2 expression and amyloid pathology. This indicates its potential as a gene therapy alternative for individuals with APP mutations (89, 90). Sun et al. (91) employed CRISPR\u2013Cas9 to modify the C-terminus of endogenous APP, markedly diminishing \u03b2-secretase cleavage and redirecting processing toward non-amyloidogenic \u03b1-secretase, thus decreasing A\u03b2 generation.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR\u2013Cas in Alzheimer's disease", "text": "Previous studies were conducted to reduce the A\u03b2 utilizing CRISPR technique the knockout of PSEN1 and PSEN2 genes (92\u201395). In addition, researchers effectively introduced this mutation into brain cells utilizing Cas9 nickase\u2013deaminase, significantly reducing A\u03b2 generation (96). Simultaneously, Song et al. (97) discovered that the suppression of KIBRA (Kidney and Brain expressed protein) in HT22 cells inhibited proliferation and induced death, whereas treatment with A\u03b21\u201342 oligomers was administered (96, 98). Prime-Base editing was used to treat AD via utilizing CRISPR to correct the gene defects, and the results shows reduction in A\u03b2 peptide accumulation. Several studies have been performed in vitro utilizing CRISPR\u2013Cas9 technology to modify and editing the APOE gene, as this method has the potential to convert high-risk APOE4 alleles into protective alleles such as APOE3 or APOE2 to reduce the AD risk (99).\n\nIn addition to constructing AD cell models, CRISPR\u2013Cas9 technology can also be utilized to develop AD animal models to improve and correct the cognitive function in AD. Serneels, T'Syen (100) Developed a novel model utilizing the CRISPR\u2013Cas9 technique to construct a humanized A sequence in the APP gene of both species (100). Another study utilized animal models in vivo to improve cognitive function and correct APP and BACE1 genes in AD. It demonstrated that the CRISPR system can suppress A\u03b2, having a significant effect on both pathology and cognitive function (101\u2013103).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR\u2013Cas in Alzheimer's disease", "text": "Furthermore, CRISPR-based gene editing techniques, particularly base editing and prime editing, exhibit the capacity to rectify harmful AD mutations without inducing double-strand DNA breaks. The genetic modification executed by base editors led to significant enhancements in tau pathology and cognitive recovery in the examined mice (58, 104, 105). APOE possesses various variations, including APOE2, APOE3, APOE3r, and APOE4. Komor et al. (59), the transfer of APOE3 to APOE4 in mouse astrocytes was achieved by changing Cincodon158 to T using the CRISPR\u2013Cas9 system, indicating that point mutations were corrected by the CRISPR\u2013Cas9 approach (59).\n\nPrime-Base editing serves as a genome-editing technique that offers extensive genetic modification possibilities via mutation repair without inducing DSBs to address AD. In vivo research focused on AD-related applications needs greater exploration (104\u2013106), such as a study by Park et al. (103) via utilizing Cas9 nanocomplexes with sgRNAs targeting the tyrosine hydroxylase and BACE1 genes were administered into mouse primary neural cells to evaluate the efficacy of Cas9 nanocomplexes. Furthermore, Takalo et al. (107) utilizing the CRISPR\u2013Cas9 gene editing technique and evaluating the preventive effects of the Plc2-P522R variant.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR\u2013Cas in Alzheimer's disease", "text": "Patient-derived induced pluripotent stem cells (iPSCs) have garnered growing interest due to their distinctive physiological characteristics. AD mutant genes, such as PSEN1, PSEN2, and APP, in iPSC-derived neurons provide exceptional platforms for elucidating disease mechanisms and formulating novel therapeutic strategies. Studies demonstrated that gene editing using CRISPR\u2013Cas9 can mitigate the impairments linked to AD by addressing synaptic dysfunction and altered signaling pathways (108\u2013111). Researchers have created gene-edited induced pluripotent stem cell models to investigate genetic mutations associated with AD, as well as knockout genes implicated in Alzheimer's pathogenesis, including ATP-binding cassette subfamily A member 7 (ABCA7) (112\u2013114). The integration of animal models and CRISPR-edited iPSCs enhances scientists' understanding of AD risk factors and therapeutic options (115). CRISPR-engineered iPSC and organoid models have elucidated critical mechanisms in AD, including lipid metabolism (APOE), endosomal dysfunction (APP/PSEN1), electrophysiological abnormalities (PSEN2), and neuroimmune interactions (APOE Christchurch). These human-relevant systems are highly effective for identifying disease pathways and evaluating treatment alternatives (116\u2013119). Stem cell models integrated with gene editing technologies facilitate the development of patient-specific therapies, paving the way for clinical applications.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR\u2013Cas in Alzheimer's disease", "text": "Stem cell-derived models, in conjunction with gene editing, exhibit potential to enhance the development of tailored AD therapies. Table 2 Illiterates applications of CRISPR\u2013Cas in AD. Utilizing CRISPR, scientists acquire enhanced insights into AD progression and develop successful therapeutic approaches.\n\nAD, Alzheimer's disease; CRISPR, clustered regularly interspaced short palindromic repeats; Cas9, CRISPR-associated protein 9; A\u03b2, amyloid-\u03b2 peptide; iPSC, induced pluripotent stem cell; KO, knockout; HDR, homology-directed repair; RNP, ribonucleoprotein complex; AAV, adeno-associated virus; BE, base editing/base editor; PE, prime editing; pegRNA, prime-editing guide RNA; APP, amyloid precursor protein; BACE1, \u03b2-site amyloid precursor protein cleaving enzyme 1; APOE, apolipoprotein E; TREM2, triggering receptor expressed on myeloid cells 2; MAPT, microtubule-associated protein tau gene; PSEN1/PSEN2, presenilin 1 and 2; SORL1, sortilin-related receptor 1; 2D, two-dimensional cell culture; 3D, three-dimensional culture.\n\nJia et al. (120) created a CRISPR-powered aptasensor that detects A\u03b2 biomarkers via fluorescence detection via CRISPR-Cas12a. There is currently little study on the use of CRISPR-Cas13 and CRISPR-Cas3 in AD diagnosis. Although its potential in diagnostic domains has not been thoroughly investigated, Ca3\u2032s primary DNA deletion function remains its most prominent feature.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR/Cas in Parkinson's disease", "text": "PD, the second prevalent neurodegenerative ailment, is marked by the degradation of dopaminergic neurons, which results in particular motor symptoms such as rigidity, bradykinesia, and tremor (121). After the loss of neuronal, the Lewy bodies appeared with cytoplasmic inclusions, which are mostly composed of aggregates of \u03b1-synuclein protein misfolded and perhaps diffusion in a prion-like shape among interconnected areas synoptically (122). PD can be classified into two forms: sporadic and familial. The familial form accounts for only 10%\u221215% of all PD cases. The hereditary versions are attributed to mutations in multiple genes, including SNCA, parkin (PRKN), PTEN-induced kinase 1 (PINK1), DJ-1/Parkinson's disease protein 7 (PARK7), and LRRK2. SNCA and LRRK2 mutations cause autosomal-dominant types of PD, while PRKN, PINK1, and DJ-1 mutations cause autosomal recessive variants (123). Although the origins of sporadic forms are currently unknown, certain genetic and environmental risk factors have been discovered. For instance, genetic polymorphisms in the LRRK2 and SNCA genes elevate the likelihood of developing sporadic PD (124).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR/Cas in Parkinson's disease", "text": "Between 1992 and 2021, global PD cases escalated from 450,000 to 1.34 million, with crude incidence rates increasing from 8.19 to 16.92 per 100,000 and age-standardized incidence rates rising from 11.54 to 15.63 per 100,000. Incidence rates increased significantly in persons aged 60 and beyond, reaching a zenith in those aged 85 and above. Forecasts indicate that by 2030, worldwide PD cases will total 1.93 million, with an age-standardized incidence rate of 27 per 100,000 (125).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR/Cas in Parkinson's disease", "text": "PD is a ND mostly attributed to genetic abnormalities and environmental stresses (126). Genetic alterations in critical genes such as SNCA, LRRK2, PARK2, PINK1, and Glucocerebrosidase 1 (GBA1) are central to the disorder. Mutations in SNCA result in the improper folding of SNCA protein, leading to aggregation into Lewy bodies, which are indicative of PD in the brain (127). Mutations in LRRK2 that enhance gene function result in increased kinase activity, which is detrimental to neurons. Mutations in PARK2 and PINK1 disrupt mitochondrial quality control, leading to mitochondrial malfunction, poor mitophagy, oxidative stress, and ultimately, the demise of dopaminergic neurons (128). Neurons exhibit increased toxicity when mutations in the LRRK2 gene lead to the enzyme acquiring excessive kinase activity. The PARK2 gene regulates parkin activity, which acts as an E3 ubiquitin ligase; however, its dysfunction results in the buildup of protein damage (128). Neuronal damage exacerbates when GBA1 mutations induce mitochondrial malfunction and elevate oxidative stress inside cells (129, 130). Genetic abnormalities alter protein degradation pathways, impair mitochondrial function, and interfere with oxidative stress responses, ultimately leading to PD symptoms through the degeneration of dopaminergic neurons in the substantia nigra (131). Impaired mitochondria result in heightened oxidative stress and diminished ATP synthesis, causing neuronal damage. The development of PD occurs due to mutations in PINK1 and PRKN that impair mitochondrial quality control mechanisms (131). PD neuroinflammation occurs when activated microglia secrete tumor necrosis factor-alpha (TNF-\u03b1), interleukin (IL)-1\u03b2, and reactive oxygen species, which concurrently harm dopaminergic neurons. The stimulation of microglial cells by misfolded SNCA protein initiates a persistent inflammatory response. Extracellular immune cells, in conjunction with activated astrocytes, facilitate the progression of PD (132).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR/Cas in Parkinson's disease", "text": "CRISPR\u2013Cas techniques have proven essential in elucidating PD pathophysiology by facilitating the precise alteration of genes associated with dopaminergic neuron susceptibility (133). CRISPR-mediated knock-in and knock-out mice of LRRK2, PINK1, and PARK have clarified abnormalities in mitochondrial quality control and dysregulation of autophagy (134, 135). CRISPR correction of LRRK2-G2019S in patient-derived iPSCs reinstates dopaminergic differentiation and mitochondrial function, underscoring its therapeutic significance. CRISPRi has been utilized to diminish SNCA overexpression, hence decreasing aggregation and related toxicity (136). Novel therapeutic approaches encompass AAV-mediated CRISPR\u2013Cas systems aimed at SNCA regulatory elements and base editors intended to rectify pathogenic point mutations; nonetheless, challenges related to delivery efficiency and off-target effects persist as significant obstacles (134).\n\nCRISPR\u2013Cas9 technologies, encompassing knockout, base editing, and gene correction, facilitate our comprehension of biological mechanisms and preserve phenotypes in models pertinent to PD. They possess significant potential for future gene therapy as they target genetic factors such as LRRK2, PINK1, and PARKIN, thereby aiding in the reduction of neuroinflammation and neuronal degeneration (137\u2013140).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR/Cas in Parkinson's disease", "text": "Recently, CRISPR\u2013Cas9 genome editing has been employed in a laboratory to rectify mutations responsible for PD and to elucidate their pathogenic mechanisms. For example, employing CRISPR\u2013Cas9 to eliminate the A53T-SNCA mutation resulted in a significant reduction in both SNCA mRNA expression and its protein levels (137, 141). Beligni et al. (142) demonstrated that CRISPR\u2013Cas9 could rectify the LRRK2 p.Gly2019Ser (G2019S) mutation in patient-derived cells. This indicates its potential as a therapeutic agent to restore normal kinase function (143). A separate work employed CRISPR\u2013Cas9 to entirely eliminate PINK1 in THP-1 cells, enhancing our understanding of PIN1\u2032s function in maintaining mitochondrial integrity and the progression of PD (142). A recent work applied by (142) addresses the role of PINK1 in PD by employing CRISPR\u2013Cas9 to generate PINK1 knock-out THP-1 cells. The PINK1 knockout exhibited diminished mitochondrial complex I activity alongside alterations in Tau protein expression patterns. In PINK1 knock-out macrophages, pro-inflammatory cytokines such as IL-6, IL-1\u03b2, and IL-23 were raised, while the lipid droplet count per cell decreased, indicating mitochondrial dysfunction and immune system pathological alterations (142).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR/Cas in Parkinson's disease", "text": "Tokuda et al. expanded upon these findings by employing HDR to correct LRRK2 G2019S and truncate the LRRK2 kinase domain in marmoset embryonic stem cells and iPSCs. This rendered non-human primate cellular models applicable (144). In pertinent research, elevating glial cell line-derived neurotrophic factor (GDNF) levels in MPTP-treated neuronal cells via CRISPR\u2013Cas9 was associated with reduced monoamine oxidase B (MAO-B) activity, diminished production of reactive oxygen species, and enhanced cell survival under neurotoxic stress (145). The in vitro applications of CRISPR\u2013Cas9, including gene deletion, point mutation correction, gene knockout, and gene upregulation, demonstrate its versatility and efficacy as a tool for investigating the biology of PD and developing gene-based therapies.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR/Cas in Parkinson's disease", "text": "Researchers have employed CRISPR\u2013Cas9 to develop large-animal models (in vivo) of PD by introducing deleterious SNCA mutations into existing brain systems (146). Zhu et al. (146) employed CRISPR\u2013Cas9 and SCNT to generate Guangxi Bama minipigs with the PD-associated E46K, H50Q, and G51D mutations in SNCA. Despite the presence of mutant SNCA in the early animals, they exhibited neither dopaminergic neurodegeneration nor Lewy body pathology at 3 months, highlighting the potential and limitations of this large-animal paradigm. Employing dCas9, CRISPRi effectively reduced SNCA mRNA and its protein levels in animal models (147). PINK1 and PARKIN constitute a pivotal signaling axis that significantly regulates the mitochondrial autophagy process in dopaminergic neurons. In PINK1 null Drosophila, the silencing of the ubiquitin c-terminal hydrolase L1 (UCHL1) gene using RNA interference can ameliorate the pathology associated with PD. The absence of UCH deubiquitination enhances mitochondrial autophagy by stimulating the expression of AMP-activated protein kinase (AMPK) and Unc-51\u2013like autophagy-activating kinase 1 (ULK1) (148).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR/Cas in Parkinson's disease", "text": "Researchers have developed the mitochondrial complex I (MCI)-Park model in mice using CRISPR Cas9 to generate dopamine neurons deficient in NADH dehydrogenase [ubiquinone] iron\u2013sulfur protein 2 (NDUSF2), which encodes mitochondrial complex I. Mice deficient in NDUSF2 exhibited neurodegenerative alterations (149). Point mutations in the vacuolar protein sorting 35 gene (VPS35) are linked to autosomal dominant late-onset PD (PARK17). The homozygous deletion of Vps35 using CRISPR Cas9 led to a survival disadvantage and a marked decrease in dopamine release in the caudate putamen of adult homozygous Vps35 mutant mice (150). Peroxidized phospholipids resulting from ferroptosis have been linked to the beginning of certain PD. Patatin-like phospholipase domain-containing protein 9 (PNPLA9), functioning as a hydrolytic enzyme, exhibits a preference for hydrolyzing peroxidized phospholipids. In mice, the ablation of PNPLA9 led to the advancement of Parkinsonian motor abnormalities and the accumulation of peroxidized phospholipids (151). This represents a promising method for gene repression in vivo PD models.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR/Cas in Parkinson's disease", "text": "Researchers have investigated the application of CRISPR techniques to activate genes. Narv\u00e1ez-P\u00e9rez et al. employed a pseudo-lentivirus to deliver a CRISPR/SAM system to rat astrocytes, enhancing the expression of tyrosine hydroxylase and resulting in dopamine production. When these altered astrocytes were introduced into a rat model exhibiting 6-hydroxydopamine (6-OHDA)-induced injury, they improved motor function (138). This demonstrated an innovative approach to treating PD. Increased CRISPR-Cas9 genome editing has produced mouse models that eliminate NDUSF2 (mitochondrial complex I) or knock out VPS35 and Pnpla9 (148\u2013152). This induces neurodegeneration and motor dysfunction akin to the manifestations of PD. Conversely, the ablation of PRKN, PINK1, and DJ-1 thrice in murine models did not induce neurodegeneration, indicating that some species exhibit greater resilience and that the genetic modeling of PD is complex (153).\n\niPSC models derived from individuals with PD are valuable for investigating genetic variations associated with the disorder, such as LRRK2, SNCA, PARKIN, and ubiquinol\u2013cytochrome c reductase core protein 1 (UQCRC1) (154\u2013156). Neural stem cells (NSCs) harboring the LRRK2 p.G2019S mutation are predisposed to proteasome stress, nuclear envelope dysfunction, clonal proliferation issues, and challenges in neural differentiation (157).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR/Cas in Parkinson's disease", "text": "Moreover, rectifying this mutation using knock-in techniques restores the phenotype to normal, indicating that nuclear morphology may serve as a biomarker for PD. LRRK2 p.G2019S disrupts mitochondrial mitophagy by impairing the Miro\u2013PINK1/PARKIN pathway, hence hindering cellular respiration and metabolism (158). Similarly, NSCs derived from individuals with PARK2 mutations differentiate into neurons exhibiting elevated oxidative stress, activated Nrf2 pathways, and abnormal mitochondrial morphology and dysregulated homeostasis (159, 160).\n\nResearch on dopaminergic SH-SY5Y cells with UQCRC1 mutations, observed in both familial and sporadic PD, indicates that axonal degradation occurs and mitochondrial respiratory chain function ceases. These iPSC-derived cellular models demonstrate how genetic risk factors influence pathways governing proteostasis, mitochondrial integrity, oxidative equilibrium, and cellular morphology (161). They demonstrate the utility of CRISPR\u2013Cas9-mediated gene repair or knockout in restoring cellular function, thereby enhancing the understanding of underlying mechanisms and identifying potential therapeutic targets in PD.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR/Cas in Parkinson's disease", "text": "CRISPR-modified PD organoids replicate dopaminergic neuron degeneration, oxidative stress, and \u03b1-synuclein pathology, offering a three-dimensional framework for longitudinal investigations. Three-dimensional human midbrain organoids generated from CRISPR-modified induced pluripotent stem cells augment the physiological relevance of these models. These organoids have a spatial organization akin to the human substantia nigra, with dopamine-producing neurons, astrocytes, and nascent microglial-like populations. CRISPR-induced PD mutations in organoids expedite SNCA fibrillization, impair neuromelanin production, cause lysosomal dysfunction, and lead to progressive neuronal loss, reflecting clinical illness with temporal advancement. These 3D systems enable longitudinal examination of synapse degeneration, neuroinflammation, and intercellular communication processes inadequately represented in 2D cultures (162).\n\nBiomedical applications illustrate the usefulness of CRISPR-Cas12 technology, which has shown promise in research for treating NDs, including PD. Healthcare research continues to develop and still under research, studies on the applicability of CRISPR-Cas12 for PD treatment, utilizing the existing CRISPR technology still limited.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR/Cas in Parkinson's disease", "text": "Researchers have discovered intriguing applications for CRISPR-Cas3\u2032s RNA-targeting DNA editing technologies in treating NDs, including PD. CRISPR systems based on DNA editing fail to target single-stranded RNA, but Cas13 can precisely control gene expression using RNA targets that do not alter genomic sequences (163). The application on PD and CRISPR-Cas13 still limited and shown in Table 3.\n\nPD, Parkinson's disease; CRISPR, clustered regularly interspaced short palindromic repeats; Cas9, CRISPR-associated protein 9; iPSC, induced pluripotent stem cell; KO, knockout; HDR, homology-directed repair; CRISPRi, CRISPR interference; dCas9, catalytically inactive Cas9; KRAB, Kr\u00fcppel-associated box transcriptional repression domain; ABE, adenine base editor; CBE, cytosine base editor; RNP, ribonucleoprotein complex; AAV, adeno-associated virus; HDAdV, helper-dependent adenoviral vector; ROS, reactive oxygen species; SNCA, \u03b1-synuclein gene; LRRK2, leucine-rich repeat kinase 2; GBA, glucocerebrosidase gene; PRKN (PARKIN), parkin gene; PINK1, PTEN-induced kinase 1; mtDNA, mitochondrial DNA.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas in Huntington's disease", "text": "HD is a progressive ND, an autosomal dominant disease. The condition is attributed to an enlarged cytosine-guanine-adenine (CAG) repeat in the HTT gene located on chromosome 4 (4p16.3), resulting in a mutant huntingtin (mHTT) protein characterized by an elongated polyglutamine tract (164\u2013166). The deleterious gain-of-function characteristics of mHTT initiate a series of molecular disruptions that ultimately result in the specific destruction of medium spiny neurons within the striatum. At the molecular level, mHTT impairs transcriptional control by anomalously interacting with transcription factors, including CREB-binding protein (CBP), resulting in extensive transcriptional suppression and modified neuronal identity. Moreover, mHTT disrupts axonal transport by obstructing microtubule-associated motor proteins (dynein/dynactin), leading to impaired trafficking of vesicles, mitochondria, and neurotrophic signals such as brain-derived neurotrophic factor (BDNF) (167). Individuals with optimal health typically possess 16\u201320 CAG repeats, however, those afflicted with HD may have over 40 repetitions. Extended repeats are associated with earlier onset and more severe symptoms (164). The misfolded mHTT protein forms toxic aggregates that disrupt critical cellular activities such as transcription, energy metabolism, axonal transport, and apoptosis. This ultimately results in the demise of striatal and cortical neurons. In the clinic, HD manifests in middle age with issues related to movement, cognition, and mental health (168). The studies detected the mitochondrial dysfunction is a primary characteristic, marked by diminished respiratory chain activity, compromised calcium buffering, elevated oxidative stress, and increased vulnerability to apoptosis. mHTT also undermines autophagy and proteostasis by obstructing autophagosome cargo loading, impairing ubiquitin\u2013proteasome functionality, and facilitating the formation of insoluble aggregates. Synaptic dysfunction manifests early via altered NMDA receptor activation, excitotoxicity, and compromised corticostriatal connection. Recent data indicates that non-cell-autonomous mechanisms, including as impaired astrocytic glutamate absorption and microglial activation, contribute to the acceleration of neuronal death. These interrelated pathways collectively characterize HD as a progressive multisystem condition influenced by proteotoxicity, transcriptional dysregulation, compromised energy metabolism, and synaptic susceptibility (169, 170). A 2012 meta-analysis revealed significant regional variation in the prevalence of HD, with elevated rates in Europe, North America, and Australia (5.7 per 100,000) compared to diminished rates in Asian populations (0.4 per 100,000) (171).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas in Huntington's disease", "text": "HD pathophysiology comes from the misfolding of mHTT into aggregates that impair cellular function, ultimately leading to neuronal cell death (172). The protein interferes with transcriptional regulators, leading to aberrant gene expression throughout the organism (173). Mitochondrial dysfunction, along with elevated oxidative stress, induces neuronal injury by diminishing cellular energy levels (174). Concurrent aberrant glutamate activity induces neuronal death pathways. The mHTT protein induces malfunction in the proteasome and autophagy processes, leading to the accumulation of toxic proteins (175). Neuronal damage increases due to chronic neuroinflammation caused by activated microglia and astrocytes in the nervous system (176). The prospective applications of CRISPR-Cas technology for the treatment of HD are significant due to its capacity for precise genetic material alteration. Gene therapy employing targeted CRISPR-Cas9 serves as a prospective treatment strategy for disease management by directly correcting genetic mutations at their origin (164, 177). Various studies have been demonstrated the potential of CRISR/Cas tool in HD application and theoretical aspects such as gene editing, selective targeting of mutant alleles, and excision of expanded cag repeats (178\u2013180). Shin et al. undertook a study with the goal of enhancing the specificity of alleles. They employed a customized approach using CRISPR\u2013Cas9 that selectively targeted specific alleles by modifying PAM-altering single nucleotide polymorphism (SNP). This approach focused on identifying CRISPR\u2013Cas9 sites that were unique to each patient and utilized a comprehensive understanding of the haplotype structure of the HTT gene. The objective was to deactivate the mutated HTT allele for a specific haplotype specifically (181).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas in Huntington's disease", "text": "Researchers have employed CRISPR\u2013Cas9 genome editing to create isogenic cell models (in vitro) of HD by precisely modifying the CAG repeat region in the HTT gene. CRISPR has been used to allele-specific targeting and genetic correction, Allele-specific excision of the mutant HTT, Shin employed SNP-guided dual-CRISPR\u2013Cas9 targeting to excise approximately 44 kb from the mHTT allele in patient-derived cells. This eliminated mutant mRNA and protein while preserving the wild-type allele (181). In a separate investigation, Yang et al. showed that the application of CRISPR\u2013Cas9 in HD140Q-KI mice to remove HTT in a non-allele-specific way can successfully and permanently eradicate the harmful effects of polyglutamine expansion on neurons in the adult brain. The study population showed a significant reduction in reactive astrocytes and a notable improvement in motor impairment (182). Another study utilized CRISPR to CAG-repeat excision, such as a study introduced 150 CAG repeats into exon 1 of HEK293T cells to create clones resembling the condition, perhaps facilitating the investigation of disease mechanisms (183). In pertinent research, scholars employed Cas9 ribonucleoprotein complexes and single-stranded oligodeoxynucleotide donors to create HEK293T lines containing 41, 53, and 84 CAG repeats. This facilitated the examination of alterations in cellular phenotypes associated with varying repeat lengths and contributed to the development of novel treatments (184). Another approach employed Cas9 nickases and guide RNAs to excise CAG repeats, resulting in frameshifts that activated premature stop codons and inhibited mHTT production. This illustrates how accurate genome editing could reinstate the typical allele configuration (185). In transgenic BacHD mice, Monteys et al. (186) incorporated a modified human HD allele and targeted human mHTT exon 1. BacHD animals treated with rAAV.spCas9 and rAAV.sgHD1B/i3 showed decreased HTT mRNA levels in their right hemisphere but not in their uninfected left (186). Ekman et al. (187) revealed effectively inhibited the expression of the human mutant HTT gene in R6/2 mice that carried human mHTT exon 1 with roughly 115\u2013150 CAG repeats in 2019 using AAV-mediated CRISPR-Cas9. This study showed that R6/2 mice's lifespan and motor deficits were enhanced by inhibiting HTT expression. In conclusion, it has been shown in various animal models that HD may be effectively treated by employing CRISPR-Cas9 to eliminate mHTT gene expression (187).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas in Huntington's disease", "text": "CRISPR-based gene-silencing techniques have been employed to reduce the expression of mHTT with the direct deletion of repeat expansions. CRISPRi exclusively inhibited mHTT transcription in fibroblasts derived from HD patients, without affecting wild-type HTT (188). Previous work employed CasRx (RNA-targeting Cas13) to downregulate HTT exon 1 in HEK293T cells, HD 140Q knock-in mice, and HD knock-in pigs. This decreased mHTT levels, diminished gliosis, and decelerated neurodegeneration (189). Ekman et al. (187) revealed effectively inhibited the expression of the human mHTT gene in R6/2 mice that carried human mHTT exon 1 with roughly 115\u2013150 CAG repeats in 2019 using AAV-mediated CRISPR-Cas9. This study showed that R6/2 mice's lifespan and motor deficits were enhanced by inhibiting HTT expression. It has been shown in various animal models that HD may be effectively treated by employing CRISPR-Cas9 to eliminate mHTT gene expression (187).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas in Huntington's disease", "text": "Oura et al. (190) addressed the target range limits of WT-SpCas9, scientists engineered SpCas9-NG, which detects NGN PAMs. They used SpCas9-NG to precisely contract HD CAG repeat tracts in mouse embryonic stem cells. The procedure of rectifying mutations restored distinctive characteristics as well as health conditions in both neurons derived from modified stem cells and the animals produced by these cells. This study reveals that SpCas9-NG is a powerful way to treating enlarged CAG repeats and other disease variants that are difficult to target (190). Another study via Merienne et al. (191) found the development of the KamiCas9 self-inactivating CRISPR\u2013Cas9 system, which enables high-efficiency Cas9 expression during transitory intervals. Using their technique on adult mouse neuronal and glial cells, researchers were able to inactivate the HTT gene while concurrently improving HD disease markers. The KamiCas9 system made it possible to reduce unwanted edits when evaluating human iPSC-derived cells since it had less off-target effects than typical constitutive Cas9 systems. The study shows how self-inactivating CRISPR\u2013Cas9 technologies could become a feasible therapy option for NDs (191). Ultimately, allele-specific CRISPR\u2013Cas9 editing utilizing SNP-generated PAM sites facilitated the targeted silencing of mHTT by nonsense-mediated degradation in approximately 20% of European HD patients. This demonstrated exceptional accuracy and little off-target effects following GUIDE-Seq validation. This research demonstrates the versatility of CRISPR approaches in investigating the etiology of HD and formulating novel therapies. They can be utilized for genome editing, gene silencing, and allele-specific strategies (192).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas in Huntington's disease", "text": "CRISPR-based techniques have been shown to be effective in vivo for HD by reducing levels of mHTT in the brain. Bunting and Donaldson demonstrated that employing antisense oligonucleotides to inhibit MSH3 activity reduced HTT CAG repeat expansions in HD mouse models. This indicates that gene-targeting methodologies may alter the trajectory of the disease (193). Similarly, administering CRISPR\u2013Cas9 to striatal neurons in HD140Q-knockin mice significantly reduced endogenous mHTT, inhibited aggregation formation, and ameliorated early symptoms of neurodegeneration and motor dysfunction, all without causing neuronal damage (182). In the prevalent R6/2 mouse model, CRISPR\u2013Cas9-induced deletion of HTT resulted in reduced protein inclusions, enhanced motor function, and an extended longevity. This indicates that targeted genome editing may be an effective approach for treating HD (103, 186, 187, 191).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas in Huntington's disease", "text": "Enhanced editing precision has significantly improved safety and efficacy. Base editing was utilized in vivo to disrupt caspase-6 cleavage sites within the HTT gene. This resulted in mHTT isoforms that exhibit resistance to proteolytic processing. This approach effectively reduced the detrimental aggregation of huntingtin and safeguarded against neuronal loss in the striatum and brain of HD animal models (187). In vivo investigations, encompassing antisense, CRISPR\u2013Cas9 gene disruption, and base editing, demonstrate that reducing or altering mHTT expression can decelerate the advancement of neuropathological disorders and enhance functional outcomes. This indicates that CRISPR possesses the potential to serve as a disease-modifying instrument for HD (194).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas in Huntington's disease", "text": "The integration of iPSCs with CRISPR\u2013Cas technology has advanced HD research by creating precise, isogenic models and revitalizing cells. Scientists developed homozygous HEK293T cell lines with varying amounts of CAG repeats in the HTT gene to facilitate mechanistic investigations and drug screening (184). Researchers have employed CRISPR\u2013Cas9 to rectify pathogenic CAG expansions in iPSCs, effectively reversing initial indicators of mitochondrial dysfunction in HD-derived neural stem cells, including respiratory issues, elevated reactive oxygen species, and calcium dysregulation (195\u2013198). The results indicate that the selective elimination of CAG repeats can restore mitochondrial function and maintain cellular equilibrium. CRISPR\u2013Cas9 editing facilitates allele-specific disruption, repeat excision, and gene repair in both in vitro and in vivo environments. Outstanding issues remain, including its efficacy, long-term safety, and performance on alternative genes. This gene-editing technique possesses significant potential to alter the trajectory of HD (199). Table 4 illustrates some HD applications and consequences.\n\nApplications of CRISPR\u2013Cas in Huntington's disease.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas in Huntington's disease", "text": "HD, Huntington's disease; HTT, huntingtin gene; mHTT, mutant huntingtin; CRISPR, clustered regularly interspaced short palindromic repeats; Cas9, CRISPR-associated protein 9; iPSC, induced pluripotent stem cell; PAM, protospacer adjacent motif; SNP, single nucleotide polymorphism; PAS, PAM-altering SNP; KO, knockout; HDR, homology-directed repair; RNP, ribonucleoprotein complex; AAV, adeno-associated virus; PE, prime editing; pegRNA, prime-editing guide RNA; PE-CORE, prime-editing correction of repeat expansions; ABE, adenine base editor; CBE, cytosine base editor; CAG/CAA, trinucleotide codons encoding glutamine; 2D, two-dimensional cell culture; 3D, three-dimensional culture; ROS, reactive oxygen species.\n\nRecent advancements indicate that CRISPR-Cas13 systems may be effective in treating HD by targeting RNA transcripts associated with the disorder. Morelli and Wu (152) conducted a notable study that advanced an enhanced CRISPR-Cas13d system known as Cas13d-CAGEX. This treatment led to enduring enhancements in motor function, reduced neuronal degeneration, and sustained therapeutic effects with minimal adverse effects observed for up to 8 months post-injection (200).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas in Huntington's disease", "text": "The efficient application of CRISPR-Cas13d demonstrates its considerable potential as an RNA-targeting tool for HD and possibly other dominant genetic disorders. Cas13d represents a safer alternative for prolonged gene regulation as it specifically inhibits pathogenic transcripts without altering genomic sequences. This differs from DNA-targeting techniques. Numerous studies employ CRISPR-Cas13 in Huntington's disease, although few utilize CRISPR-Cas12. No studies have utilized CRISPR-Cas3 in HD research, indicating that these methods remain underexplored for neurodegenerative therapeutics.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas in amyotrophic lateral sclerosis", "text": "ALS is a progressive ND characterized by the degeneration of both upper and lower motor neurons. This results in muscular weakening, atrophy, paralysis, and ultimately respiratory failure (201\u2013203). Epidemiological studies indicate that incidence significantly varies by geography, with rates between 0.26 and 23.46 cases per 100,000 person-years, whereas prevalence ranges from 1.57 to 11.80 instances per 100,000 individuals across various countries (203, 204). Approximately 75% of ALS cases occur sporadically, although around 25% are familial and associated with over 135 genetic loci (205\u2013208). A defining characteristic of ALS is the cytoplasmic mis localization and aggregation of TDP-43, which impairs RNA splicing, transport, and stability, resulting in extensive transcriptome dysregulation (202, 209). Mutant superoxide dismutase 1 (SOD1) and fused in sarcoma (FUS) intensify cellular stress through protein misfolding, oxidative injury, and impaired DNA damage repair. Intrinsic neuronal abnormalities are exacerbated by non-cell autonomous mechanisms, such as microglial activation, astrocytic glutamate toxicity due to diminished excitatory amino acid transporter 2 (EAAT2) expression, mitochondrial dysfunction, and compromised axonal transport. These mechanisms collectively begin a self-sustaining loop of neuroinflammation, energy depletion, and synaptic degeneration, ultimately resulting in motor neuron death (210, 211). The hexanucleotide (GGGGCC) expansion in C9orf72 is the predominant mutation among these. It induces toxic RNA foci and dipeptide repeat protein accumulation in around 40% of familial cases and 7% of sporadic cases (212\u2013214). TAR DNA-binding protein (TARDBP) mutations disrupt TDP-43 protein homeostasis (4% familial, 1% sporadic) (215, 216), while SOD1 mutations lead to oxidase dysregulation and protein aggregation (217, 218) in 12% of familial cases and 1%\u22122% of sporadic cases. Moreover, alterations in FUS, heterogeneous nuclear ribonucleoprotein A1 (HNRNPA1), profilin 1 (PFN1), and kinesin family member 5A (KIF5A) influence RNA metabolism, axonal transport, and cytoskeletal architecture, ultimately resulting in motor neuron degeneration in ALS (219\u2013223).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas in amyotrophic lateral sclerosis", "text": "The CRISPR\u2013Cas genome-editing technology offers a significant method for developing targeted ALS therapies through precise genetic modifications (224). The protentional applications of CRSPR/Cas in ALS disease was gene silencing and correction, disease modeling, gene therapy delivery (16, 224\u2013226) (Figure 3). In vivo CRISPR techniques have been employed to directly target pathogenic ALS mutations in animal models, yielding quantifiable functional improvements and molecular restoration. AAV-mediated CRISPR\u2013Cas9 targeting of mutant SOD1 decreased mutant SOD1 levels of proteins in the spinal cord and enhanced motor function and survival in transgenic mice, illustrating that somatic gene disruption can alleviate disease symptoms. Likewise, the excision of the pathogenic GGGGCC hexanucleotide repeat expansion (HRE) in C9orf72 by paired CRISPR\u2013Cas9 guides diminished RNA foci and dipeptide repeat proteins, therefore ameliorating molecular abnormalities in various transgenic mice models (227). Recently, RNA-targeting CRISPR systems (Cas13/CasRx) designed for high fidelity have been utilized in vivo to diminish harmful repetitive RNAs and subsequent dipeptide repeat proteins (DPR) toxicity, providing a non-DNA-cleaving therapeutic approach that may mitigate irreversible on-target genetic concerns. These investigations demonstrate the feasibility of in vivo CRISPR interventions to reverse fundamental ALS processes, while also emphasizing the problems of delivery, immunogenicity, and product purity (editing byproducts) that persist for clinical application (227).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas in amyotrophic lateral sclerosis", "text": "Schematic illustration shows the CRISPR-Cas9 uses in the investigation and treatment of Amyotrophic Lateral Sclerosis (ALS). (A) Traditional disease modeling involves obtaining diseased brain tissue biopsies to establish patient-specific primary cell lines. (B) CRISPR/Cas9-based disease modeling utilizes somatic reprogramming to create iPSCs, followed by gene editing (e.g., homologous recombination or deletion of G4C2 repeats) to investigate phenotypic changes in differentiated neurons. The figure was designed with BioRender (http://biorender.com/). CRISPR, clustered regularly interspaced short palindromic repeats; Cas9, CRISPR-associated protein 9; ALS, amyotrophic lateral sclerosis; iPSCs, induced pluripotent stem cells; qPCR, quantitative polymerase chain reaction ; G4C2, guanine\u2014cytosine hexanucleotide repeat; MiSeq, Illumina sequencing platform.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas in amyotrophic lateral sclerosis", "text": "In vitro CRISPR modification of human cells, particularly patient-derived fibroblasts and iPSC-derived motor neurons, has proved essential for elucidating ALS pathophysiology and confirming targets. The CRISPR\u2013Cas9-mediated removal or allelic modification of C9orf72 repetitions in patient-derived iPSC neurons diminishes RNA foci and DPR accumulation, while reinstating nucleocytoplasmic transport and cellular survival. CRISPR-induced knockouts of SOD1, FUS, or TARDBP in human cellular models replicate key mitochondrial, RNA metabolism, and proteostasis deficiencies, facilitating mechanistic comparisons among genotypes. The in vitro application of RNA-targeting Cas13 devices enables the direct reduction of hazardous repetitive RNAs in patient cells, facilitating swift assessments of guide efficacy and collateral activity prior to in vivo experimentation. Cell-based research is essential for determining dosage, assessing off-target effects, and developing biomarkers (228).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas in amyotrophic lateral sclerosis", "text": "CRISPR\u2013Cas9 technology has demonstrated its ability to rectify mutations associated with ALS in iPSC in vitro models derived from patient cells. For example, employing Cas9 along with a co-transfected donor template to rectify the SOD1 A272C mutation in fibroblast-derived iPSCs restored the wild-type sequence (229). Similarly, rectifying the SOD1 E100G mutation in iPSCs using homology-directed repair not only increased the quantity of motor neurons, the size of the soma, and the formation of neurites, but also diminished cell mortality post-differentiation (230). In C9orf72-associated ALS, employing CRISPR\u2013CasRx to diminish sense and antisense repeat transcripts in HEK293T cells and patient-derived iPSC neurons decreased RNA foci and dipeptide repeat proteins, thereby providing protection against glutamate-induced excitotoxicity. Furthermore, utilizing Cas9 HDR to accurately excise the deleterious HRE from C9orf72 maintained the gene's integrity, restored expression and methylation levels to baseline, reduced intron retention, and eliminated detrimental repeat artifacts in motor neurons derived from iPSCs. All of these in vitro experiments demonstrate that CRISPR-based editing is a versatile and promising approach to address the primary genetic causes of ALS (231). The generation of SOD1+/A272C iPSCs and FUS+/G1566A iPSCs from primary fibroblasts of two ALS individuals, one possessing a heterozygous A272C mutation in the SOD1 gene and the other a heterozygous G1566A mutation in the FUS gene. The CRISPR-Cas9 approach effectively rectified the SOD1+/A272C and FUS+/G1566A mutations in ALS iPSCs utilizing a donated plasmid or single-stranded oligodeoxynucleotide (ssODN) as the repair template, resulting in restored ALS iPSCs that exhibited normal pluripotency. Furthermore, SOD1+/A272C and repaired iPSCs were prompted to differentiate into motor neurons, resulting in the identification of 899 aberrant transcripts using RNA sequencing. This work may assist in identifying biomarkers and pathways for ALS and provides novel evidence for ALS therapy (229).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas in amyotrophic lateral sclerosis", "text": "Base editing and CRISPR\u2013Cas9 are two efficacious techniques for investigating and addressing ALS, particularly in vivo animals. An essential approach involves utilizing AAV9 to administer CRISPR components to transgenic ALS mice with the gain-of-function SOD1 G93A mutation. Initial endeavors to employ S. aureus Cas9 (SaCas9) and gRNAs for enhancing survival and muscle preservation yielded limited success; hence, alternative strategies were explored. Base editors, including cytidine or adenine deaminase linked to a nickase Cas, enable accurate single-nucleotide modifications without generating double-strand breaks an advantageous characteristic for rectifying or deactivating point mutations in familial ALS. A pivotal proof-of-concept study using adenine base editing to inactivate a mutant SOD1 allele in vivo, enhancing motor performance in an ALS mouse model. This demonstrated that base editing can effectively diminish hazardous protein levels and yield phenotypic advantages while circumventing several consequences associated with DSBs. Base editing is limited by editable sequence contexts and bystander edits, while the delivery of base-editor cargo to motor neurons presents a translational bottleneck; yet, these results demonstrate substantial potential for therapy for point-mutation ALS. Researchers employed a dual AAV9 split-intein methodology to circumvent the limitations of vector size, enabling the creation of specific nonsense mutations in SOD1 (227, 232, 233). We must exercise caution in simultaneous initiatives aimed at targeting FUS and TARDBP mutations to avoid the detrimental consequences of entirely eliminating a gene (234, 235). Despite these challenges, employing CRISPR to reduce mutant SOD1 in G93A-SOD1 mice significantly improved their motor functions, postponed illness onset by 37%, and extended their longevity by 25%. This indicates that gene editing possesses therapeutic potential (227).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas in amyotrophic lateral sclerosis", "text": "CRISPR\u2013Cas9 has been effectively employed to rectify faulty expansions in the C9orf72 gene, the most common genetic contributor to ALS. This pertains to SOD1-related ALS. Multiple groups have utilized paired guide RNAs around the hexanucleotide repeats to induce excision by NHEJ in iPSCs derived from patients. They have successfully eliminated up to 11% of the target DNA. Unanticipated deletions occurred; yet, C9orf72 expression remained constant (236\u2013238). To enhance fidelity, normal-length repeats were included by HDR techniques, restoring typical methylation and transcription profiles. Neurons derived from these modified iPSCs exhibited a reduced number of RNA foci, diminished dipeptide repeat protein aggregates, and enhanced stress resilience (231, 239). SOD1 participates in the detoxification of superoxide radicals, and mutations in SOD1 lead to enhanced harmful activity (209, 240). The G93A-SOD1 mouse model possesses roughly 25 tandem repeat copies of the human SOD1G93A transgene, demonstrating signs of amyotrophy and dyskinesia observed in ALS patients (241). CRISPR modification of iPSCs has emerged as a fundamental element in ALS research. By producing isogenic iPSC pairs (mutant vs. corrected control), researchers can associate genotype with phenotype while minimizing background interference. The differentiation into motor neurons and three-dimensional spinal/brain organoids facilitates the replication of TDP-43 mislocalization, RNA foci, DPR synthesis, mitochondrial dysfunction, axonal transport abnormalities, and non-cell-autonomous contributions from astrocytes and microglia found in ALS. These CRISPR-iPSC platforms are extensively utilized for mechanistic analysis, high-throughput drug screening, and preclinical assessment of gene-editing methodologies, including the evaluation of AAV/SNP-guided allele selectivity, base editors, and Cas13 guides. iPSC platforms are essential for precise off-target and transcriptome-wide RNA-editing evaluations before animal or clinical use (226).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas in amyotrophic lateral sclerosis", "text": "Recent advancements in CRISPR-based RNA-targeting methodologies suggest their potential utility in the treatment of ALS. Zhang et al. (2021) in their review article discussed an enhanced variant of CRISPR-Cas13d, termed CasRx, which can degrade both sense and antisense C9ORF72 repeat RNAs while preserving normal gene expression. Utilizing AAV vectors, CasRx was administered to HEK cells and patient-derived iPSC neurons in vitro, as well as to transgenic animal models in vivo. This resulted in significant reductions in RNA foci, DPRs, and glutamate-induced excitotoxicity (163). In another study, researchers employed Cas13 and Cas7-11 effectors to target ataxin-2 in models of TDP-43 proteinopathy. Direct injection of the Cas13 system into mice effectively reduced TDP-43 aggregation and the location of stress granules. This resulted in improved behavior, increased longevity, and reduced neurodegeneration in the mice. In comparison to previous Cas7-11 tools (242), Cas13 variants exhibiting enhanced fidelity demonstrated superior transcriptome specificity. The results indicate that RNA-targeting CRISPR systems possess significant potential as a therapeutic approach for ALS, as they can target critical disease characteristics. Table 5 illustrates CRISPR techniques in ALS.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-Cas in amyotrophic lateral sclerosis", "text": "ALS, amyotrophic lateral sclerosis; CRISPR, clustered regularly interspaced short palindromic repeats; Cas, CRISPR-associated (e.g., Cas9, Cas13); SOD1\u2227G93A, SOD1 glycine-to-alanine substitution at codon 93 (G93A) model; iPSC, induced pluripotent stem cell; MN(s), motor neuron(s); HDR, homology-directed repair; KO, knock-out; KI, knock-in; KD, knockdown; CRISPRi, CRISPR interference; ABE, adenine base editor; CBE, cytosine base editor; DSB(s), double-strand break(s); AAV, adeno-associated virus (e.g., AAV9; AAV-PHP variants); RNP, ribonucleoprotein (Cas protein + guide RNA complex); ASO, antisense oligonucleotide; RNAi, RNA interference; DPR(s), dipeptide repeat protein(s); CasRx/Cas13d, Cas13d (also called CasRx) RNA-targeting CRISPR effector; PE, prime editing; pegRNA, prime-editing guide RNA; PBS, primer-binding site; PAM, protospacer adjacent motif; v-ATPase, vacuolar-type H+-ATPase; CNS, central nervous system; LNP(s), lipid nanoparticle(s).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "The efficient and selective delivery to certain cell types is the primary practical limitation for CNS-targeted CRISPR therapies. AAV vectors provide strong neuronal tropism, sustained transgene expression, and established manufacturing processes, rendering them the primary tool for in vivo CNS editing; however, their restricted packaging capacity hinders the delivery of larger editors (such as prime editors and certain base-editor constructs), and continuous Cas expression from AAV may elevate off-target activity and immunogenicity (243). Lipid nanoparticles (LNPs) offer a supplementary, transient-expression approach for delivering mRNA or RNPs, thereby minimizing the editor's exposure duration. Recent advancements in LNP formulations have facilitated systemic and targeted delivery in preclinical models; however, efficient BBB penetration and neuronal specificity continue to be areas requiring further optimization (244). Engineered extracellular vesicles/exosomes represent a novel option characterized by low immunogenicity and potential for targeted cell delivery; ongoing research emphasizes scalable loading techniques, surface modification for tropism, and comprehensive biodistribution characterization (245). A pragmatic hybrid approach such as compact editors encapsulated in AAV for localized distribution, LNPs for temporary systemic administration, or exosome-based vehicles.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "Investigating the delivery of the CRISPR\u2013Cas system for the treatment of NDs necessitates the resolution of significant technical challenges to ensure safety and efficacy (Figure 4). The CNS is safeguarded by the BBB, which serves as a selective permeability barrier to limit the entry of chemicals from the bloodstream into brain tissue (246, 247). The protective function of the BBB presents significant challenges for administering CRISPR\u2013Cas9-based therapies for neurodegenerative diseases. The dimensions of the CRISPR\u2013Cas9 complex hinder its ability to traverse the BBB due to its substantial size. Passive diffusion seems to be the only process by which small lipophilic molecules traverse the barrier, but active transport is essential for the movement of bigger entities. The delivery of CRISPR\u2013Cas9 tools encounters obstacles resulting from their dimensional size (248, 249). The BBB inhibits the ingress of external substances into the brain, encompassing pathogens, and toxins (250). The CRISPR\u2013Cas9 complexes are unable to traverse the BBB due to its specialized mechanisms for critical nutrients and chemicals, which do not accommodate the transit of big macromolecules. Owing to the restrictive properties of the BBB, researchers must develop specialized delivery systems that facilitate the successful passage of gene-editing tools past this barrier (251).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "The schematic diagram shows the analysis of the challenges in implementing CRISPR\u2013Cas9 technology for neurological illnesses and approaches to mitigate these challenges. The figure was designed with BioRender (http://Biorender.com/). CRISPR, clustered regularly interspaced short palindromic repeats; Cas9, CRISPR-associated protein 9.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "Researchers employ AAVs as vectors for delivering CRISPR components due to their outstanding efficacy and low safety risks. AAVs possess limited capacity for packing materials, complicating the simultaneous delivery of Cas9 proteins and their guide RNAs. The utilization of AAV vectors for the delivery of CRISPR components encounters numerous constraints, as capsids elicit immunological responses, leading to off-target mutations, and AAV distribution may be detrimental to cellular DNA (252, 253). Although viral vectors are successful, they elicit immunological responses that may provoke undesirable consequences in the body. Research has demonstrated that viral vectors induce unwanted mutations and severe immune responses, rendering them unacceptably hazardous for live human use (254). Non-viral delivery approaches utilizing nanoparticles offer safer options, although they encounter distinct challenges. The effectiveness of BBB transport and neuron-specific targeting continue to be significant obstacles hindering successful research. The reliable transport and regulated release of CRISPR components require enhanced optimization for success (255, 256).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "The strategies overcome challenges for BBB, engineered variants of AAVs have demonstrated enhanced capabilities for penetrating the BBB, as researchers have tailored them for gene therapy applications aimed at treating neurodegenerative diseases (Figure 4). Researchers have modified AAV capsid proteins to enhance their affinity for endothelium receptors, hence facilitating receptor-mediated transcytosis. Experimental results demonstrated that the Cre recombinase-based AAV targeted evolution (CREATE)-derived AAV-PHP.B variant surpassed AAV9 in CNS transduction efficacy by a factor of 40 following intravenous administration in adult mice. Researchers enhanced AAV-PHP.eB to achieve improved CNS transduction and BBB penetration efficacy following its first creation. Administration of AAV-PHP.eB via the intravenous method in adult mice resulted in neuronal transduction of 69% of cortical cells and 55% of striatal cells with a vector genome dosage of 1 \u00d7 1011. Engineered AAVs can function as effective non-invasive delivery methods for genetic therapies targeting NDs (257, 258).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "Researchers are developing nanoparticles to stabilize CRISPR\u2013Cas9 components and ensure their safe transit across the BBB. The administration of mRNA-based therapeutics to the brain is effectively achieved via LNPs that facilitate the passage of genetic material across BBB (249, 251, 259, 260). The functionalization of LNPs with transferrin enhances their ability to traverse the BBB. Brain endothelial cells exhibit elevated numbers of transferrin receptors, which utilize their inherent iron transport capability for receptor-mediated transcytosis of transferrin-conjugated nanoparticles. This technique enables the delivery of medicinal medicines to the brain. A document delineates nanoparticle targeting of the brain by transferrin nanoparticle conjugation (261). Exosomes, as natural extracellular vesicles, exhibit enhanced brain-targeting capabilities when suitably modified. The incorporation of transferrin or specific peptides onto exosomes facilitates their binding to receptors located on the BBB for targeted CNS delivery. Modified exosomes displaying transferrin on their surface enhance the accumulation of these vesicles in brain tissues, facilitating more effective administration of therapeutic agents. Ongoing research indicates that exosomes exhibit enhanced efficacy for CNS distribution when modified with ApoB to interact with low-density lipoprotein receptors (LDLR) (262).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "Therapeutic medicines are delivered via intracerebral injection straight into specific brain regions for optimal distribution. This delivery technique improves the localized dispersion of CRISPR\u2013Cas9 components, potentially increasing their concentration at the target region. The administration of therapeutic agents by intracerebral delivery enables healthcare professionals to accurately inject these substances into targeted brain regions, ensuring optimal distribution. This technique facilitates local delivery of CRISPR\u2013Cas9 components, enhancing distribution and potentially increasing their concentration at the target region (263). Another study employed biodegradable PEGylated nanocapsules to provide Cas9-sgRNA ribonucleoproteins (RNPs) during intracerebral injections into the mouse striatum. This therapeutic approach achieved precise genome alteration in medium spiny neurons with minimal stimulation of glial cells, ensuring safety (264). The direct delivery methods present intriguing strategies for CRISPR\u2013Cas9-mediated therapeutic interventions in NDs. Further issues regarding surgical invasiveness, tissue damage mitigation, and precise targeting criteria must be addressed to improve the clinical application of these techniques.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "Another strategy was utilized PepFect14, in conjunction with a portion of the rabies virus glycoprotein, was utilized to create cell-penetrating peptides that associate with CRISPR\u2013Cas9 components for the targeting of neuronal cells. The approach emphasizes receptor-mediated transcytosis to facilitate the passage of gene-editing tools across the BBB for transport (260, 265\u2013267).\n\nLong-term safety issues associated with CRISPR encompass not just off-target mutations but also significant on-target deletions, insertions, inversions, chromosomal translocations, and other intricate structural changes that conventional PCR techniques may overlook. Identifying these occurrences necessitates genome-wide, high-resolution methodologies, including long-read sequencing and capture-based assays. Strategies for risk mitigation encompass the temporary administration of CRISPR components (RNP or mRNA), the utilization of high-fidelity or double-strand break-free editors (nickase, base, or prime editors), and meticulous optimization of guide RNA, bolstered by empirical off-target analysis. Comprehensive preclinical genotoxicity assessment and prolonged molecular and clinical surveillance, with established cessation criteria, are crucial for initial human studies (268).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "The occurrence of off-target cleavage has the potential to lead to unfavorable mutations and cytotoxicity, which may disturb the regular functioning of genes (269). The off-target effects linked with sgRNA and Cas9 pose significant concerns regarding the CRISPR\u2013Cas method (Figure 4). While Cas9 can tolerate variations in the distribution and quantity of mismatches in DNA, the brevity of sgRNA (about 17\u201324 bp) heightens the likelihood of off-target effects. The presence of similar sequences or sequences with minor variations elsewhere in the genome can lead to mistargeting due to slight variances in the sgRNA (270\u2013272). Mis-targeting may result in severe implications for patients, including the development of oncogenic mutations (273). CRISPR\u2013Cas9 randomly alters tumor suppressor genes and activates oncogenes, potentially leading to cancer development. Off-target effects occur with a significant frequency of 50% every study, raising considerable concerns for therapeutic application (274). The incidence neuronal dysfunction of inadvertent genetic alterations due to off-target impacts jeopardizes critical neurological genes, potentially leading to neuronal cell disorders. Off-target action occurs regularly, highlighting the necessity for scientists to implement precise targeting techniques to mitigate these adverse outcomes (275, 276).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "Creating sgRNA with high specificity and an allowable base mismatch to another sequence reduces the likelihood of mis-targeting sgRNA. The insufficient understanding of CNS biomarkers is a significant obstacle to effective cell targeting and signaling responses, yet advancements in CSF biomarkers, brain imaging, and varied vector modeling are expected to enhance specificity (277).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "An optimal sgRNA length, minimal mismatches, and optimum GC content can enhance the targeting efficiency of CRISPR-Cas9 technology. Ren et al. (269) demonstrated that sgRNA sequences ranging from 17 to 20 bp have comparable mutation efficiencies. Fu et al. (278) discovered that shortened gRNA sequences of 17 or 18 bp enhance specificity, while sgRNA sequences with a GC concentration of 40%\u221260% are advantageous for augmenting target effects (269, 279). Another strategy to reduce off-target effects is to select a low GC content of sgRNA; decreased GC content correlates with diminished stability of nucleic acid hybrids. Besides screening for distinct genomic sequence targets, modifying Cas9 enzymes to create nickases and fusing dCas9 with the FokI catalytic domain has diminished the incidence of off-target effects in CRISPR\u2013Cas9 applications (273). The sgRNA specificity, particularly with fewer than three mismatches within 10\u201312 base pairs close to the PAM site and more than two mismatches with the off-target site, can enhance targeting efficacy (275, 280). Additionally, preventing the emergence of sgRNA bulges at 12 bp next to the PAM site, generating chimeric sgRNA, substituting portions of the RNA sequence with a DNA sequence, and employing numerous sgRNAs directed at the same gene may also diminish off-target effects (280\u2013283). Various variants of CRISPR\u2013Cas9 could affect and raise the target accuracy of CRISPR technology, engineered a high-fidelity version 1 of SpCas9 (SpCas9-HF1) by Kleinstiver et al. (284). In comparison to wildtype SpCas9, SpCas9-HF1 has similar targeting efficacy. Moreover, SpCas9-HF1 possesses the ability to diminish non-specific DNA interactions, rendering genome-wide break capture and focused sequencing techniques incapable of identifying all or nearly all target activities (284). Developed an \u201cenhanced specificity\u201d SpCas9 variant 1.1 (eSpCas9 1.1). In comparison to wildtype SpCas9, eSpCas9 1.1 demonstrates robust on-target cleavage and significantly reduced effects beyond the target (285). Developed an innovative hyper-accurate CRISP/Cas9 variant (HypaCas9) that demonstrated elevated genome-wide specificity without compromising on-target activity. Moreover, SpCas9 nickase, Sniper-Cas9, and xCas9 variants can reduce off-target effects in genome editing to differing extents (276, 286).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "Both base editing and prime editing variants of CRISPR based genome editing technologies facilitate precise DNA changes by eliminating the formation of DSBs, hence mitigating the risk associated with unintended insertions or deletions (indels). Effective methodologies are now available to mitigate the unexpected consequences that conventional CRISPR\u2013Cas9 editing may induce in DNA sequences (287). Base editors provide genomic location-specific DNA base changes directly and irreversibly, while avoiding the creation of double-strand breaks. The approach mitigates the danger of generating indels due to its architecture. The research indicates that base editors induce unintended alterations in DNA and RNA due to their off-target activity. Scientific research indicates that deaminase enzymes employed in CBEs and ABEs result in significant unintentional genomic modifications due to their interaction with various DNA bases (287, 288). Investigative research indicates that prime editing effectuates precise modifications via an alteration of reverse transcriptase and a compromised Cas9 endonuclease, directed by pegRNA (289). A research study demonstrated the efficacy of prime editing in introducing the AD-associated APOE4 mutation into human induced iPSCs with high precision and productivity, resulting in editing success rates of 28.2% in iPSCs and 42.3% in HEK293T cells. The approach exhibited no indication of significant unwanted direct alterations at the target location and no remarkable off-target modification effects, so rendering it appropriate for development (99).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "The CRISPR\u2013Cas9 genome editing technique encounters significant challenges due to off-target effects, which can result in unintended alterations to the genetic material. Bioinformatics tools have been created to enhance sgRNA design and assess off-target activity. DeepCRISPR and GUIDE-seq serve as crucial tools that augment CRISPR-based applications by enhancing their specificity and safety (50, 290).\n\nDeepCRISPR employs deep learning technology to create an optimization framework that forecasts target-binding efficacy and undesirable DNA interactions of sgRNA. The integration of nucleotide sequences and epigenetic characteristics across several human cell types via DeepCRISPR yields a comprehensive assessment of sgRNA efficacy. This method enables scientists to identify sgRNAs that demonstrate robust on-target efficacy while minimizing off-target effects, thereby enhancing the precision of CRISPR\u2013Cas9 editing (291). GUIDE-seq is an experimental method that enables genome-wide identification of double-stranded breaks caused by CRISPR-Cas nucleases through a sequence-based technique. The sequencing of double-stranded oligodeoxynucleotides incorporated into double-strand breaks enables GUIDE-seq to produce comprehensive off-target site data. The procurement of this data allows researchers to observe unintentional genetic alterations that transpire throughout CRISPR-based initiatives (50, 292).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "Both the adaptive and innate immune responses present significant challenges for repeated or systemic CRISPR therapy. Documented pre-existing humoral and cellular immunity to AAV capsids and frequently utilized Cas orthologs (SpCas9, SaCas9) can neutralize vectors or induce T-cell-mediated clearance of edited cells; additionally, sustained expression of bacterial nucleases extends the period for immune recognition (293). Effective mitigation strategies, substantiated by preclinical and early clinical studies, encompass the utilization of less common AAV serotypes or engineered capsids, transient immunosuppression during dosing, the administration of editors as mRNA or RNPs (via LNPs or alternative transient carriers) to minimize antigen exposure, and the creation of Cas variants featuring diminished immunogenic epitopes or humanized sequences. Thorough immune profiling in pertinent animal models and patient samples is essential for formulating appropriate therapeutic regimens (294).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "Importantly, although pre-existing humoral and cellular immunity to commonly used Cas orthologs (e.g., SpCas9/SaCas9) has been documented in human populations (295), published clinical experience with CRISPR\u2013Cas9 therapies to date has not reported clinically meaningful Cas9-directed immune toxicities (296). For example, in the first systemic in vivo CRISPR\u2013Cas9 trials targeting transthyretin (TTR) (LPN delivery of Cas9 mRNA and sgRNA; NCT04601051), reported treatment-related events were primarily transient infusion-related reactions (and transient liver-enzyme elevations in some participants), without persistent immune-mediated sequelae attributed specifically to Cas9 (297). In parallel, ex vivo CRISPR\u2013Cas9 edited autologous hematopoietic stem and progenitor cells products (e.g., Exagamglogene autotemcel) have shown safety profiles largely consistent with conditioning/transplant procedures rather than editor immunogenicity, which is biologically plausible given the transient exposure to Cas9 during cell manufacturing (298).\n\nNevertheless, ongoing and future trials, particularly those requiring CNS delivery and/or potential redosing, should continue systematic immune surveillance (anti-Cas antibodies, T-cell responses, cytokine profiles) to define risk more precisely in neurological indications.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "Individuals possessing preexisting immunity to Cas9 derived from bacterial sources such as Staphylococcus aureus and Streptococcus pyogenes diminish the efficacy of CRISPR\u2013Cas9 gene editing therapy (273, 299). The research analysis identified SaCas9 antibodies in 78% of participants and SpCas9 (S. pyogenes Cas9) antibodies in 58%. Laboratory analyses revealed that SaCas9 antibodies were present in 78% of examined donors, whilst SpCas9 antibodies were found in 67% of individuals. Research findings indicated a significant number of T cells in the adult human population that interacted with SpCas9 (300, 301). Patients' previous immune responses often swiftly eliminate cells via Cas9 expression, diminishing therapeutic efficacy and potentially inciting inflammatory reactions or negative outcomes. The resolution of this issue is crucial for physicians to appropriately utilize CRISPR\u2013Cas9 technology in medical applications (27, 300, 302).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "Furthermore, the delivery mechanisms of CRISPR systems can stimulate immunological responses and affect neural cells. Viral vectors are frequently employed to transport CRISPR constructs across the BBB (303, 304). Nonetheless, viral transmission results in prolonged CRISPR expression, which is detrimental to neuronal cells and modifies in phenotypes of neuronal (303). Furthermore, chronic production of Cas9 induces a cytotoxic immune system response that eliminates genetically modified cells. The elimination of changed neurons in the brain may result in detrimental effects, as neuronal regeneration is restricted (294). The implementation of CRISPR\u2013Cas9 technology is hindered by the immunological response triggered by the Cas9 protein or delivering vectors in the host (300). The gene therapy vectors utilizing AAV frequently employed in clinical applications elicit a cytotoxic T-cell response, impacting both their safety and efficacy results (Figure 4). AAV capsid proteins introduced into the body are subjected to natural degradation processes, enabling major histocompatibility complex MHC class I molecules to present these components on the surfaces of transduced cells. The activation of cytotoxic T lymphocytes (CTLs) transpires via this presentation pathway, leading to cellular death and a subsequent decrease in transgene expression (305\u2013307). Three categories of T-cell responses in patients are contingent upon critical factors from AAV capsids, vector dose, and the patients' immunological backgrounds, including their natural exposure to wild-type AAV (305). Comprehending and mitigating these immunological barriers is essential for improving the safety and efficacy of AAV-mediated gene treatments (306).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "Various strategies have been devised to diminish the immunogenicity of CRISPR-Cas9 systems for therapeutic applications. Acquiring CRISPR from non-pathogenic bacterial strains offers a strategy to mitigate potential risks associated with pre-existing immune responses. Studies indicate that Cas9 derived from Geobacillus stearothermophilus (GsCas9) exhibits superior stability in human plasma compared to Streptococcus pyogenes Cas9 (SpCas9), suggesting its potential as a less immunogenic method for genome editing (299, 308). Another strategy is to target modifications of Cas9 immunogenic epitopes by engineering techniques yield less detectable proteins that preserve their editing capabilities. Researchers identified significant T-cell antigenic sites in SaCas9, after which other modifications were introduced to reduce immunogenicity (302, 309). Proposed alterations in these engineered versions resulted in a diminished T-cell response while maintaining their inherent nuclease activity (310, 311). The delivery of CRISPR-Cas9 components via mRNA results in a temporary impact that mitigates immune responses, hence enhancing the safety of genome editing therapies (312). The utilization of Cas9 mRNA facilitates temporal protein expression, hence minimizing prolonged exposure and decreasing vulnerability to immune responses. Target cells efficiently acquire Cas9 mRNA and sgRNA via LNPs, which maintain transient protein expression and mitigate undesirable immunological responses (313). This approach yields superior outcomes compared to traditional viral vectors, as it minimizes genetic changes and allows for numerous distribution opportunities due to its diminished inflammatory response. Research indicates that employing LNP to deliver Cas9 mRNA facilitates genome editing throughout many administrations with less immune response upon delivery (314). Modified nucleoside addition to mRNA molecules is extensively employed by scientific researchers as an established technique for developing safer CRISPR-based therapy. Researchers found that including pseudouridine and 5-methylcytosine into Cas9 mRNA significantly diminished the immunological response, hence enhancing Cas9 protein synthesis (315). Research has demonstrated that substituting uridine with N1-methylpseudouridine in mRNA vaccines diminishes immune responses while enhancing protein synthesis (316). The therapeutic efficacy improves as these changes allow mRNA to evade immune recognition, hence reducing unfavorable immunological responses (317).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "Administering short-term immunosuppressive regimens with CRISPR\u2013Cas9 delivery may reduce immune responses; nonetheless, this strategy is contentious and requires comprehensive safety assessments (318). Research has shown that the immune system can identify and respond to CRISPR elements, resulting in inflammation and diminished effectiveness of gene editing (302). Research demonstrates that Cas9 expression in murine muscle tissue led to lymphocyte infiltration, implying an immunological response to the exogenous protein (319).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "CRISPR-based therapeutics for NDs presents significant ethical and regulatory issues, especially when utilized solely on somatic cells. Critical concerns encompass securing informed consent for permanent genomic modifications, equitable access to expensive therapies, and meticulously assessing the risks and benefits of irreversible operations in the CNS. First-in-human trials necessitate stringent supervision, particularly when employing novel delivery technologies. Despite widespread rejection of germline editing, its public discourse continues to influence trust in somatic editing. Robust community involvement, autonomous ethical assessment, and compliance with international standards are needed. Regulators are progressively mandating comprehensive preclinical evidence about safety, immunological responses, and off-target effects, in addition to dependable manufacturing and long-term monitoring strategies. Engaging bioethicists and patient representatives in trial design enhances the quality of consent and public acceptance (320).\n\nCRISPR gene-editing in neurological research raises significant ethical issues owing to its substantial influence on cognitive abilities and identity, possible unforeseen consequences, and ramifications for subsequent generations.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "Ethical considerations are crucial, considering the fragility of several patients, and acquiring informed approval. Consent may be more complex due to cognitive problems. Regulatory obstacles in guaranteeing the safety of experimental therapies, while accelerating approvals for life-threatening disorders, contribute to the complexity (321).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "The application of gene-editing technologies in the brain poses risks of off-target effects that may disrupt normal neural processes, potentially resulting in alterations in behavior or cognitive capacities. The intricacy of the human brain renders such risks perilous, since little alterations from a brain computer interface could lead to substantial neurological repercussions that may not become apparent until later in life. These warrants worry if the targeted modification involves germline cells, as they transmit genetic mutations to subsequent generations, hence amplifying the potential repercussions of such changes indefinitely (322). The intentional alteration of DNA in human embryos, gametes, and zygotes for the aim of disease prevention has sparked significant ethical controversies (323, 324). The major ethical dilemma arises from the capacity to engineer \u201cdesigner babies\u201d by DNA alteration, which pursues medicinal advantages while also striving to cultivate enhanced inherited traits, so provoking concerns over eugenics and inequality (325). Opponents argue that these methods would create communities with genetically modified elites, exacerbating existing societal disparities (323). Ethical considerations are crucial prerequisites prior to the application of CRISPR in humans. Genome editing in clinical contexts is still restricted to somatic cells, as this poses a lower risk of unethical exploitation compared to germline modification (326). Germline editing raises ethical dilemmas because of its incalculable implications over generations (323). Modifications made during the embryonic period are heritable, indicating they will be passed on to subsequent generations. Such erratic alterations pose risks that impact particular people and future generations within the human gene pool (327). The uncertainties associated with mutations and unanticipated adverse effects from these therapies raise questions regarding their safety and efficacy (328). Germline editing encounters substantial difficulties due to the complexities of obtaining consent. Individuals who have not yet been born cannot consent to genetic alterations performed prior to their birth. Scientists encounter an ethical quandary over their authority to implement irreversible alterations on individual's incapable of providing consent for such adjustments. Individuals subjected to germline editing experience a restriction of their autonomy rights, so limiting their capacity to make decisions regarding significant life-altering issues (329, 330).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "However, the practice of germline gene editing holds potential for preventing specific genetic abnormalities of the nervous system, although it engenders significant ethical issues. Ethical frameworks and comprehensive evaluations must be established to address concerns around designer babies, unforeseen consequences, issues of consent, and potential misuse prior to the ethical application of such technology.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "In addition, the ethical considerations for genome editing fundamentally differ between somatic and germline operations. Somatic editing, which focuses on non-reproductive cells, aligns with conventional therapeutic ethics when risks, benefits, and informed consent are adequately addressed (331). In contrast, germline or heritable editing presents intergenerational, societal, and justice issues, as modifications would be passed to future generations and impact individuals unable to provide consent (332). International expert organizations and policy evaluations consistently advise against permitting clinical germline heritable genome editing outside of stringent research frameworks at this time, asserting that any future deliberation necessitates exceptional evidence of safety, efficacy, and widespread societal agreement. The updated wording encapsulates these viewpoints and emphasizes the practical ramifications for researchers and sponsors: prioritize robust somatic-cell methodologies for imminent clinical application while fostering responsible, transparent research about long-term ethical considerations (332).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "A verified translational pathway commences with mechanism validation in pertinent models (isogenic human cell structures, organoids, and various animal species), coupled with stringent safety assays (empirical genome-wide off-target mapping, long-read structural variant detection, transcriptomic profiling, and immunogenicity testing) (333). Investigational New Drug/Clinical Trial Application IND/CTA enabling packages must encompass biodistribution, vector shedding, reproductive toxicity, and dose-ranging investigations. Manufacturing must adhere to good manufacturing practice (GMP) standards and incorporate confirmed potency assays associated with the editing method; for viral vectors, capsid characterization and release assays are crucial. Initial clinical trials must utilize careful, incremental dose escalation, stringent molecular monitoring of editing results in accessible tissues, clinical indicators for target engagement, and established stopping criteria. Sponsors must ultimately engage in long-term follow-up (multi-year, disease-specific) to identify delayed side effects and gather enduring efficacy data. These measures correspond with the recommendations of international commissions and recent translational assessments (334).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "The brain's CRISPR technologies raise pressing issues related to rational agency and voluntariness. The dilemma of who should provide permission for a person's personality to change arises if cognitive qualities can be altered by gene editing. This is particularly difficult when a subject is unable to make educated decisions for various reasons, such as mental impairments (335, 336).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "Ethical considerations around brain somatic gene editing arise due to the potential for unintended genetic modifications that may be inherited by subsequent generations. The targeted modifications of non-reproductive cells may indirectly induce changes in germline cells, potentially resulting in irreversible genetic abnormalities. Somatic gene editing raises ethical dilemmas about permission, as future generations are unable to provide assent for such alterations (331, 337). Although somatic edits are generally non-heritable, there is apprehension that unintended alterations might be transmitted to future generations if specific editing inadvertently impacts germline cells or influences mechanisms affecting genetic material in incompletely understood ways. This issue highlights ethical dilemmas concerning the limits of human control over gene editing, particularly in relation to somatic therapies designed to treat NDs (338\u2013340). The difficulty in obtaining consent for brain somatic gene editing remains significant within children and cognitively challenged patient demographics. An ethical dilemma arises since stakeholders involved in gene editing interventions may be unaware of the various risks and genetic alterations that could come from these procedures (341, 342). Somatic gene editing in the brain may result in unintended consequences owing to the intricacy of neural circuitry and cerebral function. The ethical dilemma pertains to the potential risks of modifying the brain's genetic composition without a comprehensive understanding of the long-term safety of such operations. Even slight, inadvertent alterations in genetic material may result in significant NDs or unexpected psychological diseases (343, 344).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "Human gene editing utilizing CRISPR has raised significant concerns, particularly regarding germline modifications (345). Current guidelines, including those from the national institutes of health (NIH) and World Health Organization (WHO), emphasize restricting CRISPR treatments to somatic cells in adults to prevent heritable alterations that may inadvertently affect subsequent generations (Figure 4) (346, 347). By focusing on non-reproductive cells, somatic gene editing mitigates any ethical dilemmas associated with unintended genetic alterations (331). The formulation of these regulations primarily relies on comprehensive risk evaluations and continuous monitoring to avert the misuse of gene-editing technology (348). The NIH has instituted stringent standards for clinical research utilizing gene-edited technology, necessitating comprehensive assessment by institutional ethics committees (344, 349). Various nations implement distinct policies; in China, research on CRISPR technology progresses with fairly lenient regulations, whereas Western countries enforce stringent restrictions governing human gene editing (350, 351).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges and strategies for applying CRISPR\u2013Cas in treating neurodegenerative disorders", "text": "Reaching an international agreement on CRISPR guidelines is vitally vital given the global nature of scientific collaboration (352). The World Medical Association (WMA) has recommendations meant to guide members in creating consistent standards for CRISPR research (353). These requirements underscore the necessity of global control in order to stop unethical behavior and guarantee responsible and safe evolution of gene-edited technology. Harmonizing guidelines will assist in stopping \u201cethics shopping,\u201d in which researchers may undertake experiments in countries with more lax laws, therefore avoiding more rigorous ethical rules elsewhere (354, 355). Moreover, a combined worldwide framework would help to improve cooperation, enable more effective monitoring, and advance equity in access to contemporary medical treatments (349). Robust ethical norms and international cooperation mitigate the risks associated with CRISPR technology and ensure its appropriate use for societal benefit.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Delivery mechanisms", "text": "The efficient and selective delivery to certain cell types is the primary practical limitation for CNS-targeted CRISPR therapies. AAV vectors provide strong neuronal tropism, sustained transgene expression, and established manufacturing processes, rendering them the primary tool for in vivo CNS editing; however, their restricted packaging capacity hinders the delivery of larger editors (such as prime editors and certain base-editor constructs), and continuous Cas expression from AAV may elevate off-target activity and immunogenicity (243). Lipid nanoparticles (LNPs) offer a supplementary, transient-expression approach for delivering mRNA or RNPs, thereby minimizing the editor's exposure duration. Recent advancements in LNP formulations have facilitated systemic and targeted delivery in preclinical models; however, efficient BBB penetration and neuronal specificity continue to be areas requiring further optimization (244). Engineered extracellular vesicles/exosomes represent a novel option characterized by low immunogenicity and potential for targeted cell delivery; ongoing research emphasizes scalable loading techniques, surface modification for tropism, and comprehensive biodistribution characterization (245). A pragmatic hybrid approach such as compact editors encapsulated in AAV for localized distribution, LNPs for temporary systemic administration, or exosome-based vehicles.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Delivery mechanisms", "text": "Investigating the delivery of the CRISPR\u2013Cas system for the treatment of NDs necessitates the resolution of significant technical challenges to ensure safety and efficacy (Figure 4). The CNS is safeguarded by the BBB, which serves as a selective permeability barrier to limit the entry of chemicals from the bloodstream into brain tissue (246, 247). The protective function of the BBB presents significant challenges for administering CRISPR\u2013Cas9-based therapies for neurodegenerative diseases. The dimensions of the CRISPR\u2013Cas9 complex hinder its ability to traverse the BBB due to its substantial size. Passive diffusion seems to be the only process by which small lipophilic molecules traverse the barrier, but active transport is essential for the movement of bigger entities. The delivery of CRISPR\u2013Cas9 tools encounters obstacles resulting from their dimensional size (248, 249). The BBB inhibits the ingress of external substances into the brain, encompassing pathogens, and toxins (250). The CRISPR\u2013Cas9 complexes are unable to traverse the BBB due to its specialized mechanisms for critical nutrients and chemicals, which do not accommodate the transit of big macromolecules. Owing to the restrictive properties of the BBB, researchers must develop specialized delivery systems that facilitate the successful passage of gene-editing tools past this barrier (251).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Delivery mechanisms", "text": "The schematic diagram shows the analysis of the challenges in implementing CRISPR\u2013Cas9 technology for neurological illnesses and approaches to mitigate these challenges. The figure was designed with BioRender (http://Biorender.com/). CRISPR, clustered regularly interspaced short palindromic repeats; Cas9, CRISPR-associated protein 9.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Delivery mechanisms", "text": "Researchers employ AAVs as vectors for delivering CRISPR components due to their outstanding efficacy and low safety risks. AAVs possess limited capacity for packing materials, complicating the simultaneous delivery of Cas9 proteins and their guide RNAs. The utilization of AAV vectors for the delivery of CRISPR components encounters numerous constraints, as capsids elicit immunological responses, leading to off-target mutations, and AAV distribution may be detrimental to cellular DNA (252, 253). Although viral vectors are successful, they elicit immunological responses that may provoke undesirable consequences in the body. Research has demonstrated that viral vectors induce unwanted mutations and severe immune responses, rendering them unacceptably hazardous for live human use (254). Non-viral delivery approaches utilizing nanoparticles offer safer options, although they encounter distinct challenges. The effectiveness of BBB transport and neuron-specific targeting continue to be significant obstacles hindering successful research. The reliable transport and regulated release of CRISPR components require enhanced optimization for success (255, 256).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Delivery mechanisms", "text": "The strategies overcome challenges for BBB, engineered variants of AAVs have demonstrated enhanced capabilities for penetrating the BBB, as researchers have tailored them for gene therapy applications aimed at treating neurodegenerative diseases (Figure 4). Researchers have modified AAV capsid proteins to enhance their affinity for endothelium receptors, hence facilitating receptor-mediated transcytosis. Experimental results demonstrated that the Cre recombinase-based AAV targeted evolution (CREATE)-derived AAV-PHP.B variant surpassed AAV9 in CNS transduction efficacy by a factor of 40 following intravenous administration in adult mice. Researchers enhanced AAV-PHP.eB to achieve improved CNS transduction and BBB penetration efficacy following its first creation. Administration of AAV-PHP.eB via the intravenous method in adult mice resulted in neuronal transduction of 69% of cortical cells and 55% of striatal cells with a vector genome dosage of 1 \u00d7 1011. Engineered AAVs can function as effective non-invasive delivery methods for genetic therapies targeting NDs (257, 258).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Delivery mechanisms", "text": "Researchers are developing nanoparticles to stabilize CRISPR\u2013Cas9 components and ensure their safe transit across the BBB. The administration of mRNA-based therapeutics to the brain is effectively achieved via LNPs that facilitate the passage of genetic material across BBB (249, 251, 259, 260). The functionalization of LNPs with transferrin enhances their ability to traverse the BBB. Brain endothelial cells exhibit elevated numbers of transferrin receptors, which utilize their inherent iron transport capability for receptor-mediated transcytosis of transferrin-conjugated nanoparticles. This technique enables the delivery of medicinal medicines to the brain. A document delineates nanoparticle targeting of the brain by transferrin nanoparticle conjugation (261). Exosomes, as natural extracellular vesicles, exhibit enhanced brain-targeting capabilities when suitably modified. The incorporation of transferrin or specific peptides onto exosomes facilitates their binding to receptors located on the BBB for targeted CNS delivery. Modified exosomes displaying transferrin on their surface enhance the accumulation of these vesicles in brain tissues, facilitating more effective administration of therapeutic agents. Ongoing research indicates that exosomes exhibit enhanced efficacy for CNS distribution when modified with ApoB to interact with low-density lipoprotein receptors (LDLR) (262).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Delivery mechanisms", "text": "Therapeutic medicines are delivered via intracerebral injection straight into specific brain regions for optimal distribution. This delivery technique improves the localized dispersion of CRISPR\u2013Cas9 components, potentially increasing their concentration at the target region. The administration of therapeutic agents by intracerebral delivery enables healthcare professionals to accurately inject these substances into targeted brain regions, ensuring optimal distribution. This technique facilitates local delivery of CRISPR\u2013Cas9 components, enhancing distribution and potentially increasing their concentration at the target region (263). Another study employed biodegradable PEGylated nanocapsules to provide Cas9-sgRNA ribonucleoproteins (RNPs) during intracerebral injections into the mouse striatum. This therapeutic approach achieved precise genome alteration in medium spiny neurons with minimal stimulation of glial cells, ensuring safety (264). The direct delivery methods present intriguing strategies for CRISPR\u2013Cas9-mediated therapeutic interventions in NDs. Further issues regarding surgical invasiveness, tissue damage mitigation, and precise targeting criteria must be addressed to improve the clinical application of these techniques.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Delivery mechanisms", "text": "Another strategy was utilized PepFect14, in conjunction with a portion of the rabies virus glycoprotein, was utilized to create cell-penetrating peptides that associate with CRISPR\u2013Cas9 components for the targeting of neuronal cells. The approach emphasizes receptor-mediated transcytosis to facilitate the passage of gene-editing tools across the BBB for transport (260, 265\u2013267).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenge", "text": "Investigating the delivery of the CRISPR\u2013Cas system for the treatment of NDs necessitates the resolution of significant technical challenges to ensure safety and efficacy (Figure 4). The CNS is safeguarded by the BBB, which serves as a selective permeability barrier to limit the entry of chemicals from the bloodstream into brain tissue (246, 247). The protective function of the BBB presents significant challenges for administering CRISPR\u2013Cas9-based therapies for neurodegenerative diseases. The dimensions of the CRISPR\u2013Cas9 complex hinder its ability to traverse the BBB due to its substantial size. Passive diffusion seems to be the only process by which small lipophilic molecules traverse the barrier, but active transport is essential for the movement of bigger entities. The delivery of CRISPR\u2013Cas9 tools encounters obstacles resulting from their dimensional size (248, 249). The BBB inhibits the ingress of external substances into the brain, encompassing pathogens, and toxins (250). The CRISPR\u2013Cas9 complexes are unable to traverse the BBB due to its specialized mechanisms for critical nutrients and chemicals, which do not accommodate the transit of big macromolecules. Owing to the restrictive properties of the BBB, researchers must develop specialized delivery systems that facilitate the successful passage of gene-editing tools past this barrier (251).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenge", "text": "The schematic diagram shows the analysis of the challenges in implementing CRISPR\u2013Cas9 technology for neurological illnesses and approaches to mitigate these challenges. The figure was designed with BioRender (http://Biorender.com/). CRISPR, clustered regularly interspaced short palindromic repeats; Cas9, CRISPR-associated protein 9.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenge", "text": "Researchers employ AAVs as vectors for delivering CRISPR components due to their outstanding efficacy and low safety risks. AAVs possess limited capacity for packing materials, complicating the simultaneous delivery of Cas9 proteins and their guide RNAs. The utilization of AAV vectors for the delivery of CRISPR components encounters numerous constraints, as capsids elicit immunological responses, leading to off-target mutations, and AAV distribution may be detrimental to cellular DNA (252, 253). Although viral vectors are successful, they elicit immunological responses that may provoke undesirable consequences in the body. Research has demonstrated that viral vectors induce unwanted mutations and severe immune responses, rendering them unacceptably hazardous for live human use (254). Non-viral delivery approaches utilizing nanoparticles offer safer options, although they encounter distinct challenges. The effectiveness of BBB transport and neuron-specific targeting continue to be significant obstacles hindering successful research. The reliable transport and regulated release of CRISPR components require enhanced optimization for success (255, 256).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Potential strategy", "text": "The strategies overcome challenges for BBB, engineered variants of AAVs have demonstrated enhanced capabilities for penetrating the BBB, as researchers have tailored them for gene therapy applications aimed at treating neurodegenerative diseases (Figure 4). Researchers have modified AAV capsid proteins to enhance their affinity for endothelium receptors, hence facilitating receptor-mediated transcytosis. Experimental results demonstrated that the Cre recombinase-based AAV targeted evolution (CREATE)-derived AAV-PHP.B variant surpassed AAV9 in CNS transduction efficacy by a factor of 40 following intravenous administration in adult mice. Researchers enhanced AAV-PHP.eB to achieve improved CNS transduction and BBB penetration efficacy following its first creation. Administration of AAV-PHP.eB via the intravenous method in adult mice resulted in neuronal transduction of 69% of cortical cells and 55% of striatal cells with a vector genome dosage of 1 \u00d7 1011. Engineered AAVs can function as effective non-invasive delivery methods for genetic therapies targeting NDs (257, 258).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Potential strategy", "text": "Researchers are developing nanoparticles to stabilize CRISPR\u2013Cas9 components and ensure their safe transit across the BBB. The administration of mRNA-based therapeutics to the brain is effectively achieved via LNPs that facilitate the passage of genetic material across BBB (249, 251, 259, 260). The functionalization of LNPs with transferrin enhances their ability to traverse the BBB. Brain endothelial cells exhibit elevated numbers of transferrin receptors, which utilize their inherent iron transport capability for receptor-mediated transcytosis of transferrin-conjugated nanoparticles. This technique enables the delivery of medicinal medicines to the brain. A document delineates nanoparticle targeting of the brain by transferrin nanoparticle conjugation (261). Exosomes, as natural extracellular vesicles, exhibit enhanced brain-targeting capabilities when suitably modified. The incorporation of transferrin or specific peptides onto exosomes facilitates their binding to receptors located on the BBB for targeted CNS delivery. Modified exosomes displaying transferrin on their surface enhance the accumulation of these vesicles in brain tissues, facilitating more effective administration of therapeutic agents. Ongoing research indicates that exosomes exhibit enhanced efficacy for CNS distribution when modified with ApoB to interact with low-density lipoprotein receptors (LDLR) (262).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Potential strategy", "text": "Therapeutic medicines are delivered via intracerebral injection straight into specific brain regions for optimal distribution. This delivery technique improves the localized dispersion of CRISPR\u2013Cas9 components, potentially increasing their concentration at the target region. The administration of therapeutic agents by intracerebral delivery enables healthcare professionals to accurately inject these substances into targeted brain regions, ensuring optimal distribution. This technique facilitates local delivery of CRISPR\u2013Cas9 components, enhancing distribution and potentially increasing their concentration at the target region (263). Another study employed biodegradable PEGylated nanocapsules to provide Cas9-sgRNA ribonucleoproteins (RNPs) during intracerebral injections into the mouse striatum. This therapeutic approach achieved precise genome alteration in medium spiny neurons with minimal stimulation of glial cells, ensuring safety (264). The direct delivery methods present intriguing strategies for CRISPR\u2013Cas9-mediated therapeutic interventions in NDs. Further issues regarding surgical invasiveness, tissue damage mitigation, and precise targeting criteria must be addressed to improve the clinical application of these techniques.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Potential strategy", "text": "Another strategy was utilized PepFect14, in conjunction with a portion of the rabies virus glycoprotein, was utilized to create cell-penetrating peptides that associate with CRISPR\u2013Cas9 components for the targeting of neuronal cells. The approach emphasizes receptor-mediated transcytosis to facilitate the passage of gene-editing tools across the BBB for transport (260, 265\u2013267).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Off-target effects", "text": "Long-term safety issues associated with CRISPR encompass not just off-target mutations but also significant on-target deletions, insertions, inversions, chromosomal translocations, and other intricate structural changes that conventional PCR techniques may overlook. Identifying these occurrences necessitates genome-wide, high-resolution methodologies, including long-read sequencing and capture-based assays. Strategies for risk mitigation encompass the temporary administration of CRISPR components (RNP or mRNA), the utilization of high-fidelity or double-strand break-free editors (nickase, base, or prime editors), and meticulous optimization of guide RNA, bolstered by empirical off-target analysis. Comprehensive preclinical genotoxicity assessment and prolonged molecular and clinical surveillance, with established cessation criteria, are crucial for initial human studies (268).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Off-target effects", "text": "The occurrence of off-target cleavage has the potential to lead to unfavorable mutations and cytotoxicity, which may disturb the regular functioning of genes (269). The off-target effects linked with sgRNA and Cas9 pose significant concerns regarding the CRISPR\u2013Cas method (Figure 4). While Cas9 can tolerate variations in the distribution and quantity of mismatches in DNA, the brevity of sgRNA (about 17\u201324 bp) heightens the likelihood of off-target effects. The presence of similar sequences or sequences with minor variations elsewhere in the genome can lead to mistargeting due to slight variances in the sgRNA (270\u2013272). Mis-targeting may result in severe implications for patients, including the development of oncogenic mutations (273). CRISPR\u2013Cas9 randomly alters tumor suppressor genes and activates oncogenes, potentially leading to cancer development. Off-target effects occur with a significant frequency of 50% every study, raising considerable concerns for therapeutic application (274). The incidence neuronal dysfunction of inadvertent genetic alterations due to off-target impacts jeopardizes critical neurological genes, potentially leading to neuronal cell disorders. Off-target action occurs regularly, highlighting the necessity for scientists to implement precise targeting techniques to mitigate these adverse outcomes (275, 276).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Off-target effects", "text": "Creating sgRNA with high specificity and an allowable base mismatch to another sequence reduces the likelihood of mis-targeting sgRNA. The insufficient understanding of CNS biomarkers is a significant obstacle to effective cell targeting and signaling responses, yet advancements in CSF biomarkers, brain imaging, and varied vector modeling are expected to enhance specificity (277).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Off-target effects", "text": "An optimal sgRNA length, minimal mismatches, and optimum GC content can enhance the targeting efficiency of CRISPR-Cas9 technology. Ren et al. (269) demonstrated that sgRNA sequences ranging from 17 to 20 bp have comparable mutation efficiencies. Fu et al. (278) discovered that shortened gRNA sequences of 17 or 18 bp enhance specificity, while sgRNA sequences with a GC concentration of 40%\u221260% are advantageous for augmenting target effects (269, 279). Another strategy to reduce off-target effects is to select a low GC content of sgRNA; decreased GC content correlates with diminished stability of nucleic acid hybrids. Besides screening for distinct genomic sequence targets, modifying Cas9 enzymes to create nickases and fusing dCas9 with the FokI catalytic domain has diminished the incidence of off-target effects in CRISPR\u2013Cas9 applications (273). The sgRNA specificity, particularly with fewer than three mismatches within 10\u201312 base pairs close to the PAM site and more than two mismatches with the off-target site, can enhance targeting efficacy (275, 280). Additionally, preventing the emergence of sgRNA bulges at 12 bp next to the PAM site, generating chimeric sgRNA, substituting portions of the RNA sequence with a DNA sequence, and employing numerous sgRNAs directed at the same gene may also diminish off-target effects (280\u2013283). Various variants of CRISPR\u2013Cas9 could affect and raise the target accuracy of CRISPR technology, engineered a high-fidelity version 1 of SpCas9 (SpCas9-HF1) by Kleinstiver et al. (284). In comparison to wildtype SpCas9, SpCas9-HF1 has similar targeting efficacy. Moreover, SpCas9-HF1 possesses the ability to diminish non-specific DNA interactions, rendering genome-wide break capture and focused sequencing techniques incapable of identifying all or nearly all target activities (284). Developed an \u201cenhanced specificity\u201d SpCas9 variant 1.1 (eSpCas9 1.1). In comparison to wildtype SpCas9, eSpCas9 1.1 demonstrates robust on-target cleavage and significantly reduced effects beyond the target (285). Developed an innovative hyper-accurate CRISP/Cas9 variant (HypaCas9) that demonstrated elevated genome-wide specificity without compromising on-target activity. Moreover, SpCas9 nickase, Sniper-Cas9, and xCas9 variants can reduce off-target effects in genome editing to differing extents (276, 286).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Off-target effects", "text": "Both base editing and prime editing variants of CRISPR based genome editing technologies facilitate precise DNA changes by eliminating the formation of DSBs, hence mitigating the risk associated with unintended insertions or deletions (indels). Effective methodologies are now available to mitigate the unexpected consequences that conventional CRISPR\u2013Cas9 editing may induce in DNA sequences (287). Base editors provide genomic location-specific DNA base changes directly and irreversibly, while avoiding the creation of double-strand breaks. The approach mitigates the danger of generating indels due to its architecture. The research indicates that base editors induce unintended alterations in DNA and RNA due to their off-target activity. Scientific research indicates that deaminase enzymes employed in CBEs and ABEs result in significant unintentional genomic modifications due to their interaction with various DNA bases (287, 288). Investigative research indicates that prime editing effectuates precise modifications via an alteration of reverse transcriptase and a compromised Cas9 endonuclease, directed by pegRNA (289). A research study demonstrated the efficacy of prime editing in introducing the AD-associated APOE4 mutation into human induced iPSCs with high precision and productivity, resulting in editing success rates of 28.2% in iPSCs and 42.3% in HEK293T cells. The approach exhibited no indication of significant unwanted direct alterations at the target location and no remarkable off-target modification effects, so rendering it appropriate for development (99).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Off-target effects", "text": "The CRISPR\u2013Cas9 genome editing technique encounters significant challenges due to off-target effects, which can result in unintended alterations to the genetic material. Bioinformatics tools have been created to enhance sgRNA design and assess off-target activity. DeepCRISPR and GUIDE-seq serve as crucial tools that augment CRISPR-based applications by enhancing their specificity and safety (50, 290).\n\nDeepCRISPR employs deep learning technology to create an optimization framework that forecasts target-binding efficacy and undesirable DNA interactions of sgRNA. The integration of nucleotide sequences and epigenetic characteristics across several human cell types via DeepCRISPR yields a comprehensive assessment of sgRNA efficacy. This method enables scientists to identify sgRNAs that demonstrate robust on-target efficacy while minimizing off-target effects, thereby enhancing the precision of CRISPR\u2013Cas9 editing (291). GUIDE-seq is an experimental method that enables genome-wide identification of double-stranded breaks caused by CRISPR-Cas nucleases through a sequence-based technique. The sequencing of double-stranded oligodeoxynucleotides incorporated into double-strand breaks enables GUIDE-seq to produce comprehensive off-target site data. The procurement of this data allows researchers to observe unintentional genetic alterations that transpire throughout CRISPR-based initiatives (50, 292).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenge", "text": "The occurrence of off-target cleavage has the potential to lead to unfavorable mutations and cytotoxicity, which may disturb the regular functioning of genes (269). The off-target effects linked with sgRNA and Cas9 pose significant concerns regarding the CRISPR\u2013Cas method (Figure 4). While Cas9 can tolerate variations in the distribution and quantity of mismatches in DNA, the brevity of sgRNA (about 17\u201324 bp) heightens the likelihood of off-target effects. The presence of similar sequences or sequences with minor variations elsewhere in the genome can lead to mistargeting due to slight variances in the sgRNA (270\u2013272). Mis-targeting may result in severe implications for patients, including the development of oncogenic mutations (273). CRISPR\u2013Cas9 randomly alters tumor suppressor genes and activates oncogenes, potentially leading to cancer development. Off-target effects occur with a significant frequency of 50% every study, raising considerable concerns for therapeutic application (274). The incidence neuronal dysfunction of inadvertent genetic alterations due to off-target impacts jeopardizes critical neurological genes, potentially leading to neuronal cell disorders. Off-target action occurs regularly, highlighting the necessity for scientists to implement precise targeting techniques to mitigate these adverse outcomes (275, 276).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenge", "text": "Creating sgRNA with high specificity and an allowable base mismatch to another sequence reduces the likelihood of mis-targeting sgRNA. The insufficient understanding of CNS biomarkers is a significant obstacle to effective cell targeting and signaling responses, yet advancements in CSF biomarkers, brain imaging, and varied vector modeling are expected to enhance specificity (277).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Potential strategy", "text": "An optimal sgRNA length, minimal mismatches, and optimum GC content can enhance the targeting efficiency of CRISPR-Cas9 technology. Ren et al. (269) demonstrated that sgRNA sequences ranging from 17 to 20 bp have comparable mutation efficiencies. Fu et al. (278) discovered that shortened gRNA sequences of 17 or 18 bp enhance specificity, while sgRNA sequences with a GC concentration of 40%\u221260% are advantageous for augmenting target effects (269, 279). Another strategy to reduce off-target effects is to select a low GC content of sgRNA; decreased GC content correlates with diminished stability of nucleic acid hybrids. Besides screening for distinct genomic sequence targets, modifying Cas9 enzymes to create nickases and fusing dCas9 with the FokI catalytic domain has diminished the incidence of off-target effects in CRISPR\u2013Cas9 applications (273). The sgRNA specificity, particularly with fewer than three mismatches within 10\u201312 base pairs close to the PAM site and more than two mismatches with the off-target site, can enhance targeting efficacy (275, 280). Additionally, preventing the emergence of sgRNA bulges at 12 bp next to the PAM site, generating chimeric sgRNA, substituting portions of the RNA sequence with a DNA sequence, and employing numerous sgRNAs directed at the same gene may also diminish off-target effects (280\u2013283). Various variants of CRISPR\u2013Cas9 could affect and raise the target accuracy of CRISPR technology, engineered a high-fidelity version 1 of SpCas9 (SpCas9-HF1) by Kleinstiver et al. (284). In comparison to wildtype SpCas9, SpCas9-HF1 has similar targeting efficacy. Moreover, SpCas9-HF1 possesses the ability to diminish non-specific DNA interactions, rendering genome-wide break capture and focused sequencing techniques incapable of identifying all or nearly all target activities (284). Developed an \u201cenhanced specificity\u201d SpCas9 variant 1.1 (eSpCas9 1.1). In comparison to wildtype SpCas9, eSpCas9 1.1 demonstrates robust on-target cleavage and significantly reduced effects beyond the target (285). Developed an innovative hyper-accurate CRISP/Cas9 variant (HypaCas9) that demonstrated elevated genome-wide specificity without compromising on-target activity. Moreover, SpCas9 nickase, Sniper-Cas9, and xCas9 variants can reduce off-target effects in genome editing to differing extents (276, 286).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Potential strategy", "text": "Both base editing and prime editing variants of CRISPR based genome editing technologies facilitate precise DNA changes by eliminating the formation of DSBs, hence mitigating the risk associated with unintended insertions or deletions (indels). Effective methodologies are now available to mitigate the unexpected consequences that conventional CRISPR\u2013Cas9 editing may induce in DNA sequences (287). Base editors provide genomic location-specific DNA base changes directly and irreversibly, while avoiding the creation of double-strand breaks. The approach mitigates the danger of generating indels due to its architecture. The research indicates that base editors induce unintended alterations in DNA and RNA due to their off-target activity. Scientific research indicates that deaminase enzymes employed in CBEs and ABEs result in significant unintentional genomic modifications due to their interaction with various DNA bases (287, 288). Investigative research indicates that prime editing effectuates precise modifications via an alteration of reverse transcriptase and a compromised Cas9 endonuclease, directed by pegRNA (289). A research study demonstrated the efficacy of prime editing in introducing the AD-associated APOE4 mutation into human induced iPSCs with high precision and productivity, resulting in editing success rates of 28.2% in iPSCs and 42.3% in HEK293T cells. The approach exhibited no indication of significant unwanted direct alterations at the target location and no remarkable off-target modification effects, so rendering it appropriate for development (99).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Potential strategy", "text": "The CRISPR\u2013Cas9 genome editing technique encounters significant challenges due to off-target effects, which can result in unintended alterations to the genetic material. Bioinformatics tools have been created to enhance sgRNA design and assess off-target activity. DeepCRISPR and GUIDE-seq serve as crucial tools that augment CRISPR-based applications by enhancing their specificity and safety (50, 290).\n\nDeepCRISPR employs deep learning technology to create an optimization framework that forecasts target-binding efficacy and undesirable DNA interactions of sgRNA. The integration of nucleotide sequences and epigenetic characteristics across several human cell types via DeepCRISPR yields a comprehensive assessment of sgRNA efficacy. This method enables scientists to identify sgRNAs that demonstrate robust on-target efficacy while minimizing off-target effects, thereby enhancing the precision of CRISPR\u2013Cas9 editing (291). GUIDE-seq is an experimental method that enables genome-wide identification of double-stranded breaks caused by CRISPR-Cas nucleases through a sequence-based technique. The sequencing of double-stranded oligodeoxynucleotides incorporated into double-strand breaks enables GUIDE-seq to produce comprehensive off-target site data. The procurement of this data allows researchers to observe unintentional genetic alterations that transpire throughout CRISPR-based initiatives (50, 292).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Immune response", "text": "Both the adaptive and innate immune responses present significant challenges for repeated or systemic CRISPR therapy. Documented pre-existing humoral and cellular immunity to AAV capsids and frequently utilized Cas orthologs (SpCas9, SaCas9) can neutralize vectors or induce T-cell-mediated clearance of edited cells; additionally, sustained expression of bacterial nucleases extends the period for immune recognition (293). Effective mitigation strategies, substantiated by preclinical and early clinical studies, encompass the utilization of less common AAV serotypes or engineered capsids, transient immunosuppression during dosing, the administration of editors as mRNA or RNPs (via LNPs or alternative transient carriers) to minimize antigen exposure, and the creation of Cas variants featuring diminished immunogenic epitopes or humanized sequences. Thorough immune profiling in pertinent animal models and patient samples is essential for formulating appropriate therapeutic regimens (294).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Immune response", "text": "Importantly, although pre-existing humoral and cellular immunity to commonly used Cas orthologs (e.g., SpCas9/SaCas9) has been documented in human populations (295), published clinical experience with CRISPR\u2013Cas9 therapies to date has not reported clinically meaningful Cas9-directed immune toxicities (296). For example, in the first systemic in vivo CRISPR\u2013Cas9 trials targeting transthyretin (TTR) (LPN delivery of Cas9 mRNA and sgRNA; NCT04601051), reported treatment-related events were primarily transient infusion-related reactions (and transient liver-enzyme elevations in some participants), without persistent immune-mediated sequelae attributed specifically to Cas9 (297). In parallel, ex vivo CRISPR\u2013Cas9 edited autologous hematopoietic stem and progenitor cells products (e.g., Exagamglogene autotemcel) have shown safety profiles largely consistent with conditioning/transplant procedures rather than editor immunogenicity, which is biologically plausible given the transient exposure to Cas9 during cell manufacturing (298).\n\nNevertheless, ongoing and future trials, particularly those requiring CNS delivery and/or potential redosing, should continue systematic immune surveillance (anti-Cas antibodies, T-cell responses, cytokine profiles) to define risk more precisely in neurological indications.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Immune response", "text": "Individuals possessing preexisting immunity to Cas9 derived from bacterial sources such as Staphylococcus aureus and Streptococcus pyogenes diminish the efficacy of CRISPR\u2013Cas9 gene editing therapy (273, 299). The research analysis identified SaCas9 antibodies in 78% of participants and SpCas9 (S. pyogenes Cas9) antibodies in 58%. Laboratory analyses revealed that SaCas9 antibodies were present in 78% of examined donors, whilst SpCas9 antibodies were found in 67% of individuals. Research findings indicated a significant number of T cells in the adult human population that interacted with SpCas9 (300, 301). Patients' previous immune responses often swiftly eliminate cells via Cas9 expression, diminishing therapeutic efficacy and potentially inciting inflammatory reactions or negative outcomes. The resolution of this issue is crucial for physicians to appropriately utilize CRISPR\u2013Cas9 technology in medical applications (27, 300, 302).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Immune response", "text": "Furthermore, the delivery mechanisms of CRISPR systems can stimulate immunological responses and affect neural cells. Viral vectors are frequently employed to transport CRISPR constructs across the BBB (303, 304). Nonetheless, viral transmission results in prolonged CRISPR expression, which is detrimental to neuronal cells and modifies in phenotypes of neuronal (303). Furthermore, chronic production of Cas9 induces a cytotoxic immune system response that eliminates genetically modified cells. The elimination of changed neurons in the brain may result in detrimental effects, as neuronal regeneration is restricted (294). The implementation of CRISPR\u2013Cas9 technology is hindered by the immunological response triggered by the Cas9 protein or delivering vectors in the host (300). The gene therapy vectors utilizing AAV frequently employed in clinical applications elicit a cytotoxic T-cell response, impacting both their safety and efficacy results (Figure 4). AAV capsid proteins introduced into the body are subjected to natural degradation processes, enabling major histocompatibility complex MHC class I molecules to present these components on the surfaces of transduced cells. The activation of cytotoxic T lymphocytes (CTLs) transpires via this presentation pathway, leading to cellular death and a subsequent decrease in transgene expression (305\u2013307). Three categories of T-cell responses in patients are contingent upon critical factors from AAV capsids, vector dose, and the patients' immunological backgrounds, including their natural exposure to wild-type AAV (305). Comprehending and mitigating these immunological barriers is essential for improving the safety and efficacy of AAV-mediated gene treatments (306).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Immune response", "text": "Various strategies have been devised to diminish the immunogenicity of CRISPR-Cas9 systems for therapeutic applications. Acquiring CRISPR from non-pathogenic bacterial strains offers a strategy to mitigate potential risks associated with pre-existing immune responses. Studies indicate that Cas9 derived from Geobacillus stearothermophilus (GsCas9) exhibits superior stability in human plasma compared to Streptococcus pyogenes Cas9 (SpCas9), suggesting its potential as a less immunogenic method for genome editing (299, 308). Another strategy is to target modifications of Cas9 immunogenic epitopes by engineering techniques yield less detectable proteins that preserve their editing capabilities. Researchers identified significant T-cell antigenic sites in SaCas9, after which other modifications were introduced to reduce immunogenicity (302, 309). Proposed alterations in these engineered versions resulted in a diminished T-cell response while maintaining their inherent nuclease activity (310, 311). The delivery of CRISPR-Cas9 components via mRNA results in a temporary impact that mitigates immune responses, hence enhancing the safety of genome editing therapies (312). The utilization of Cas9 mRNA facilitates temporal protein expression, hence minimizing prolonged exposure and decreasing vulnerability to immune responses. Target cells efficiently acquire Cas9 mRNA and sgRNA via LNPs, which maintain transient protein expression and mitigate undesirable immunological responses (313). This approach yields superior outcomes compared to traditional viral vectors, as it minimizes genetic changes and allows for numerous distribution opportunities due to its diminished inflammatory response. Research indicates that employing LNP to deliver Cas9 mRNA facilitates genome editing throughout many administrations with less immune response upon delivery (314). Modified nucleoside addition to mRNA molecules is extensively employed by scientific researchers as an established technique for developing safer CRISPR-based therapy. Researchers found that including pseudouridine and 5-methylcytosine into Cas9 mRNA significantly diminished the immunological response, hence enhancing Cas9 protein synthesis (315). Research has demonstrated that substituting uridine with N1-methylpseudouridine in mRNA vaccines diminishes immune responses while enhancing protein synthesis (316). The therapeutic efficacy improves as these changes allow mRNA to evade immune recognition, hence reducing unfavorable immunological responses (317).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Immune response", "text": "Administering short-term immunosuppressive regimens with CRISPR\u2013Cas9 delivery may reduce immune responses; nonetheless, this strategy is contentious and requires comprehensive safety assessments (318). Research has shown that the immune system can identify and respond to CRISPR elements, resulting in inflammation and diminished effectiveness of gene editing (302). Research demonstrates that Cas9 expression in murine muscle tissue led to lymphocyte infiltration, implying an immunological response to the exogenous protein (319).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenge", "text": "Individuals possessing preexisting immunity to Cas9 derived from bacterial sources such as Staphylococcus aureus and Streptococcus pyogenes diminish the efficacy of CRISPR\u2013Cas9 gene editing therapy (273, 299). The research analysis identified SaCas9 antibodies in 78% of participants and SpCas9 (S. pyogenes Cas9) antibodies in 58%. Laboratory analyses revealed that SaCas9 antibodies were present in 78% of examined donors, whilst SpCas9 antibodies were found in 67% of individuals. Research findings indicated a significant number of T cells in the adult human population that interacted with SpCas9 (300, 301). Patients' previous immune responses often swiftly eliminate cells via Cas9 expression, diminishing therapeutic efficacy and potentially inciting inflammatory reactions or negative outcomes. The resolution of this issue is crucial for physicians to appropriately utilize CRISPR\u2013Cas9 technology in medical applications (27, 300, 302).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenge", "text": "Furthermore, the delivery mechanisms of CRISPR systems can stimulate immunological responses and affect neural cells. Viral vectors are frequently employed to transport CRISPR constructs across the BBB (303, 304). Nonetheless, viral transmission results in prolonged CRISPR expression, which is detrimental to neuronal cells and modifies in phenotypes of neuronal (303). Furthermore, chronic production of Cas9 induces a cytotoxic immune system response that eliminates genetically modified cells. The elimination of changed neurons in the brain may result in detrimental effects, as neuronal regeneration is restricted (294). The implementation of CRISPR\u2013Cas9 technology is hindered by the immunological response triggered by the Cas9 protein or delivering vectors in the host (300). The gene therapy vectors utilizing AAV frequently employed in clinical applications elicit a cytotoxic T-cell response, impacting both their safety and efficacy results (Figure 4). AAV capsid proteins introduced into the body are subjected to natural degradation processes, enabling major histocompatibility complex MHC class I molecules to present these components on the surfaces of transduced cells. The activation of cytotoxic T lymphocytes (CTLs) transpires via this presentation pathway, leading to cellular death and a subsequent decrease in transgene expression (305\u2013307). Three categories of T-cell responses in patients are contingent upon critical factors from AAV capsids, vector dose, and the patients' immunological backgrounds, including their natural exposure to wild-type AAV (305). Comprehending and mitigating these immunological barriers is essential for improving the safety and efficacy of AAV-mediated gene treatments (306).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Potential strategy", "text": "Various strategies have been devised to diminish the immunogenicity of CRISPR-Cas9 systems for therapeutic applications. Acquiring CRISPR from non-pathogenic bacterial strains offers a strategy to mitigate potential risks associated with pre-existing immune responses. Studies indicate that Cas9 derived from Geobacillus stearothermophilus (GsCas9) exhibits superior stability in human plasma compared to Streptococcus pyogenes Cas9 (SpCas9), suggesting its potential as a less immunogenic method for genome editing (299, 308). Another strategy is to target modifications of Cas9 immunogenic epitopes by engineering techniques yield less detectable proteins that preserve their editing capabilities. Researchers identified significant T-cell antigenic sites in SaCas9, after which other modifications were introduced to reduce immunogenicity (302, 309). Proposed alterations in these engineered versions resulted in a diminished T-cell response while maintaining their inherent nuclease activity (310, 311). The delivery of CRISPR-Cas9 components via mRNA results in a temporary impact that mitigates immune responses, hence enhancing the safety of genome editing therapies (312). The utilization of Cas9 mRNA facilitates temporal protein expression, hence minimizing prolonged exposure and decreasing vulnerability to immune responses. Target cells efficiently acquire Cas9 mRNA and sgRNA via LNPs, which maintain transient protein expression and mitigate undesirable immunological responses (313). This approach yields superior outcomes compared to traditional viral vectors, as it minimizes genetic changes and allows for numerous distribution opportunities due to its diminished inflammatory response. Research indicates that employing LNP to deliver Cas9 mRNA facilitates genome editing throughout many administrations with less immune response upon delivery (314). Modified nucleoside addition to mRNA molecules is extensively employed by scientific researchers as an established technique for developing safer CRISPR-based therapy. Researchers found that including pseudouridine and 5-methylcytosine into Cas9 mRNA significantly diminished the immunological response, hence enhancing Cas9 protein synthesis (315). Research has demonstrated that substituting uridine with N1-methylpseudouridine in mRNA vaccines diminishes immune responses while enhancing protein synthesis (316). The therapeutic efficacy improves as these changes allow mRNA to evade immune recognition, hence reducing unfavorable immunological responses (317).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Potential strategy", "text": "Administering short-term immunosuppressive regimens with CRISPR\u2013Cas9 delivery may reduce immune responses; nonetheless, this strategy is contentious and requires comprehensive safety assessments (318). Research has shown that the immune system can identify and respond to CRISPR elements, resulting in inflammation and diminished effectiveness of gene editing (302). Research demonstrates that Cas9 expression in murine muscle tissue led to lymphocyte infiltration, implying an immunological response to the exogenous protein (319).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Ethical and regulatory challenges", "text": "CRISPR-based therapeutics for NDs presents significant ethical and regulatory issues, especially when utilized solely on somatic cells. Critical concerns encompass securing informed consent for permanent genomic modifications, equitable access to expensive therapies, and meticulously assessing the risks and benefits of irreversible operations in the CNS. First-in-human trials necessitate stringent supervision, particularly when employing novel delivery technologies. Despite widespread rejection of germline editing, its public discourse continues to influence trust in somatic editing. Robust community involvement, autonomous ethical assessment, and compliance with international standards are needed. Regulators are progressively mandating comprehensive preclinical evidence about safety, immunological responses, and off-target effects, in addition to dependable manufacturing and long-term monitoring strategies. Engaging bioethicists and patient representatives in trial design enhances the quality of consent and public acceptance (320).\n\nCRISPR gene-editing in neurological research raises significant ethical issues owing to its substantial influence on cognitive abilities and identity, possible unforeseen consequences, and ramifications for subsequent generations.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Ethical and regulatory challenges", "text": "Ethical considerations are crucial, considering the fragility of several patients, and acquiring informed approval. Consent may be more complex due to cognitive problems. Regulatory obstacles in guaranteeing the safety of experimental therapies, while accelerating approvals for life-threatening disorders, contribute to the complexity (321).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Ethical and regulatory challenges", "text": "The application of gene-editing technologies in the brain poses risks of off-target effects that may disrupt normal neural processes, potentially resulting in alterations in behavior or cognitive capacities. The intricacy of the human brain renders such risks perilous, since little alterations from a brain computer interface could lead to substantial neurological repercussions that may not become apparent until later in life. These warrants worry if the targeted modification involves germline cells, as they transmit genetic mutations to subsequent generations, hence amplifying the potential repercussions of such changes indefinitely (322). The intentional alteration of DNA in human embryos, gametes, and zygotes for the aim of disease prevention has sparked significant ethical controversies (323, 324). The major ethical dilemma arises from the capacity to engineer \u201cdesigner babies\u201d by DNA alteration, which pursues medicinal advantages while also striving to cultivate enhanced inherited traits, so provoking concerns over eugenics and inequality (325). Opponents argue that these methods would create communities with genetically modified elites, exacerbating existing societal disparities (323). Ethical considerations are crucial prerequisites prior to the application of CRISPR in humans. Genome editing in clinical contexts is still restricted to somatic cells, as this poses a lower risk of unethical exploitation compared to germline modification (326). Germline editing raises ethical dilemmas because of its incalculable implications over generations (323). Modifications made during the embryonic period are heritable, indicating they will be passed on to subsequent generations. Such erratic alterations pose risks that impact particular people and future generations within the human gene pool (327). The uncertainties associated with mutations and unanticipated adverse effects from these therapies raise questions regarding their safety and efficacy (328). Germline editing encounters substantial difficulties due to the complexities of obtaining consent. Individuals who have not yet been born cannot consent to genetic alterations performed prior to their birth. Scientists encounter an ethical quandary over their authority to implement irreversible alterations on individual's incapable of providing consent for such adjustments. Individuals subjected to germline editing experience a restriction of their autonomy rights, so limiting their capacity to make decisions regarding significant life-altering issues (329, 330).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Ethical and regulatory challenges", "text": "However, the practice of germline gene editing holds potential for preventing specific genetic abnormalities of the nervous system, although it engenders significant ethical issues. Ethical frameworks and comprehensive evaluations must be established to address concerns around designer babies, unforeseen consequences, issues of consent, and potential misuse prior to the ethical application of such technology.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Ethical and regulatory challenges", "text": "In addition, the ethical considerations for genome editing fundamentally differ between somatic and germline operations. Somatic editing, which focuses on non-reproductive cells, aligns with conventional therapeutic ethics when risks, benefits, and informed consent are adequately addressed (331). In contrast, germline or heritable editing presents intergenerational, societal, and justice issues, as modifications would be passed to future generations and impact individuals unable to provide consent (332). International expert organizations and policy evaluations consistently advise against permitting clinical germline heritable genome editing outside of stringent research frameworks at this time, asserting that any future deliberation necessitates exceptional evidence of safety, efficacy, and widespread societal agreement. The updated wording encapsulates these viewpoints and emphasizes the practical ramifications for researchers and sponsors: prioritize robust somatic-cell methodologies for imminent clinical application while fostering responsible, transparent research about long-term ethical considerations (332).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Ethical and regulatory challenges", "text": "A verified translational pathway commences with mechanism validation in pertinent models (isogenic human cell structures, organoids, and various animal species), coupled with stringent safety assays (empirical genome-wide off-target mapping, long-read structural variant detection, transcriptomic profiling, and immunogenicity testing) (333). Investigational New Drug/Clinical Trial Application IND/CTA enabling packages must encompass biodistribution, vector shedding, reproductive toxicity, and dose-ranging investigations. Manufacturing must adhere to good manufacturing practice (GMP) standards and incorporate confirmed potency assays associated with the editing method; for viral vectors, capsid characterization and release assays are crucial. Initial clinical trials must utilize careful, incremental dose escalation, stringent molecular monitoring of editing results in accessible tissues, clinical indicators for target engagement, and established stopping criteria. Sponsors must ultimately engage in long-term follow-up (multi-year, disease-specific) to identify delayed side effects and gather enduring efficacy data. These measures correspond with the recommendations of international commissions and recent translational assessments (334).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Ethical and regulatory challenges", "text": "The brain's CRISPR technologies raise pressing issues related to rational agency and voluntariness. The dilemma of who should provide permission for a person's personality to change arises if cognitive qualities can be altered by gene editing. This is particularly difficult when a subject is unable to make educated decisions for various reasons, such as mental impairments (335, 336).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Ethical and regulatory challenges", "text": "Ethical considerations around brain somatic gene editing arise due to the potential for unintended genetic modifications that may be inherited by subsequent generations. The targeted modifications of non-reproductive cells may indirectly induce changes in germline cells, potentially resulting in irreversible genetic abnormalities. Somatic gene editing raises ethical dilemmas about permission, as future generations are unable to provide assent for such alterations (331, 337). Although somatic edits are generally non-heritable, there is apprehension that unintended alterations might be transmitted to future generations if specific editing inadvertently impacts germline cells or influences mechanisms affecting genetic material in incompletely understood ways. This issue highlights ethical dilemmas concerning the limits of human control over gene editing, particularly in relation to somatic therapies designed to treat NDs (338\u2013340). The difficulty in obtaining consent for brain somatic gene editing remains significant within children and cognitively challenged patient demographics. An ethical dilemma arises since stakeholders involved in gene editing interventions may be unaware of the various risks and genetic alterations that could come from these procedures (341, 342). Somatic gene editing in the brain may result in unintended consequences owing to the intricacy of neural circuitry and cerebral function. The ethical dilemma pertains to the potential risks of modifying the brain's genetic composition without a comprehensive understanding of the long-term safety of such operations. Even slight, inadvertent alterations in genetic material may result in significant NDs or unexpected psychological diseases (343, 344).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Ethical and regulatory challenges", "text": "Human gene editing utilizing CRISPR has raised significant concerns, particularly regarding germline modifications (345). Current guidelines, including those from the national institutes of health (NIH) and World Health Organization (WHO), emphasize restricting CRISPR treatments to somatic cells in adults to prevent heritable alterations that may inadvertently affect subsequent generations (Figure 4) (346, 347). By focusing on non-reproductive cells, somatic gene editing mitigates any ethical dilemmas associated with unintended genetic alterations (331). The formulation of these regulations primarily relies on comprehensive risk evaluations and continuous monitoring to avert the misuse of gene-editing technology (348). The NIH has instituted stringent standards for clinical research utilizing gene-edited technology, necessitating comprehensive assessment by institutional ethics committees (344, 349). Various nations implement distinct policies; in China, research on CRISPR technology progresses with fairly lenient regulations, whereas Western countries enforce stringent restrictions governing human gene editing (350, 351).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Ethical and regulatory challenges", "text": "Reaching an international agreement on CRISPR guidelines is vitally vital given the global nature of scientific collaboration (352). The World Medical Association (WMA) has recommendations meant to guide members in creating consistent standards for CRISPR research (353). These requirements underscore the necessity of global control in order to stop unethical behavior and guarantee responsible and safe evolution of gene-edited technology. Harmonizing guidelines will assist in stopping \u201cethics shopping,\u201d in which researchers may undertake experiments in countries with more lax laws, therefore avoiding more rigorous ethical rules elsewhere (354, 355). Moreover, a combined worldwide framework would help to improve cooperation, enable more effective monitoring, and advance equity in access to contemporary medical treatments (349). Robust ethical norms and international cooperation mitigate the risks associated with CRISPR technology and ensure its appropriate use for societal benefit.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenge", "text": "CRISPR gene-editing in neurological research raises significant ethical issues owing to its substantial influence on cognitive abilities and identity, possible unforeseen consequences, and ramifications for subsequent generations.\n\nEthical considerations are crucial, considering the fragility of several patients, and acquiring informed approval. Consent may be more complex due to cognitive problems. Regulatory obstacles in guaranteeing the safety of experimental therapies, while accelerating approvals for life-threatening disorders, contribute to the complexity (321).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenge", "text": "The application of gene-editing technologies in the brain poses risks of off-target effects that may disrupt normal neural processes, potentially resulting in alterations in behavior or cognitive capacities. The intricacy of the human brain renders such risks perilous, since little alterations from a brain computer interface could lead to substantial neurological repercussions that may not become apparent until later in life. These warrants worry if the targeted modification involves germline cells, as they transmit genetic mutations to subsequent generations, hence amplifying the potential repercussions of such changes indefinitely (322). The intentional alteration of DNA in human embryos, gametes, and zygotes for the aim of disease prevention has sparked significant ethical controversies (323, 324). The major ethical dilemma arises from the capacity to engineer \u201cdesigner babies\u201d by DNA alteration, which pursues medicinal advantages while also striving to cultivate enhanced inherited traits, so provoking concerns over eugenics and inequality (325). Opponents argue that these methods would create communities with genetically modified elites, exacerbating existing societal disparities (323). Ethical considerations are crucial prerequisites prior to the application of CRISPR in humans. Genome editing in clinical contexts is still restricted to somatic cells, as this poses a lower risk of unethical exploitation compared to germline modification (326). Germline editing raises ethical dilemmas because of its incalculable implications over generations (323). Modifications made during the embryonic period are heritable, indicating they will be passed on to subsequent generations. Such erratic alterations pose risks that impact particular people and future generations within the human gene pool (327). The uncertainties associated with mutations and unanticipated adverse effects from these therapies raise questions regarding their safety and efficacy (328). Germline editing encounters substantial difficulties due to the complexities of obtaining consent. Individuals who have not yet been born cannot consent to genetic alterations performed prior to their birth. Scientists encounter an ethical quandary over their authority to implement irreversible alterations on individual's incapable of providing consent for such adjustments. Individuals subjected to germline editing experience a restriction of their autonomy rights, so limiting their capacity to make decisions regarding significant life-altering issues (329, 330).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenge", "text": "However, the practice of germline gene editing holds potential for preventing specific genetic abnormalities of the nervous system, although it engenders significant ethical issues. Ethical frameworks and comprehensive evaluations must be established to address concerns around designer babies, unforeseen consequences, issues of consent, and potential misuse prior to the ethical application of such technology.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenge", "text": "In addition, the ethical considerations for genome editing fundamentally differ between somatic and germline operations. Somatic editing, which focuses on non-reproductive cells, aligns with conventional therapeutic ethics when risks, benefits, and informed consent are adequately addressed (331). In contrast, germline or heritable editing presents intergenerational, societal, and justice issues, as modifications would be passed to future generations and impact individuals unable to provide consent (332). International expert organizations and policy evaluations consistently advise against permitting clinical germline heritable genome editing outside of stringent research frameworks at this time, asserting that any future deliberation necessitates exceptional evidence of safety, efficacy, and widespread societal agreement. The updated wording encapsulates these viewpoints and emphasizes the practical ramifications for researchers and sponsors: prioritize robust somatic-cell methodologies for imminent clinical application while fostering responsible, transparent research about long-term ethical considerations (332).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenge", "text": "A verified translational pathway commences with mechanism validation in pertinent models (isogenic human cell structures, organoids, and various animal species), coupled with stringent safety assays (empirical genome-wide off-target mapping, long-read structural variant detection, transcriptomic profiling, and immunogenicity testing) (333). Investigational New Drug/Clinical Trial Application IND/CTA enabling packages must encompass biodistribution, vector shedding, reproductive toxicity, and dose-ranging investigations. Manufacturing must adhere to good manufacturing practice (GMP) standards and incorporate confirmed potency assays associated with the editing method; for viral vectors, capsid characterization and release assays are crucial. Initial clinical trials must utilize careful, incremental dose escalation, stringent molecular monitoring of editing results in accessible tissues, clinical indicators for target engagement, and established stopping criteria. Sponsors must ultimately engage in long-term follow-up (multi-year, disease-specific) to identify delayed side effects and gather enduring efficacy data. These measures correspond with the recommendations of international commissions and recent translational assessments (334).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenge", "text": "The brain's CRISPR technologies raise pressing issues related to rational agency and voluntariness. The dilemma of who should provide permission for a person's personality to change arises if cognitive qualities can be altered by gene editing. This is particularly difficult when a subject is unable to make educated decisions for various reasons, such as mental impairments (335, 336).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenge", "text": "Ethical considerations around brain somatic gene editing arise due to the potential for unintended genetic modifications that may be inherited by subsequent generations. The targeted modifications of non-reproductive cells may indirectly induce changes in germline cells, potentially resulting in irreversible genetic abnormalities. Somatic gene editing raises ethical dilemmas about permission, as future generations are unable to provide assent for such alterations (331, 337). Although somatic edits are generally non-heritable, there is apprehension that unintended alterations might be transmitted to future generations if specific editing inadvertently impacts germline cells or influences mechanisms affecting genetic material in incompletely understood ways. This issue highlights ethical dilemmas concerning the limits of human control over gene editing, particularly in relation to somatic therapies designed to treat NDs (338\u2013340). The difficulty in obtaining consent for brain somatic gene editing remains significant within children and cognitively challenged patient demographics. An ethical dilemma arises since stakeholders involved in gene editing interventions may be unaware of the various risks and genetic alterations that could come from these procedures (341, 342). Somatic gene editing in the brain may result in unintended consequences owing to the intricacy of neural circuitry and cerebral function. The ethical dilemma pertains to the potential risks of modifying the brain's genetic composition without a comprehensive understanding of the long-term safety of such operations. Even slight, inadvertent alterations in genetic material may result in significant NDs or unexpected psychological diseases (343, 344).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Potential strategy", "text": "Human gene editing utilizing CRISPR has raised significant concerns, particularly regarding germline modifications (345). Current guidelines, including those from the national institutes of health (NIH) and World Health Organization (WHO), emphasize restricting CRISPR treatments to somatic cells in adults to prevent heritable alterations that may inadvertently affect subsequent generations (Figure 4) (346, 347). By focusing on non-reproductive cells, somatic gene editing mitigates any ethical dilemmas associated with unintended genetic alterations (331). The formulation of these regulations primarily relies on comprehensive risk evaluations and continuous monitoring to avert the misuse of gene-editing technology (348). The NIH has instituted stringent standards for clinical research utilizing gene-edited technology, necessitating comprehensive assessment by institutional ethics committees (344, 349). Various nations implement distinct policies; in China, research on CRISPR technology progresses with fairly lenient regulations, whereas Western countries enforce stringent restrictions governing human gene editing (350, 351).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Potential strategy", "text": "Reaching an international agreement on CRISPR guidelines is vitally vital given the global nature of scientific collaboration (352). The World Medical Association (WMA) has recommendations meant to guide members in creating consistent standards for CRISPR research (353). These requirements underscore the necessity of global control in order to stop unethical behavior and guarantee responsible and safe evolution of gene-edited technology. Harmonizing guidelines will assist in stopping \u201cethics shopping,\u201d in which researchers may undertake experiments in countries with more lax laws, therefore avoiding more rigorous ethical rules elsewhere (354, 355). Moreover, a combined worldwide framework would help to improve cooperation, enable more effective monitoring, and advance equity in access to contemporary medical treatments (349). Robust ethical norms and international cooperation mitigate the risks associated with CRISPR technology and ensure its appropriate use for societal benefit.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-based therapeutics for neurological disorders in clinical trials", "text": "The neurological disease clinical trials provide distinct obstacles owing to the complexity and variability of these conditions. A major problem is patient recruitment, as identifying appropriate individuals with certain NDs might be arduous. Moreover, the advancement of NDs differs significantly among people, complicating the establishment of standardized trial endpoints (321). No clinical trials utilizing CRISPR for NDs have commenced yet. The majority of current clinical trials focus on cancer treatment and immunotherapy. CRISPR-mediated modification of T-cell antigen receptors (CAR-T technology) is being investigated for the treatment of multiple myeloma, sarcoma, and melanoma in the United States (356). A new clinical CAR-T trial has commenced in China for lymphoma, B-cell leukemia, invasive bladder cancer, and metastatic non-small-cell lung carcinoma (NSCLC) (357).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "CRISPR-based therapeutics for neurological disorders in clinical trials", "text": "Some clinical trials have been recorded in http://clinicaltrials.gov applied on neurological disease such as NCT06615206 is an interventional study, until now there is not results resisted, utilized HG204 is a CRISPR RNA-editing to treat MECP2 duplication syndrome. Not completed NCT06025032 early phase I the objective of the trial is to ascertain the safety and efficacy of HG205 as a CRISPR\u2013Cas13 RNA base-editing therapy for treating hearing loss attributed to the p.Q829X mutation in the otoferlin (OTOF) gene, NCT05740761 this trial is observational type utilizing CRISPR\u2013Cas9-based gene editing combined with AAV-based delivery for correcting of the common methyl-CpG-binding protein 2 (MeCP2) mutations both in vitro and in vivo. Other trails had conducted on neurological diseases NCT05683860 this is an open-label extension study designed to assess the safety, tolerability, pharmacokinetics, pharmacodynamics, and clinical effects of WVE-004 in adult patients diagnosed with ALS, frontotemporal dementia (FTD), or a mixed ALS/FTD phenotype possessing a confirmed mutation in the C9orf72 gene. Patients must have successfully completed the Phase 1b/2a WVE-004-001 study to participate in the research. NCT05032196 this is a Phase 1b/2a multicenter, randomized, double-blind, placebo-controlled trial designed to assess the safety, tolerability, pharmacokinetics, and pharmacodynamics of WVE-003 in adult patients with early-manifest HD who possess the specific SNP. NCT04931862 is a Phase 1b/2a multicenter, randomized, double-blind, placebo-controlled trial designed to assess the safety, tolerability, pharmacokinetics, and pharmacodynamics of intrathecal WVE-004 in adult patients with C9orf72-associated ALS or FTD. Patients must possess a documented mutation (GGGGCC [G4C2] repeat expansion) in the initial intronic region of the C9orf72 gene and be diagnosed with ALS or FTD to participate in the study.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Future directions and emerging trends", "text": "CRISPR\u2013Cas9 exhibits considerable potential; yet, its recent discovery and implementation in humans limit its utility in clinical investigations. The utilization of CRISPR\u2013Cas in neurodegenerative diseases presents numerous significant hurdles, each necessitating inventive solutions. Comprehending CRISPR\u2013Cas9 delivery systems. CRISPR\u2013Cas has developed significantly within a relatively short period. The technology has facilitated numerous achievements, and there are elevated prospects for additional advancements in the forthcoming years (247). Current advancements in neurodegenerative disease therapies utilizing CRISPR\u2013Cas demonstrate encouraging developments, particularly in the pursuit of safer and more efficacious molecular interventions aimed at mutant genes associated with these disorders. These advancements include enhanced editing tools such as Cas9 and CRISPR diagnostics variations, as well as novel techniques, including base and prime editing for comparatively accurate gene alterations. These precision procedures are crucial for neurodegenerative disorders, as even minor inaccuracies can have significant consequences due to the brain's functioning (358).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Future directions and emerging trends", "text": "Several intriguing avenues may expedite translation in the future. CRISPR-based diagnostics, such as specific high-sensitivity enzymatic reporter locking (SHERLOCK) and DETECTR, provide rapid and sensitive identification of disease biomarkers and may enhance therapeutic platforms for patient stratification and therapy response monitoring (359, 360). AI-assisted sgRNA design and machine learning models for forecasting off-target effects, structural variant risk, and cell type specificity can enhance the safety and efficacy of next-generation editing (361). Progress in patient-derived organoids of the brain and assembloids facilitates mechanistic investigations of intricate neurodegenerative processes, encompassing neuronal\u2013glial interactions and circuit-level dysfunction, while offering scalable platforms for the validation of CRISPR therapeutics prior to animal and human trials (362, 363).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Future directions and emerging trends", "text": "Although CRISPR\u2013Cas possesses considerable promise for the treatment of various genetic disorders, issues concerning off-target effects and the efficient delivery of the CRISPR\u2013Cas genome editing apparatus must be resolved to facilitate its implementation in clinical settings, particularly for neurodegenerative diseases. The primary problem in creating theragnostic for brain injury is site-specific delivery. Instances have arisen throughout the years in which the transport of nanoparticles (CRISPR\u2013Gold, magneto-electric nanoparticles) has been effectively utilized (303, 364). This technique facilitates genomic modifications, encompassing the deletion of extensive nucleotide sequences, homologous recombination, insertion/deletion point mutations, and transcriptional regulation of certain genetic regions. Owing to its functional characteristics and resilience to epigenetic alterations, CRISPR\u2013Cas9 has emerged as the most advantageous method for genome editing in human disease models, both in vivo and in vitro (365). Nonetheless, as previously discussion, the primary barrier of effective gene therapy for neurodegenerative diseases is the inadequate comprehension of its pathogenic mechanisms. Nonetheless, with a growing quantity of research aimed at elucidating the molecular causes and formulating gene therapies for this disease indicates that CRISPR\u2013Cas9 technology has considerable promise for enhancing our comprehension of NDs and for effective future interventions targeting specific genes.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Future directions and emerging trends", "text": "Future applications of CRISPR\u2013Cas focus on precise, allele-specific editing to selectively target mutant alleles responsible for autosomal dominant NDs such as HD and fAD. Advancements in base editors and prime editing facilitate single-nucleotide modifications without inducing double-stranded breaks, hence reducing off-target effects and enhancing safety profiles (54). In preclinical models of HD, the disruption of the mHTT gene with CRISPR\u2013Cas9 has shown promising results, including improved motor performance and reduced neurotoxicity (186, 366).\n\nRecent preclinical studies indicate that genome- and transcriptome-targeted CRISPR approaches can significantly alter ALS biology in vivo and in human cells. Specifically, in vivo base-editing or nuclease techniques that reduce mutant SOD1 expression have demonstrated functional advantages and prolonged survival in murine models, thereby providing robust therapeutic proof-of-concept for gene-specific familial ALS (233). Similarly, the excision of the C9ORF72 GGGGCC repeat expansion via Cas9 diminishes RNA foci and DPRs in cellular and murine models, whereas RNA-targeting CRISPR (Cas13/CasRx) can reduce toxic repeat RNAs/DPRs without modifying DNA providing a complementary, less-permanent approach for patients (228).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Future directions and emerging trends", "text": "Implementing these advancements in clinical settings will necessitate concurrent advancements in four domains. The primary obstacle is the secure and effective delivery to motor neurons (spinal cord and pertinent brain regions) at a therapeutic scale: innovative AAV capsids and non-viral methods (LNPs, nanoparticles, intrathecal or intraparenchymal regional techniques) must be refined for biodistribution, dosage, and immunogenicity in large animals prior to human trials (367). Secondly, it is imperative to minimize genotoxic risk by incorporating base and prime editors that circumvent double-strand breaks, utilizing transient delivery formats (such as RNPs, self-inactivating vectors, and split-intein systems), and conducting thorough, unbiased genome- and transcriptome-wide off-target profiling within preclinical packages (233).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Future directions and emerging trends", "text": "Third, the clinical translation is enhanced by established regulatory and biomarker precedents: the approval and regulatory experience with antisense oligonucleotide therapy for SOD1-ALS (tofersen) illustrates that reducing a disease protein and exhibiting biomarker effects (e.g., neurofilament light) can facilitate accelerated pathways while long-term clinical benefits are validated in confirmatory studies. This presents three practical implications for CRISPR strategies: (a) initial trials should incorporate robust molecular biomarkers (Cerebrospinal fluid (CSF)/plasma neurofilament and target RNA/protein measurements), (b) patient selection must closely correspond with the genomic target (e.g., SOD1 or chromosome 9 open reading frame 72 (C9ORF72) carriers), and (c) early regulatory engagement in development is essential (368).\n\nOne of the biggest challenges still is effective and focused in vivo distribution of CRISPR components. Developed to improve transport over the BBB to reach particular neural populations are innovations include AAV vectors, LNPs, and modified exosomes (224, 369\u2013371). To fix mutations in SOD1 in ALS mice, for instance, a dual AAV system has been employed to deliver CRISPR\u2013Cas9, hence displaying longer longevity (232). Furthermore, in development are ligand-conjugated nanoparticles to increase CNS cell-type selectivity (372).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Future directions and emerging trends", "text": "CRISPR/dCas9, in conjunction with transcriptional repressors or activators, facilitates epigenetic modulation of gene expression without cleaving DNA. This approach offers a reversible and perhaps safer alternative, especially for multifactorial conditions like PD or sporadic AD. Research has demonstrated that in PD models, dCas9-mediated activation of neuroprotective genes (e.g., BDNF) can preserve dopaminergic neurons (373).\n\nThe advancement of human-relevant disease models is being revolutionized by CRISPR\u2013Cas technology. Modifying disease-related mutations in iPSCs enables them to emulate the pathogenic characteristics of neurodegeneration. Novel therapeutic targets and disease development regulators are being identified using genome-wide CRISPR screening (374). These tools facilitate the identification of significant genetic associations and cellular pathways involved in disease mechanisms.\n\nThe integration of CRISPR-based methodologies with other treatment strategies such as RNA interference (RNAi), antisense oligonucleotides (ASOs), and small molecule pharmaceuticals is a significant trend. This combination may mitigate CRISPR-induced risks while simultaneously enhancing therapeutic efficacy. In AD models, targeting both APP and BACE1 genes has demonstrated a synergistic reduction in \u03b2-amyloid levels (375, 376).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Treatment of genetic disorders", "text": "Although CRISPR\u2013Cas possesses considerable promise for the treatment of various genetic disorders, issues concerning off-target effects and the efficient delivery of the CRISPR\u2013Cas genome editing apparatus must be resolved to facilitate its implementation in clinical settings, particularly for neurodegenerative diseases. The primary problem in creating theragnostic for brain injury is site-specific delivery. Instances have arisen throughout the years in which the transport of nanoparticles (CRISPR\u2013Gold, magneto-electric nanoparticles) has been effectively utilized (303, 364). This technique facilitates genomic modifications, encompassing the deletion of extensive nucleotide sequences, homologous recombination, insertion/deletion point mutations, and transcriptional regulation of certain genetic regions. Owing to its functional characteristics and resilience to epigenetic alterations, CRISPR\u2013Cas9 has emerged as the most advantageous method for genome editing in human disease models, both in vivo and in vitro (365). Nonetheless, as previously discussion, the primary barrier of effective gene therapy for neurodegenerative diseases is the inadequate comprehension of its pathogenic mechanisms. Nonetheless, with a growing quantity of research aimed at elucidating the molecular causes and formulating gene therapies for this disease indicates that CRISPR\u2013Cas9 technology has considerable promise for enhancing our comprehension of NDs and for effective future interventions targeting specific genes.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Future applications", "text": "Future applications of CRISPR\u2013Cas focus on precise, allele-specific editing to selectively target mutant alleles responsible for autosomal dominant NDs such as HD and fAD. Advancements in base editors and prime editing facilitate single-nucleotide modifications without inducing double-stranded breaks, hence reducing off-target effects and enhancing safety profiles (54). In preclinical models of HD, the disruption of the mHTT gene with CRISPR\u2013Cas9 has shown promising results, including improved motor performance and reduced neurotoxicity (186, 366).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Current findings to potential clinical translation", "text": "Recent preclinical studies indicate that genome- and transcriptome-targeted CRISPR approaches can significantly alter ALS biology in vivo and in human cells. Specifically, in vivo base-editing or nuclease techniques that reduce mutant SOD1 expression have demonstrated functional advantages and prolonged survival in murine models, thereby providing robust therapeutic proof-of-concept for gene-specific familial ALS (233). Similarly, the excision of the C9ORF72 GGGGCC repeat expansion via Cas9 diminishes RNA foci and DPRs in cellular and murine models, whereas RNA-targeting CRISPR (Cas13/CasRx) can reduce toxic repeat RNAs/DPRs without modifying DNA providing a complementary, less-permanent approach for patients (228).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Current findings to potential clinical translation", "text": "Implementing these advancements in clinical settings will necessitate concurrent advancements in four domains. The primary obstacle is the secure and effective delivery to motor neurons (spinal cord and pertinent brain regions) at a therapeutic scale: innovative AAV capsids and non-viral methods (LNPs, nanoparticles, intrathecal or intraparenchymal regional techniques) must be refined for biodistribution, dosage, and immunogenicity in large animals prior to human trials (367). Secondly, it is imperative to minimize genotoxic risk by incorporating base and prime editors that circumvent double-strand breaks, utilizing transient delivery formats (such as RNPs, self-inactivating vectors, and split-intein systems), and conducting thorough, unbiased genome- and transcriptome-wide off-target profiling within preclinical packages (233).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Current findings to potential clinical translation", "text": "Third, the clinical translation is enhanced by established regulatory and biomarker precedents: the approval and regulatory experience with antisense oligonucleotide therapy for SOD1-ALS (tofersen) illustrates that reducing a disease protein and exhibiting biomarker effects (e.g., neurofilament light) can facilitate accelerated pathways while long-term clinical benefits are validated in confirmatory studies. This presents three practical implications for CRISPR strategies: (a) initial trials should incorporate robust molecular biomarkers (Cerebrospinal fluid (CSF)/plasma neurofilament and target RNA/protein measurements), (b) patient selection must closely correspond with the genomic target (e.g., SOD1 or chromosome 9 open reading frame 72 (C9ORF72) carriers), and (c) early regulatory engagement in development is essential (368).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges", "text": "One of the biggest challenges still is effective and focused in vivo distribution of CRISPR components. Developed to improve transport over the BBB to reach particular neural populations are innovations include AAV vectors, LNPs, and modified exosomes (224, 369\u2013371). To fix mutations in SOD1 in ALS mice, for instance, a dual AAV system has been employed to deliver CRISPR\u2013Cas9, hence displaying longer longevity (232). Furthermore, in development are ligand-conjugated nanoparticles to increase CNS cell-type selectivity (372).\n\nCRISPR/dCas9, in conjunction with transcriptional repressors or activators, facilitates epigenetic modulation of gene expression without cleaving DNA. This approach offers a reversible and perhaps safer alternative, especially for multifactorial conditions like PD or sporadic AD. Research has demonstrated that in PD models, dCas9-mediated activation of neuroprotective genes (e.g., BDNF) can preserve dopaminergic neurons (373).\n\nThe advancement of human-relevant disease models is being revolutionized by CRISPR\u2013Cas technology. Modifying disease-related mutations in iPSCs enables them to emulate the pathogenic characteristics of neurodegeneration. Novel therapeutic targets and disease development regulators are being identified using genome-wide CRISPR screening (374). These tools facilitate the identification of significant genetic associations and cellular pathways involved in disease mechanisms.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Challenges", "text": "The integration of CRISPR-based methodologies with other treatment strategies such as RNA interference (RNAi), antisense oligonucleotides (ASOs), and small molecule pharmaceuticals is a significant trend. This combination may mitigate CRISPR-induced risks while simultaneously enhancing therapeutic efficacy. In AD models, targeting both APP and BACE1 genes has demonstrated a synergistic reduction in \u03b2-amyloid levels (375, 376).", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Conclusion", "text": "CRISPR\u2013Cas technology possesses the capacity to transform the treatment of neurodegenerative illnesses through precise, targeted, and adaptable modifications to the genome. Preclinical research indicates that it may rectify deleterious mutations, attenuate detrimental gene expressions, and restore cellular function in conditions such as AD, PD, HD, and ALS. Furthermore, CRISPR-based diagnostic instruments and iPSC-derived models have significantly enhanced our understanding of disease mechanisms. Nonetheless, transforming these advancements into practical therapies is exceptionally challenging, particularly regarding their passage through the BBB, minimizing off-target genetic consequences, and stimulating the immune system. Ethical considerations and long-term safety concerns must be addressed for germline editing. Addressing these issues necessitates a multidisciplinary strategy encompassing biomedical innovation, novel drug delivery methods, and governmental oversight. As the science advances, it will be crucial to develop next-generation CRISPR tools, including base and prime editors, high-fidelity Cas enzymes, and RNA-targeting systems, to progress toward safe and effective therapeutics. Ultimately, CRISPR\u2013Cas represents a significant scientific breakthrough, offering potential for the development of novel therapies for now incurable neurodegenerative diseases.", "title": "" }, { "pmid": 41674784, "pmcid": "PMC12887895", "section": "Conclusion", "text": "CRISPR-based genome editing is transforming approaches to comprehend and address NDs by the precise alteration of genes and pathways linked to these conditions. Recent advancements underscore the potential of integrating high-fidelity CRISPR systems with AI-optimized gRNA to improve specificity and therapeutic safety. Concurrent advancements in patient-derived organoid models offer more physiologically pertinent platforms for evaluating gene-editing results and forecasting clinical behavior. Advancing the area necessitates the incorporation of CRISPR diagnostics, scalable delivery methods, and comprehensive ethical frameworks to convert these technologies into safe and effective medicines. Collectively, these advancements highlight a dramatic path toward individualized, mechanism-driven therapies for intricate NDs.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Results", "text": "A common neuropathological hallmark of polyQ diseases is the accumulation of mutant proteins in neuronal nuclei, where they form detergent-resistant, p62-positive inclusions. To model polyQ aggregation, we used the C-terminal segment of ataxin-3, which harbors the expanded polyQ tract that causes spinocerebellar ataxia 3 and has been shown in prior studies to be highly aggregation prone (13, 14, 15, 16, 17). To create a model amenable to high-throughput analysis, we developed an FRET-based reporter system, which has previously proven useful for detecting aggregation of different disease-associated proteins by flow cytometry (18, 19, 20, 21). We selected the highly efficient FRET pair mNeonGreen and mScarlet (22).", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Results", "text": "We used a lentiviral backbone containing a Tet-On 3G promoter to generate constructs for doxycycline-inducible expression of nuclear-localized mNeonGreen or mScarlet, each N-terminally fused to a C-terminal ataxin-3 fragment with 79 glutamines (Q79). To generate a stable monoclonal cell line, the constructs were packaged into lentivirus and codelivered into human embryonic kidney 293T (HEK293T) cells containing CRISPRi machinery (Fig. 1A). At 5 days after transduction, we incubated the cells with doxycycline overnight, flow sorted for cells containing both fluorescent proteins, and plated cells by serial dilution to isolate a clonal cell line (Fig. 1A). We selected a clone that we henceforth refer to as the \u201cnuclear localization signal (NLS)-FRET-Q79\u201d reporter line. This line was selected among other clones based on flow cytometry showing a tight cluster of cells expressing both fluorescent proteins and the emergence of a population of cells with a higher FRET signal that coincided with the appearance of microscopic nuclear puncta.Figure 1An inducible cell-based model of nuclear, detergent-resistant, p62-positive polyglutamine (polyQ) protein aggregates monitored by an FRET-based reporter.A, schematic of lentiviral constructs for doxycycline-inducible expression of nuclear-localized fluorescent proteins fused to a C-terminal ataxin-3 with 79 glutamines. HEK293T cells engineered with CRISPRi machinery were transduced with these constructs, and clones expressing both fluorescent proteins were selected to generate the NLS-FRET-Q79 cell line. B, fluorescence imaging of NLS-FRET-Q79 cells at 1 day and 5 days of doxycycline. C, immunofluorescence staining for endogenous p62 in NLS-FRET-Q79 cells after 5 days of doxycycline. D, fluorescence imaging of NLS-FRET-Q79 cells at 5 days of doxycycline before and after treatment with detergent, in the same field of view. E, schematic illustrating how polyQ aggregation gives rise to a \u201cFRET-high\u201d population observable by flow cytometry. F, flow cytometry plots at 1 day or 5 days of doxycycline, before and after treatment with detergent. All scale bars represent 10 \u03bcm. CRISPRi, CRISPR interference; HEK293T, human embryonic kidney 293T cell line; NLS, nuclear localization signal.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Results", "text": "An inducible cell-based model of nuclear, detergent-resistant, p62-positive polyglutamine (polyQ) protein aggregates monitored by an FRET-based reporter.A, schematic of lentiviral constructs for doxycycline-inducible expression of nuclear-localized fluorescent proteins fused to a C-terminal ataxin-3 with 79 glutamines. HEK293T cells engineered with CRISPRi machinery were transduced with these constructs, and clones expressing both fluorescent proteins were selected to generate the NLS-FRET-Q79 cell line. B, fluorescence imaging of NLS-FRET-Q79 cells at 1 day and 5 days of doxycycline. C, immunofluorescence staining for endogenous p62 in NLS-FRET-Q79 cells after 5 days of doxycycline. D, fluorescence imaging of NLS-FRET-Q79 cells at 5 days of doxycycline before and after treatment with detergent, in the same field of view. E, schematic illustrating how polyQ aggregation gives rise to a \u201cFRET-high\u201d population observable by flow cytometry. F, flow cytometry plots at 1 day or 5 days of doxycycline, before and after treatment with detergent. All scale bars represent 10 \u03bcm. CRISPRi, CRISPR interference; HEK293T, human embryonic kidney 293T cell line; NLS, nuclear localization signal.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Results", "text": "In the absence of doxycycline, no expression of fluorescent proteins was detected in the NLS-FRET-Q79 cells. One day after doxycycline induction, most cells exhibited diffuse nuclear fluorescence, whereas a subset displayed small nuclear puncta (Fig. 1B). By 5 days of induction, these puncta became more prominent and frequent. The puncta contained both mNeonGreen and mScarlet signals, consistent with polyQ-dependent nucleation and coaggregation. Immunofluorescence staining revealed that the nuclear puncta colocalized with endogenous p62/SQSTM1 (Fig. 1C), a hallmark of polyQ inclusions in human disease. Furthermore, treatment of live cells with 0.5% Triton X-100 resulted in the complete loss of diffuse fluorescence, whereas the nuclear puncta remained intact (Fig. 1D), indicating that these aggregates are detergent resistant. We did not observe any evident toxicity from inducing expression of the FRET reporter (not shown).", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Results", "text": "Aggregation of mNeonGreen and mScarlet is predicted to increase FRET efficiency because of their proximity within polyQ aggregates (Fig. 1E). After 1 day of doxycycline induction, flow cytometry analysis of FRET intensity versus donor mNeonGreen signal shows a dominant single cluster of cells with a small emerging population with increased FRET signal, but nearly all fluorescence was lost upon detergent treatment (Fig. 1F). In contrast, after 5 days of doxycycline induction, a more distinct population of cells emerged with an even higher FRET signal that persisted even after detergent treatment, consistent with the presence of polyQ aggregates seen by microscopy at this time point. We refer to these two populations as \u201cFRET-high\u201d and \u201cFRET-low,\u201d corresponding to aggregated and nonaggregated states, respectively. To confirm that the observed aggregation is polyQ repeat length dependent, we generated plasmids using a nonpathogenic Q24 length. Transient cotransfection of the FRET pair plasmids containing Q79, but not Q24, led to the formation of nuclear puncta that coincided with an increased FRET signal by flow cytometry, which was also resistant to detergent treatment (Fig. S1). Together, these data demonstrate that the NLS-FRET-Q79 cell line recapitulates key features of nuclear polyQ aggregation observed in human disease and provides a robust platform for high-throughput detection of aggregate-containing cells by flow cytometry.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Results", "text": "To identify molecular chaperones that modulate polyQ aggregation, we performed a CRISPRi screen in the NLS-FRET-Q79 cell line using a library of 2103 single guide RNAs (sgRNAs) targeting 356 genes encoding all known molecular chaperones and cochaperones (screen workflow shown in Fig. 2A). The screen revealed several candidate modifiers of aggregation, including Hsp70 chaperones, members of the Hsp40/DNAJ family of cochaperones, and several proteasomal subunits (Fig. 2B). The top hit whose knockdown increased the FRET signal was HSPA8, a constitutively expressed and broadly acting Hsp70 family member involved in ATP-dependent protein folding. In addition, several subunits of the proteasome, including PSMG4, PSMD4, PSMC1, and PSMD2, were among the top hits, where knockdown increased FRET signal. To validate this finding, we treated cells with the proteasome inhibitor carfilzomib, which led to an increase in the FRET-high population (Fig. 2C).Figure 2A CRISPRi screen of molecular chaperones identifies DNAJC7 as a modifier of polyQ aggregation.A, workflow of CRISPRi screen to identify molecular chaperone modifiers of polyQ aggregation in NLS-FRET-Q79 cells. Created with BioRender. B, volcano plot CRISPRi screen results, highlighting DNAJ protein and proteasomal subunit hits. C, flow cytometry plots of NLS-FRET-Q79 cells at 4 days of doxycycline with or without proteasomal inhibitor (carfilzomib). D and E, Western blot (D) and quantification (E) of DNAJC7 levels in NLS-FRET-Q79 cells expressing nontargeting control (NTC) sgRNAs or sgRNAs targeting DNAJC7. Each lane represents lysate from an independent well; bars and error bars indicate mean \u00b1 SD, and points represent n = 3 wells. Signal intensities were normalized to total protein and then to sgNTC#1. F, results of flow cytometry measuring fold change in the fraction of FRET-high cells after 5 days of doxycycline in sgRNA+ cells transduced with sgNTC or sgDNAJC7. Data are presented as fold change relative to the mean of the sgNTCs for each independent experiment. Bars and error bars represent mean \u00b1 SD, and points represent n = 3 independent experiments. \u2217\u2217p < 0.05, one-sample t test compared with sgNTCs (fold change = 1.0). CRISPRi, CRISPR interference. NLS, nuclear localization signal; polyQ, polyglutamine; sgRNA, single guide RNA.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Results", "text": "A CRISPRi screen of molecular chaperones identifies DNAJC7 as a modifier of polyQ aggregation.A, workflow of CRISPRi screen to identify molecular chaperone modifiers of polyQ aggregation in NLS-FRET-Q79 cells. Created with BioRender. B, volcano plot CRISPRi screen results, highlighting DNAJ protein and proteasomal subunit hits. C, flow cytometry plots of NLS-FRET-Q79 cells at 4 days of doxycycline with or without proteasomal inhibitor (carfilzomib). D and E, Western blot (D) and quantification (E) of DNAJC7 levels in NLS-FRET-Q79 cells expressing nontargeting control (NTC) sgRNAs or sgRNAs targeting DNAJC7. Each lane represents lysate from an independent well; bars and error bars indicate mean \u00b1 SD, and points represent n = 3 wells. Signal intensities were normalized to total protein and then to sgNTC#1. F, results of flow cytometry measuring fold change in the fraction of FRET-high cells after 5 days of doxycycline in sgRNA+ cells transduced with sgNTC or sgDNAJC7. Data are presented as fold change relative to the mean of the sgNTCs for each independent experiment. Bars and error bars represent mean \u00b1 SD, and points represent n = 3 independent experiments. \u2217\u2217p < 0.05, one-sample t test compared with sgNTCs (fold change = 1.0). CRISPRi, CRISPR interference. NLS, nuclear localization signal; polyQ, polyglutamine; sgRNA, single guide RNA.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Results", "text": "Whereas Hsp70s and the proteasome play broad, nonspecific roles in protein quality control, the diverse Hsp40/DNAJ family of nearly 50 cochaperones is thought to confer substrate specificity to Hsp70-mediated refolding (23, 24). Among the top DNAJ family hits in our screen, where knockdown increased polyQ aggregation were DNAJB6, DNAJB1, DNAJC24, and DNAJC7, whereas knockdown of DNAJA1 decreased aggregation. These findings are consistent with prior literature: DNAJB6 and DNAJB1 are well-established suppressors of polyQ aggregation (7, 8, 25, 26, 27), and DNAJA1 knockout cells demonstrate reduced polyQ aggregation (28). Thus, our screen independently recovers known modifiers of polyQ aggregation, validating the robustness of the approach.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Results", "text": "Importantly, in addition to known suppressors, the screen identified DNAJC7 and DNAJC24 as previously unexplored hits that significantly increased polyQ aggregation when knocked down. Between these two genes, we prioritized DNAJC7 for further investigation based on several lines of evidence. First, it is highly expressed in the brain (29). Second, it was recently reported to interact with or modify the aggregation of other neurodegenerative disease\u2013associated proteins, including Tau and TAR DNA-binding protein 43 (TDP-43) (30, 31, 32, 33). Third, loss-of-function mutations in DNAJC7 have been linked to amyotrophic lateral sclerosis (34, 35). Finally, it has been previously identified within polyQ inclusions in mouse neuroblastoma cell models (36, 37). In contrast, while DNAJC24 was a stronger hit, it has low brain expression and no known links to neurological disease.\n\nTo validate the effect of DNAJC7 knockdown, we generated NLS-FRET-Q79 lines stably expressing either two different nontargeting control (NTC) sgRNAs or sgRNAs targeting DNAJC7. Western blotting confirmed efficient knockdown of DNAJC7 protein (Fig. 2, D and E). Importantly, flow cytometry showed a significant increase in the fraction of FRET-high cells upon DNAJC7 knockdown (Fig. 2F, gating strategy in Fig. S2). These findings confirmed the predictive capacity of the screen and that DNAJC7 is an endogenous suppressor of polyQ aggregation, at least in the context of this FRET reporter system.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Results", "text": "To determine whether DNAJC7 modifies aggregation in other polyQ disease contexts, and to exclude the possibility that its effects are specific to the nucleus or the specific FRET pair fluorescent proteins, we generated an independent reporter CRISPRi cell line expressing doxycycline-inducible enhanced GFP (eGFP)\u2013tagged Huntingtin exon 1 with 72 glutamines (GFP-HTTex1-Q72) (Fig. 3A). Unlike the NLS-FRET-Q79 construct, Huntingtin exon 1 (HTTex1) naturally contains an N-terminal nuclear export signal, resulting in primarily cytoplasmic aggregation in HEK293T cells (38, 39). After 7 days of doxycycline induction, we observed detergent-resistant cytoplasmic GFP+ inclusions in a small subset (\u223c1\u20132%) of cells (Fig. 3B). CRISPRi knockdown of DNAJC7 in the GFP-HTTex1-Q72 cell line significantly increased the fraction of detergent-insoluble GFP+aggregates, as measured by flow cytometry (Fig. 3C, gating strategy Fig. S2), indicating that DNAJC7 suppresses aggregation in the cytoplasm of a different polyQ protein.Figure 3DNAJC7 suppresses mutant Huntingtin exon 1 (HTTex1) aggregation and colocalizes with inclusions.A, schematic of a lentiviral construct encoding a doxycycline-inducible eGFP-tagged fragment of HTTex1 containing 72 glutamines (GFP-HTTex1-Q72) used to generate a monoclonal HEK293T CRISPRi cell line. B, fluorescence micrograph of GFP-HTTex1-Q72 cells at 7 days of doxycycline treatment before and after treatment with Triton X-100. C, results of flow cytometry experiments measuring the fraction of total events containing detergent-resistant GFP+ cells comparing those stably transduced with NTC or DNAJC7-targeting sgRNAs and normalized to the mean of NTCs. Data represent mean \u00b1 SD from n = 4 independent experiments. \u2217\u2217p < 0.05, one-sample t test compared with sgNTCs (fold change = 1.0). D, fluorescence micrographs of HEK293T cells 48 h after transient cotransfection of GFP-HTTex1-Q72 and either mTagBFP2 (BFP) alone or BFP-tagged DNAJC7. Arrows indicate cytoplasmic aggregates. E, flow cytometry plotting GFP-height versus GFP-width (pulse shape analysis) of HEK293T cells transfected with GFP-HTTex1-Q72 for 48 h. The detergent-resistant population of cells is designated as aggregate positive (Agg+). F, results of flow cytometry experiments measuring Agg+ cells in HEK293T cells 48 h after transient cotransfection with GFP-HTTex1-Q72 and either BFP or BFP-DNAJC7. Bars and error bars represent mean \u00b1 SD from n = 4 independent biological replicates (indicated by color), with each performed in technical triplicate. \u2217p < 0.05 by Student\u2019s t test using mean values of technical replicates. The scale bars represent 10 \u03bcm. CRISPRi, CRISPR interference; eGFP, enhanced GFP; HEK293T, human embryonic kidney 293T cell line; NTC, nontargeting control; sgRNA, single guide RNA.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Results", "text": "DNAJC7 suppresses mutant Huntingtin exon 1 (HTTex1) aggregation and colocalizes with inclusions.A, schematic of a lentiviral construct encoding a doxycycline-inducible eGFP-tagged fragment of HTTex1 containing 72 glutamines (GFP-HTTex1-Q72) used to generate a monoclonal HEK293T CRISPRi cell line. B, fluorescence micrograph of GFP-HTTex1-Q72 cells at 7 days of doxycycline treatment before and after treatment with Triton X-100. C, results of flow cytometry experiments measuring the fraction of total events containing detergent-resistant GFP+ cells comparing those stably transduced with NTC or DNAJC7-targeting sgRNAs and normalized to the mean of NTCs. Data represent mean \u00b1 SD from n = 4 independent experiments. \u2217\u2217p < 0.05, one-sample t test compared with sgNTCs (fold change = 1.0). D, fluorescence micrographs of HEK293T cells 48 h after transient cotransfection of GFP-HTTex1-Q72 and either mTagBFP2 (BFP) alone or BFP-tagged DNAJC7. Arrows indicate cytoplasmic aggregates. E, flow cytometry plotting GFP-height versus GFP-width (pulse shape analysis) of HEK293T cells transfected with GFP-HTTex1-Q72 for 48 h. The detergent-resistant population of cells is designated as aggregate positive (Agg+). F, results of flow cytometry experiments measuring Agg+ cells in HEK293T cells 48 h after transient cotransfection with GFP-HTTex1-Q72 and either BFP or BFP-DNAJC7. Bars and error bars represent mean \u00b1 SD from n = 4 independent biological replicates (indicated by color), with each performed in technical triplicate. \u2217p < 0.05 by Student\u2019s t test using mean values of technical replicates. The scale bars represent 10 \u03bcm. CRISPRi, CRISPR interference; eGFP, enhanced GFP; HEK293T, human embryonic kidney 293T cell line; NTC, nontargeting control; sgRNA, single guide RNA.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Results", "text": "We next sought to determine whether DNAJC7 colocalizes with HTTex1 aggregates. However, attempts to visualize endogenous DNAJC7 by immunostaining were unsuccessful because of the lack of a suitable antibody. To address this, we performed transient cotransfection experiments in HEK293T cells using GFP-HTTex1-Q72 together with either mTagBFP2 (BFP) alone or BFP-tagged DNAJC7. As expected, GFP-HTTex1-Q72 formed cytoplasmic aggregates. Notably, BFP-DNAJC7, but not BFP alone, colocalized with a subset of these aggregates (Fig. 3D), suggesting an interaction between DNAJC7 and aggregated HTTex1.\n\nTo assess whether DNAJC7 overexpression could reduce HTTex1 aggregation, we used flow cytometry pulse-shape analysis, which distinguishes aggregates based on a characteristic high-intensity and narrow-width fluorescence signal (27, 40). At 48 h post-transfection, coinciding with the appearance of aggregates by microscopy, we observed a distinct population of cells with high and narrow GFP signal, designated as Agg+ (aggregate-positive), which persisted following detergent treatment (Fig. 3E). Compared with BFP control, overexpression of BFP-DNAJC7 significantly reduced the fraction of Agg+ cells (Fig. 3F), indicating that DNAJC7 suppresses aggregation of mutant HTTex1 when overexpressed. BFP-DNAJC7 did not reduce the total number of GFP-positive cells compared with BFP (Fig. S3), indicating similar transfection efficiencies.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Results", "text": "Given the emerging role of DNAJC7 as a modifier of multiple disease\u2013associated protein aggregates, we next sought to examine its relevance in a distinct repeat expansion disorder. We focused on NIID, which is caused by an expanded polyG tract encoded within the upstream ORF of the NOTCH2NLC (uN2C) gene and, like the polyQ diseases, is pathologically characterized by the presence of p62-positive nuclear inclusions (41, 42).", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Results", "text": "Using a parallel approach to our polyQ model, we developed a clonal HEK293T CRISPRi cell line containing a doxycycline-inducible, nuclear-localized FRET reporter expressing uN2C containing 100 glycines (designated NLS-FRET-G100) (Fig. 4A). Five days after doxycycline induction, a subset of cells exhibited nuclear puncta positive for both fluoroscent proteins, which colocalized with p62 (Fig. 4B). These puncta were resistant to Triton X-100 treatment, indicating that they form detergent-insoluble aggregates (Fig. 4C). Flow cytometry revealed a progressive increase in the FRET-high population over time, which was maintained after detergent treatment (Fig. 4D). Inhibition of the proteasome with carfilzomib further increased the FRET-high population (Fig. 4E). By transient transfection experiments, we confirmed that uN2C-G100, but not uN2C-G12, expression led to the formation of detergent-resistant FRET-high cells (Fig. S4).Figure 4An inducible model of polyglycine (polyG) aggregation reveals detergent-resistant, p62-positive nuclear inclusions, and seeding activity.A, schematic of lentiviral constructs for generating the NLS-FRET-G100 cell line expressing the upstream ORF (uORF) of the NOTCH2NLC gene with a polyG tract of 100 residues. B, immunofluorescence staining for p62 of NLS-FRET-G100 cells at 5 days of doxycycline. C, fluorescence imaging of NLS-FRET-G100 cells at 5 days of doxycycline before and after detergent treatment, in the same field of view. D, flow cytometry plots at 1 and 5 days of doxycycline treatment and the latter before and after detergent treatment. E, flow cytometry plots of NLS-FRET-G100 cells at 4 days of doxycycline with or without proteasomal inhibitor (carfilzomib). All scale bars represent 10 \u03bcm. NLS, nuclear localization signal.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Results", "text": "An inducible model of polyglycine (polyG) aggregation reveals detergent-resistant, p62-positive nuclear inclusions, and seeding activity.A, schematic of lentiviral constructs for generating the NLS-FRET-G100 cell line expressing the upstream ORF (uORF) of the NOTCH2NLC gene with a polyG tract of 100 residues. B, immunofluorescence staining for p62 of NLS-FRET-G100 cells at 5 days of doxycycline. C, fluorescence imaging of NLS-FRET-G100 cells at 5 days of doxycycline before and after detergent treatment, in the same field of view. D, flow cytometry plots at 1 and 5 days of doxycycline treatment and the latter before and after detergent treatment. E, flow cytometry plots of NLS-FRET-G100 cells at 4 days of doxycycline with or without proteasomal inhibitor (carfilzomib). All scale bars represent 10 \u03bcm. NLS, nuclear localization signal.\n\nTogether, these findings establish NLS-FRET-G100 as a robust live-cell reporter of nuclear polyG aggregation that recapitulates key pathological features of NIID and enables quantitative, high-throughput analysis.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Results", "text": "To determine whether aggregate formation by each FRET reporter is driven by the specific amino acid repeat rather than nonspecific aggregation of fluorescent tags or other aggregation-prone proteins, we cotransfected cells with different combinations of mScarlet- or mNeonGreen-tagged Q24 or Q79 constructs, uN2C-G12 or uN2C-G100 constructs, as well as GFP-HTTex1-Q25 or -Q72. The NLS was removed from all constructs to promote predominantly cytosolic localization, as HTTex1 is known to localize primarily to the cytoplasm. We observed that HTTex1-Q72 inclusions sequestered both Q24 and Q79 but not uN2C-G12 proteins (Fig. 5). Conversely, uN2C-G100 inclusions sequestered uN2C-G12 proteins despite the short repeat length but did not recruit Q24, HTTex1-Q25, or HTTex1-Q72 proteins. In all cases, the fluorescent proteins alone were not recruited into inclusions. The sequestration of proteins containing nonexpanded polyQ repeats by polyQ aggregates has been reported previously (17, 43, 44, 45, 46). Recent work has similarly shown sequestration of endogenous polyG-containing proteins, such as FAM98B, into polyG aggregates (47). Interestingly, in cells containing both HTTex1-Q72 and uN2C-G100 inclusions, aggregates were most frequently adjacent to each other, whereas only some showed partial overlap. Similar spatial segregation was reported previously between HTTex1 and poly(glycine\u2013alanine)dipeptide repeat aggregates (48). In sum, these findings confirmed aggregation in the FRET reporter is driven in a homotypic amino acid repeat\u2013dependent manner and can lead to sequestration of other proteins containing even relatively short repeats of the same amino acid.Figure 5PolyQ and polyG proteins undergo independent, repeat-dependent aggregation. Fluorescence micrographs of HEK293T cells cotransfected with the indicated combinations of fluorescently tagged polyQ and polyG proteins. The scale bar represents 10 \u03bcm. HEK293T, human embryonic kidney 293 cell line; polyG, polyglucine; polyQ, polyglutamine.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Results", "text": "PolyQ and polyG proteins undergo independent, repeat-dependent aggregation. Fluorescence micrographs of HEK293T cells cotransfected with the indicated combinations of fluorescently tagged polyQ and polyG proteins. The scale bar represents 10 \u03bcm. HEK293T, human embryonic kidney 293 cell line; polyG, polyglucine; polyQ, polyglutamine.\n\nWe also tested whether polyQ or polyG aggregates from homogenized cells or tissues could seed aggregation of the FRET reporter. NLS-FRET-Q79 cells transfected with homogenates of NLS-FRET-Q79 cells or from cortical tissue of transgenic HD mice led to an increase in FRET-high cells (Fig. S5). We used 22-week-old mice of the R6/1 HD mouse model, which expresses HTTex1 with approximately 115 glutamines and exhibits widespread aggregation of mutant HTT throughout the brain beginning as early as 8 weeks of age (49, 50). Immunostaining the contralateral hemispheres of these mice confirmed frequent p62-positive nuclear inclusions in the cortex (Fig. S6). NLS-FRET-G100 cell homogenates did not increase FRET-high cells in the NLS-FRET-Q79 line but robustly increased it in the NLS-FRET-G100 cells. These results indicate that the FRET reporter lines could potentially be used as a \u201cbiosensor\u201d to test the presence of polyQ aggregates of other cells, tissues, or biofluids, as for FRET-based aggregation reporters of Tau and TDP-43 (19, 20, 21, 51, 52, 53).", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Results", "text": "Using the same CRISPRi screening approach as with the NLS-FRET-Q79 cell line, we performed a molecular chaperone screen in the NLS-FRET-G100 reporter line to identify modifiers of polyG aggregation (Fig. 6A). Surprisingly, this screen revealed relatively few significant hits. Notably, key polyQ modifiers, such as DNAJC7, DNAJB6, DNAJB1, and HSPA8, were not identified as hits in the polyG screen. A direct comparison of gene scores between the polyQ and polyG screens revealed minimal overlap in chaperone modifiers (Fig. 6B), suggesting distinct mechanisms of proteostasis regulation between polyQ and polyG aggregates.Figure 6Chaperone CRISPRi screening reveals limited suppression of polyG aggregation despite effects by DNAJC7 overexpression.A, volcano plot showing results of a CRISPRi screen for molecular chaperone modifiers of polyG aggregation in the NLS-FRET-G100 cell line. B, pairwise comparison of gene scores between the polyG screen (from A) and the polyQ screen (Fig. 2B). C, results of flow cytometry experiments measuring fold change in the fraction of FRET-high cells after 5 days of doxycycline in sgRNA+ cells transduced with NTC or DNAJC7. Data represent mean \u00b1 SD from n = 3 independent experiments. D, representative fluorescence micrographs of HEK293T cells transiently cotransfected with NLS-mScarlet- and NLS-mNeonGreen-tagged uN2C-G100 together with either BFP or BFP-DNAJC7; only the mNeonGreen channel is shown. The scale bar represents 10 \u03bcm. E, flow cytometry quantification of the percentage of FRET-high cells relative to total cells expressing both mNeonGreen and mScarlet. Bars indicate mean \u00b1 SD from n = 3 independent experiments. \u2217p < 0.05 by Student\u2019s t test. CRISPRi, CRISPR interference; HEK293T, human embryonic kidney 293T cell line; NLS, nuclear localization signal; NTC, nontargeting control; polyG, polyglucine; polyQ, polyglutamine; sgRNA, single guide RNA.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Results", "text": "Chaperone CRISPRi screening reveals limited suppression of polyG aggregation despite effects by DNAJC7 overexpression.A, volcano plot showing results of a CRISPRi screen for molecular chaperone modifiers of polyG aggregation in the NLS-FRET-G100 cell line. B, pairwise comparison of gene scores between the polyG screen (from A) and the polyQ screen (Fig. 2B). C, results of flow cytometry experiments measuring fold change in the fraction of FRET-high cells after 5 days of doxycycline in sgRNA+ cells transduced with NTC or DNAJC7. Data represent mean \u00b1 SD from n = 3 independent experiments. D, representative fluorescence micrographs of HEK293T cells transiently cotransfected with NLS-mScarlet- and NLS-mNeonGreen-tagged uN2C-G100 together with either BFP or BFP-DNAJC7; only the mNeonGreen channel is shown. The scale bar represents 10 \u03bcm. E, flow cytometry quantification of the percentage of FRET-high cells relative to total cells expressing both mNeonGreen and mScarlet. Bars indicate mean \u00b1 SD from n = 3 independent experiments. \u2217p < 0.05 by Student\u2019s t test. CRISPRi, CRISPR interference; HEK293T, human embryonic kidney 293T cell line; NLS, nuclear localization signal; NTC, nontargeting control; polyG, polyglucine; polyQ, polyglutamine; sgRNA, single guide RNA.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Results", "text": "Among the few shared hits was OGT (O-GlcNAc transferase), which encodes O-GlcNAc transferase. OGT knockdown was among the top hits in both screens, and we independently validated that its depletion significantly increases the fraction of FRET-high cells in both the NLS-FRET-Q79 and NLS-FRET-G100 lines (Fig. S7), confirming the reliability of the polyG screen and confirming active CRISPRi machinery. In contrast, direct knockdown of DNAJC7 had no significant effect on the proportion of FRET-high cells in the NLS-FRET-G100 model (Fig. 6C).\n\nTo determine whether this difference was associated with a lack of DNAJC7 colocalization with polyG inclusions, we transiently transfected cells with NLS- and mScarlet- and mNeonGreen-tagged uN2C-G100 together with either BFP or BFP-DNAJC7, as described above. Similar to observations with GFP-HTTex1-Q72 inclusions, a subset of polyG inclusions colocalized with BFP-DNAJC7 (Fig. 6D). Consistent with this finding, flow cytometry revealed a significant reduction in the fraction of FRET-high cells (Fig. 6E). We repeated these experiments using a non-nuclear mScarlet/mNeonGreen-tagged uN2C-G100 and observed the same results (Fig. S8).", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Results", "text": "To exclude the possibility that the observed effects of DNAJC7 knockdown were due to clonal artifacts of the selected FRET reporter lines, we examined one additional polyQ and multiple other polyG monoclonal cell lines. Consistent with results obtained with the NLS-FRET-Q79 and HTTex1-Q72 lines, DNAJC7 knockdown robustly increased aggregation in the polyQ line (Fig. S9). In contrast, analysis of three additional uN2C-G100 monoclonal lines with comparable FRET signal properties, two of which lacked nuclear localization, revealed no effect of DNAJC7 knockdown on aggregation in any clone. Together, these data support that polyG aggregation is insensitive to DNAJC7 knockdown, whereas DNAJC7 overexpression can reduce polyG aggregation.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "A FRET-based reporter of polyQ protein aggregation in the nucleus", "text": "A common neuropathological hallmark of polyQ diseases is the accumulation of mutant proteins in neuronal nuclei, where they form detergent-resistant, p62-positive inclusions. To model polyQ aggregation, we used the C-terminal segment of ataxin-3, which harbors the expanded polyQ tract that causes spinocerebellar ataxia 3 and has been shown in prior studies to be highly aggregation prone (13, 14, 15, 16, 17). To create a model amenable to high-throughput analysis, we developed an FRET-based reporter system, which has previously proven useful for detecting aggregation of different disease-associated proteins by flow cytometry (18, 19, 20, 21). We selected the highly efficient FRET pair mNeonGreen and mScarlet (22).", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "A FRET-based reporter of polyQ protein aggregation in the nucleus", "text": "We used a lentiviral backbone containing a Tet-On 3G promoter to generate constructs for doxycycline-inducible expression of nuclear-localized mNeonGreen or mScarlet, each N-terminally fused to a C-terminal ataxin-3 fragment with 79 glutamines (Q79). To generate a stable monoclonal cell line, the constructs were packaged into lentivirus and codelivered into human embryonic kidney 293T (HEK293T) cells containing CRISPRi machinery (Fig. 1A). At 5 days after transduction, we incubated the cells with doxycycline overnight, flow sorted for cells containing both fluorescent proteins, and plated cells by serial dilution to isolate a clonal cell line (Fig. 1A). We selected a clone that we henceforth refer to as the \u201cnuclear localization signal (NLS)-FRET-Q79\u201d reporter line. This line was selected among other clones based on flow cytometry showing a tight cluster of cells expressing both fluorescent proteins and the emergence of a population of cells with a higher FRET signal that coincided with the appearance of microscopic nuclear puncta.Figure 1An inducible cell-based model of nuclear, detergent-resistant, p62-positive polyglutamine (polyQ) protein aggregates monitored by an FRET-based reporter.A, schematic of lentiviral constructs for doxycycline-inducible expression of nuclear-localized fluorescent proteins fused to a C-terminal ataxin-3 with 79 glutamines. HEK293T cells engineered with CRISPRi machinery were transduced with these constructs, and clones expressing both fluorescent proteins were selected to generate the NLS-FRET-Q79 cell line. B, fluorescence imaging of NLS-FRET-Q79 cells at 1 day and 5 days of doxycycline. C, immunofluorescence staining for endogenous p62 in NLS-FRET-Q79 cells after 5 days of doxycycline. D, fluorescence imaging of NLS-FRET-Q79 cells at 5 days of doxycycline before and after treatment with detergent, in the same field of view. E, schematic illustrating how polyQ aggregation gives rise to a \u201cFRET-high\u201d population observable by flow cytometry. F, flow cytometry plots at 1 day or 5 days of doxycycline, before and after treatment with detergent. All scale bars represent 10 \u03bcm. CRISPRi, CRISPR interference; HEK293T, human embryonic kidney 293T cell line; NLS, nuclear localization signal.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "A FRET-based reporter of polyQ protein aggregation in the nucleus", "text": "An inducible cell-based model of nuclear, detergent-resistant, p62-positive polyglutamine (polyQ) protein aggregates monitored by an FRET-based reporter.A, schematic of lentiviral constructs for doxycycline-inducible expression of nuclear-localized fluorescent proteins fused to a C-terminal ataxin-3 with 79 glutamines. HEK293T cells engineered with CRISPRi machinery were transduced with these constructs, and clones expressing both fluorescent proteins were selected to generate the NLS-FRET-Q79 cell line. B, fluorescence imaging of NLS-FRET-Q79 cells at 1 day and 5 days of doxycycline. C, immunofluorescence staining for endogenous p62 in NLS-FRET-Q79 cells after 5 days of doxycycline. D, fluorescence imaging of NLS-FRET-Q79 cells at 5 days of doxycycline before and after treatment with detergent, in the same field of view. E, schematic illustrating how polyQ aggregation gives rise to a \u201cFRET-high\u201d population observable by flow cytometry. F, flow cytometry plots at 1 day or 5 days of doxycycline, before and after treatment with detergent. All scale bars represent 10 \u03bcm. CRISPRi, CRISPR interference; HEK293T, human embryonic kidney 293T cell line; NLS, nuclear localization signal.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "A FRET-based reporter of polyQ protein aggregation in the nucleus", "text": "In the absence of doxycycline, no expression of fluorescent proteins was detected in the NLS-FRET-Q79 cells. One day after doxycycline induction, most cells exhibited diffuse nuclear fluorescence, whereas a subset displayed small nuclear puncta (Fig. 1B). By 5 days of induction, these puncta became more prominent and frequent. The puncta contained both mNeonGreen and mScarlet signals, consistent with polyQ-dependent nucleation and coaggregation. Immunofluorescence staining revealed that the nuclear puncta colocalized with endogenous p62/SQSTM1 (Fig. 1C), a hallmark of polyQ inclusions in human disease. Furthermore, treatment of live cells with 0.5% Triton X-100 resulted in the complete loss of diffuse fluorescence, whereas the nuclear puncta remained intact (Fig. 1D), indicating that these aggregates are detergent resistant. We did not observe any evident toxicity from inducing expression of the FRET reporter (not shown).", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "A FRET-based reporter of polyQ protein aggregation in the nucleus", "text": "Aggregation of mNeonGreen and mScarlet is predicted to increase FRET efficiency because of their proximity within polyQ aggregates (Fig. 1E). After 1 day of doxycycline induction, flow cytometry analysis of FRET intensity versus donor mNeonGreen signal shows a dominant single cluster of cells with a small emerging population with increased FRET signal, but nearly all fluorescence was lost upon detergent treatment (Fig. 1F). In contrast, after 5 days of doxycycline induction, a more distinct population of cells emerged with an even higher FRET signal that persisted even after detergent treatment, consistent with the presence of polyQ aggregates seen by microscopy at this time point. We refer to these two populations as \u201cFRET-high\u201d and \u201cFRET-low,\u201d corresponding to aggregated and nonaggregated states, respectively. To confirm that the observed aggregation is polyQ repeat length dependent, we generated plasmids using a nonpathogenic Q24 length. Transient cotransfection of the FRET pair plasmids containing Q79, but not Q24, led to the formation of nuclear puncta that coincided with an increased FRET signal by flow cytometry, which was also resistant to detergent treatment (Fig. S1). Together, these data demonstrate that the NLS-FRET-Q79 cell line recapitulates key features of nuclear polyQ aggregation observed in human disease and provides a robust platform for high-throughput detection of aggregate-containing cells by flow cytometry.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "A CRISPRi screen identifies DNAJC7 as a suppressor of polyQ aggregation", "text": "To identify molecular chaperones that modulate polyQ aggregation, we performed a CRISPRi screen in the NLS-FRET-Q79 cell line using a library of 2103 single guide RNAs (sgRNAs) targeting 356 genes encoding all known molecular chaperones and cochaperones (screen workflow shown in Fig. 2A). The screen revealed several candidate modifiers of aggregation, including Hsp70 chaperones, members of the Hsp40/DNAJ family of cochaperones, and several proteasomal subunits (Fig. 2B). The top hit whose knockdown increased the FRET signal was HSPA8, a constitutively expressed and broadly acting Hsp70 family member involved in ATP-dependent protein folding. In addition, several subunits of the proteasome, including PSMG4, PSMD4, PSMC1, and PSMD2, were among the top hits, where knockdown increased FRET signal. To validate this finding, we treated cells with the proteasome inhibitor carfilzomib, which led to an increase in the FRET-high population (Fig. 2C).Figure 2A CRISPRi screen of molecular chaperones identifies DNAJC7 as a modifier of polyQ aggregation.A, workflow of CRISPRi screen to identify molecular chaperone modifiers of polyQ aggregation in NLS-FRET-Q79 cells. Created with BioRender. B, volcano plot CRISPRi screen results, highlighting DNAJ protein and proteasomal subunit hits. C, flow cytometry plots of NLS-FRET-Q79 cells at 4 days of doxycycline with or without proteasomal inhibitor (carfilzomib). D and E, Western blot (D) and quantification (E) of DNAJC7 levels in NLS-FRET-Q79 cells expressing nontargeting control (NTC) sgRNAs or sgRNAs targeting DNAJC7. Each lane represents lysate from an independent well; bars and error bars indicate mean \u00b1 SD, and points represent n = 3 wells. Signal intensities were normalized to total protein and then to sgNTC#1. F, results of flow cytometry measuring fold change in the fraction of FRET-high cells after 5 days of doxycycline in sgRNA+ cells transduced with sgNTC or sgDNAJC7. Data are presented as fold change relative to the mean of the sgNTCs for each independent experiment. Bars and error bars represent mean \u00b1 SD, and points represent n = 3 independent experiments. \u2217\u2217p < 0.05, one-sample t test compared with sgNTCs (fold change = 1.0). CRISPRi, CRISPR interference. NLS, nuclear localization signal; polyQ, polyglutamine; sgRNA, single guide RNA.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "A CRISPRi screen identifies DNAJC7 as a suppressor of polyQ aggregation", "text": "A CRISPRi screen of molecular chaperones identifies DNAJC7 as a modifier of polyQ aggregation.A, workflow of CRISPRi screen to identify molecular chaperone modifiers of polyQ aggregation in NLS-FRET-Q79 cells. Created with BioRender. B, volcano plot CRISPRi screen results, highlighting DNAJ protein and proteasomal subunit hits. C, flow cytometry plots of NLS-FRET-Q79 cells at 4 days of doxycycline with or without proteasomal inhibitor (carfilzomib). D and E, Western blot (D) and quantification (E) of DNAJC7 levels in NLS-FRET-Q79 cells expressing nontargeting control (NTC) sgRNAs or sgRNAs targeting DNAJC7. Each lane represents lysate from an independent well; bars and error bars indicate mean \u00b1 SD, and points represent n = 3 wells. Signal intensities were normalized to total protein and then to sgNTC#1. F, results of flow cytometry measuring fold change in the fraction of FRET-high cells after 5 days of doxycycline in sgRNA+ cells transduced with sgNTC or sgDNAJC7. Data are presented as fold change relative to the mean of the sgNTCs for each independent experiment. Bars and error bars represent mean \u00b1 SD, and points represent n = 3 independent experiments. \u2217\u2217p < 0.05, one-sample t test compared with sgNTCs (fold change = 1.0). CRISPRi, CRISPR interference. NLS, nuclear localization signal; polyQ, polyglutamine; sgRNA, single guide RNA.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "A CRISPRi screen identifies DNAJC7 as a suppressor of polyQ aggregation", "text": "Whereas Hsp70s and the proteasome play broad, nonspecific roles in protein quality control, the diverse Hsp40/DNAJ family of nearly 50 cochaperones is thought to confer substrate specificity to Hsp70-mediated refolding (23, 24). Among the top DNAJ family hits in our screen, where knockdown increased polyQ aggregation were DNAJB6, DNAJB1, DNAJC24, and DNAJC7, whereas knockdown of DNAJA1 decreased aggregation. These findings are consistent with prior literature: DNAJB6 and DNAJB1 are well-established suppressors of polyQ aggregation (7, 8, 25, 26, 27), and DNAJA1 knockout cells demonstrate reduced polyQ aggregation (28). Thus, our screen independently recovers known modifiers of polyQ aggregation, validating the robustness of the approach.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "A CRISPRi screen identifies DNAJC7 as a suppressor of polyQ aggregation", "text": "Importantly, in addition to known suppressors, the screen identified DNAJC7 and DNAJC24 as previously unexplored hits that significantly increased polyQ aggregation when knocked down. Between these two genes, we prioritized DNAJC7 for further investigation based on several lines of evidence. First, it is highly expressed in the brain (29). Second, it was recently reported to interact with or modify the aggregation of other neurodegenerative disease\u2013associated proteins, including Tau and TAR DNA-binding protein 43 (TDP-43) (30, 31, 32, 33). Third, loss-of-function mutations in DNAJC7 have been linked to amyotrophic lateral sclerosis (34, 35). Finally, it has been previously identified within polyQ inclusions in mouse neuroblastoma cell models (36, 37). In contrast, while DNAJC24 was a stronger hit, it has low brain expression and no known links to neurological disease.\n\nTo validate the effect of DNAJC7 knockdown, we generated NLS-FRET-Q79 lines stably expressing either two different nontargeting control (NTC) sgRNAs or sgRNAs targeting DNAJC7. Western blotting confirmed efficient knockdown of DNAJC7 protein (Fig. 2, D and E). Importantly, flow cytometry showed a significant increase in the fraction of FRET-high cells upon DNAJC7 knockdown (Fig. 2F, gating strategy in Fig. S2). These findings confirmed the predictive capacity of the screen and that DNAJC7 is an endogenous suppressor of polyQ aggregation, at least in the context of this FRET reporter system.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "DNAJC7 suppresses mutant HTT exon 1 aggregation and colocalizes with inclusions in cells", "text": "To determine whether DNAJC7 modifies aggregation in other polyQ disease contexts, and to exclude the possibility that its effects are specific to the nucleus or the specific FRET pair fluorescent proteins, we generated an independent reporter CRISPRi cell line expressing doxycycline-inducible enhanced GFP (eGFP)\u2013tagged Huntingtin exon 1 with 72 glutamines (GFP-HTTex1-Q72) (Fig. 3A). Unlike the NLS-FRET-Q79 construct, Huntingtin exon 1 (HTTex1) naturally contains an N-terminal nuclear export signal, resulting in primarily cytoplasmic aggregation in HEK293T cells (38, 39). After 7 days of doxycycline induction, we observed detergent-resistant cytoplasmic GFP+ inclusions in a small subset (\u223c1\u20132%) of cells (Fig. 3B). CRISPRi knockdown of DNAJC7 in the GFP-HTTex1-Q72 cell line significantly increased the fraction of detergent-insoluble GFP+aggregates, as measured by flow cytometry (Fig. 3C, gating strategy Fig. S2), indicating that DNAJC7 suppresses aggregation in the cytoplasm of a different polyQ protein.Figure 3DNAJC7 suppresses mutant Huntingtin exon 1 (HTTex1) aggregation and colocalizes with inclusions.A, schematic of a lentiviral construct encoding a doxycycline-inducible eGFP-tagged fragment of HTTex1 containing 72 glutamines (GFP-HTTex1-Q72) used to generate a monoclonal HEK293T CRISPRi cell line. B, fluorescence micrograph of GFP-HTTex1-Q72 cells at 7 days of doxycycline treatment before and after treatment with Triton X-100. C, results of flow cytometry experiments measuring the fraction of total events containing detergent-resistant GFP+ cells comparing those stably transduced with NTC or DNAJC7-targeting sgRNAs and normalized to the mean of NTCs. Data represent mean \u00b1 SD from n = 4 independent experiments. \u2217\u2217p < 0.05, one-sample t test compared with sgNTCs (fold change = 1.0). D, fluorescence micrographs of HEK293T cells 48 h after transient cotransfection of GFP-HTTex1-Q72 and either mTagBFP2 (BFP) alone or BFP-tagged DNAJC7. Arrows indicate cytoplasmic aggregates. E, flow cytometry plotting GFP-height versus GFP-width (pulse shape analysis) of HEK293T cells transfected with GFP-HTTex1-Q72 for 48 h. The detergent-resistant population of cells is designated as aggregate positive (Agg+). F, results of flow cytometry experiments measuring Agg+ cells in HEK293T cells 48 h after transient cotransfection with GFP-HTTex1-Q72 and either BFP or BFP-DNAJC7. Bars and error bars represent mean \u00b1 SD from n = 4 independent biological replicates (indicated by color), with each performed in technical triplicate. \u2217p < 0.05 by Student\u2019s t test using mean values of technical replicates. The scale bars represent 10 \u03bcm. CRISPRi, CRISPR interference; eGFP, enhanced GFP; HEK293T, human embryonic kidney 293T cell line; NTC, nontargeting control; sgRNA, single guide RNA.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "DNAJC7 suppresses mutant HTT exon 1 aggregation and colocalizes with inclusions in cells", "text": "DNAJC7 suppresses mutant Huntingtin exon 1 (HTTex1) aggregation and colocalizes with inclusions.A, schematic of a lentiviral construct encoding a doxycycline-inducible eGFP-tagged fragment of HTTex1 containing 72 glutamines (GFP-HTTex1-Q72) used to generate a monoclonal HEK293T CRISPRi cell line. B, fluorescence micrograph of GFP-HTTex1-Q72 cells at 7 days of doxycycline treatment before and after treatment with Triton X-100. C, results of flow cytometry experiments measuring the fraction of total events containing detergent-resistant GFP+ cells comparing those stably transduced with NTC or DNAJC7-targeting sgRNAs and normalized to the mean of NTCs. Data represent mean \u00b1 SD from n = 4 independent experiments. \u2217\u2217p < 0.05, one-sample t test compared with sgNTCs (fold change = 1.0). D, fluorescence micrographs of HEK293T cells 48 h after transient cotransfection of GFP-HTTex1-Q72 and either mTagBFP2 (BFP) alone or BFP-tagged DNAJC7. Arrows indicate cytoplasmic aggregates. E, flow cytometry plotting GFP-height versus GFP-width (pulse shape analysis) of HEK293T cells transfected with GFP-HTTex1-Q72 for 48 h. The detergent-resistant population of cells is designated as aggregate positive (Agg+). F, results of flow cytometry experiments measuring Agg+ cells in HEK293T cells 48 h after transient cotransfection with GFP-HTTex1-Q72 and either BFP or BFP-DNAJC7. Bars and error bars represent mean \u00b1 SD from n = 4 independent biological replicates (indicated by color), with each performed in technical triplicate. \u2217p < 0.05 by Student\u2019s t test using mean values of technical replicates. The scale bars represent 10 \u03bcm. CRISPRi, CRISPR interference; eGFP, enhanced GFP; HEK293T, human embryonic kidney 293T cell line; NTC, nontargeting control; sgRNA, single guide RNA.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "DNAJC7 suppresses mutant HTT exon 1 aggregation and colocalizes with inclusions in cells", "text": "We next sought to determine whether DNAJC7 colocalizes with HTTex1 aggregates. However, attempts to visualize endogenous DNAJC7 by immunostaining were unsuccessful because of the lack of a suitable antibody. To address this, we performed transient cotransfection experiments in HEK293T cells using GFP-HTTex1-Q72 together with either mTagBFP2 (BFP) alone or BFP-tagged DNAJC7. As expected, GFP-HTTex1-Q72 formed cytoplasmic aggregates. Notably, BFP-DNAJC7, but not BFP alone, colocalized with a subset of these aggregates (Fig. 3D), suggesting an interaction between DNAJC7 and aggregated HTTex1.\n\nTo assess whether DNAJC7 overexpression could reduce HTTex1 aggregation, we used flow cytometry pulse-shape analysis, which distinguishes aggregates based on a characteristic high-intensity and narrow-width fluorescence signal (27, 40). At 48 h post-transfection, coinciding with the appearance of aggregates by microscopy, we observed a distinct population of cells with high and narrow GFP signal, designated as Agg+ (aggregate-positive), which persisted following detergent treatment (Fig. 3E). Compared with BFP control, overexpression of BFP-DNAJC7 significantly reduced the fraction of Agg+ cells (Fig. 3F), indicating that DNAJC7 suppresses aggregation of mutant HTTex1 when overexpressed. BFP-DNAJC7 did not reduce the total number of GFP-positive cells compared with BFP (Fig. S3), indicating similar transfection efficiencies.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "An FRET-based reporter for polyG aggregation reveals detergent-resistant, p62-positive nuclear inclusions", "text": "Given the emerging role of DNAJC7 as a modifier of multiple disease\u2013associated protein aggregates, we next sought to examine its relevance in a distinct repeat expansion disorder. We focused on NIID, which is caused by an expanded polyG tract encoded within the upstream ORF of the NOTCH2NLC (uN2C) gene and, like the polyQ diseases, is pathologically characterized by the presence of p62-positive nuclear inclusions (41, 42).", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "An FRET-based reporter for polyG aggregation reveals detergent-resistant, p62-positive nuclear inclusions", "text": "Using a parallel approach to our polyQ model, we developed a clonal HEK293T CRISPRi cell line containing a doxycycline-inducible, nuclear-localized FRET reporter expressing uN2C containing 100 glycines (designated NLS-FRET-G100) (Fig. 4A). Five days after doxycycline induction, a subset of cells exhibited nuclear puncta positive for both fluoroscent proteins, which colocalized with p62 (Fig. 4B). These puncta were resistant to Triton X-100 treatment, indicating that they form detergent-insoluble aggregates (Fig. 4C). Flow cytometry revealed a progressive increase in the FRET-high population over time, which was maintained after detergent treatment (Fig. 4D). Inhibition of the proteasome with carfilzomib further increased the FRET-high population (Fig. 4E). By transient transfection experiments, we confirmed that uN2C-G100, but not uN2C-G12, expression led to the formation of detergent-resistant FRET-high cells (Fig. S4).Figure 4An inducible model of polyglycine (polyG) aggregation reveals detergent-resistant, p62-positive nuclear inclusions, and seeding activity.A, schematic of lentiviral constructs for generating the NLS-FRET-G100 cell line expressing the upstream ORF (uORF) of the NOTCH2NLC gene with a polyG tract of 100 residues. B, immunofluorescence staining for p62 of NLS-FRET-G100 cells at 5 days of doxycycline. C, fluorescence imaging of NLS-FRET-G100 cells at 5 days of doxycycline before and after detergent treatment, in the same field of view. D, flow cytometry plots at 1 and 5 days of doxycycline treatment and the latter before and after detergent treatment. E, flow cytometry plots of NLS-FRET-G100 cells at 4 days of doxycycline with or without proteasomal inhibitor (carfilzomib). All scale bars represent 10 \u03bcm. NLS, nuclear localization signal.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "An FRET-based reporter for polyG aggregation reveals detergent-resistant, p62-positive nuclear inclusions", "text": "An inducible model of polyglycine (polyG) aggregation reveals detergent-resistant, p62-positive nuclear inclusions, and seeding activity.A, schematic of lentiviral constructs for generating the NLS-FRET-G100 cell line expressing the upstream ORF (uORF) of the NOTCH2NLC gene with a polyG tract of 100 residues. B, immunofluorescence staining for p62 of NLS-FRET-G100 cells at 5 days of doxycycline. C, fluorescence imaging of NLS-FRET-G100 cells at 5 days of doxycycline before and after detergent treatment, in the same field of view. D, flow cytometry plots at 1 and 5 days of doxycycline treatment and the latter before and after detergent treatment. E, flow cytometry plots of NLS-FRET-G100 cells at 4 days of doxycycline with or without proteasomal inhibitor (carfilzomib). All scale bars represent 10 \u03bcm. NLS, nuclear localization signal.\n\nTogether, these findings establish NLS-FRET-G100 as a robust live-cell reporter of nuclear polyG aggregation that recapitulates key pathological features of NIID and enables quantitative, high-throughput analysis.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "PolyQ and polyG proteins aggregate independently in a homotypic repeat\u2013dependent manner", "text": "To determine whether aggregate formation by each FRET reporter is driven by the specific amino acid repeat rather than nonspecific aggregation of fluorescent tags or other aggregation-prone proteins, we cotransfected cells with different combinations of mScarlet- or mNeonGreen-tagged Q24 or Q79 constructs, uN2C-G12 or uN2C-G100 constructs, as well as GFP-HTTex1-Q25 or -Q72. The NLS was removed from all constructs to promote predominantly cytosolic localization, as HTTex1 is known to localize primarily to the cytoplasm. We observed that HTTex1-Q72 inclusions sequestered both Q24 and Q79 but not uN2C-G12 proteins (Fig. 5). Conversely, uN2C-G100 inclusions sequestered uN2C-G12 proteins despite the short repeat length but did not recruit Q24, HTTex1-Q25, or HTTex1-Q72 proteins. In all cases, the fluorescent proteins alone were not recruited into inclusions. The sequestration of proteins containing nonexpanded polyQ repeats by polyQ aggregates has been reported previously (17, 43, 44, 45, 46). Recent work has similarly shown sequestration of endogenous polyG-containing proteins, such as FAM98B, into polyG aggregates (47). Interestingly, in cells containing both HTTex1-Q72 and uN2C-G100 inclusions, aggregates were most frequently adjacent to each other, whereas only some showed partial overlap. Similar spatial segregation was reported previously between HTTex1 and poly(glycine\u2013alanine)dipeptide repeat aggregates (48). In sum, these findings confirmed aggregation in the FRET reporter is driven in a homotypic amino acid repeat\u2013dependent manner and can lead to sequestration of other proteins containing even relatively short repeats of the same amino acid.Figure 5PolyQ and polyG proteins undergo independent, repeat-dependent aggregation. Fluorescence micrographs of HEK293T cells cotransfected with the indicated combinations of fluorescently tagged polyQ and polyG proteins. The scale bar represents 10 \u03bcm. HEK293T, human embryonic kidney 293 cell line; polyG, polyglucine; polyQ, polyglutamine.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "PolyQ and polyG proteins aggregate independently in a homotypic repeat\u2013dependent manner", "text": "PolyQ and polyG proteins undergo independent, repeat-dependent aggregation. Fluorescence micrographs of HEK293T cells cotransfected with the indicated combinations of fluorescently tagged polyQ and polyG proteins. The scale bar represents 10 \u03bcm. HEK293T, human embryonic kidney 293 cell line; polyG, polyglucine; polyQ, polyglutamine.\n\nWe also tested whether polyQ or polyG aggregates from homogenized cells or tissues could seed aggregation of the FRET reporter. NLS-FRET-Q79 cells transfected with homogenates of NLS-FRET-Q79 cells or from cortical tissue of transgenic HD mice led to an increase in FRET-high cells (Fig. S5). We used 22-week-old mice of the R6/1 HD mouse model, which expresses HTTex1 with approximately 115 glutamines and exhibits widespread aggregation of mutant HTT throughout the brain beginning as early as 8 weeks of age (49, 50). Immunostaining the contralateral hemispheres of these mice confirmed frequent p62-positive nuclear inclusions in the cortex (Fig. S6). NLS-FRET-G100 cell homogenates did not increase FRET-high cells in the NLS-FRET-Q79 line but robustly increased it in the NLS-FRET-G100 cells. These results indicate that the FRET reporter lines could potentially be used as a \u201cbiosensor\u201d to test the presence of polyQ aggregates of other cells, tissues, or biofluids, as for FRET-based aggregation reporters of Tau and TDP-43 (19, 20, 21, 51, 52, 53).", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Chaperone screening of polyG aggregation shows few modifiers, but polyG aggregation is suppressed by DNAJC7 overexpression", "text": "Using the same CRISPRi screening approach as with the NLS-FRET-Q79 cell line, we performed a molecular chaperone screen in the NLS-FRET-G100 reporter line to identify modifiers of polyG aggregation (Fig. 6A). Surprisingly, this screen revealed relatively few significant hits. Notably, key polyQ modifiers, such as DNAJC7, DNAJB6, DNAJB1, and HSPA8, were not identified as hits in the polyG screen. A direct comparison of gene scores between the polyQ and polyG screens revealed minimal overlap in chaperone modifiers (Fig. 6B), suggesting distinct mechanisms of proteostasis regulation between polyQ and polyG aggregates.Figure 6Chaperone CRISPRi screening reveals limited suppression of polyG aggregation despite effects by DNAJC7 overexpression.A, volcano plot showing results of a CRISPRi screen for molecular chaperone modifiers of polyG aggregation in the NLS-FRET-G100 cell line. B, pairwise comparison of gene scores between the polyG screen (from A) and the polyQ screen (Fig. 2B). C, results of flow cytometry experiments measuring fold change in the fraction of FRET-high cells after 5 days of doxycycline in sgRNA+ cells transduced with NTC or DNAJC7. Data represent mean \u00b1 SD from n = 3 independent experiments. D, representative fluorescence micrographs of HEK293T cells transiently cotransfected with NLS-mScarlet- and NLS-mNeonGreen-tagged uN2C-G100 together with either BFP or BFP-DNAJC7; only the mNeonGreen channel is shown. The scale bar represents 10 \u03bcm. E, flow cytometry quantification of the percentage of FRET-high cells relative to total cells expressing both mNeonGreen and mScarlet. Bars indicate mean \u00b1 SD from n = 3 independent experiments. \u2217p < 0.05 by Student\u2019s t test. CRISPRi, CRISPR interference; HEK293T, human embryonic kidney 293T cell line; NLS, nuclear localization signal; NTC, nontargeting control; polyG, polyglucine; polyQ, polyglutamine; sgRNA, single guide RNA.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Chaperone screening of polyG aggregation shows few modifiers, but polyG aggregation is suppressed by DNAJC7 overexpression", "text": "Chaperone CRISPRi screening reveals limited suppression of polyG aggregation despite effects by DNAJC7 overexpression.A, volcano plot showing results of a CRISPRi screen for molecular chaperone modifiers of polyG aggregation in the NLS-FRET-G100 cell line. B, pairwise comparison of gene scores between the polyG screen (from A) and the polyQ screen (Fig. 2B). C, results of flow cytometry experiments measuring fold change in the fraction of FRET-high cells after 5 days of doxycycline in sgRNA+ cells transduced with NTC or DNAJC7. Data represent mean \u00b1 SD from n = 3 independent experiments. D, representative fluorescence micrographs of HEK293T cells transiently cotransfected with NLS-mScarlet- and NLS-mNeonGreen-tagged uN2C-G100 together with either BFP or BFP-DNAJC7; only the mNeonGreen channel is shown. The scale bar represents 10 \u03bcm. E, flow cytometry quantification of the percentage of FRET-high cells relative to total cells expressing both mNeonGreen and mScarlet. Bars indicate mean \u00b1 SD from n = 3 independent experiments. \u2217p < 0.05 by Student\u2019s t test. CRISPRi, CRISPR interference; HEK293T, human embryonic kidney 293T cell line; NLS, nuclear localization signal; NTC, nontargeting control; polyG, polyglucine; polyQ, polyglutamine; sgRNA, single guide RNA.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Chaperone screening of polyG aggregation shows few modifiers, but polyG aggregation is suppressed by DNAJC7 overexpression", "text": "Among the few shared hits was OGT (O-GlcNAc transferase), which encodes O-GlcNAc transferase. OGT knockdown was among the top hits in both screens, and we independently validated that its depletion significantly increases the fraction of FRET-high cells in both the NLS-FRET-Q79 and NLS-FRET-G100 lines (Fig. S7), confirming the reliability of the polyG screen and confirming active CRISPRi machinery. In contrast, direct knockdown of DNAJC7 had no significant effect on the proportion of FRET-high cells in the NLS-FRET-G100 model (Fig. 6C).\n\nTo determine whether this difference was associated with a lack of DNAJC7 colocalization with polyG inclusions, we transiently transfected cells with NLS- and mScarlet- and mNeonGreen-tagged uN2C-G100 together with either BFP or BFP-DNAJC7, as described above. Similar to observations with GFP-HTTex1-Q72 inclusions, a subset of polyG inclusions colocalized with BFP-DNAJC7 (Fig. 6D). Consistent with this finding, flow cytometry revealed a significant reduction in the fraction of FRET-high cells (Fig. 6E). We repeated these experiments using a non-nuclear mScarlet/mNeonGreen-tagged uN2C-G100 and observed the same results (Fig. S8).", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Chaperone screening of polyG aggregation shows few modifiers, but polyG aggregation is suppressed by DNAJC7 overexpression", "text": "To exclude the possibility that the observed effects of DNAJC7 knockdown were due to clonal artifacts of the selected FRET reporter lines, we examined one additional polyQ and multiple other polyG monoclonal cell lines. Consistent with results obtained with the NLS-FRET-Q79 and HTTex1-Q72 lines, DNAJC7 knockdown robustly increased aggregation in the polyQ line (Fig. S9). In contrast, analysis of three additional uN2C-G100 monoclonal lines with comparable FRET signal properties, two of which lacked nuclear localization, revealed no effect of DNAJC7 knockdown on aggregation in any clone. Together, these data support that polyG aggregation is insensitive to DNAJC7 knockdown, whereas DNAJC7 overexpression can reduce polyG aggregation.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Discussion", "text": "In this study, we developed inducible FRET\u2013based reporter systems for polyQ and polyG aggregation that recapitulate features of human disease and enable dynamic monitoring of aggregation. Importantly, these reporters allow for high-throughput flow cytometry\u2013based genetic screens, which we used to systematically test all known molecular chaperones to identify modifiers of aggregation. Beyond screening applications, these models also provide a valuable platform for exploring the relatively understudied nuclear proteostasis pathways, and they could serve as biosensors to assess seeding activity of mouse or human brain homogenates or other biospecimens.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Discussion", "text": "By taking a comprehensive approach, we were able to identify for the first time the Hsp40 cochaperone DNAJC7 as a potent modifier of polyQ protein aggregation and further found that DNAJC7 overexpression suppresses both polyQ and polyG aggregation. A major next step is to understand the functional impact of DNAJC7 in the central nervous system, particularly given its high expression among Hsp40 family members. DNAJC7 appears to be especially important in motor neurons, supported by its genetic link to amyotrophic lateral sclerosis and recent evidence that loss of DNAJC7 in induced pluripotent stem cell\u2013derived motor neurons increases susceptibility to proteotoxic stress, in part because of impaired HSF1 signaling (54). However, its role may extend more broadly across central nervous system diseases. Prior studies have shown that DNAJC7 interacts with other aggregation-prone proteins, including Tau and TDP-43, with Hou et al. (31) demonstrating that mutant tau coprecipitates with anti-DNAJC7 antibodies in a transgenic mouse model of frontotemporal dementia. These findings suggest that DNAJC7 may act as a general regulator of protein aggregation in neurodegenerative disease. Our study was limited by the availability of an antibody that could reliably immunostain DNAJC7 to test its colocalization in mouse or human brain tissues to further validate this finding.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Discussion", "text": "The growing number of possible DNAJC7 substrates raises questions regarding the basis of its broad substrate selectivity. As previously demonstrated for DNAJB6, DNAJC7 may preferentially bind to fibrillar or amyloid-like structures, a biophysical state shared by tau and polyQ aggregates (55, 56, 57). To date, the amyloidogenic potential of polyG aggregates is not defined. Future studies using live-cell imaging approaches, such as fluorescence recovery after photobleaching, along with biochemical characterization, could help determine if an amyloidogenic state is required for DNAJC7 binding, as well as the dynamics of its recruitment. It is notable that DNAJC7 is structurally distinct among Hsp40 family members, as it contains multiple tetratricopeptide repeat domains in addition to the canonical J-domain. Beyond the shared J-domain, DNAJC7 has no significant sequence homology with DNAJB6 or other Hsp40s, suggesting that it may recognize and engage substrates through a different mechanism. Although we observe colocalization between DNAJC7 and polyQ/G protein aggregates in cells, it remains unclear whether this reflects direct binding to misfolded species or occurs through intermediary factors. As previously shown with tau (31, 58), biochemical reconstitution studies using purified proteins will be critical to determine whether DNAJC7 directly recognizes misfolded polyQ and polyG aggregates and to dissect the specific contribution of its tetratricopeptide repeat domains to client binding and chaperone activity.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Discussion", "text": "The basis for the relative insensitivity of polyG aggregation to knockdown of endogenous molecular chaperones, despite clear colocalization with DNAJC7 and reduced aggregation upon DNAJC7 overexpression, remains unclear. Discrepancies between knockdown and overexpression phenotypes are well documented (59, 60). DNAJC7 can seemingly engage polyG aggregates, consistent with recent proteomic analyses of the insoluble fraction of uN2C-polyG immunoprecipitated proteins that identified DNAJC7 as being significantly enriched (47). One possibility is that endogenous levels of DNAJC7 may be insufficient to counteract the robust aggregation propensity of polyG proteins. Indeed, we observed that the formation of detergent-resistant, FRET-high aggregates in polyG reporter lines was markedly more robust than in comparable polyQ reporters. Testing shorter, yet aggregation-prone, polyG repeats that more closely match the aggregation properties of the polyQ reporters could reveal an effect. Alternatively, differences in aggregation kinetics between polyG and polyQ proteins may mean that the time points examined in this study were not optimal to capture the effects of chaperone depletion.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Discussion", "text": "The FRET-based reporter lines described here are monoclonal cell lines that exhibit spontaneous protein aggregation, even in the absence of exogenous seeding. Despite being clonal, there is a seemingly stochastic emergence of microscopically visible aggregates in only a subset of cells, the basis of which remains unclear and represents an intriguing avenue for investigation. To an extent, this phenomenon parallels observations in human disease and transgenic mouse models, where visible aggregates are present in only a subset of neurons within otherwise homogeneous populations. In our FRET reporter lines, cells lacking detectable aggregates nonetheless retain the capacity to aggregate, as aggregation can be accelerated by genetic perturbations, proteasome inhibition, or exogenous seeding. Together, these features make the reporter lines a tractable system for interrogating intrinsic cellular factors that govern spontaneous aggregate formation. In addition, given the relatively low fraction of FRET-high cells in the polyQ reporter (typically single-digit percentages), ongoing efforts include increasing expression levels, expanding repeat length, and slowing cell proliferation to enhance aggregation and improve the throughput of large-scale genetic screens for modifiers.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Discussion", "text": "Beyond DNAJC7, our CRISPRi screens uncovered several additional candidate modifiers of protein aggregation that merit further investigation. One of the most prominent was OGT, the enzyme responsible for O-GlcNAcylation, whose knockdown markedly increased aggregation in both the polyQ and polyG models. O-GlcNAcylation is a dynamic post-translational modification that regulates a wide array of proteins and cellular processes, including in the brain (61, 62, 63, 64), with our findings suggesting a potential role in proteostasis. Another notable hit specific to polyQ protein aggregates was BAG6, a chaperone \u201choldase\u201d required for ubiquitin-mediated degradation of newly synthesized misfolded polypeptides that has previously been shown to modulate the aggregation of TDP-43 fragments (65, 66, 67). In addition, hits unique to the polyG screen included members of the peptidyl-prolyl isomerase family, such as PPIG and PPIL1, which have been implicated in modulating tau aggregation (68). These findings highlight the potential of our platform to identify both shared and distinct regulators of aggregation across different disease-relevant proteins.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Discussion", "text": "In summary, this work establishes a robust and scalable platform for studying the aggregation of polyQ and polyG proteins, leveraging inducible FRET\u2013based reporters that enable high-throughput genetic screening in live cells. Our CRISPRi screens provide new insights into the molecular chaperone network and its role in regulating aggregation, uncovering both shared and distinct modifiers between two different repeat expansion disorders. Given the scalability of this system, ongoing efforts to extend these screens to a genome-wide level are expected to uncover additional regulators, offering a deeper understanding of proteostasis mechanisms and informing the development of targeted therapeutic strategies for neurodegenerative diseases.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Experimental procedures", "text": "All mice were maintained according to the National Institutes of Health guidelines, and all procedures used in this study were approved by the University of California, San Francisco (UCSF) Institutional Animal Care and Use Committee. Mice were housed on a 12-h light/dark cycle at 22 to 25 \u00b0C, 50% to 60% humidity, and had food and water provided ad libitum. The mice used in this study were R6/1 mouse models of HD (B6.Cg-Tg(HDexon1)61Gpb/J, Research Resource Identifier: IMSR_JAX:006471), which are crossed onto a homozygous background for Cre-inducible CRISPRi machinery (B6;129S6-Gt(ROSA)26Sortm2(CAG-cas9\u2217/ZNF10\u2217)Gers/J, Research Resource Identifier: IMSR_JAX:033066). For genotyping HD mice, DNA was purified from ear punches following the manufacturer\u2019s instructions (New England Biolabs, T3010S). The eluted DNA (1 \u03bcl) was used for PCR amplification using primers and cycling conditions detailed in Table S1.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Experimental procedures", "text": "A list of all plasmids used in this study with information on their main expression elements is provided in Table S1. A map of each plasmid in a GenBank file format is provided in a Supporting File. The FRET-based aggregation reporters were constructed on a lentiviral pLEX-TetOne backbone kindly provided by Michael Ward (National Institutes of Health) for doxycycline-inducible expression under a Tet-ON 3G promoter. C-terminal ataxin-3 (amino acid 257 to the C terminus) containing a 24 or 79 consecutive glutamine stretch was cloned by PCR from pEGFP-Ataxin3Q28 or Q84, respectively (69) (Addgene #22122 and #22123, gifts from Henry Paulson). HTTex1 with Q25 and Q72 repeats were cloned from pGW1-HTTN586 constructs (70), kindly provided by Steven Finkbeiner (Gladstone Institutes). The upstream ORF of the NOTCH2NLC encoding 12 or 100 glycines (42) (Addgene #224355 and #224356, gifts from Nicolas Charlet-Berguerand) was cloned by restriction digestion into the pLEX-TetOne backbone. mNeonGreen, mScarlet, and eGFP were cloned by PCR. A 2\u00d7 c-myc NLS and the mTagBFP2 (BFP) sequences were PCR amplified from pMK1334 (60). The complementary DNA for DNAJC7 was kindly provided by Jason Gestwicki (UCSF). BFP and BFP fused to DNAJC7 were cloned by PCR into pMK1200 (71), a lentiviral backbone with an EF1\u03b1 promoter.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Experimental procedures", "text": "Transient transfection experiments that used GFP-HTTex1-Q72 and 3\u00d7 FLAG-tagged constructs were built on an adeno-associated virus backbone with an EF1\u03b1 promoter (72) (Addgene #55636, a gift from Karl Deisseroth).\n\nThe CRISPRi molecular chaperone sgRNA library used in this study was described previously (73). Individual sgRNAs for validation studies were selected and cloned into the pMK1334 lentiviral plasmid backbone between BstXI and BlpI by annealing and ligating annealed oligonucleotides. The sgRNA protospacer sequences are provided in Table S1.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Experimental procedures", "text": "Lentiviral packaging was performed as previously described (74). Briefly, HEK293T cells were seeded in complete Dulbecco's modified Eagle's medium (DMEM) to reach approximately 70% confluency the following day. For transfection, third-generation lentiviral packaging plasmids (pRSV, pMDL, and pVSV-G) were mixed at a 1:1:1 mass ratio (lentiviral pack-mix) and combined with an equal mass of the transfer plasmid. The DNA mixture was diluted in Opti-MEM and complexed with polyethylenimine PEI (Polysciences, 23966) at a 3:1 PEI:DNA mass ratio. After 15 min of incubation at room temperature, the transfection mixture was added dropwise to the cells. Conditioned media were collected 48 h post-transfection and filter-sterilized using a Millex-GV syringe filter unit (Millipore, SLGV033RB). Lentivirus was then precipitated using the Alstem Lentivirus Concentration Kit (VC100) according to the manufacturer's instructions and resuspended in 1\u00d7 Dulbecco\u2019s PBS (DPBS) (Sigma\u2013Aldrich, D8537).\n\nFor pooled sgRNA library packaging, 15 \u03bcg of sgRNA library DNA and 15 \u03bcg of lentiviral pack-mix were transfected into HEK293T cells plated in a 15-cm plate format. The precipitated virus was resuspended in 5 ml of 1\u00d7 DPBS. For individual sgRNAs cloned into the pMK1334 backbone, 1 \u03bcg of the transfer DNA and 1 \u03bcg of lentiviral pack-mix were transfected onto HEK293T cells plated in a 6-well (35 mm) plate format. The resulting virus was resuspended in 200 \u03bcl of 1\u00d7 DPBS.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Experimental procedures", "text": "All cells were maintained in a tissue culture incubator (37 \u00b0C, 5% CO2) and checked regularly for mycoplasma contamination. HEK293T cells were cultured in DMEM supplemented with 10% fetal bovine serum (Seradigm, 89510-186), penicillin\u2013streptomycin (Gibco, 15140122), and l-glutamine (Life Technologies, 25030081).\n\nThe doxycycline-inducible NLS-FRET-Q79 and NLS-FRET-G100 cell lines were made on the \u201ccXG284\u201d HEK293T cell line that has stably integrated CRISPRi machinery (dCas9-BFP-KRAB) in the CLYBL locus (75). The cells were transduced with lentivirus, and, after at least 72 h and in the presence of 2 \u03bcg/ml doxycycline, the cells were dissociated and sorted for those expressing both mNeonGreen and mScarlet using a BD FACSAria FusionCell Sorter. Monoclonal lines were obtained by plating cells at limiting dilution, followed by screening for cells that showed the highest frequency of inclusions by microscopy and by the corresponding presence of a population of cells with a higher FRET signal. The GFP-HTTex1-Q72 monoclonal cell line was generated similarly but used HEK293T cells expressing dCas9-BFP-KRAB introduced by random integration of lentivirus.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Experimental procedures", "text": "All transfections of HEK293T cells were conducted the next day after plating into wells by diluting plasmids into Opti-MEM containing PEI at a 3:1 PEI:DNA mass ratio, incubated at room temperature for 15 min, and dispensed dropwise over the cells. The well format and amount of DNA transfected for the different experiments are detailed below.\n\nLive-cell fluorescence imaging was performed using an ECHO Revolve microscope with a 20\u00d7 objective. To assess detergent sensitivity, cells were first imaged under baseline conditions. Triton X-100 (5% stock in 1\u00d7 DPBS) was then added directly to the well to achieve a final concentration of 0.5%, followed by a 1-min incubation. The same field of view was subsequently reimaged to assess detergent-resistant fluorescence.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Experimental procedures", "text": "For immunocytochemistry of p62, cells were fixed at room temperature for 10 min with 4% paraformaldehyde (PFA; Electron Microscopy Sciences, 15710), diluted in 1\u00d7 DPBS, then briefly rinsed with 1\u00d7 DPBS. Cells were permeabilized and blocked for 10 min in blocking solution consisting of 1\u00d7 DPBS with 0.1% Triton X-100 and 5% normal goat serum. Primary antibody against p62 (clone D5L7G, Cell Signaling, 88588) was diluted 1:1000 in blocking solution and incubated overnight at 4 \u00b0C. The next day, cells were washed three times with 1\u00d7 DPBS and incubated with a secondary antibody, goat anti-mouse Alexa Fluor 647, for 1 h at room temperature (ThermoFisher, A32728). Nuclei were counterstained with Hoechst 33342 (ThermoFisher, 5553141) at a 1:2000 dilution. p62 antibody specificity was validated by confirming the absence of signal with sgRNA-mediated knockdown.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Experimental procedures", "text": "For immunostaining of mouse brain tissue, the right hemispheres were drop-fixed in 4% PFA overnight at 4 \u00b0C, then cryoprotected in 30% sucrose prepared in 1\u00d7 DPBS for at least 24 h. Brains were sectioned at 40 \u03bcm thickness, and free-floating sections were blocked in 1\u00d7 DPBS containing 0.3% Triton X-100 and 5% normal goat serum for 1 h at room temperature. Sections were then incubated overnight at 4 \u00b0C with a rodent-specific anti-p62 antibody (clone D6M5X, Cell Signaling, 23214) at a 1:500 dilution in blocking buffer. The next day, slices were washed three times for 10 min each in 1\u00d7 DPBS and incubated with goat anti-rabbit Alexa Fluor 488 secondary antibody (ThermoFisher, A11008) for 1 h at room temperature. Nuclei were counterstained with Hoechst 33342 (1:2000 dilution in 1\u00d7 DPBS) for 5 min, followed by three additional 10-min washes in fresh 1\u00d7 DPBS. Sections were then mounted onto microscope slides (Fisher Scientific, 12-550-143) and coverslipped using ProLong Gold antifade mounting medium (Invitrogen, P36930).\n\nTo examine expression and colocalization of polyQ and polyG aggregates with different fluorescent tags, cells plated in 96-well format were transfected with 200 ng of each indicated polyQ, polyG, or control plasmid with the addition of doxycycline to 2 \u03bcg/ml, followed by imaging at 48 h to 72 h.\n\nFluorescent images of all immunostained cells and tissues were acquired on the ECHO Revolve on a 20\u00d7 objective.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Experimental procedures", "text": "For screening with the chaperone pooled sgRNA library, 15 million cells containing the FRET reporter were transduced with 1 ml of lentivirus and plated on a T175 flask (day 0). On day 2, with \u223c30% of the cells showing BFP positivity, the cells were passaged and replated with 2 \u03bcg/ml puromycin (Gibco, A1113803). This was repeated on day 5. On day 7, with >70% of cells showing BFP positivity, the cells were passaged, and 20 million cells were plated without puromycin and with the addition of 2 \u03bcg/ml doxycycline. The cells were regularly passaged until day 5. The cells were dissociated with trypsin, resuspended in complete DMEM, and sorted using the BD FACSAria Fusion Cell Sorter into FRET-low (\u223c6 million cells) and FRET-high (\u223c2 million cells). Sorting and genomic DNA (gDNA) isolation for the polyQ chaperone screen was performed twice (two separate times from the same starting population of library-transduced cells). Sorting for the polyG chaperone screen was performed once. gDNA was isolated using a Monarch gDNA extraction kit according to the manufacturer's protocols. sgRNA-encoding regions were then amplified, followed by sequencing of the protospacers by Illumina NextSeq2000 as recently described (76).", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Experimental procedures", "text": "The generation of knockdown phenotypes, p values, and gene scores to identify the hits used bioinformatics pipelines that we recently described (76), including \u201csgcount\u201d (https://github.com/noamteyssier/sgcount) and \u201ccrispr_screen\u201d (https://github.com/noamteyssier/crispr_screen/). In brief, raw sequencing reads were aligned to a custom reference file containing the CRISPRi chaperone library protospacer sequences using \u201csgcount,\u201d generating sgRNA count matrices for each sample. Using \u201ccrispr_screen,\u201d these counts were then normalized, and p values were calculated for individual sgRNAs based on differential abundance between FRET-low and FRET-high. Gene-level phenotypes, including knockdown phenotypes, p values, and false discovery rates, were derived using the Robust Rank Aggregation algorithm (77). The sgRNA count matrices and gene-level phenotypes are provided in Table S2.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Animals", "text": "All mice were maintained according to the National Institutes of Health guidelines, and all procedures used in this study were approved by the University of California, San Francisco (UCSF) Institutional Animal Care and Use Committee. Mice were housed on a 12-h light/dark cycle at 22 to 25 \u00b0C, 50% to 60% humidity, and had food and water provided ad libitum. The mice used in this study were R6/1 mouse models of HD (B6.Cg-Tg(HDexon1)61Gpb/J, Research Resource Identifier: IMSR_JAX:006471), which are crossed onto a homozygous background for Cre-inducible CRISPRi machinery (B6;129S6-Gt(ROSA)26Sortm2(CAG-cas9\u2217/ZNF10\u2217)Gers/J, Research Resource Identifier: IMSR_JAX:033066). For genotyping HD mice, DNA was purified from ear punches following the manufacturer\u2019s instructions (New England Biolabs, T3010S). The eluted DNA (1 \u03bcl) was used for PCR amplification using primers and cycling conditions detailed in Table S1.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Plasmid construction and lentivirus packaging", "text": "A list of all plasmids used in this study with information on their main expression elements is provided in Table S1. A map of each plasmid in a GenBank file format is provided in a Supporting File. The FRET-based aggregation reporters were constructed on a lentiviral pLEX-TetOne backbone kindly provided by Michael Ward (National Institutes of Health) for doxycycline-inducible expression under a Tet-ON 3G promoter. C-terminal ataxin-3 (amino acid 257 to the C terminus) containing a 24 or 79 consecutive glutamine stretch was cloned by PCR from pEGFP-Ataxin3Q28 or Q84, respectively (69) (Addgene #22122 and #22123, gifts from Henry Paulson). HTTex1 with Q25 and Q72 repeats were cloned from pGW1-HTTN586 constructs (70), kindly provided by Steven Finkbeiner (Gladstone Institutes). The upstream ORF of the NOTCH2NLC encoding 12 or 100 glycines (42) (Addgene #224355 and #224356, gifts from Nicolas Charlet-Berguerand) was cloned by restriction digestion into the pLEX-TetOne backbone. mNeonGreen, mScarlet, and eGFP were cloned by PCR. A 2\u00d7 c-myc NLS and the mTagBFP2 (BFP) sequences were PCR amplified from pMK1334 (60). The complementary DNA for DNAJC7 was kindly provided by Jason Gestwicki (UCSF). BFP and BFP fused to DNAJC7 were cloned by PCR into pMK1200 (71), a lentiviral backbone with an EF1\u03b1 promoter.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Plasmid construction and lentivirus packaging", "text": "Transient transfection experiments that used GFP-HTTex1-Q72 and 3\u00d7 FLAG-tagged constructs were built on an adeno-associated virus backbone with an EF1\u03b1 promoter (72) (Addgene #55636, a gift from Karl Deisseroth).\n\nThe CRISPRi molecular chaperone sgRNA library used in this study was described previously (73). Individual sgRNAs for validation studies were selected and cloned into the pMK1334 lentiviral plasmid backbone between BstXI and BlpI by annealing and ligating annealed oligonucleotides. The sgRNA protospacer sequences are provided in Table S1.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Plasmid construction and lentivirus packaging", "text": "Lentiviral packaging was performed as previously described (74). Briefly, HEK293T cells were seeded in complete Dulbecco's modified Eagle's medium (DMEM) to reach approximately 70% confluency the following day. For transfection, third-generation lentiviral packaging plasmids (pRSV, pMDL, and pVSV-G) were mixed at a 1:1:1 mass ratio (lentiviral pack-mix) and combined with an equal mass of the transfer plasmid. The DNA mixture was diluted in Opti-MEM and complexed with polyethylenimine PEI (Polysciences, 23966) at a 3:1 PEI:DNA mass ratio. After 15 min of incubation at room temperature, the transfection mixture was added dropwise to the cells. Conditioned media were collected 48 h post-transfection and filter-sterilized using a Millex-GV syringe filter unit (Millipore, SLGV033RB). Lentivirus was then precipitated using the Alstem Lentivirus Concentration Kit (VC100) according to the manufacturer's instructions and resuspended in 1\u00d7 Dulbecco\u2019s PBS (DPBS) (Sigma\u2013Aldrich, D8537).\n\nFor pooled sgRNA library packaging, 15 \u03bcg of sgRNA library DNA and 15 \u03bcg of lentiviral pack-mix were transfected into HEK293T cells plated in a 15-cm plate format. The precipitated virus was resuspended in 5 ml of 1\u00d7 DPBS. For individual sgRNAs cloned into the pMK1334 backbone, 1 \u03bcg of the transfer DNA and 1 \u03bcg of lentiviral pack-mix were transfected onto HEK293T cells plated in a 6-well (35 mm) plate format. The resulting virus was resuspended in 200 \u03bcl of 1\u00d7 DPBS.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Cell culture and cell line generation", "text": "All cells were maintained in a tissue culture incubator (37 \u00b0C, 5% CO2) and checked regularly for mycoplasma contamination. HEK293T cells were cultured in DMEM supplemented with 10% fetal bovine serum (Seradigm, 89510-186), penicillin\u2013streptomycin (Gibco, 15140122), and l-glutamine (Life Technologies, 25030081).\n\nThe doxycycline-inducible NLS-FRET-Q79 and NLS-FRET-G100 cell lines were made on the \u201ccXG284\u201d HEK293T cell line that has stably integrated CRISPRi machinery (dCas9-BFP-KRAB) in the CLYBL locus (75). The cells were transduced with lentivirus, and, after at least 72 h and in the presence of 2 \u03bcg/ml doxycycline, the cells were dissociated and sorted for those expressing both mNeonGreen and mScarlet using a BD FACSAria FusionCell Sorter. Monoclonal lines were obtained by plating cells at limiting dilution, followed by screening for cells that showed the highest frequency of inclusions by microscopy and by the corresponding presence of a population of cells with a higher FRET signal. The GFP-HTTex1-Q72 monoclonal cell line was generated similarly but used HEK293T cells expressing dCas9-BFP-KRAB introduced by random integration of lentivirus.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Transfections, fluorescence imaging, and immunofluorescence", "text": "All transfections of HEK293T cells were conducted the next day after plating into wells by diluting plasmids into Opti-MEM containing PEI at a 3:1 PEI:DNA mass ratio, incubated at room temperature for 15 min, and dispensed dropwise over the cells. The well format and amount of DNA transfected for the different experiments are detailed below.\n\nLive-cell fluorescence imaging was performed using an ECHO Revolve microscope with a 20\u00d7 objective. To assess detergent sensitivity, cells were first imaged under baseline conditions. Triton X-100 (5% stock in 1\u00d7 DPBS) was then added directly to the well to achieve a final concentration of 0.5%, followed by a 1-min incubation. The same field of view was subsequently reimaged to assess detergent-resistant fluorescence.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Transfections, fluorescence imaging, and immunofluorescence", "text": "For immunocytochemistry of p62, cells were fixed at room temperature for 10 min with 4% paraformaldehyde (PFA; Electron Microscopy Sciences, 15710), diluted in 1\u00d7 DPBS, then briefly rinsed with 1\u00d7 DPBS. Cells were permeabilized and blocked for 10 min in blocking solution consisting of 1\u00d7 DPBS with 0.1% Triton X-100 and 5% normal goat serum. Primary antibody against p62 (clone D5L7G, Cell Signaling, 88588) was diluted 1:1000 in blocking solution and incubated overnight at 4 \u00b0C. The next day, cells were washed three times with 1\u00d7 DPBS and incubated with a secondary antibody, goat anti-mouse Alexa Fluor 647, for 1 h at room temperature (ThermoFisher, A32728). Nuclei were counterstained with Hoechst 33342 (ThermoFisher, 5553141) at a 1:2000 dilution. p62 antibody specificity was validated by confirming the absence of signal with sgRNA-mediated knockdown.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Transfections, fluorescence imaging, and immunofluorescence", "text": "For immunostaining of mouse brain tissue, the right hemispheres were drop-fixed in 4% PFA overnight at 4 \u00b0C, then cryoprotected in 30% sucrose prepared in 1\u00d7 DPBS for at least 24 h. Brains were sectioned at 40 \u03bcm thickness, and free-floating sections were blocked in 1\u00d7 DPBS containing 0.3% Triton X-100 and 5% normal goat serum for 1 h at room temperature. Sections were then incubated overnight at 4 \u00b0C with a rodent-specific anti-p62 antibody (clone D6M5X, Cell Signaling, 23214) at a 1:500 dilution in blocking buffer. The next day, slices were washed three times for 10 min each in 1\u00d7 DPBS and incubated with goat anti-rabbit Alexa Fluor 488 secondary antibody (ThermoFisher, A11008) for 1 h at room temperature. Nuclei were counterstained with Hoechst 33342 (1:2000 dilution in 1\u00d7 DPBS) for 5 min, followed by three additional 10-min washes in fresh 1\u00d7 DPBS. Sections were then mounted onto microscope slides (Fisher Scientific, 12-550-143) and coverslipped using ProLong Gold antifade mounting medium (Invitrogen, P36930).\n\nTo examine expression and colocalization of polyQ and polyG aggregates with different fluorescent tags, cells plated in 96-well format were transfected with 200 ng of each indicated polyQ, polyG, or control plasmid with the addition of doxycycline to 2 \u03bcg/ml, followed by imaging at 48 h to 72 h.\n\nFluorescent images of all immunostained cells and tissues were acquired on the ECHO Revolve on a 20\u00d7 objective.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Primary CRISPRi screen and analysis", "text": "For screening with the chaperone pooled sgRNA library, 15 million cells containing the FRET reporter were transduced with 1 ml of lentivirus and plated on a T175 flask (day 0). On day 2, with \u223c30% of the cells showing BFP positivity, the cells were passaged and replated with 2 \u03bcg/ml puromycin (Gibco, A1113803). This was repeated on day 5. On day 7, with >70% of cells showing BFP positivity, the cells were passaged, and 20 million cells were plated without puromycin and with the addition of 2 \u03bcg/ml doxycycline. The cells were regularly passaged until day 5. The cells were dissociated with trypsin, resuspended in complete DMEM, and sorted using the BD FACSAria Fusion Cell Sorter into FRET-low (\u223c6 million cells) and FRET-high (\u223c2 million cells). Sorting and genomic DNA (gDNA) isolation for the polyQ chaperone screen was performed twice (two separate times from the same starting population of library-transduced cells). Sorting for the polyG chaperone screen was performed once. gDNA was isolated using a Monarch gDNA extraction kit according to the manufacturer's protocols. sgRNA-encoding regions were then amplified, followed by sequencing of the protospacers by Illumina NextSeq2000 as recently described (76).", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Primary CRISPRi screen and analysis", "text": "The generation of knockdown phenotypes, p values, and gene scores to identify the hits used bioinformatics pipelines that we recently described (76), including \u201csgcount\u201d (https://github.com/noamteyssier/sgcount) and \u201ccrispr_screen\u201d (https://github.com/noamteyssier/crispr_screen/). In brief, raw sequencing reads were aligned to a custom reference file containing the CRISPRi chaperone library protospacer sequences using \u201csgcount,\u201d generating sgRNA count matrices for each sample. Using \u201ccrispr_screen,\u201d these counts were then normalized, and p values were calculated for individual sgRNAs based on differential abundance between FRET-low and FRET-high. Gene-level phenotypes, including knockdown phenotypes, p values, and false discovery rates, were derived using the Robust Rank Aggregation algorithm (77). The sgRNA count matrices and gene-level phenotypes are provided in Table S2.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Secondary assays based on flow cytometry", "text": "NLS-FRET-Q79 or NLS-FRET-G100 cell lines were seeded (300,000 cells/well) into a 6-well format and transduced with lentivirus packaged from sgRNA-containing pMK1334. After at least 72 h, reporter expression was induced by doxycycline and analyzed after 5 days by flow cytometry using a BD FACS Fortessa, specifically examining BFP+ (sgRNA-containing) cells. Biological replicates of the experiments were from passaging the transduced cells. Gating strategies for representative experiments are shown in Fig. S5.\n\nFor assessing detergent-resistant aggregates by flow cytometry, we first acquired 10,000 events to determine baseline fluorescence intensity. We then added 5% Triton X-100 diluted in 1\u00d7 DPBS directly into the tube to a final concentration of 0.5% Triton X-100, gently swirled the tube, and performed flow cytometry to collect another 10,000 events.\n\nDetergent-resistant GFP+ aggregates in the eGFP-HTTex1-Q72 cell line, shown in Figure 4, were evaluated by flow cytometry after 7 days of doxycycline treatment. The relative fold change in the fraction of FRET-high cells was obtained by normalizing to the sgRNA NTC#1 condition for each experiment.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Secondary assays based on flow cytometry", "text": "To test the effect of BFP-DNAJC7 overexpression on GFP-HTTex1-Q72 aggregates, HEK293T cells (seeded the day before at 1 \u00d7 105 cells/well in a 24-well plate) were transfected with 500 ng of EF1\u237a-driven BFP or BFP-tagged DNAJC7 and 250 ng of GFP-HTTex1-Q72 or only 250 ng of GFP-HTTex1-Q72. After a 15 min incubation, the mixture was placed dropwise over the cells of one well. After 48 h, we obtained fluorescence micrographs, followed by dissociating the cells for flow cytometry. We used pulse shape analysis (plotting GFP-height versus GFP-width) to quantify aggregates.\n\nTo test the effect of BFP-DNAJC7 overexpression on polyG FRET-high aggregates, HEK293T cells were seeded as above and transfected with plasmids (250 ng each) expressing doxycycline-inducible mScarlet- and mNeonGreen-tagged uN2C-G12 or uN2C-G100 (with or without an NLS) and cotransfected with BFP or BFP-DNAJC7. Doxycycline (2 \u03bcg/ml) was added at the time of transfection. Imaging to examine colocalization and flow cytometry for FRET was conducted 48 h after transfection.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Secondary assays based on flow cytometry", "text": "NLS-FRET-Q79 cells containing 2 \u03bcg/ml doxycycline were grown in a 10-cm plate for 5 days. The cells were dissociated with trypsin, spun down at 200g for 10 min, and resuspended in 200 \u03bcl 1\u00d7 DPBS. For tissue homogenates, we used fresh-frozen left hemispheres of 22-week-old R6/1 male transgenic mice or age-matched nontransgenic mice. All mice are homozygous for conditional CRISPRi machinery. The right hemispheres of the same mice were drop-fixed in 4% PFA, cryoprotected, and immunostained as described above. The left hemispheres were frozen directly in dry ice and stored at \u221280 \u00b0C. A portion of the fresh-frozen cortex was placed in 200 \u03bcl 1\u00d7 DPBS and homogenized in a 1.5 ml microfuge tube. The crude homogenates from cells or tissues were briefly sonicated using a Q700 Sonicator (QSonica Sonicators) with a 1/16\u201d microtip probe at amplitude 10 for 10 s. The samples were centrifuged at 20,000g for 15 min at 4 \u00b0C, followed by the collection of the supernatant into a new tube. The concentrations of the homogenates were estimated by Nanodrop absorbance at 280 nm, using 1 absorbance = 1 mg/ml. Homogenates (100 \u03bcg) were diluted into 50 \u03bcl Opti-MEM containing 2 \u03bcl PEI (1 mg/ml) and dispensed dropwise over NLS-FRET-Q79 cells plated in a 24-well format that had been treated with 2 \u03bcg/ml of doxycycline for 1 day.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Secondary assays based on flow cytometry", "text": "To evaluate DNAJC7 knockdown, NLS-FRET-Q79 cells were transduced with lentivirus-containing sgRNAs targeting DNAJC7 or NTCs for 48 h. sgRNA-expressing cells were selected using 2 \u03bcg/ml puromycin for 3 days, followed by an additional 2 days of recovery. At 7 days post-transduction, cells were plated in a 12-well format to confluency, using three wells per condition. The following day, the cells were lysed with 100 \u03bcl of radioimmunoprecipitation assay with protease inhibitors, sonicated briefly, and centrifuged at 21,000g for 15 min. The supernatant was collected and combined with LDS buffer (Invitrogen, B0007) and 50 mM DTT (Cell Signaling, 7016) and boiled for 5 min.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Secondary assays based on flow cytometry", "text": "Each lysate (3 \u03bcl) was resolved on a 4% to 20% Tris\u2013glycine gel (Bio-Rad, 4561096), transferred to a 0.2 \u03bcm nitrocellulose membrane (Bio-Rad, 1620112) using a Trans-Blot Turbo, and blocked for 1 h with Intercept Blocking Buffer (LicorBio) before being incubated with primary antibody at 4\u00b0C overnight in the same buffer. The next day, the secondary antibody was incubated for 1 h at room temperature. Membranes were washed 3 \u00d7 5 min with 1x Tris-buffered saline with 0.1% Tween. Primary antibodies include rabbit anti-DNAJC7 (Proteintech, 11090-1-AP, 1:2000 dilution) and mouse anti-GAPDH (Proteintech, 60004-1-Ig; 1:2000 dilution). Secondary antibodies are IRDye 800CW Goat anti-Rabbit (LicorBio, 926-32211) and IRDye 700CW Goat anti-Mouse (LicorBio, 926-68070). To assess total resolved protein, the blot was stripped with 1\u00d7 NewBlot IR Stripping Buffer (LicorBio, 928-40028) for 15 min, washed for 5 min with 1\u00d7 Tris-buffered saline three times, and stained with Revert 700 Total Protein Stain (LicorBio, 926-11011) for 10 min. Membranes were imaged on the Licor Odyssey, and band densities were quantified using ImageJ software.\n\nAll bar graphs, scatter plots, and volcano plots were generated using R. Statistical tests were performed in R as indicated in the figure legends, except in the CRISPR screening analysis, which is described separately.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Transfecting cell or tissue homogenates for seeding aggregation", "text": "NLS-FRET-Q79 cells containing 2 \u03bcg/ml doxycycline were grown in a 10-cm plate for 5 days. The cells were dissociated with trypsin, spun down at 200g for 10 min, and resuspended in 200 \u03bcl 1\u00d7 DPBS. For tissue homogenates, we used fresh-frozen left hemispheres of 22-week-old R6/1 male transgenic mice or age-matched nontransgenic mice. All mice are homozygous for conditional CRISPRi machinery. The right hemispheres of the same mice were drop-fixed in 4% PFA, cryoprotected, and immunostained as described above. The left hemispheres were frozen directly in dry ice and stored at \u221280 \u00b0C. A portion of the fresh-frozen cortex was placed in 200 \u03bcl 1\u00d7 DPBS and homogenized in a 1.5 ml microfuge tube. The crude homogenates from cells or tissues were briefly sonicated using a Q700 Sonicator (QSonica Sonicators) with a 1/16\u201d microtip probe at amplitude 10 for 10 s. The samples were centrifuged at 20,000g for 15 min at 4 \u00b0C, followed by the collection of the supernatant into a new tube. The concentrations of the homogenates were estimated by Nanodrop absorbance at 280 nm, using 1 absorbance = 1 mg/ml. Homogenates (100 \u03bcg) were diluted into 50 \u03bcl Opti-MEM containing 2 \u03bcl PEI (1 mg/ml) and dispensed dropwise over NLS-FRET-Q79 cells plated in a 24-well format that had been treated with 2 \u03bcg/ml of doxycycline for 1 day.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Immunoprecipitation and Western blot", "text": "To evaluate DNAJC7 knockdown, NLS-FRET-Q79 cells were transduced with lentivirus-containing sgRNAs targeting DNAJC7 or NTCs for 48 h. sgRNA-expressing cells were selected using 2 \u03bcg/ml puromycin for 3 days, followed by an additional 2 days of recovery. At 7 days post-transduction, cells were plated in a 12-well format to confluency, using three wells per condition. The following day, the cells were lysed with 100 \u03bcl of radioimmunoprecipitation assay with protease inhibitors, sonicated briefly, and centrifuged at 21,000g for 15 min. The supernatant was collected and combined with LDS buffer (Invitrogen, B0007) and 50 mM DTT (Cell Signaling, 7016) and boiled for 5 min.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Immunoprecipitation and Western blot", "text": "Each lysate (3 \u03bcl) was resolved on a 4% to 20% Tris\u2013glycine gel (Bio-Rad, 4561096), transferred to a 0.2 \u03bcm nitrocellulose membrane (Bio-Rad, 1620112) using a Trans-Blot Turbo, and blocked for 1 h with Intercept Blocking Buffer (LicorBio) before being incubated with primary antibody at 4\u00b0C overnight in the same buffer. The next day, the secondary antibody was incubated for 1 h at room temperature. Membranes were washed 3 \u00d7 5 min with 1x Tris-buffered saline with 0.1% Tween. Primary antibodies include rabbit anti-DNAJC7 (Proteintech, 11090-1-AP, 1:2000 dilution) and mouse anti-GAPDH (Proteintech, 60004-1-Ig; 1:2000 dilution). Secondary antibodies are IRDye 800CW Goat anti-Rabbit (LicorBio, 926-32211) and IRDye 700CW Goat anti-Mouse (LicorBio, 926-68070). To assess total resolved protein, the blot was stripped with 1\u00d7 NewBlot IR Stripping Buffer (LicorBio, 928-40028) for 15 min, washed for 5 min with 1\u00d7 Tris-buffered saline three times, and stained with Revert 700 Total Protein Stain (LicorBio, 926-11011) for 10 min. Membranes were imaged on the Licor Odyssey, and band densities were quantified using ImageJ software.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Graphing and statistical analysis", "text": "All bar graphs, scatter plots, and volcano plots were generated using R. Statistical tests were performed in R as indicated in the figure legends, except in the CRISPR screening analysis, which is described separately.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Data availability", "text": "All data and plasmids are available from the corresponding author (to B. R.). The sequencing (fastq.gz) files from the CRISPRi screens are available at the UCSF Dryad data repository (https://doi.org/10.5061/dryad.0gb5mkmdh). There are no restrictions on data availability.", "title": "" }, { "pmid": 41708002, "pmcid": "PMC12993198", "section": "Conflict of interest", "text": "B. R. and M. K. have filed a patent application on in vivo screening methods. M. K. is an inventor on US Patent 11254933 related to CRISPRi and CRISPRa screening, a coscientific founder of Montara Therapeutics, and serves on the Scientific Advisory Boards of Engine Biosciences, Alector, and Montara Therapeutics, and is an advisor to Modulo Bio and Theseus Therapeutics.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "INTRODUCTION", "text": "This year, nervous system disorders passed cardiovascular diseases and cancer and became the most prevalent illness of humanity (1), with neurodegenerative disorders as the major unmet medical and social needs in a rapidly aging population worldwide. In age-dependent, progressive and fatal disorders such as Alzheimer\u2019s disease, Parkinson\u2019s disease, amyotrophic lateral sclerosis, and Huntington\u2019s disease (HD), the pathogenesis is extremely complex from molecule, cell, neural circuit, brain regional degeneration to behavior and symptoms. This provided numerous potential targets that attracted tremendous efforts from research groups and pharmaceutical industry to develop intervention strategies. However, to date, effective treatments to prevent the onset and progression of these diseases are elusive. All of these diseases have a strong genetic component, varying from substantial percentages to monogenic with near 100% penetrance. Many of gene mutations have been identified as the direct causes for specific disorders (2), and some are shared between hereditary and sporadic cases. Because gene mutations in these diseases are the most upstream events, strategies to correct mutations, delete disease genes, or add back absent genes are being pursued actively with huge potential.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "INTRODUCTION", "text": "Among these neurodegenerative diseases, HD is monogenic with near full penetrance by mHTT. HD is autosomal dominant and caused by a CAG repeat expansion (\u226536 repeats) in the first exon of the huntingtin gene (3, 4), which encodes a toxic gain-of-function mHTT protein with an expanded polyglutamine (polyQ) tract in the N-terminal region. mHTT can cause a wide range of cellular dysfunctions, which lastly lead to specific neuronal loss, mainly in medium spiny neurons in striatum and pyramidal neurons in cortex (5). Because mHTT is the cause of HD and the severity of HD symptoms is linked to mHTT expression (6, 7), reducing mHTT is a valid and direct therapeutic strategy for patients with HD.\n\nHTT lowering has been explored by multiple methods, including increasing protein degradation by enhancing proteasome or autophagy functions (8), as well as degrading RNA via antisense oligonucleotides (9\u201311), RNA interference (12\u201318), and CRISPR-Cas13 (19, 20). These strategies have been evaluated for safety, tolerability, and efficacy in preclinical studies and are being actively explored in clinical trials (21) (www.clinicaltrials.gov, identifier NCT02519036, NCT04120493). Furthermore, allele-selectively targeting HTT CAG repeats or single-nucleotide polymorphism has been attempted by zinc-finger nuclease approach (22\u201324) and CRISPR-based approach (20, 25\u201328).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "INTRODUCTION", "text": "CRISPR-Cas9, conventional (29\u201332) or self-inactivating (33\u201337) forms, are powerful tools for correcting or deleting disease-causing mutated genes. Specifically for HD, the most upstream mHTT lowering could be achieved by CRISPR-Cas9\u2013mediated DNA targeting. Therapies targeting HTT DNA are expected to ameliorate all aspects of HD. Recent studies have shown the feasibility of CRISPR-Cas9\u2013mediated HTT gene disruption in cell models (25), mouse models (26, 27, 34, 38, 39), and large mammals (40). Several of above studies (9, 12, 18, 38, 39, 41), along with previous findings (42\u201345), suggested that significant reduction of wild-type HTT in adult mouse, nonhuman primates, and human might be tolerated. To date, none of these CRISPR-Cas9\u2013based studies entered clinical trials yet.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "INTRODUCTION", "text": "To better push forward the CRISPR-Cas9\u2013mediated gene editing into clinical trials, we developed a self-inactivating CRISPR-Cas9 system to limit Cas9 expression only transiently for maximal safety in HD therapeutics. To verify the feasibility of CRISPR-Cas9\u2013mediated gene therapy for HD, we tested a robust and suitable mouse model BAC226Q that expresses full-length human mHTT with a stable 226 mixed CAG-CAA repeats (46), within the highest repeat number of 250 found in human patient (47). BAC226Q recapitulates many aspects of HD pathology and symptoms, including age-dependent and progressive neuropathology, cytosolic and nuclear mHTT aggregation in striatal and cortical neurons, motor deficits, weight loss, and shortened life span (46). Unlike HdhQ111 or Q140 KI that can be used to study somatic expansion as an important aspect of HD pathogenesis (48, 49), BAC226Q does not model neuronal-selective somatic CAG expansion. Instead, all cells in BAC226Q mice carry uniformly long CAG-CAA repeats, representing the end status of somatic expansion beyond the 150-CAG threshold for toxicity (50). However, for its polyQ length high stability, robust, and consistent pathological recapitulations, BAC226Q is an appropriate model for testing gene therapy.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "INTRODUCTION", "text": "Both conventional and self-inactivating CRISPR-Cas9 efficiently and permanently disrupted the human mHTT transgene, leading to 60 to 90% reduction of mutant HTT protein and 90% of cytosolic aggregates and nuclear inclusions, which led to significant rescue of the abovementioned deficits in BAC226Q mice. We also tested the effectiveness of gene editing at various mouse ages corresponding to human patient presymptomatic, onset, and symptomatic disease stages.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "RESULTS", "text": "For complete human mHTT elimination, single guide RNAs (sgRNAs) HTTg1 and HTTg2 were designed to target the human HTT N-terminal 17\u2013amino acid domain (N17)\u2013coding DNA sequence before CAG repeats (Fig. 1A). N17 is a conserved region across most of the species and is shown to play an important role in mHTT-induced pathogenesis (51). For in vivo experiments, cytomegalovirus (CMV)\u2013driven SaCas9 and U6\u2013driven HTT sgRNA were inserted into a single adeno-associated virus (AAV) vector (Fig. 1B). In human embryonic kidney (HEK) 293T cells expressing human HTT-exon 1 fragment with 20Q or 120Q and in-frame N-terminal enhanced green fluorescent protein fusion (EGFP-hHTT-20Q or EGFP-hHTT-120Q), both sgRNAs resulted in robust HTT reduction, with the better performing SaCas9-HTTg1 decreasing HTT by 70 to 80%, and therefore chosen for future experiments (Fig. 1, C to K).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "RESULTS", "text": "(A) Schematic representation of targeted human HTT exon 1. sgRNAs HTTg1 and HTTg2 were designed to target the N17 domain\u2013coding DNA sequence, which is located upstream of the CAG repeats. PAM, protospacer-adjacent motif. (B) Schematic representation of AAV-SaCas9-HTT sgRNA vector. U6-driven sgRNA and a CMV-driven Cas9 from S. aureus tagged with HA were packaged into a single AAV vector of 4.5 kb. ITR, inverted terminal repeat; NLS, nuclear localization signal sequence; 3\u00d7HA, three tandem repeats of the human influenza hemagglutinin (HA) tag. (C to K) Validation of human HTT editing in human embryonic kidney (HEK) 293T cells expressing EGFP-hHTT-20Q or EGFP-hHTT-120Q reporter system. (C) SaCas9-HTTg1 and SaCas9-HTTg2 reduced EGFP-fusion protein expression and aggregation in cells, whereas the control SaCas9-sgRNA did not. Scale bar, 100 \u03bcm. [(D) and (E)] Quantitative analyses of EGFP-20Q (D) or EGFP-120Q (E) fluorescence intensity. n\u00a0=\u00a03 or 4; **P\u00a0=\u00a00.0083 (20Q), **P\u00a0=\u00a00.0087 (120Q), and ***P\u00a0=\u00a00.0009 (120Q). [(F) to (K)] Western blot detecting EGFP and SaCas9 (anti-HA) protein level in EGFP-hHTT-20Q [(F) to (H)] or EGFP-hHTT-120Q [(I) to (K)] reporter system. [(F) and (I)] SaCas9-HTTg1 and SaCas9-HTTg2 reduced EGFP-hHTT protein expression compared to control SaCas9-sgRNA with equivalent SaCas9 expression level. [(G), (H), (J), and (K)] Quantification of relative EGFP [(G) and (J)] and SaCas9 [(H) and (K)] protein expression. n\u00a0=\u00a03 per group; **P =\u00a00.0087 (20Q) and **P\u00a0=\u00a00.0069 (120Q). Data were shown as means\u00a0\u00b1\u00a0SEM. One-way analysis of variance (ANOVA) with Tukey\u2019s multiple comparisons test.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "RESULTS", "text": "To evaluate the effects of CRISPR-Cas9\u2013mediated mHTT gene editing in vivo, we used BAC226Q expressing full-length human mHTT with 226 CAG-CAA repeats (46). This mouse model recapitulates key aspects of age-dependent and progressive HD-like pathology and phenotypes and, therefore, a powerful tool for testing gene-modifying therapeutics.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "RESULTS", "text": "A longitudinal study of biochemistry, cellular pathology, and behavioral tests was designed in BAC226Q mice with AAV-SaCas9-HTTg1 (Fig. 2A). To evaluate AAV delivery system, we first confirmed that AAV9 stereotactic injection into striatum and motor cortex, the two major HD pathological regions, yielded highly efficient neuronal transduction without inducing immunoreactivity (fig. S1, A to H). Subsequently, AAV-SaCas9-HTTg1 was injected into mouse brains at the age of 1 month, resulting in successful deletion of mHTT gene and lifelong 60 to 90% reduction of mHTT protein (Fig. 2, B and C). In injected areas of striatum and cortex, AAV-SaCas9-HTTg1 reduced mHTT near completion (Fig. 2D). Furthermore, among cells effectively transduced by AAV-SaCas9-HTTg1, 94 to 96% mHTT was eliminated (fig. S2, A to D). We further investigated the consequence of effective gene editing on mHTT neuropil aggregates and nuclear inclusions, cardinal HD pathological hallmarks (52), which were accurately recapitulated in BAC226Q mice (46). By 4 months of age, BAC226Q mice exhibit clear cytoplasmic and nuclear mHTT aggregation in striatum and mainly cytoplasmic aggregation in cortex. As pathology progresses to 7 months, mHTT nuclear inclusions become more prominent (Fig. 2, E to L; and fig. S3, A and B). In 4-month-old BAC226Q mice, AAV-SaCas9-HTTg1 reduced total mHTT aggregation signal in striatal and cortical neurons by 90 and 55%, respectively (Fig. 2, E to H). At 7 months, the rescue effects remained robust, with total mHTT reduction of 86% in striatum and 48% in cortex (Fig. 2, I to L). We further investigated the reduction of cytoplasmic and nuclear mHTT aggregates in number and size separately and observed the same robust reduction (fig. S3, A to J). The percentage of nucleus with mHTT aggregates was both reduced in striatum and cortex (fig. S3, K to L). AAV-SaCas9-HTTg1 did not change glial fibrillary acidic protein (GFAP) immunoreactivity in BAC226Q mice (fig. S4, A and B). In conclusion, these results confirmed a sustained and remarkable reduction of mHTT protein and aggregates in vivo for lifetime, just by a single dose of AAV-SaCas9-HTTg1.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "RESULTS", "text": "(A) Experimental timeline and corresponding analyses. (B to D) Significant reduction of mHTT protein in 11-month-old BAC226Q striatum and cortex injected with AAV-SaCas9-HTTg1. Non\u2013Tg-SaCas9, nontransgenic mice injected with AAV-SaCas9; HD-SaCas9, BAC226Q injected with AAV-SaCas9; HD-SaCas9-HTTg1, BAC226Q injected with AAV-SaCas9-HTTg1. [(B) and (C)] Western blot and quantification of sustained decrease of striatal mHTT in HD-SaCas9-HTTg1. 1C2 antibody detected mHTT protein. \u03b2-Actin was a loading control. n\u00a0=\u00a03 for non\u2013Tg-SaCas9 and n\u00a0=\u00a04 for HD-SaCas9 and HD-SaCas9-HTTg1; *P\u00a0=\u00a00.0104. (D) Immunohistochemistry of robust mHTT decrease in striatal virus\u2013diffused areas. S830 antibody detected mHTT. AAV9-EGFP was a control. Dotted lines: AAV9 diffusion boundary. On the right are enlarged figures. Scale bars, 2000 \u03bcm. (E to L) Significant decrease of S830-positive mHTT aggregates in HD-SaCas9-HTTg1. [(E), (F), (I), and (J)] Immunofluorescent histology of striatum [(E) and (I)] and primary motor cortex [(F) and (J)] at 4 [(E) and (F)] and 7 months [(I) and (J)]; boxed areas were enlarged on the right. Cytoplasmic aggregates (arrows) and nuclear inclusion (arrowheads). S830 and 4\u2032,6-diamidino-2-phenylindole (DAPI) detected mHTT (green) and nucleus (blue). Scale bars, 50 \u03bcm. [(G), (H), (K), and (L)] Quantification of mHTT aggregation signal in striatum [(G) and (K)] and primary motor cortex [(H) and (L)] at 4 [(G) and (H)] and 7 months [(K) and (L)]. Aggregation signal was analyzed by particle analysis in ImageJ and calculated by the number of aggregates \u00d7 integrated density. n\u00a0=\u00a05 per group; 4 months: ***P\u00a0<\u00a00.0001 (non\u2013Tg-SaCas9 versus HD-SaCas9, striatum); ***P\u00a0=\u00a00.0002 (HD-SaCas9 versus HD-SaCas9-HTTg1, striatum); **P\u00a0=\u00a00.0011; *P\u00a0=\u00a00.0495; not significant (n.s.),\u00a00.7886 (striatum); and n.s.,\u00a00.1176 (cortex); 7 months: ***P\u00a0<\u00a00.0001 (striatum); n.s.,\u00a00.1211 (striatum); ***P\u00a0=\u00a00.0002 (cortex); *P\u00a0=\u00a00.0265 (non\u2013Tg-SaCas9 versus HD-SaCas9-HTTg1, cortex); and *P\u00a0=\u00a00.0402 (HD-SaCas9 versus HD-SaCas9-HTTg1, cortex). Data were shown as means\u00a0\u00b1\u00a0SEM. One-way ANOVA with Tukey\u2019s multiple comparisons test.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "RESULTS", "text": "Because AAV-SaCas9-HTTg1 effectively deleted mHTT gene and decreased mHTT protein and aggregates, we set out to investigate whether gene editing rescues cardinal HD-like phenotypes such as age-dependent motor deficits, weight loss, and shortened life span in BAC226Q mice (46). BAC226Q mice exhibit normal motor function until 2 to 3 months of age, followed by severe hyperactivity and chorea-like movement from 3 to 4 months and a subsequent reduction in motor activity from 6 to 7 months. These deficits and rescue effects of AAV-SaCas9-HTTg1 were observed in home cage behavior (movie S1) and analyzed by catwalk, rotarod, and open-field tasks. In catwalk task, BAC226Q mice developed progressive gait irregularities reflected by scrambled step sequences and footprint area reductions (Fig. 3, A to E). Severe scrambled step sequences at 6 months were fully rescued by AAV-SaCas9-HTTg1 (Fig. 3, A and B, and movie S2). Decreased footprint areas at 2.5, 4, and 6 months in BAC226Q mice were significantly ameliorated by AAV-SaCas9-HTTg1 (Fig. 3, C to E). In rotarod task, BAC226Q mice exhibited strong deficits in motor coordination and balance measured by latency to fall, which were significantly improved in HD-SaCas9-HTTg1 mice (Fig. 3F). In open-field test, nontransgenic control mice traveled across the cage, whereas BAC226Q mice circled in periphery. This deficit was reversed in HD-SaCas9-HTTg1 mice (Fig. 3G). Furthermore, markedly increased total travel distance in BAC226Q mice was attenuated by about 30 to 35% (Fig. 3H).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "RESULTS", "text": "(A to E) Improved gait performance in HD-SaCas9-HTTg1 analyzed by catwalk. (A) Representative footprints of non\u2013Tg-SaCas9 (top), HD-SaCas9 (middle), and HD-SaCas9-HTTg1 (bottom) at 6 months. (B) Significant increase of step sequence regularity index in 6-month HD-SaCas9-HTTg1. n\u00a0=\u00a010 per group; *P\u00a0=\u00a00.0309 (HD-SaCas9 versus HD-SaCas9-HTTg1). [(C) to (E)] Footprint area quantification. (C) Similar baselines at 2.5 months. [(D) and (E)] Deficits in HD-SaCas9 were significantly rescued in HD-SaCas9-HTTg1 at 4 months (*P\u00a0=\u00a00.0231, RH) and 6 months (**P\u00a0=\u00a00.0010, RH; ***P\u00a0<\u00a00.001, LH). n\u00a0=\u00a09 or 10 per group. RF, right front; RH, right hind; LF, left front; LH, left hind. (F) Improved coordination and balance ability in rotarod test. n\u00a0=\u00a09 or 10 per group; HD-SaCas9-HTTg1 versus HD-SaCas9: *P\u00a0=\u00a00.0202 (12 weeks), **P\u00a0=\u00a00.0126 (13 weeks), **P\u00a0=\u00a00.0035 (14 weeks), and ***P\u00a0<\u00a00.001 (15 and 16 weeks). (G and H) Rescue of locomotor performance in HD-SaCas9-HTTg1 in open-field test. (G) Representative heatmaps exhibiting cumulative time and location for 6-month-old non\u2013Tg-SaCas9 (left), HD-SaCas9 (middle), and HD-SaCas9-HTTg1 (right). (H) Total travel distance at 2.5, 4, and 6 months. n\u00a0=\u00a09 or 10 per group; HD-SaCas9-HTTg1 versus HD-SaCas9: *P\u00a0=\u00a00.0109 (4 months) and *P\u00a0=\u00a00.0456 (6 months). (I) Weight loss attenuation in HD-SaCas9-HTTg1. Non\u2013Tg-SaCas9 (n\u00a0=\u00a027), HD-SaCas9 (n\u00a0=\u00a026), and BAC226Q-HTTg1 (n\u00a0=\u00a026); *P\u00a0<\u00a00.05 and **P\u00a0<\u00a00.01. (J) Complete rescue of survival time by gene editing. The survival of HD-SaCas9-HTTg1 was significantly improved compared with HD-SaCas9 and reached to a level consistent with that of non\u2013Tg-SaCas9. Non\u2013Tg-SaCas9 (n\u00a0=\u00a017), HD-SaCas9 (n\u00a0=\u00a017), and HD-SaCas9-HTTg1 (n\u00a0=\u00a021); **P\u00a0=\u00a00.0061 (HD-SaCas9-HTTg1 versus HD-SaCas9), P\u00a0=\u00a00.7434 (HD-SaCas9-HTTg1 versus non\u2013Tg-SaCas9), and **P\u00a0=\u00a00.0025 (HD-SaCas9 versus non\u2013Tg-SaCas9). Data were shown as means\u00a0\u00b1\u00a0SEM. One-way ANOVA with Tukey\u2019s multiple comparisons test [(B) and (H)], two-way ANOVA with Tukey\u2019s multiple comparisons test [(C) to (F) and (I)], and log-rank (Mantel-Cox) test (J).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "RESULTS", "text": "BAC226Q mouse model recapitulates HD-like age-dependent progressive weight loss and shortened life span as in patients with HD (46). AAV-SaCas9-HTTg1 attenuated body weight loss by 5 to 22% (Fig. 3I) and promoted life span markedly (Fig. 3J). Together, significant and long-term rescue of motor deficits, weight loss, and shortened life span were obtained if gene editing eliminated mHTT in 1-month-old BAC226Q mice before any pathology and motor deficits were observed.\n\nAn important question relevant to clinical treatment is at which disease stages gene editing is still effective (53). BAC226Q mice accurately recapitulated the age-dependent progression from normal development to phenotype onset and worsening and severe end stages. Therefore, we investigated the rescue effects of gene editing in BAC226Q at 1, 4, and 7 months of age, which are before, at, and well after the onset of pathological and behavioral abnormalities (Fig. 4A).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "RESULTS", "text": "(A) Experimental timeline in HD mice and corresponding phenotypes at different time points. (B to I) Significant reduction of mEM48-positive mHTT aggregates (green) in the striatum from 4-month-injected [(B) to (E)] and 7-month-injected HD mice [(F) to (I)]. Mouse brain samples were collected 4 months postinjection. [(B) and (F)] Representative figures showing that mHTT aggregation was decreased in 4-month-injected HD-SaCas9-HTTg1 (B) and 7-month-injected HD-SaCas9-HTTg1 (F). mHTT (green) and SaCas9 (red) were detected by mEM48 and anti-HA antibody, respectively. Scale bars, 100 \u03bcm. Enlarged figures were shown on the right side. Scale bars, 20 \u03bcm. [(C) to (E) and (G) to (I)] Quantification of mHTT aggregation signal [(C) and (G)], aggregate number [(D) and (H)], and aggregate size [(E) and (I)] in the striatum in 4-month-injected HD-SaCas9-HTTg1 mice [(C) to (E)] and 7-month-injected HD-SaCas9-HTTg1 mice [(G) to (I)]. n\u00a0=\u00a03 to 6 per group; ***P\u00a0<\u00a00.0001 [(C) to (E)], **P\u00a0=\u00a00.0031 (D), ***P =\u00a00.0006 [(G), HD-no inj-7M versus HD-SaCas9-7M inj-11M], ***P\u00a0<\u00a00.0001 [(G), HD-SaCas9-7M inj-11M versus HD-SaCas9-HTTg1-7M inj-11M], ***P =\u00a00.0007 (H), **P\u00a0=\u00a00.0082 (I), and *P\u00a0=\u00a00.0363 (I). Data were shown as means\u00a0\u00b1\u00a0SEM. One-way ANOVA with Tukey\u2019s multiple comparisons test. n.s., not significant.\n\nGene editing at the first time point of 1 month in prephenotypic mice yielded strong benefits as shown above (Figs. 2 and 3 and fig. S2 and S3).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "RESULTS", "text": "The second time point of gene editing was at 4 months when BAC226Q mice start to exhibit notable mHTT aggregation and motor behavior abnormalities. We injected AAV-SaCas9-HTTg1 at 4 months and examined the rescue effects of brain pathology and behavior throughout the life span of mice. At 8 months, AAV-SaCas9-HTTg1 reduced overall mHTT aggregation signal by 91% (Fig. 4, B and C). Further detailed analyses revealed that the reduction of aggregate number and size were 80 and 62%, respectively (Fig. 4, D and E). The marked 10-fold increase of aggregations from 4 to 8 months in BAC226Q was completely abolished by AAV-SaCas9-HTTg1 (Fig. 4, B to E). The similar mHTT aggregation reduction was maintained at 11 months (fig. S5, A and B). In motor behavioral tests, gait irregularities at 6 months (fig. S6, A and B, and movie S3) and locomotor activity at 8 months (fig. S6, C and D) were both rescued, although slightly less effectively than that by 1-month injection.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "RESULTS", "text": "The third time point of gene editing was at 7 months when BAC226Q mice show severe motor deficits and pathology such as bigger and more predominant mHTT nuclear inclusions. This intervention still robustly reduced overall mHTT aggregation signal by 74%, aggregation number by 60%, and aggregation size by 32% (Fig. 4, F to I). Similar to the 4-month injection data, AAV-SaCas9-HTTg1 completely abolished the onefold increase of aggregations from 7 to 11 months in BAC226Q (Fig. 4, F to I; and fig. S5, C and D). Consistent with the robust rescue in pathology, BAC226Q postinjection survival time was greatly increased (fig. S6E). In open-field test, locomotor abnormality was not rescued (fig. S6F). Gait irregularities measured by catwalk test could not be performed due to severe motor deficits. It is important that these data collectively demonstrated that gene editing at different pathogenic time points resulted in significant mHTT reduction and recovery of life span in all cases and rescue of motor deficits in early injections.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "RESULTS", "text": "To address the clinically relevant and general concern that conventional long-term Cas9 expression in patients, potentially for decades, may cause cumulative off-target effects and acquired immune responses against Cas9 harboring cells, it will be greatly advantageous to develop a strategy that eliminates both the target gene and subsequently Cas9 itself. Here, we report such a self-inactivating system that carries AAV-SaCas9-HTTg1 and an additional AAV-sgRNA to target Cas9 (gCas9), so that Cas9 sequence will be destroyed after HTT disruption (Fig. 5, A and B).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "RESULTS", "text": "(A) Schematic representation of dual vectors for self-inactivating system. In AAV-SaCas9-HTTg1 vector, U6 drives HTTg1 to target human HTT and CMV drives SaCas9 gene. In AAV-EGFP-gCas9 vector, U6 drives gCas9 to target SaCas9 and CMV drives EGFP as a reporter. (B) Intraparenchymal delivery of dual AAV vectors into 1-month-old mouse striatum and primary motor cortex. (C to N) Self-inactivating AAV-SaCas9-HTTg1/gCas9 robustly decreased SaCas9 and reduced mHTT aggregation as effectively as the conventional AAV-SaCas9-HTTg1. Representative immunofluorescence images of mEM48-labeled mHTT aggregation and HA-labeled SaCas9 in striatum of BAC226Q at 4 months (C), 7 months (D), and 11 months (E). Scale bars, 50 \u03bcm. [(F) and (G)] Quantification of SaCas9 fluorescence intensity (F) and mHTT aggregation number (G) in (C). n\u00a0=\u00a03 per group; SaCas9: ***P\u00a0<\u00a00.0001; mHTT: ***P\u00a0<\u00a00.0001 (HD-SaCas9 versus HD-SaCas9-HTTg1); ***P\u00a0<\u00a00.0001 (HD-SaCas9 versus HD-SaCas9-HTTg1/gCas9); and not significant (n.s.),\u00a00.9679. [(H) and (I)] Quantification of SaCas9 fluorescence intensity (H) and mHTT aggregation number (I) in (D). n\u00a0=\u00a03 to 5 per group; SaCas9: ***P\u00a0<\u00a00.0001; mHTT: ***P\u00a0<\u00a00.0001 (HD-SaCas9 versus HD-SaCas9-HTTg1); ***P\u00a0<\u00a00.0001 (HD-SaCas9 versus HD-SaCas9-HTTg1/gCas9); and n.s.,\u00a00.9987. [(J) and (K)] Quantification of SaCas9 fluorescence intensity (J) and mHTT aggregation number (K) in (E). n\u00a0=\u00a03 per group; SaCas9: ***P\u00a0=\u00a00.0007; mHTT: ***P\u00a0=\u00a00.0002 (HD-SaCas9 versus HD-SaCas9-HTTg1); ***P\u00a0<\u00a00.0002 (HD-SaCas9 versus HD-SaCas9-HTTg1/gCas9); n.s.,\u00a00.9096. (L) Immunohistochemistry of HA-labeled SaCas9 (top) and S830-labeled mHTT (bottom) in 7-month-old BAC226Q. Boxed areas were enlarged on the right. Scale bar, 2000 \u03bcm. (M) Quantification of SaCas9 reduction in (L). n\u00a0=\u00a04 per group; ***P\u00a0<\u00a00.0001. IOD, integrated optical density. (N) Quantification of striatal mHTT aggregation number in boxed areas of (L). n\u00a0=\u00a03 or 4 per group; ***P\u00a0<\u00a00.0001 (HD-SaCas9 versus HD-SaCas9-HTTg1); ***P\u00a0<\u00a00.0001 (HD-SaCas9 versus HD-SaCas9-HTTg1/gCas9); and n.s.,\u00a00.5473. Data were shown as means\u00a0\u00b1\u00a0SEM. One-way ANOVA with Tukey\u2019s multiple comparisons test [(F) to (K) and (N)] and unpaired two-tailed t test (M).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "RESULTS", "text": "To select the most effective gCas9, we used HEK293T cells expressing EGFP-hHTT-20Q/120Q to test 10 gCas9 against different SaCas9 coding sequences (fig. S7A and table S1). The results showed that gCas9-1 exhibited the highest efficiency in eliminating Cas9 without impairing mHTT reduction (fig. S7, B to L) and thus was used for subsequent in vivo experiments.\n\nWe then tested in BAC226Q mice. Similar to the conventional gene editing, the self-inactivating AAV-SaCas9-HTTg1/gCas9 effectively reduced mHTT by 94% at 4 months, 92% at 7 months, and 92% at 11 months (Fig. 5, C to N; and fig. S8, A to F). Meanwhile, AAV-SaCas9-HTTg1/gCas9 reduced above 90% Cas9 by 1 week postinjection (fig. S8, G and H) and lasted throughout life span (Fig. 5, C to N; and fig. S8, A to F). In mice injected with AAV9-SaCas9-HTTg1/AAV9-EGFP-gCas9, 93% of green fluorescent protein (GFP)\u2013positive cells expressed neither SaCas9 nor mHTT (fig. S9, A and B). These data demonstrated that the self-inactivating system can efficiently eliminate SaCas9 without affecting the clearance of mHTT and that Cas9-mediated gene editing events in neurons happen very rapidly well within 1 week postinjection, after which Cas9 is no longer needed.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "RESULTS", "text": "Further analyses revealed that AAV-SaCas9-HTTg1/gCas9\u2013 and AAV-SaCas9-HTTg1\u2013injected mice showed equivalent rescue of step sequence irregularity (Fig. 6, A and B, and movie S4), footprint area reduction (Fig. 6C) in catwalk task, locomotor hyperactivity in open-field test (Fig. 6D), latency to fall in rotarod test (Fig. 6E), and shortened life span (Fig. 6F).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "RESULTS", "text": "(A to C) Self-inactivating AAV-SaCas9-HTTg1/gCas9 improved gait performance in catwalk analysis of BAC226Q mice at 6 months. (A) Representative footprints of non\u2013Tg-SaCas9, HD-SaCas9, HD-SaCas9-HTTg1, and HD injected with AAV-SaCas9-HTTg1/gCas9 (HD-SaCas9-HTTg1/gCas9) from top to bottom. (B) Increased step sequence regularity index in HD-SaCas9-HTTg1/gCas9 and HD-SaCas9-HTTg1. n\u00a0=\u00a010 per group; *P\u00a0=\u00a00.0387 (HD-SaCas9 versus HD-SaCas9-HTTg1); *P\u00a0=\u00a00.0164 (HD-SaCas9 versus HD-SaCas9-HTTg1/gCas9); and not significant (n.s.),\u00a00.9844. (C) Quantification of footprint area in gait analysis. n\u00a0=\u00a010 per group; ***P\u00a0=\u00a00.0007 (LH); **P\u00a0=\u00a00.0010 (RH); **P\u00a0=\u00a00.0057 (LH); *P\u00a0=\u00a00.0136 (RH); n.s.,\u00a00.8595 (RH); and n.s.,\u00a00.9319 (LH). RF, right front; RH, right hind; LF, left front; LH, left hind. (D) Rescue of locomotor activity in HD-SaCas9-HTTg1/gCas9 and HD-SaCas9-HTTg1 in open-field test. n\u00a0=\u00a015 per group; 4 months: ***P\u00a0<\u00a00.0001 (HD-SaCas9 versus HD-SaCas9-HTTg1); ***P\u00a0<\u00a00.0001 (HD-SaCas9 versus HD-SaCas9-HTTg1/gCas9); and n.s.,\u00a00.8905; 7 months: ***P\u00a0=\u00a00.0003 (HD-SaCas9 versus HD-SaCas9-HTTg1); ***P\u00a0<\u00a00.0001 (HD-SaCas9 versus HD-SaCas9-HTTg1/gCas9); and n.s.,\u00a00.9207; 11 months: n.s.,\u00a00.1145 (HD-SaCas9 versus HD-SaCas9-HTTg1); n.s.,\u00a00.0746 (HD-SaCas9 versus HD-SaCas9-HTTg1/gCas9); and n.s.,\u00a00.9980 (HD-SaCas9-HTTg1 versus HD-SaCas9-HTTg1/gCas9). (E) Improvement of coordination and balance ability in HD-SaCas9-HTTg1/gCas9 and HD-SaCas9-HTTg1 in rotarod test. n\u00a0=\u00a015 per group; HD-SaCas9 versus HD-SaCas9-HTTg1/gCas9: **P\u00a0=\u00a00.0065 (15 weeks), ***P =\u00a00.0006 (16 weeks), and ***P\u00a0<\u00a00.0001 (17 to 20 weeks); HD-SaCas9 versus HD-SaCas9-HTTg1/gCas9: ###P\u00a0<\u00a00.0001 (15 to 20 weeks). (F) Compared to HD-SaCas9, the survival time of HD-SaCas9-HTTg1/gCas9 was significantly extended and reached to a level consistent with that of HD-SaCas9-HTTg1. Non\u2013Tg-SaCas9 (n\u00a0=\u00a08), HD-SaCas9 (n\u00a0=\u00a010), HD-SaCas9-HTTg1 (n\u00a0=\u00a010), and HD-SaCas9-HTTg1/gCas9 (n\u00a0=\u00a09); ***P\u00a0<\u00a00.0001 (HD-SaCas9 versus HD-SaCas9-HTTg1), ***P\u00a0<\u00a00.0001 (HD-SaCas9 versus HD-SaCas9-HTTg1/gCas9), and P\u00a0=\u00a00.7174 (HD-SaCas9-HTTg1 versus HTTg1/gCas9). Data were shown as means\u00a0\u00b1\u00a0SEM. One-way ANOVA with Tukey\u2019s multiple comparisons test (B), two-way ANOVA with Tukey\u2019s multiple comparisons test [(C) to (E)], and log-rank (Mantel-Cox) test (F).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "RESULTS", "text": "Transcriptomic dysregulation is a hallmark in patients with HD and animal models (54). We investigated transcriptomes and found that, compared to nontransgenic mice, BAC226Q had signature changes that were partially reversed by conventional and self-inactivating gene editing (fig. S10A). The distribution of differentially expressed genes (DEGs) across groups was also visualized in the volcano plot (fig. S10, B to D, and table S2). In comparison to the differential gene expression observed between BAC226Q and non-Tg mice, the transcriptional differences between BAC226Q-HTTg1 or BAC226Q-HTTg1/gCas9 and non-Tg were reduced, more prominent for the down-regulated genes. We further analyzed how the 245 up-regulated and 452 down-regulated genes in BAC226Q compared to non-Tg changed in the BAC226Q-HTTg1 and BAC226Q-HTTg1/gCas9 groups. The results indicated that the gene editing therapy could partially reverse the dysregulated genes observed in HD mice (fig. S10, E and F). Gene Ontology enrichment analysis was also performed on the DEGs in the BAC226Q-SaCas9 control, BAC226Q-SaCas9-HTTg1, and BAC226Q-SaCas9-HTTg1/gCas9 groups in comparison with those in non\u2013Tg-SaCas9 mice. Both the conventional CRISPR-Cas9 and the self-inactivating systems showed comparable efficacies in ameliorating transcriptomic-level dysregulation. The number and expression levels of genes involved in important pathways were robustly improved in gene-edited mice. Noticeably, most prominent changes were enriched in genes for postsynaptic structure and function, ion channel activity, and transmembrane transporters (fig. S10G).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "RESULTS", "text": "Therefore, we succeeded in developing a self-inactivating system that eliminated mHTT gene and protein, rescued deficits in motor functions, and life span with the same high efficiency as the conventional CRISPR-Cas9 and that abolished Cas9 after a transient expression. This system could be highly beneficial for clinical applications.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "RESULTS", "text": "BAC226Q contains human mHTT transgene and endogenous mouse Htt gene. To investigate the long-term cleavage effects on human mHTT gene, mouse brain genomic DNA was extracted at 11 months. Compared with control BAC226Q mice, the human mHTT gene sequence in BAC226Q-SaCas9-HTTg1 had been edited exactly from 3 base pairs (bp) upstream of PAM (Fig. 7A), consistent with previous findings (55). To compare the specificity of self-inactivating and conventional CRISPR systems, we applied primer extension\u2013mediated sequencing (PEM-seq), a method for detecting editing efficiency and off-target hotspots simultaneously because it can identify CRISPR-Cas9 editing events including indels, large deletions, as well as genome-wide translocations with high sensitivity (56). In human HTT on-target site, the editing events were predominantly small indels, about 90% of which resulted in frameshift and premature stop codon (Fig. 7, B and C). In both strategies, no off-target events were found except for the predictable mouse Htt gene, which shares 20/21 sequence homology with human HTT gene target site (Fig. 7D). Even with 20/21 sequence homology, a 95% identical base pairs, HTTg1 only yielded off-target translocations with very low frequencies of 0.012\u00a0\u00b1\u00a00.001% in 2-month-old and 0.017\u00a0\u00b1\u00a00.004% in 12-month-old BAC226Q mouse brain striatum. HTTg1/gCas9 further significantly reduced HTTg1-induced off-target effects in 12-month-old mouse by 2.5-fold (Fig. 7E). For more clinical relevance, the off-target effect for HTTg1 was also investigated in human cell lines. In human induced pluripotent stem cells (iPSCs), neither HTTg1 nor HTTg1/gCas9 induced any off-target mismatches (Fig. 7F). In HEK293T cells, HTTg1 resulted only one off-target editing in chromosome 1 22,620,422-22,620,442 locus. The self-inactivating HTTg1/gCas9 completely eliminated all off-target editing (Fig. 7, G and H). It is worth noting that gCas9 per se did not induce any off-target effect in mouse brain and human cell line. Together, these data confirmed the successful editing by SaCas9-HTTg1 and SaCas9-HTTg1/gCas9 on human HTT gene and suggested that self-inactivating CRISPR-Cas9 may eliminate off-target effects and thus provide maximal safety for long-term therapeutic use.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "RESULTS", "text": "(A) Sanger sequencing of human HTT exon 1 amplicons in control HD (top) and HD-SaCas9-HTTg1 mouse (bottom). The genomic DNA editing by HTTg1 was demonstrated by the scrambled nucleotide signals in the bottom panel. (B to H) Gene editing specificity detected by PEM-seq. [(B) and (C)] Indel patterns (B) and frequency (C) on human HTT in 12-month-old BAC226Q. Blue, guide RNA target site; purple, PAM sequence; red, insertions; blue dashed line, deletions. n\u00a0=\u00a09. (D) Circos plot for off-target site in BAC226Q. Arrows indicate the on-target site of human HTT gene and the only off-target site of mouse Htt gene. The curved line connects the on-target and off-target site. Blue, guide RNA target site; purple, PAM sequence; red, mismatch. (E) Low frequencies calculated as off-target translocations over total editing events in the striatum of 2- and 12-month-old BAC226Q. n\u00a0=\u00a08 for HD-SaCas9-HTTg1 (2 months), n\u00a0=\u00a05 for HD-SaCas9-HTTg1/gCas9 (2 and 12 months), and n\u00a0=\u00a06 for HD-SaCas9-HTTg1 (12 months); *P\u00a0=\u00a00.0135. (F) No off-target mismatch calculated as off-target translocations over total editing events in human iPSCs. n\u00a0=\u00a03. [(G) and (H)] Off-target frequencies calculated as off-target translocations over total editing events (G) and circos plot (H) in HEK293T cells. Arrows indicate the on-target site of human HTT gene and off-target site on chromosome 1. The curved line connects the on-target and off-target site. Blue, guide RNA target site; purple, PAM sequence; red, mismatch. n\u00a0=\u00a03 for SaCas9 and HTTg1/gCas9 and n\u00a0=\u00a04 for HTTg1. Data were shown as means\u00a0\u00b1\u00a0SEM. Two-way ANOVA with Bonferroni\u2019s multiple comparisons test (E) and one-way ANOVA with Tukey\u2019s multiple comparisons test (G). n.s., not significant.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "CRISPR-Cas9 design and validation of the cleavage of human ", "text": "For complete human mHTT elimination, single guide RNAs (sgRNAs) HTTg1 and HTTg2 were designed to target the human HTT N-terminal 17\u2013amino acid domain (N17)\u2013coding DNA sequence before CAG repeats (Fig. 1A). N17 is a conserved region across most of the species and is shown to play an important role in mHTT-induced pathogenesis (51). For in vivo experiments, cytomegalovirus (CMV)\u2013driven SaCas9 and U6\u2013driven HTT sgRNA were inserted into a single adeno-associated virus (AAV) vector (Fig. 1B). In human embryonic kidney (HEK) 293T cells expressing human HTT-exon 1 fragment with 20Q or 120Q and in-frame N-terminal enhanced green fluorescent protein fusion (EGFP-hHTT-20Q or EGFP-hHTT-120Q), both sgRNAs resulted in robust HTT reduction, with the better performing SaCas9-HTTg1 decreasing HTT by 70 to 80%, and therefore chosen for future experiments (Fig. 1, C to K).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "CRISPR-Cas9 design and validation of the cleavage of human ", "text": "(A) Schematic representation of targeted human HTT exon 1. sgRNAs HTTg1 and HTTg2 were designed to target the N17 domain\u2013coding DNA sequence, which is located upstream of the CAG repeats. PAM, protospacer-adjacent motif. (B) Schematic representation of AAV-SaCas9-HTT sgRNA vector. U6-driven sgRNA and a CMV-driven Cas9 from S. aureus tagged with HA were packaged into a single AAV vector of 4.5 kb. ITR, inverted terminal repeat; NLS, nuclear localization signal sequence; 3\u00d7HA, three tandem repeats of the human influenza hemagglutinin (HA) tag. (C to K) Validation of human HTT editing in human embryonic kidney (HEK) 293T cells expressing EGFP-hHTT-20Q or EGFP-hHTT-120Q reporter system. (C) SaCas9-HTTg1 and SaCas9-HTTg2 reduced EGFP-fusion protein expression and aggregation in cells, whereas the control SaCas9-sgRNA did not. Scale bar, 100 \u03bcm. [(D) and (E)] Quantitative analyses of EGFP-20Q (D) or EGFP-120Q (E) fluorescence intensity. n\u00a0=\u00a03 or 4; **P\u00a0=\u00a00.0083 (20Q), **P\u00a0=\u00a00.0087 (120Q), and ***P\u00a0=\u00a00.0009 (120Q). [(F) to (K)] Western blot detecting EGFP and SaCas9 (anti-HA) protein level in EGFP-hHTT-20Q [(F) to (H)] or EGFP-hHTT-120Q [(I) to (K)] reporter system. [(F) and (I)] SaCas9-HTTg1 and SaCas9-HTTg2 reduced EGFP-hHTT protein expression compared to control SaCas9-sgRNA with equivalent SaCas9 expression level. [(G), (H), (J), and (K)] Quantification of relative EGFP [(G) and (J)] and SaCas9 [(H) and (K)] protein expression. n\u00a0=\u00a03 per group; **P =\u00a00.0087 (20Q) and **P\u00a0=\u00a00.0069 (120Q). Data were shown as means\u00a0\u00b1\u00a0SEM. One-way analysis of variance (ANOVA) with Tukey\u2019s multiple comparisons test.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "SaCas9-HTTg1\u2013mediated human mHTT disruption in BAC226Q mice", "text": "To evaluate the effects of CRISPR-Cas9\u2013mediated mHTT gene editing in vivo, we used BAC226Q expressing full-length human mHTT with 226 CAG-CAA repeats (46). This mouse model recapitulates key aspects of age-dependent and progressive HD-like pathology and phenotypes and, therefore, a powerful tool for testing gene-modifying therapeutics.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "SaCas9-HTTg1\u2013mediated human mHTT disruption in BAC226Q mice", "text": "A longitudinal study of biochemistry, cellular pathology, and behavioral tests was designed in BAC226Q mice with AAV-SaCas9-HTTg1 (Fig. 2A). To evaluate AAV delivery system, we first confirmed that AAV9 stereotactic injection into striatum and motor cortex, the two major HD pathological regions, yielded highly efficient neuronal transduction without inducing immunoreactivity (fig. S1, A to H). Subsequently, AAV-SaCas9-HTTg1 was injected into mouse brains at the age of 1 month, resulting in successful deletion of mHTT gene and lifelong 60 to 90% reduction of mHTT protein (Fig. 2, B and C). In injected areas of striatum and cortex, AAV-SaCas9-HTTg1 reduced mHTT near completion (Fig. 2D). Furthermore, among cells effectively transduced by AAV-SaCas9-HTTg1, 94 to 96% mHTT was eliminated (fig. S2, A to D). We further investigated the consequence of effective gene editing on mHTT neuropil aggregates and nuclear inclusions, cardinal HD pathological hallmarks (52), which were accurately recapitulated in BAC226Q mice (46). By 4 months of age, BAC226Q mice exhibit clear cytoplasmic and nuclear mHTT aggregation in striatum and mainly cytoplasmic aggregation in cortex. As pathology progresses to 7 months, mHTT nuclear inclusions become more prominent (Fig. 2, E to L; and fig. S3, A and B). In 4-month-old BAC226Q mice, AAV-SaCas9-HTTg1 reduced total mHTT aggregation signal in striatal and cortical neurons by 90 and 55%, respectively (Fig. 2, E to H). At 7 months, the rescue effects remained robust, with total mHTT reduction of 86% in striatum and 48% in cortex (Fig. 2, I to L). We further investigated the reduction of cytoplasmic and nuclear mHTT aggregates in number and size separately and observed the same robust reduction (fig. S3, A to J). The percentage of nucleus with mHTT aggregates was both reduced in striatum and cortex (fig. S3, K to L). AAV-SaCas9-HTTg1 did not change glial fibrillary acidic protein (GFAP) immunoreactivity in BAC226Q mice (fig. S4, A and B). In conclusion, these results confirmed a sustained and remarkable reduction of mHTT protein and aggregates in vivo for lifetime, just by a single dose of AAV-SaCas9-HTTg1.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "SaCas9-HTTg1\u2013mediated human mHTT disruption in BAC226Q mice", "text": "(A) Experimental timeline and corresponding analyses. (B to D) Significant reduction of mHTT protein in 11-month-old BAC226Q striatum and cortex injected with AAV-SaCas9-HTTg1. Non\u2013Tg-SaCas9, nontransgenic mice injected with AAV-SaCas9; HD-SaCas9, BAC226Q injected with AAV-SaCas9; HD-SaCas9-HTTg1, BAC226Q injected with AAV-SaCas9-HTTg1. [(B) and (C)] Western blot and quantification of sustained decrease of striatal mHTT in HD-SaCas9-HTTg1. 1C2 antibody detected mHTT protein. \u03b2-Actin was a loading control. n\u00a0=\u00a03 for non\u2013Tg-SaCas9 and n\u00a0=\u00a04 for HD-SaCas9 and HD-SaCas9-HTTg1; *P\u00a0=\u00a00.0104. (D) Immunohistochemistry of robust mHTT decrease in striatal virus\u2013diffused areas. S830 antibody detected mHTT. AAV9-EGFP was a control. Dotted lines: AAV9 diffusion boundary. On the right are enlarged figures. Scale bars, 2000 \u03bcm. (E to L) Significant decrease of S830-positive mHTT aggregates in HD-SaCas9-HTTg1. [(E), (F), (I), and (J)] Immunofluorescent histology of striatum [(E) and (I)] and primary motor cortex [(F) and (J)] at 4 [(E) and (F)] and 7 months [(I) and (J)]; boxed areas were enlarged on the right. Cytoplasmic aggregates (arrows) and nuclear inclusion (arrowheads). S830 and 4\u2032,6-diamidino-2-phenylindole (DAPI) detected mHTT (green) and nucleus (blue). Scale bars, 50 \u03bcm. [(G), (H), (K), and (L)] Quantification of mHTT aggregation signal in striatum [(G) and (K)] and primary motor cortex [(H) and (L)] at 4 [(G) and (H)] and 7 months [(K) and (L)]. Aggregation signal was analyzed by particle analysis in ImageJ and calculated by the number of aggregates \u00d7 integrated density. n\u00a0=\u00a05 per group; 4 months: ***P\u00a0<\u00a00.0001 (non\u2013Tg-SaCas9 versus HD-SaCas9, striatum); ***P\u00a0=\u00a00.0002 (HD-SaCas9 versus HD-SaCas9-HTTg1, striatum); **P\u00a0=\u00a00.0011; *P\u00a0=\u00a00.0495; not significant (n.s.),\u00a00.7886 (striatum); and n.s.,\u00a00.1176 (cortex); 7 months: ***P\u00a0<\u00a00.0001 (striatum); n.s.,\u00a00.1211 (striatum); ***P\u00a0=\u00a00.0002 (cortex); *P\u00a0=\u00a00.0265 (non\u2013Tg-SaCas9 versus HD-SaCas9-HTTg1, cortex); and *P\u00a0=\u00a00.0402 (HD-SaCas9 versus HD-SaCas9-HTTg1, cortex). Data were shown as means\u00a0\u00b1\u00a0SEM. One-way ANOVA with Tukey\u2019s multiple comparisons test.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Robust and long-lasting rescue of motor deficits, weight loss, and life span by CRISPR-Cas9\u2013mediated m", "text": "Because AAV-SaCas9-HTTg1 effectively deleted mHTT gene and decreased mHTT protein and aggregates, we set out to investigate whether gene editing rescues cardinal HD-like phenotypes such as age-dependent motor deficits, weight loss, and shortened life span in BAC226Q mice (46). BAC226Q mice exhibit normal motor function until 2 to 3 months of age, followed by severe hyperactivity and chorea-like movement from 3 to 4 months and a subsequent reduction in motor activity from 6 to 7 months. These deficits and rescue effects of AAV-SaCas9-HTTg1 were observed in home cage behavior (movie S1) and analyzed by catwalk, rotarod, and open-field tasks. In catwalk task, BAC226Q mice developed progressive gait irregularities reflected by scrambled step sequences and footprint area reductions (Fig. 3, A to E). Severe scrambled step sequences at 6 months were fully rescued by AAV-SaCas9-HTTg1 (Fig. 3, A and B, and movie S2). Decreased footprint areas at 2.5, 4, and 6 months in BAC226Q mice were significantly ameliorated by AAV-SaCas9-HTTg1 (Fig. 3, C to E). In rotarod task, BAC226Q mice exhibited strong deficits in motor coordination and balance measured by latency to fall, which were significantly improved in HD-SaCas9-HTTg1 mice (Fig. 3F). In open-field test, nontransgenic control mice traveled across the cage, whereas BAC226Q mice circled in periphery. This deficit was reversed in HD-SaCas9-HTTg1 mice (Fig. 3G). Furthermore, markedly increased total travel distance in BAC226Q mice was attenuated by about 30 to 35% (Fig. 3H).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Robust and long-lasting rescue of motor deficits, weight loss, and life span by CRISPR-Cas9\u2013mediated m", "text": "(A to E) Improved gait performance in HD-SaCas9-HTTg1 analyzed by catwalk. (A) Representative footprints of non\u2013Tg-SaCas9 (top), HD-SaCas9 (middle), and HD-SaCas9-HTTg1 (bottom) at 6 months. (B) Significant increase of step sequence regularity index in 6-month HD-SaCas9-HTTg1. n\u00a0=\u00a010 per group; *P\u00a0=\u00a00.0309 (HD-SaCas9 versus HD-SaCas9-HTTg1). [(C) to (E)] Footprint area quantification. (C) Similar baselines at 2.5 months. [(D) and (E)] Deficits in HD-SaCas9 were significantly rescued in HD-SaCas9-HTTg1 at 4 months (*P\u00a0=\u00a00.0231, RH) and 6 months (**P\u00a0=\u00a00.0010, RH; ***P\u00a0<\u00a00.001, LH). n\u00a0=\u00a09 or 10 per group. RF, right front; RH, right hind; LF, left front; LH, left hind. (F) Improved coordination and balance ability in rotarod test. n\u00a0=\u00a09 or 10 per group; HD-SaCas9-HTTg1 versus HD-SaCas9: *P\u00a0=\u00a00.0202 (12 weeks), **P\u00a0=\u00a00.0126 (13 weeks), **P\u00a0=\u00a00.0035 (14 weeks), and ***P\u00a0<\u00a00.001 (15 and 16 weeks). (G and H) Rescue of locomotor performance in HD-SaCas9-HTTg1 in open-field test. (G) Representative heatmaps exhibiting cumulative time and location for 6-month-old non\u2013Tg-SaCas9 (left), HD-SaCas9 (middle), and HD-SaCas9-HTTg1 (right). (H) Total travel distance at 2.5, 4, and 6 months. n\u00a0=\u00a09 or 10 per group; HD-SaCas9-HTTg1 versus HD-SaCas9: *P\u00a0=\u00a00.0109 (4 months) and *P\u00a0=\u00a00.0456 (6 months). (I) Weight loss attenuation in HD-SaCas9-HTTg1. Non\u2013Tg-SaCas9 (n\u00a0=\u00a027), HD-SaCas9 (n\u00a0=\u00a026), and BAC226Q-HTTg1 (n\u00a0=\u00a026); *P\u00a0<\u00a00.05 and **P\u00a0<\u00a00.01. (J) Complete rescue of survival time by gene editing. The survival of HD-SaCas9-HTTg1 was significantly improved compared with HD-SaCas9 and reached to a level consistent with that of non\u2013Tg-SaCas9. Non\u2013Tg-SaCas9 (n\u00a0=\u00a017), HD-SaCas9 (n\u00a0=\u00a017), and HD-SaCas9-HTTg1 (n\u00a0=\u00a021); **P\u00a0=\u00a00.0061 (HD-SaCas9-HTTg1 versus HD-SaCas9), P\u00a0=\u00a00.7434 (HD-SaCas9-HTTg1 versus non\u2013Tg-SaCas9), and **P\u00a0=\u00a00.0025 (HD-SaCas9 versus non\u2013Tg-SaCas9). Data were shown as means\u00a0\u00b1\u00a0SEM. One-way ANOVA with Tukey\u2019s multiple comparisons test [(B) and (H)], two-way ANOVA with Tukey\u2019s multiple comparisons test [(C) to (F) and (I)], and log-rank (Mantel-Cox) test (J).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Robust and long-lasting rescue of motor deficits, weight loss, and life span by CRISPR-Cas9\u2013mediated m", "text": "BAC226Q mouse model recapitulates HD-like age-dependent progressive weight loss and shortened life span as in patients with HD (46). AAV-SaCas9-HTTg1 attenuated body weight loss by 5 to 22% (Fig. 3I) and promoted life span markedly (Fig. 3J). Together, significant and long-term rescue of motor deficits, weight loss, and shortened life span were obtained if gene editing eliminated mHTT in 1-month-old BAC226Q mice before any pathology and motor deficits were observed.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Rescue effects of gene editing at different pathogenic time points", "text": "An important question relevant to clinical treatment is at which disease stages gene editing is still effective (53). BAC226Q mice accurately recapitulated the age-dependent progression from normal development to phenotype onset and worsening and severe end stages. Therefore, we investigated the rescue effects of gene editing in BAC226Q at 1, 4, and 7 months of age, which are before, at, and well after the onset of pathological and behavioral abnormalities (Fig. 4A).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Rescue effects of gene editing at different pathogenic time points", "text": "(A) Experimental timeline in HD mice and corresponding phenotypes at different time points. (B to I) Significant reduction of mEM48-positive mHTT aggregates (green) in the striatum from 4-month-injected [(B) to (E)] and 7-month-injected HD mice [(F) to (I)]. Mouse brain samples were collected 4 months postinjection. [(B) and (F)] Representative figures showing that mHTT aggregation was decreased in 4-month-injected HD-SaCas9-HTTg1 (B) and 7-month-injected HD-SaCas9-HTTg1 (F). mHTT (green) and SaCas9 (red) were detected by mEM48 and anti-HA antibody, respectively. Scale bars, 100 \u03bcm. Enlarged figures were shown on the right side. Scale bars, 20 \u03bcm. [(C) to (E) and (G) to (I)] Quantification of mHTT aggregation signal [(C) and (G)], aggregate number [(D) and (H)], and aggregate size [(E) and (I)] in the striatum in 4-month-injected HD-SaCas9-HTTg1 mice [(C) to (E)] and 7-month-injected HD-SaCas9-HTTg1 mice [(G) to (I)]. n\u00a0=\u00a03 to 6 per group; ***P\u00a0<\u00a00.0001 [(C) to (E)], **P\u00a0=\u00a00.0031 (D), ***P =\u00a00.0006 [(G), HD-no inj-7M versus HD-SaCas9-7M inj-11M], ***P\u00a0<\u00a00.0001 [(G), HD-SaCas9-7M inj-11M versus HD-SaCas9-HTTg1-7M inj-11M], ***P =\u00a00.0007 (H), **P\u00a0=\u00a00.0082 (I), and *P\u00a0=\u00a00.0363 (I). Data were shown as means\u00a0\u00b1\u00a0SEM. One-way ANOVA with Tukey\u2019s multiple comparisons test. n.s., not significant.\n\nGene editing at the first time point of 1 month in prephenotypic mice yielded strong benefits as shown above (Figs. 2 and 3 and fig. S2 and S3).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Rescue effects of gene editing at different pathogenic time points", "text": "The second time point of gene editing was at 4 months when BAC226Q mice start to exhibit notable mHTT aggregation and motor behavior abnormalities. We injected AAV-SaCas9-HTTg1 at 4 months and examined the rescue effects of brain pathology and behavior throughout the life span of mice. At 8 months, AAV-SaCas9-HTTg1 reduced overall mHTT aggregation signal by 91% (Fig. 4, B and C). Further detailed analyses revealed that the reduction of aggregate number and size were 80 and 62%, respectively (Fig. 4, D and E). The marked 10-fold increase of aggregations from 4 to 8 months in BAC226Q was completely abolished by AAV-SaCas9-HTTg1 (Fig. 4, B to E). The similar mHTT aggregation reduction was maintained at 11 months (fig. S5, A and B). In motor behavioral tests, gait irregularities at 6 months (fig. S6, A and B, and movie S3) and locomotor activity at 8 months (fig. S6, C and D) were both rescued, although slightly less effectively than that by 1-month injection.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Rescue effects of gene editing at different pathogenic time points", "text": "The third time point of gene editing was at 7 months when BAC226Q mice show severe motor deficits and pathology such as bigger and more predominant mHTT nuclear inclusions. This intervention still robustly reduced overall mHTT aggregation signal by 74%, aggregation number by 60%, and aggregation size by 32% (Fig. 4, F to I). Similar to the 4-month injection data, AAV-SaCas9-HTTg1 completely abolished the onefold increase of aggregations from 7 to 11 months in BAC226Q (Fig. 4, F to I; and fig. S5, C and D). Consistent with the robust rescue in pathology, BAC226Q postinjection survival time was greatly increased (fig. S6E). In open-field test, locomotor abnormality was not rescued (fig. S6F). Gait irregularities measured by catwalk test could not be performed due to severe motor deficits. It is important that these data collectively demonstrated that gene editing at different pathogenic time points resulted in significant mHTT reduction and recovery of life span in all cases and rescue of motor deficits in early injections.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Highly effective and sequential deletion of mHTT and Cas9 by self-inactivating CRISPR-Cas9 rescued deficits in motor function and life span", "text": "To address the clinically relevant and general concern that conventional long-term Cas9 expression in patients, potentially for decades, may cause cumulative off-target effects and acquired immune responses against Cas9 harboring cells, it will be greatly advantageous to develop a strategy that eliminates both the target gene and subsequently Cas9 itself. Here, we report such a self-inactivating system that carries AAV-SaCas9-HTTg1 and an additional AAV-sgRNA to target Cas9 (gCas9), so that Cas9 sequence will be destroyed after HTT disruption (Fig. 5, A and B).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Highly effective and sequential deletion of mHTT and Cas9 by self-inactivating CRISPR-Cas9 rescued deficits in motor function and life span", "text": "(A) Schematic representation of dual vectors for self-inactivating system. In AAV-SaCas9-HTTg1 vector, U6 drives HTTg1 to target human HTT and CMV drives SaCas9 gene. In AAV-EGFP-gCas9 vector, U6 drives gCas9 to target SaCas9 and CMV drives EGFP as a reporter. (B) Intraparenchymal delivery of dual AAV vectors into 1-month-old mouse striatum and primary motor cortex. (C to N) Self-inactivating AAV-SaCas9-HTTg1/gCas9 robustly decreased SaCas9 and reduced mHTT aggregation as effectively as the conventional AAV-SaCas9-HTTg1. Representative immunofluorescence images of mEM48-labeled mHTT aggregation and HA-labeled SaCas9 in striatum of BAC226Q at 4 months (C), 7 months (D), and 11 months (E). Scale bars, 50 \u03bcm. [(F) and (G)] Quantification of SaCas9 fluorescence intensity (F) and mHTT aggregation number (G) in (C). n\u00a0=\u00a03 per group; SaCas9: ***P\u00a0<\u00a00.0001; mHTT: ***P\u00a0<\u00a00.0001 (HD-SaCas9 versus HD-SaCas9-HTTg1); ***P\u00a0<\u00a00.0001 (HD-SaCas9 versus HD-SaCas9-HTTg1/gCas9); and not significant (n.s.),\u00a00.9679. [(H) and (I)] Quantification of SaCas9 fluorescence intensity (H) and mHTT aggregation number (I) in (D). n\u00a0=\u00a03 to 5 per group; SaCas9: ***P\u00a0<\u00a00.0001; mHTT: ***P\u00a0<\u00a00.0001 (HD-SaCas9 versus HD-SaCas9-HTTg1); ***P\u00a0<\u00a00.0001 (HD-SaCas9 versus HD-SaCas9-HTTg1/gCas9); and n.s.,\u00a00.9987. [(J) and (K)] Quantification of SaCas9 fluorescence intensity (J) and mHTT aggregation number (K) in (E). n\u00a0=\u00a03 per group; SaCas9: ***P\u00a0=\u00a00.0007; mHTT: ***P\u00a0=\u00a00.0002 (HD-SaCas9 versus HD-SaCas9-HTTg1); ***P\u00a0<\u00a00.0002 (HD-SaCas9 versus HD-SaCas9-HTTg1/gCas9); n.s.,\u00a00.9096. (L) Immunohistochemistry of HA-labeled SaCas9 (top) and S830-labeled mHTT (bottom) in 7-month-old BAC226Q. Boxed areas were enlarged on the right. Scale bar, 2000 \u03bcm. (M) Quantification of SaCas9 reduction in (L). n\u00a0=\u00a04 per group; ***P\u00a0<\u00a00.0001. IOD, integrated optical density. (N) Quantification of striatal mHTT aggregation number in boxed areas of (L). n\u00a0=\u00a03 or 4 per group; ***P\u00a0<\u00a00.0001 (HD-SaCas9 versus HD-SaCas9-HTTg1); ***P\u00a0<\u00a00.0001 (HD-SaCas9 versus HD-SaCas9-HTTg1/gCas9); and n.s.,\u00a00.5473. Data were shown as means\u00a0\u00b1\u00a0SEM. One-way ANOVA with Tukey\u2019s multiple comparisons test [(F) to (K) and (N)] and unpaired two-tailed t test (M).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Highly effective and sequential deletion of mHTT and Cas9 by self-inactivating CRISPR-Cas9 rescued deficits in motor function and life span", "text": "To select the most effective gCas9, we used HEK293T cells expressing EGFP-hHTT-20Q/120Q to test 10 gCas9 against different SaCas9 coding sequences (fig. S7A and table S1). The results showed that gCas9-1 exhibited the highest efficiency in eliminating Cas9 without impairing mHTT reduction (fig. S7, B to L) and thus was used for subsequent in vivo experiments.\n\nWe then tested in BAC226Q mice. Similar to the conventional gene editing, the self-inactivating AAV-SaCas9-HTTg1/gCas9 effectively reduced mHTT by 94% at 4 months, 92% at 7 months, and 92% at 11 months (Fig. 5, C to N; and fig. S8, A to F). Meanwhile, AAV-SaCas9-HTTg1/gCas9 reduced above 90% Cas9 by 1 week postinjection (fig. S8, G and H) and lasted throughout life span (Fig. 5, C to N; and fig. S8, A to F). In mice injected with AAV9-SaCas9-HTTg1/AAV9-EGFP-gCas9, 93% of green fluorescent protein (GFP)\u2013positive cells expressed neither SaCas9 nor mHTT (fig. S9, A and B). These data demonstrated that the self-inactivating system can efficiently eliminate SaCas9 without affecting the clearance of mHTT and that Cas9-mediated gene editing events in neurons happen very rapidly well within 1 week postinjection, after which Cas9 is no longer needed.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Highly effective and sequential deletion of mHTT and Cas9 by self-inactivating CRISPR-Cas9 rescued deficits in motor function and life span", "text": "Further analyses revealed that AAV-SaCas9-HTTg1/gCas9\u2013 and AAV-SaCas9-HTTg1\u2013injected mice showed equivalent rescue of step sequence irregularity (Fig. 6, A and B, and movie S4), footprint area reduction (Fig. 6C) in catwalk task, locomotor hyperactivity in open-field test (Fig. 6D), latency to fall in rotarod test (Fig. 6E), and shortened life span (Fig. 6F).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Highly effective and sequential deletion of mHTT and Cas9 by self-inactivating CRISPR-Cas9 rescued deficits in motor function and life span", "text": "(A to C) Self-inactivating AAV-SaCas9-HTTg1/gCas9 improved gait performance in catwalk analysis of BAC226Q mice at 6 months. (A) Representative footprints of non\u2013Tg-SaCas9, HD-SaCas9, HD-SaCas9-HTTg1, and HD injected with AAV-SaCas9-HTTg1/gCas9 (HD-SaCas9-HTTg1/gCas9) from top to bottom. (B) Increased step sequence regularity index in HD-SaCas9-HTTg1/gCas9 and HD-SaCas9-HTTg1. n\u00a0=\u00a010 per group; *P\u00a0=\u00a00.0387 (HD-SaCas9 versus HD-SaCas9-HTTg1); *P\u00a0=\u00a00.0164 (HD-SaCas9 versus HD-SaCas9-HTTg1/gCas9); and not significant (n.s.),\u00a00.9844. (C) Quantification of footprint area in gait analysis. n\u00a0=\u00a010 per group; ***P\u00a0=\u00a00.0007 (LH); **P\u00a0=\u00a00.0010 (RH); **P\u00a0=\u00a00.0057 (LH); *P\u00a0=\u00a00.0136 (RH); n.s.,\u00a00.8595 (RH); and n.s.,\u00a00.9319 (LH). RF, right front; RH, right hind; LF, left front; LH, left hind. (D) Rescue of locomotor activity in HD-SaCas9-HTTg1/gCas9 and HD-SaCas9-HTTg1 in open-field test. n\u00a0=\u00a015 per group; 4 months: ***P\u00a0<\u00a00.0001 (HD-SaCas9 versus HD-SaCas9-HTTg1); ***P\u00a0<\u00a00.0001 (HD-SaCas9 versus HD-SaCas9-HTTg1/gCas9); and n.s.,\u00a00.8905; 7 months: ***P\u00a0=\u00a00.0003 (HD-SaCas9 versus HD-SaCas9-HTTg1); ***P\u00a0<\u00a00.0001 (HD-SaCas9 versus HD-SaCas9-HTTg1/gCas9); and n.s.,\u00a00.9207; 11 months: n.s.,\u00a00.1145 (HD-SaCas9 versus HD-SaCas9-HTTg1); n.s.,\u00a00.0746 (HD-SaCas9 versus HD-SaCas9-HTTg1/gCas9); and n.s.,\u00a00.9980 (HD-SaCas9-HTTg1 versus HD-SaCas9-HTTg1/gCas9). (E) Improvement of coordination and balance ability in HD-SaCas9-HTTg1/gCas9 and HD-SaCas9-HTTg1 in rotarod test. n\u00a0=\u00a015 per group; HD-SaCas9 versus HD-SaCas9-HTTg1/gCas9: **P\u00a0=\u00a00.0065 (15 weeks), ***P =\u00a00.0006 (16 weeks), and ***P\u00a0<\u00a00.0001 (17 to 20 weeks); HD-SaCas9 versus HD-SaCas9-HTTg1/gCas9: ###P\u00a0<\u00a00.0001 (15 to 20 weeks). (F) Compared to HD-SaCas9, the survival time of HD-SaCas9-HTTg1/gCas9 was significantly extended and reached to a level consistent with that of HD-SaCas9-HTTg1. Non\u2013Tg-SaCas9 (n\u00a0=\u00a08), HD-SaCas9 (n\u00a0=\u00a010), HD-SaCas9-HTTg1 (n\u00a0=\u00a010), and HD-SaCas9-HTTg1/gCas9 (n\u00a0=\u00a09); ***P\u00a0<\u00a00.0001 (HD-SaCas9 versus HD-SaCas9-HTTg1), ***P\u00a0<\u00a00.0001 (HD-SaCas9 versus HD-SaCas9-HTTg1/gCas9), and P\u00a0=\u00a00.7174 (HD-SaCas9-HTTg1 versus HTTg1/gCas9). Data were shown as means\u00a0\u00b1\u00a0SEM. One-way ANOVA with Tukey\u2019s multiple comparisons test (B), two-way ANOVA with Tukey\u2019s multiple comparisons test [(C) to (E)], and log-rank (Mantel-Cox) test (F).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Highly effective and sequential deletion of mHTT and Cas9 by self-inactivating CRISPR-Cas9 rescued deficits in motor function and life span", "text": "Transcriptomic dysregulation is a hallmark in patients with HD and animal models (54). We investigated transcriptomes and found that, compared to nontransgenic mice, BAC226Q had signature changes that were partially reversed by conventional and self-inactivating gene editing (fig. S10A). The distribution of differentially expressed genes (DEGs) across groups was also visualized in the volcano plot (fig. S10, B to D, and table S2). In comparison to the differential gene expression observed between BAC226Q and non-Tg mice, the transcriptional differences between BAC226Q-HTTg1 or BAC226Q-HTTg1/gCas9 and non-Tg were reduced, more prominent for the down-regulated genes. We further analyzed how the 245 up-regulated and 452 down-regulated genes in BAC226Q compared to non-Tg changed in the BAC226Q-HTTg1 and BAC226Q-HTTg1/gCas9 groups. The results indicated that the gene editing therapy could partially reverse the dysregulated genes observed in HD mice (fig. S10, E and F). Gene Ontology enrichment analysis was also performed on the DEGs in the BAC226Q-SaCas9 control, BAC226Q-SaCas9-HTTg1, and BAC226Q-SaCas9-HTTg1/gCas9 groups in comparison with those in non\u2013Tg-SaCas9 mice. Both the conventional CRISPR-Cas9 and the self-inactivating systems showed comparable efficacies in ameliorating transcriptomic-level dysregulation. The number and expression levels of genes involved in important pathways were robustly improved in gene-edited mice. Noticeably, most prominent changes were enriched in genes for postsynaptic structure and function, ion channel activity, and transmembrane transporters (fig. S10G).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Highly effective and sequential deletion of mHTT and Cas9 by self-inactivating CRISPR-Cas9 rescued deficits in motor function and life span", "text": "Therefore, we succeeded in developing a self-inactivating system that eliminated mHTT gene and protein, rescued deficits in motor functions, and life span with the same high efficiency as the conventional CRISPR-Cas9 and that abolished Cas9 after a transient expression. This system could be highly beneficial for clinical applications.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "On and off-target effects of conventional and self-inactivating CRISPR-Cas9", "text": "BAC226Q contains human mHTT transgene and endogenous mouse Htt gene. To investigate the long-term cleavage effects on human mHTT gene, mouse brain genomic DNA was extracted at 11 months. Compared with control BAC226Q mice, the human mHTT gene sequence in BAC226Q-SaCas9-HTTg1 had been edited exactly from 3 base pairs (bp) upstream of PAM (Fig. 7A), consistent with previous findings (55). To compare the specificity of self-inactivating and conventional CRISPR systems, we applied primer extension\u2013mediated sequencing (PEM-seq), a method for detecting editing efficiency and off-target hotspots simultaneously because it can identify CRISPR-Cas9 editing events including indels, large deletions, as well as genome-wide translocations with high sensitivity (56). In human HTT on-target site, the editing events were predominantly small indels, about 90% of which resulted in frameshift and premature stop codon (Fig. 7, B and C). In both strategies, no off-target events were found except for the predictable mouse Htt gene, which shares 20/21 sequence homology with human HTT gene target site (Fig. 7D). Even with 20/21 sequence homology, a 95% identical base pairs, HTTg1 only yielded off-target translocations with very low frequencies of 0.012\u00a0\u00b1\u00a00.001% in 2-month-old and 0.017\u00a0\u00b1\u00a00.004% in 12-month-old BAC226Q mouse brain striatum. HTTg1/gCas9 further significantly reduced HTTg1-induced off-target effects in 12-month-old mouse by 2.5-fold (Fig. 7E). For more clinical relevance, the off-target effect for HTTg1 was also investigated in human cell lines. In human induced pluripotent stem cells (iPSCs), neither HTTg1 nor HTTg1/gCas9 induced any off-target mismatches (Fig. 7F). In HEK293T cells, HTTg1 resulted only one off-target editing in chromosome 1 22,620,422-22,620,442 locus. The self-inactivating HTTg1/gCas9 completely eliminated all off-target editing (Fig. 7, G and H). It is worth noting that gCas9 per se did not induce any off-target effect in mouse brain and human cell line. Together, these data confirmed the successful editing by SaCas9-HTTg1 and SaCas9-HTTg1/gCas9 on human HTT gene and suggested that self-inactivating CRISPR-Cas9 may eliminate off-target effects and thus provide maximal safety for long-term therapeutic use.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "On and off-target effects of conventional and self-inactivating CRISPR-Cas9", "text": "(A) Sanger sequencing of human HTT exon 1 amplicons in control HD (top) and HD-SaCas9-HTTg1 mouse (bottom). The genomic DNA editing by HTTg1 was demonstrated by the scrambled nucleotide signals in the bottom panel. (B to H) Gene editing specificity detected by PEM-seq. [(B) and (C)] Indel patterns (B) and frequency (C) on human HTT in 12-month-old BAC226Q. Blue, guide RNA target site; purple, PAM sequence; red, insertions; blue dashed line, deletions. n\u00a0=\u00a09. (D) Circos plot for off-target site in BAC226Q. Arrows indicate the on-target site of human HTT gene and the only off-target site of mouse Htt gene. The curved line connects the on-target and off-target site. Blue, guide RNA target site; purple, PAM sequence; red, mismatch. (E) Low frequencies calculated as off-target translocations over total editing events in the striatum of 2- and 12-month-old BAC226Q. n\u00a0=\u00a08 for HD-SaCas9-HTTg1 (2 months), n\u00a0=\u00a05 for HD-SaCas9-HTTg1/gCas9 (2 and 12 months), and n\u00a0=\u00a06 for HD-SaCas9-HTTg1 (12 months); *P\u00a0=\u00a00.0135. (F) No off-target mismatch calculated as off-target translocations over total editing events in human iPSCs. n\u00a0=\u00a03. [(G) and (H)] Off-target frequencies calculated as off-target translocations over total editing events (G) and circos plot (H) in HEK293T cells. Arrows indicate the on-target site of human HTT gene and off-target site on chromosome 1. The curved line connects the on-target and off-target site. Blue, guide RNA target site; purple, PAM sequence; red, mismatch. n\u00a0=\u00a03 for SaCas9 and HTTg1/gCas9 and n\u00a0=\u00a04 for HTTg1. Data were shown as means\u00a0\u00b1\u00a0SEM. Two-way ANOVA with Bonferroni\u2019s multiple comparisons test (E) and one-way ANOVA with Tukey\u2019s multiple comparisons test (G). n.s., not significant.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "DISCUSSION", "text": "The revolutionary gene editing tools provide the possibility of root-cause cure for genetic disorders, especially monogenic and gain-of-toxic-function diseases, by eliminating or correcting mutated genes. To achieve clinical success, several aspects have to be satisfied, including high efficiency gene editing, effective delivery of gene editing tools to the neurons in pertinent brain structures, selection of intervention time points, and elimination of off-target effects. Our study strives to reach these goals with immediate clinical relevance.\n\nHighly efficient gene editing is pivotal for disease rescue. In this study, we developed conventional and self-inactivating CRISPR gene editing systems, both of which successfully eliminated human mHTT gene, mHTT protein, and aggregation in BAC226Q HD mouse brain. This was achieved by injection coverage of about 90% of striatal and 60% of M1 cortical area. In the injected areas, AAV9 transduced above 70% of neurons, among which 95% went through successful gene editing, leading to 60 to 90% mHTT protein and 90% of mHTT aggregation reduction. This mHTT suppression rescued deficient phenotypes in HD mouse model.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "DISCUSSION", "text": "In patients with HD and animal models, there is an age-dependent and progressive mHTT accumulation of cytoplasmic aggregates and nuclear inclusions (52). This is well recapitulated in BAC226Q mice. In our early injection experiment, cytoplasmic aggregates were greatly reduced, and nuclear aggregates were almost eliminated. In later injection, the reduction of mHTT is still significant, although to a lesser degree. Injection at 4 months when BAC226Q start to show obvious cytoplasmic and nuclear mHTT aggregation, the deletion of mHTT gene not only prevented its expression but also allowed neurons to gradually degrade the existing mHTT aggregates.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "DISCUSSION", "text": "The primary neuropathology of HD occurs in the striatum and cerebral cortex (57), suggesting that we must take striatum, cortex, and their interaction into account when we develop effective therapeutic strategy through directly targeting HTT (58). Our results illustrated that striatal and cortical disruption of human mutant HTT expression in BAC226Q mice by CRISPR-Cas9 was sufficient to slow the disease progression at multiple disease time points. This result is also the basis for choosing direct local injection into striatum and cortex as the method of delivery. The direct injection method circumvents issues such as considerable dilution of AAV in intravenous or intrathecal delivery (59), challenges in crossing the blood-brain barrier (60), transduction of brain areas irrelevant to HD pathogenesis, high AAV dosage induced toxicities (61, 62), immune responses (63), and off-target effects.\n\nIn neurodegenerative diseases including HD, glia-neuron interaction has been actively studied (64\u201367). In our study, AAV9 was stereotactically injected into mouse striatum and cortex and transduced predominantly neurons and much less glia. We found that CRISPR-Cas9\u2013mediated HTT gene deletion in neuronal cells was adequate to rescue most mutant phenotypes in BAC226Q mouse.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "DISCUSSION", "text": "Because patients with HD can be in presymptomatic, onset, or symptomatic stages, the proper timing of gene editing intervention is important. Ideally, earlier mHTT gene disruption would yield better rescue benefits. However, how late can gene editing be performed and still be effective? Herein, we disrupted mHTT expression in BAC226Q mice at different time points: 1-month-old BAC226Q mice before phenotypical onset, 4-month-old BAC226Q mice of initial phenotypical onset of movement abnormalities, and 7-month-old BAC226Q mice with severe mutant phenotypes. We found that, as expected, earlier gene editing intervention has stronger therapeutic effects. However, it is exciting that gene editing at much later stages can still significantly extend their survival time. This systematic investigation on HTT gene disruption time points provides useful information on potential initiation time of treatments. Now, HD clinical trials have included early symptomatic patients, and there is an ongoing discussion whether treatments can start at stage 1 when there are biomarker changes, but not overt motor symptoms (68).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "DISCUSSION", "text": "In all clinical interventions including that for HD, a serious concern is that prolonged expression of SaCas9 may induce immune response and off-target effects (69\u201373). To address these concerns, we designed the self-inactivating CRISPR-Cas9 system where another sgRNA was introduced to delete SaCas9 in a short period of time. Comparable to the conventional CRISPR-Cas9, this self-inactivating system effectively deleted HTT gene, markedly decreased mutant HTT aggregation, and successfully ameliorated deficits in BAC226Q mice. Meanwhile, the self-inactivating system diminished the very low off-target frequency in conventional CRISPR. It is worth noting that the self-inactivating CRISPR-Cas9 could be a universal tool for maximal safety in clinical treatments.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "DISCUSSION", "text": "In summary, our study demonstrated a clinically relevant gene editing strategy to efficiently ameliorate neuropathological phenotypes and behavioral deficits in a human genomic transgenic mouse model of HD. We illustrated that both conventional and self-inactivating CRISPR-Cas9 can efficiently and permanently delete mutant HTT gene in vitro and in vivo. This significantly reduced pathogenic protein aggregation, attenuated transcriptional dysregulation, improved motor performance, alleviated weight loss, and extended the life span of HD mice. It should be noted that these robust rescue effects were achieved by eliminating human mHTT with the present of wild-type mouse Htt. In clinic setting, our strategy would knock down both the mutant and wild-type HTT. As explained in Introduction, substantial reductions of wild-type HTT were well tolerated in clinical trials and animal experiments. However, for clinical trial readiness and confidence, we are testing our system to eliminate wild-type HTT in cynomolgus macaque striatum and cortex for safety, gene editing efficiency, and AAV dosages. We hope that our CRISPR-Cas9\u2013mediated gene editing via AAV delivery is feasible and safe for HD therapy, and our self-inactivating gene editing strategy may contribute to the therapeutic development for other diseases.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "MATERIALS AND METHODS", "text": "The aim of this study was to develop a CRISPR-based gene editing as a therapy for HD with maximal safety. This study investigated the long-term rescue and off-target effects in a robust HD mouse model BAC226Q at different time points of intervention. Given the high HTT editing efficiency of CRISPR in HEK293T cell line, we therefore packaged our editing system in AAV9 and delivered it to the striatum and cortex of BAC226Q HD mice, evaluating the in vivo editing effects at different postinjection time points by sequencing, Western blot, immunostaining, and behavioral tests. The dosage of AAV vectors was determined on the basis of published papers and our pilot studies. The cell line used in this study was indicated in the place of related results shown in the text, figure legends, and Materials and Methods. Animal size was determined on the basis of the previous studies in BAC226Q mouse. Experimental end points were determined by cardinal parameters of age-dependent pathogenesis in BAC226Q. Both sexes were used in Western blot and staining studies, and only male mice were used in behavioral and sequencing studies. Animals were randomly assigned to the control and experimental groups, and researchers were blinded to genotype and treatment information during the experiment but not during analysis. Details on the use of cell line and animal in different experiments could be found in the corresponding sections of Materials and Methods. Biological replicates were indicated by n values in the figure legends. No data were excluded.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "MATERIALS AND METHODS", "text": "A single-vector AAV-SaCas9 system containing Cas9 from Staphylococcus aureus (SaCas9) and its sgRNA scaffold was obtained from Addgene (plasmid no. 61591). sgRNAs targeting human HTT and SaCas9 were designed on the basis of PAM sequence (5\u2032-NNGRRT-3\u2032) and by CRISPR RGEN Tools (www.rgenome.net/). sgRNA sequences targeting human HTT are as follows: HTTg1: 5\u2032-TGG AAA AGC TGA TGA AGG CCT-3\u2032; and HTTg2: 5\u2032-GAA AAG CTG ATG AAG GCC TTC-3\u2032. AAV-EGFP-gCas9 or AAV-mCherry-gCas9, which contain EGFP or mCherry, were constructed by replacing SaCas9 on vector AAV-SaCas9. sgRNAs targeting SaCas9 are listed in table S1.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "MATERIALS AND METHODS", "text": "Cas9 was replaced by EGFP or mCherry in the AAV vector backbone via Age I and Eco RI enzyme digestion to generate AAV-EGFP and AAV-mCherry vectors, respectively. Top and bottom strands of oligos for each sgRNA were phosphorylated and annealed into duplex. HTTg1 and HTTg2 were inserted into AAV-SaCas9 vector via Bsa I enzyme digestion. Similarly, gCas9 was cloned into the AAV-EGFP and AAV-mCherry vector through Bsa I digestion. The ligation products were transformed into Stbl3 chemically competent cells (TransGen, CD521), and plasmid DNA was extracted from cultures using the QIAprep spin miniprep kit (QIAGEN, 27104) according to the manufacturer\u2019s instructions. sgRNA sequences were verified by Sanger sequencing using the LKO.1 5\u2032 primer: 5\u2032-GAC TAT CAT ATG CTT ACC GT-3\u2032. EGFP and mCherry sequences were verified using the EGFP-forward primer: 5\u2032-CGA AGG CTA CGT CCA GGA GC-3\u2032 and the mCherry-forward primer: 5\u2032-ACA ACC GGT ATG GTG AGC AAG-3\u2032, respectively. AAV plasmids were then extracted by EndoFree plasmid maxi kit (QIAGEN, 12362) for subsequent virus production and purification.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "MATERIALS AND METHODS", "text": "HEK293T cells (American Type Culture Collection, CRL-1573) were cultured in Dulbecco\u2019s modified Eagle\u2019s medium (DMEM; Cytiva, SH3024301) supplemented with 10% (v/v) fetal bovine serum (Thermo Fisher Scientific), 1% (v/v) penicillin-streptomycin (Gibco, 21103049), and 1% (v/v) l-glutamine (Gibco, 25030081) in a humidified 5% CO2 atmosphere at 37\u00b0C. Cells were transfected with 2.5 \u03bcg of AAV-SaCas9-sgRNA vectors and 2.5 \u03bcg of pEGFPc3-human HTT exon 1-20 CAG/pEGFPc3-human HTT exon 1-120 CAG (EGFP-hHTT-20Q/120Q) at 70% confluency with Lipofectamine 2000 (Invitrogen, 11668030); in self-inactivating system, cells were cotransfected with 0.8 \u03bcg of AAV-SaCas9-HTTg1/AAV-SaCas9, 0.8 \u03bcg of AAV-mCherry-gCas9/AAV-mCherry, and 0.8 \u03bcg of EGFP-hHTT-20Q/120Q. For in vitro validation of cleavage effects, EGFP fluorescence was imaged under microscope (Carl Zeiss, Vert.A1) 24 or 48 hours after transfection, and, then, protein was harvested for Western blot. For PEM-seq assay, cells were transfected with 5 \u03bcg of AAV-SaCas9-HTTg1/AAV-SaCas9 and 5 \u03bcg of AAV-mCherry-gCas9/AAV-mCherry, and, then, mCherry-positive cells were sorted by flow cytometry (Beckman Coulter, Astrios EQ) 24 hours after transfection, followed by genomic DNA extraction.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "MATERIALS AND METHODS", "text": "Human chemical induced pluripotent stem cell (hCiPSC) line (#1217-2) was obtained from H. Deng at Peking University. hCiPSCs were cultured on Matrigel (Corning, 354230)\u2013coated plates in mTeSR Plus Medium (STEMCELL Technologies, 100-0276) in a humidified 5% CO2 atmosphere at 37\u00b0C. When reached to 80% confluence, cells were dissociated with ReLeSR (STEMCELL Technologies, 05872) and passaged with a 1:10 split ratio. Detached cell aggregates were first cultured in mTeSR Plus Medium supplemented with 10 \u03bcM Y-27632 and replaced by fresh mTeSR Plus Medium the next day. Culture medium was changed every 1 or 2 days. hCiPSCs were transfected by nucleofection with a Lonza P3 primary cell kit under CM115 program (Lonza). For each six-well, 2.5 \u03bcg of AAV-SaCas9-HTTg1/AAV-SaCas9 vectors and 2.5 \u03bcg of AAV-mCherry-gCas9/AAV-mCherry were transfected. For PEM-seq assay, mCherry-positive hCiPSCs were sorted by flow cytometry 48 hours after nucleofection. Cells were cultured in mTeSR Plus Medium supplemented with 10 \u03bcM Y-27632 for 24 hours after nucleofection and sorting and replaced by fresh mTeSR Plus Medium for further culturing.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "MATERIALS AND METHODS", "text": "Mice were housed in the specific pathogen\u2013free facility at Peking University in accordance with the guidelines. The room was controlled with 12-hour light/dark cycle, 40 to 60% humidity, and 22\u00b0\u00a0\u00b1\u00a01\u00b0C temperature. Food and water were always accessible. All mice used in this research were FVB/N (Taconic) background as BAC226Q mice in this background showed the most robust HD-like phenotypes. BAC226Q mice were bred with wild-type nontransgenic mice and confirmed by genotyping which produced a 403-bp polymerase chain reaction (PCR) fragment (forward primer: 5\u2032-GTA TAT GCT GCT GCC TGC AA-3\u2032; and reverse primer: 5\u2032-AGG GGA CAG TGT TGG TCA AG-3\u2032). All mice experimental procedures were approved by the Institutional Animal Care and Use Committee of Peking University (CFLS-LiCJ-1).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "MATERIALS AND METHODS", "text": "Mouse striatal tissue was dissected and immediately collected in ice-cold phosphate-buffered saline (PBS). Lysates were prepared by ultrasonication in radioimmunoprecipitation assay buffer [150 mM NaCl, 0.1% SDS, 1% sodium deoxycholate, 50 mM triethanolamine, and 1% NP-40 (pH 7.4), supplemented with Caspase inhibitor Boc-D-FMK (Abcam, ab142036) and Protease inhibitor cocktail (Sigma-Aldrich, 111 M4009)] using a sonic dismembrator (Fisherbrand, FB120), pulsed on for 1 s and pulsed off for 2 s, and repeated for 20 to 25 times at 20% amplitude on ice. The lysates were incubated at 4\u00b0C for 60 min, followed by centrifugation at 4\u00b0C for 20 min at 16,100g to remove insoluble components. Protein concentration was measured by bicinchoninic acid (BCA) protein assay using Pierce BCA Protein Assay Kit (Thermo Fisher Scientific, 23225), and the absorbance was detected at 562 nm. Protein samples were prepared for loading by heating in NuPAGE 4\u00d7\u00a0LDS sample buffer plus 10\u00d7\u00a0sample reducing agent (Invitrogen) according to the manufacturer\u2019s protocol and heated for 10 min at 70\u00b0C. A total amount of 90 to 120 \u03bcg of protein was loaded onto 4 to 12% NuPAGE bis-tris gels (Invitrogen, NP0321/NP0336) using Mops running buffer and wet transferred to an Immobilon-FL polyvinylidene difluoride (PVDF) membrane (Millipore, IPFL00010). Lysates from HEK293T cells were prepared in Triton X-100 buffer [150 mM NaCl, 50 mM tris-HCl (pH 7.4), 1 mM EDTA, and 1% Triton X-100]. The lysates were incubated at 4\u00b0C for 30 min, followed by centrifugation at 4\u00b0C for 20 min at 16,100g to remove insoluble components. Protein concentration was measured by BCA protein assay. Protein samples were prepared for loading by heating in 4\u00d7\u00a0SDS sample buffer and heated for 10 min at 95\u00b0C. A total amount of 30 \u03bcg of protein was loaded onto SDS\u2013polyacrylamide gel electrophoresis (PAGE) gels using running buffer and wet transferred to an Immobilon-FL PVDF membrane. Blots were incubated with Odyssey blocking buffer (LI-COR, 927-40000) for 60 min. After washing with TBST, blots were incubated with primary antibodies diluted in blocking buffer at 4\u00b0C overnight. Blots were washed three times (10 min each) and incubated with fluorescently labeled IRDye 680RD goat anti-mouse (1:10,000; LI-COR, 926-68070) or goat anti-rabbit (1:10,000; LI-COR, 926-68071) secondary antibodies in TBST for 60 min at room temperature. Protein was detected in the 700-nm channel using Odyssey CLx imager (LI-COR). The following primary antibodies were used: 1C2 (1:5000, mouse; Millipore, MAB1574), GFP (1:8000, rabbit; Abcam, ab183734), hemagglutinin (HA; 1:2000, mouse; Abcam, ab18181), and \u03b2-actin (1:2000, rabbit; Cell Signaling Technology, 4970).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "MATERIALS AND METHODS", "text": "Mice were anesthetized, perfused with fresh 4% paraformaldehyde in PBS, and postfixed overnight in the same fixative solution. Fixed brains were sliced at 40-\u03bcm thickness with a vibrating blade microtome (Leica, VT1200S). The brain slices were incubated in PBS supplemented with 0.3% (v/v) Triton X-100 for 30 min in room temperature and then blocked with 10% (v/v) goat serum (ZSGB-BIO, ZLI-9021) in PBS supplemented with 0.1 or 0.3% Triton X-100 for 60 min at room temperature. Following incubation of brain slices with primary antibodies at 4\u00b0C overnight and washes with PBS supplemented with 0.1% Triton X-100, fluor-conjugated secondary antibodies (anti-mouse, anti-rabbit, or anti-sheep Alexa Fluor 488, 555, 594, or 647; 1:1000; Thermo Fisher Scientific) were added to the samples for 60 min at room temperature. Sections were cover slipped by antifade mountant with Hoechst 33342 nuclear stain (Thermo Fisher Scientific, P36985). Images were taken by Zeiss LSM 780 or LSM 880 confocal microscope (Carl Zeiss), and whole-brain section images were taken by Axio Scan Z1 (Carl Zeiss). The following primary antibodies were used: mEM48 (1:1000, mouse; Millipore, MAB5374), S830 (1:5000, sheep, provided by G. Bates from University College London), HA (1:1000, rabbit; Cell Signaling Technology, 3724), HA (1:1000, mouse; Abcam, ab18181), GFAP (1:1000, rabbit; Abcam, ab7260), and NeuN (1:1000, mouse; Abcam, ab104224). For the quantification of S830-positive mHTT in Fig. 2 (E, F, I, and J), we performed the analysis within dorsal striatum, where AAV9 coverage is high. For the quantification of mEM48-positive mHTT in Figs. 4 (B and F) and 5 (C to E) and fig. S5 (A and C), we performed the analysis specifically within regions exhibiting clear HA staining. In fig. S2 (A and B), the proportion of mEM48-negative cells within the HA-positive population was quantified.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "MATERIALS AND METHODS", "text": "Mouse brain sections were prepared by the same procedure as the immunofluorescent staining samples. Sections were stained according to instructions provided with the ABC and DAB substrate kits (Vector Labs). Briefly, sections were first processed free-floating with blocking serum and primary antibodies. After treatment with biotinylated secondary antibody, followed by avidin-biotinylated horseradish peroxidase complexes and incubation with diaminobenzidine, sections were mounted onto slides. In addition, brain sections were dehydrated using the setup as 30% alcohol, 50% alcohol, 70% alcohol, 95% alcohol, 100% alcohol-I, and 100% alcohol-II (2 min each) and xylene-I and xylene-II (10 min each). Then, sections were cover slipped by DPX mounting medium (Sigma-Aldrich, 06522) and imaged under Axio Scan Z1 microscope. The following primary antibodies were used: S830 (1:25,000, sheep, provided by G. Bates from University College London), HA (1:1000, rabbit; Cell Signaling Technology, 3724), and GFAP (1:2000, rabbit; Abcam, ab7260). For the quantification of S830-positive mHTT in Fig. 5L and fig. S8 (A and B), we performed the analysis specifically within regions exhibiting clear HA staining.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "MATERIALS AND METHODS", "text": "HEK293T cells were plated on 10 15-cm dishes and transfected at 90% confluence by adding 5 ml of the plasmid and linear polyethylenimine (PEI; Sigma-Aldrich, 765090) mixture to each plate. The mixture was prepared with 70 \u03bcg of AAV9 vector (AAV-SaCas9-HTTg1, AAV-SaCas9, AAV-EGFP-gCas9, or AAV-EGFP), 200 \u03bcg of Ad-Helper plasmid, 70 \u03bcg of AAV-Rep/Cap plasmid, and PEI (1 \u03bcg/\u03bcl), and the PEI-to-DNA ratio was 5:1 (v/g) in our experiment. The mixture was filled to 50 ml with DMEM and mixed well. Cells were collected with culture medium 60 hours after transfection and centrifuged at 1000 rpm for 10 min. Cell pellet was resuspended in 5 ml of cell lysis buffer [150 mM NaCl and 20 mM tris (pH 8.0)].\n\nThe cell lysate was freeze-thawed three times by changing between lipid nitrogen and 37\u00b0C water bath to release virus from the destroyed cell membrane. Cell lysate was added with MgCl2 and Benzonase (Sigma-Aldrich, E8263-25k) to a final concentration of 1 mM and 250 U/ml, respectively, followed by incubation at 37\u00b0C for 15 min to dissolve the DNA/protein aggregation. The supernatant was collected after centrifugation of the cell lysate at 4000 rpm for 30 min at 4\u00b0C.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "MATERIALS AND METHODS", "text": "The viral solution was loaded on the iodixanol gradient solution for purification, and the gradient was prepared in the following order (from top to bottom): 6 ml of 17% iodixanol [5 ml of 10\u00d7\u00a0PBS, 0.05 ml of 1 M MgCl2, 0.125 ml of 1 M KCl, 10 ml of 5 M NaCl, 12.5 ml of OptiPrep (Sigma-Aldrich, D1556), and up to 50 ml of H2O], 6 ml of 25% iodixanol (5 ml of 10\u00d7\u00a0PBS, 0.05 ml of 1 M MgCl2, 0.125 ml of 1 M KCl, 20 ml of OptiPrep, and H2O up to 50 ml), 5 ml of 40% iodixanol (5 ml of 10\u00d7\u00a0PBS, 0.05 ml of 1 M MgCl2, 0.125 ml of 1 M KCl, 33.3 ml of OptiPrep, and up to 50 ml of H2O), and 4 ml of 60% iodixanol (0.05 ml of 1 M MgCl2, 0.125 ml of 1 M KCl, and 50 ml of OptiPrep). The sample was centrifuged at 53,000 rpm for 160 min at 14\u00b0C, and the viral fraction was harvested in the 40% layer with a syringe.\n\nThe viral fraction was mixed with PBS solution and 10% Poloxamer 188 solution (F188) (1:10,000; Sigma-Aldrich, P5556) and then run through the Amacon 100K filter (MilliporeSigma, UFC910008) at 3500 rpm for 20 min at 4\u00b0C to remove the iodixanol and concentrate the virus. The filtrate was removed, and the viral fraction was again filled with F188-containing PBS solution. The filter was centrifuged at 3500 rpm for 20 min at 4\u00b0C. The filtrate was discarded again, and the viral fraction was mixed with PBS. The virus was concentrated by centrifugation at 3500 rpm for 20 min at 4\u00b0C until the viral volume was about 500 \u03bcl.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "MATERIALS AND METHODS", "text": "Virus titer was measured by quantitative PCR. For AAV-SaCas9-HTTg1 and AAV-SaCas9, common U6-forward primer (5\u2032-CGA TAC AAG GCT GTT AGA GAG-3\u2032) was used, with reverse primers were AAV-HTTg1-reverse (5\u2032-AGG CCT TCA TCA GCT TTT CCA-3\u2032) and AAV-SaCas9-reverse (5\u2032-GTG GTC TCC GGT GTT TCG TCC-3\u2032), respectively. For AAV-EGFP-gCas9 and AAV-EGFP, the primers used are AAV-EGFP-forward (5\u2032-CGA AGG CTA CGT CCA GGA GC-3\u2032) and AAV-EGFP-reverse (5\u2032-CGA TGT TGT GGC GGA TCT TG-3\u2032). SDS-PAGE followed by Coomassie blue staining was performed to check the purity of the virus vectors, in which the only proteins visible in the gel were the VP1, VP2, and VP3 proteins that make up the capsid particle, migrating at ~87, 73, and 62 kDa, respectively.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "MATERIALS AND METHODS", "text": "Stereotaxic administration of AAV9 was performed on 1-, 4-, or 7-month-old mice. Mice were anesthetized with 1.5% isoflurane inhalation and stabilized in a stereotaxic instrument (RWD, 68019). Anterior-posterior (AP), medial-lateral (ML), and dorsal-ventral (DV) injection coordinates at different ages were determined empirically and calculated relative to the bregma. The injection coordinates for striatum were as follows: AP\u00a0=\u00a00.8 mm, ML\u00a0=\u00a0\u00b12.1 mm, and DV\u00a0=\u00a0\u22123.1 mm (1-month-old mice); AP\u00a0=\u00a00.7 mm, ML\u00a0=\u00a0\u00b12.0 mm, and DV\u00a0=\u00a0\u22123.1 mm (4-month-old mice); and AP\u00a0=\u00a00.6 mm, ML\u00a0=\u00a0\u00b12.1 mm, and DV\u00a0=\u00a0\u22123.1 mm (7-month-old mice). The injection coordinates for cortex were AP\u00a0=\u00a01.5 mm, ML\u00a0=\u00a0\u00b11.5 mm, and DV\u00a0=\u00a0\u22121.0 mm (mice of all ages). AAV9 was bilaterally injected into the striatum (2.5 \u03bcl) and primary motor cortex (0.5 \u03bcl) of mice. In conventional CRISP-Cas9 system, a total dose of 1.2\u00a0\u00d7\u00a01011 viral genomes (vg) of AAV-SaCas9-HTTg1 or AAV-SaCas9 was injected into each mouse brain (virus titer: 2\u00a0\u00d7\u00a01013 vg/ml). In self-inactivating system, AAV-EGFP-gCas9 or AAV-EGFP were coinjected with AAV-SaCas9-HTTg1 or AAV-SaCas9 at 1:1 molar ratio, and a total dose of 1.2\u00a0\u00d7\u00a01010 vg of virus was administrated for each mouse (virus titer: 2\u00a0\u00d7\u00a01012 vg/ml). Injection was performed through a microinjection pump (RWD, 788130)\u2013connected Hamilton syringe at an injection rate of 0.3 \u03bcl/min in striatum and 0.1 \u03bcl/min in cortex. A 1701 Hamilton microsyringe (Hamilton, 7853-01) with 33-gauge needle (Hamilton, 7803-05) was used to deliver the virus. The needle was left in place for 15 min after each injection to minimize upward flow of viral solution after raising the needle. After surgery, the mice were placed on a heated blanket to recover from the anesthetic.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "MATERIALS AND METHODS", "text": "All behavioral tests were performed during the dark phase of the light cycle and at the same time of day. Mice were handled under the same conditions by one investigator who was blind to the genotype and treatment of the mice.\n\nRotarod tests were operated on a rotarod treadmill (Med Associates Inc., ENV-574 M). Mice were trained at a constant speed of 10 rpm/min for 1 min per trial for three trials per day for 3 consecutive days. During the training trials, mice that fell were gently returned to the rotarod treadmill. Animals were tested for three trials per day for 3 consecutive days in every week, allowing at least 15 min of rest between each trial. Mice were tested with the rotarod accelerating from 5 to 40 rpm with a maximum period of 300 s. The latency to fall was recorded, and the mean of all trials was used for statistical analysis. The latency to fall was defined by either the animal falling from the rotarod or when the animal holds the rod for more than 3 cycles.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "MATERIALS AND METHODS", "text": "The open field consisted of a clear glass box with evenly distributed photobeams (27.3 cm by 27.3 cm by 20.3 cm; Med Associates Inc., ENV-510). Mice were placed in the open field over a 10-min period and kept track of locomotion. For each time points, mice were tested for three trials per day for 3 consecutive days. Average ambulatory distance or total travel distance was recorded, and the mean of all trials was used for statistical analysis. Representative figures of heatmaps showing the time and location to stay in the open field box were drawn by R (www.r-project.org/), referring to the average of nine sessions for one mouse.\n\nGait of mice was analyzed with CatWalk XT (Noldus Information Technology, Wageningen, The Netherlands). The CatWalk system consists of an enclosed walkway on a glass plate that is traversed by a mouse from one side of the walkway to the other. A completely internally reflected green light can emit only at areas where the paws make contact with the glass plate. It scatters and illuminates the paw prints. Paw prints are captured by a high-speed video camera that is positioned underneath the walkway. The video images with paw prints are used in the footprint classification and further analysis.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "MATERIALS AND METHODS", "text": "Mice were trained and adapted to walk across the 70-cm glass walkway in an unforced manner without any interruption or hesitation at least four times a day for at least 4 days before measurement. After training, mice footprints were measured three times per day on 3 consecutive days at each time point. Runs with any wall climbing, grooming, and staying on the walkway were not analyzed. Mice that failed the catwalk training were excluded from the study. An average number of three to six compliant trials made by each mouse were used for analysis. Analysis was performed using the CatWalk XT 10.6 Software. In catwalk performance, there are six-step sequence patterns of walking for mice: AA (RF-RH-LF-LH), AB (LF-RH-RF-LH), CA (RF-LF-RH-LH), CB (LF-RF-LH-RH), RA (RF-LF-LH-RH), and RB (LF-RF-RH-LH). Step sequence regularity index is a measurement of walking coordination and regular step patterns (74).\n\nThe body weight of mice was measured in grams every week after surgery.\n\nThe mice were kept in mix genotype with three to five mice per cage. After surgery, mice were monitored weekly for survival for at least 13 months.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "MATERIALS AND METHODS", "text": "Genomic DNA of mouse brain and cell line samples were extracted by a FastPure DNA isolation kit (Vazyme, DC102-01) following manufacturer\u2019s instructions and were sonicated to 300 to 500 bp. In addition, biotinylated 5\u2032-CTC AGG TTC TGC TTT TAC CTG CG-3\u2032 was used for primer extension followed by bridge adapter ligation. 5\u2032-CCG AGG CCT CCG GGG ACT GC-3\u2032 was used for nested PCR followed by tagged PCR with primers compatible for Illumina HiSeq sequencing. The raw data processing was described before (56). Editing efficiency was the percentage of indels and translocations to total events, and the off-target translocation was the percentage of junctions at off-target site to the total editing events.\n\nThe sequence of disrupted human HTT gene in BAC226Q mouse was characterized by Sanger sequencing. Mouse brain genomic DNA was extracted as previously described. Genomic DNA (100 ng) was amplified with human HTT-specific primer (hsHTT forward: 5\u2032-CTG CCG GGA CGG GTC CAA GAT-3\u2032; and hsHTT reverse: 5\u2032-TGC AGC GGC TCC TCA GCC AC-3\u2032) with the KAPA 2G Robust HotStart PCR kit (Roche, KE5507). The resulting PCR products of human HTT exon 1 fragments were purified with the QIAquick PCR purification kit (QIAGEN, 28104) according to the manufacturer\u2019s instructions and sequenced by Sanger sequencing.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "MATERIALS AND METHODS", "text": "The total stratal mRNA from 4-month-old non\u2013Tg-SaCas9, BAC226Q-SaCas9, BAC226Q-SaCas9-HTTg1, and BAC226Q-SaCas9-HTTg1/gCas9 mice were isolated by FreeZol Reagent (Vazyme, R711-01). The RNAs were sent to Beijing Novogene Technology for transcriptome sequencing and analysis. Total RNA (10 \u03bcg) for each qualified sample was used in the NEBNext Ultra RNA library prep kit (E7530L, New England Biolabs) to build the library. After cluster generation, the library preparations were sequenced on an Illumina NovaSeq platform, and 150-bp paired-end reads were generated. Index of the reference genome was built using Hisat2 v2.0.5, and paired-end clean reads were aligned to the reference genome using Hisat2 v2.0.5. Feature Counts v1.5.0-p3 was used to count the number of reads for each gene, and, then, the FPKM of each gene was calculated according to the length of the gene and reads count mapped to this gene. For differential gene analysis, the number of reads was normalized using DESeq software. The Benjamini and Hochberg\u2019s method was used for adjusting the resulting P values, and genes with absolute fold change > 1 and P < 0.05 were considered DEGs.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "MATERIALS AND METHODS", "text": "Unless otherwise mentioned, one-way analysis of variance (ANOVA) with Tukey\u2019s multiple comparisons test or two-way ANOVA with Tukey\u2019s multiple comparisons test was used for statistical comparison. Survival data were analyzed by log-rank (Mantel Cox) testing. Results were expressed as a means\u00a0\u00b1\u00a0SEM. Statistical tests were performed and graphs were generated using GraphPad Prism (GraphPad Software Inc.). Statistical significance was defined as P\u00a0<\u00a00.05. Statistical significance level was set as follows: *P\u00a0<\u00a00.05; **P\u00a0<\u00a00.01; ***P\u00a0<\u00a00.001.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Experimental design", "text": "The aim of this study was to develop a CRISPR-based gene editing as a therapy for HD with maximal safety. This study investigated the long-term rescue and off-target effects in a robust HD mouse model BAC226Q at different time points of intervention. Given the high HTT editing efficiency of CRISPR in HEK293T cell line, we therefore packaged our editing system in AAV9 and delivered it to the striatum and cortex of BAC226Q HD mice, evaluating the in vivo editing effects at different postinjection time points by sequencing, Western blot, immunostaining, and behavioral tests. The dosage of AAV vectors was determined on the basis of published papers and our pilot studies. The cell line used in this study was indicated in the place of related results shown in the text, figure legends, and Materials and Methods. Animal size was determined on the basis of the previous studies in BAC226Q mouse. Experimental end points were determined by cardinal parameters of age-dependent pathogenesis in BAC226Q. Both sexes were used in Western blot and staining studies, and only male mice were used in behavioral and sequencing studies. Animals were randomly assigned to the control and experimental groups, and researchers were blinded to genotype and treatment information during the experiment but not during analysis. Details on the use of cell line and animal in different experiments could be found in the corresponding sections of Materials and Methods. Biological replicates were indicated by n values in the figure legends. No data were excluded.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Molecular cloning", "text": "A single-vector AAV-SaCas9 system containing Cas9 from Staphylococcus aureus (SaCas9) and its sgRNA scaffold was obtained from Addgene (plasmid no. 61591). sgRNAs targeting human HTT and SaCas9 were designed on the basis of PAM sequence (5\u2032-NNGRRT-3\u2032) and by CRISPR RGEN Tools (www.rgenome.net/). sgRNA sequences targeting human HTT are as follows: HTTg1: 5\u2032-TGG AAA AGC TGA TGA AGG CCT-3\u2032; and HTTg2: 5\u2032-GAA AAG CTG ATG AAG GCC TTC-3\u2032. AAV-EGFP-gCas9 or AAV-mCherry-gCas9, which contain EGFP or mCherry, were constructed by replacing SaCas9 on vector AAV-SaCas9. sgRNAs targeting SaCas9 are listed in table S1.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Molecular cloning", "text": "Cas9 was replaced by EGFP or mCherry in the AAV vector backbone via Age I and Eco RI enzyme digestion to generate AAV-EGFP and AAV-mCherry vectors, respectively. Top and bottom strands of oligos for each sgRNA were phosphorylated and annealed into duplex. HTTg1 and HTTg2 were inserted into AAV-SaCas9 vector via Bsa I enzyme digestion. Similarly, gCas9 was cloned into the AAV-EGFP and AAV-mCherry vector through Bsa I digestion. The ligation products were transformed into Stbl3 chemically competent cells (TransGen, CD521), and plasmid DNA was extracted from cultures using the QIAprep spin miniprep kit (QIAGEN, 27104) according to the manufacturer\u2019s instructions. sgRNA sequences were verified by Sanger sequencing using the LKO.1 5\u2032 primer: 5\u2032-GAC TAT CAT ATG CTT ACC GT-3\u2032. EGFP and mCherry sequences were verified using the EGFP-forward primer: 5\u2032-CGA AGG CTA CGT CCA GGA GC-3\u2032 and the mCherry-forward primer: 5\u2032-ACA ACC GGT ATG GTG AGC AAG-3\u2032, respectively. AAV plasmids were then extracted by EndoFree plasmid maxi kit (QIAGEN, 12362) for subsequent virus production and purification.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Culture and transfection of HEK293T cell", "text": "HEK293T cells (American Type Culture Collection, CRL-1573) were cultured in Dulbecco\u2019s modified Eagle\u2019s medium (DMEM; Cytiva, SH3024301) supplemented with 10% (v/v) fetal bovine serum (Thermo Fisher Scientific), 1% (v/v) penicillin-streptomycin (Gibco, 21103049), and 1% (v/v) l-glutamine (Gibco, 25030081) in a humidified 5% CO2 atmosphere at 37\u00b0C. Cells were transfected with 2.5 \u03bcg of AAV-SaCas9-sgRNA vectors and 2.5 \u03bcg of pEGFPc3-human HTT exon 1-20 CAG/pEGFPc3-human HTT exon 1-120 CAG (EGFP-hHTT-20Q/120Q) at 70% confluency with Lipofectamine 2000 (Invitrogen, 11668030); in self-inactivating system, cells were cotransfected with 0.8 \u03bcg of AAV-SaCas9-HTTg1/AAV-SaCas9, 0.8 \u03bcg of AAV-mCherry-gCas9/AAV-mCherry, and 0.8 \u03bcg of EGFP-hHTT-20Q/120Q. For in vitro validation of cleavage effects, EGFP fluorescence was imaged under microscope (Carl Zeiss, Vert.A1) 24 or 48 hours after transfection, and, then, protein was harvested for Western blot. For PEM-seq assay, cells were transfected with 5 \u03bcg of AAV-SaCas9-HTTg1/AAV-SaCas9 and 5 \u03bcg of AAV-mCherry-gCas9/AAV-mCherry, and, then, mCherry-positive cells were sorted by flow cytometry (Beckman Coulter, Astrios EQ) 24 hours after transfection, followed by genomic DNA extraction.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Culture and nucleofection of human iPSC", "text": "Human chemical induced pluripotent stem cell (hCiPSC) line (#1217-2) was obtained from H. Deng at Peking University. hCiPSCs were cultured on Matrigel (Corning, 354230)\u2013coated plates in mTeSR Plus Medium (STEMCELL Technologies, 100-0276) in a humidified 5% CO2 atmosphere at 37\u00b0C. When reached to 80% confluence, cells were dissociated with ReLeSR (STEMCELL Technologies, 05872) and passaged with a 1:10 split ratio. Detached cell aggregates were first cultured in mTeSR Plus Medium supplemented with 10 \u03bcM Y-27632 and replaced by fresh mTeSR Plus Medium the next day. Culture medium was changed every 1 or 2 days. hCiPSCs were transfected by nucleofection with a Lonza P3 primary cell kit under CM115 program (Lonza). For each six-well, 2.5 \u03bcg of AAV-SaCas9-HTTg1/AAV-SaCas9 vectors and 2.5 \u03bcg of AAV-mCherry-gCas9/AAV-mCherry were transfected. For PEM-seq assay, mCherry-positive hCiPSCs were sorted by flow cytometry 48 hours after nucleofection. Cells were cultured in mTeSR Plus Medium supplemented with 10 \u03bcM Y-27632 for 24 hours after nucleofection and sorting and replaced by fresh mTeSR Plus Medium for further culturing.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Mice", "text": "Mice were housed in the specific pathogen\u2013free facility at Peking University in accordance with the guidelines. The room was controlled with 12-hour light/dark cycle, 40 to 60% humidity, and 22\u00b0\u00a0\u00b1\u00a01\u00b0C temperature. Food and water were always accessible. All mice used in this research were FVB/N (Taconic) background as BAC226Q mice in this background showed the most robust HD-like phenotypes. BAC226Q mice were bred with wild-type nontransgenic mice and confirmed by genotyping which produced a 403-bp polymerase chain reaction (PCR) fragment (forward primer: 5\u2032-GTA TAT GCT GCT GCC TGC AA-3\u2032; and reverse primer: 5\u2032-AGG GGA CAG TGT TGG TCA AG-3\u2032). All mice experimental procedures were approved by the Institutional Animal Care and Use Committee of Peking University (CFLS-LiCJ-1).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Western blot", "text": "Mouse striatal tissue was dissected and immediately collected in ice-cold phosphate-buffered saline (PBS). Lysates were prepared by ultrasonication in radioimmunoprecipitation assay buffer [150 mM NaCl, 0.1% SDS, 1% sodium deoxycholate, 50 mM triethanolamine, and 1% NP-40 (pH 7.4), supplemented with Caspase inhibitor Boc-D-FMK (Abcam, ab142036) and Protease inhibitor cocktail (Sigma-Aldrich, 111 M4009)] using a sonic dismembrator (Fisherbrand, FB120), pulsed on for 1 s and pulsed off for 2 s, and repeated for 20 to 25 times at 20% amplitude on ice. The lysates were incubated at 4\u00b0C for 60 min, followed by centrifugation at 4\u00b0C for 20 min at 16,100g to remove insoluble components. Protein concentration was measured by bicinchoninic acid (BCA) protein assay using Pierce BCA Protein Assay Kit (Thermo Fisher Scientific, 23225), and the absorbance was detected at 562 nm. Protein samples were prepared for loading by heating in NuPAGE 4\u00d7\u00a0LDS sample buffer plus 10\u00d7\u00a0sample reducing agent (Invitrogen) according to the manufacturer\u2019s protocol and heated for 10 min at 70\u00b0C. A total amount of 90 to 120 \u03bcg of protein was loaded onto 4 to 12% NuPAGE bis-tris gels (Invitrogen, NP0321/NP0336) using Mops running buffer and wet transferred to an Immobilon-FL polyvinylidene difluoride (PVDF) membrane (Millipore, IPFL00010). Lysates from HEK293T cells were prepared in Triton X-100 buffer [150 mM NaCl, 50 mM tris-HCl (pH 7.4), 1 mM EDTA, and 1% Triton X-100]. The lysates were incubated at 4\u00b0C for 30 min, followed by centrifugation at 4\u00b0C for 20 min at 16,100g to remove insoluble components. Protein concentration was measured by BCA protein assay. Protein samples were prepared for loading by heating in 4\u00d7\u00a0SDS sample buffer and heated for 10 min at 95\u00b0C. A total amount of 30 \u03bcg of protein was loaded onto SDS\u2013polyacrylamide gel electrophoresis (PAGE) gels using running buffer and wet transferred to an Immobilon-FL PVDF membrane. Blots were incubated with Odyssey blocking buffer (LI-COR, 927-40000) for 60 min. After washing with TBST, blots were incubated with primary antibodies diluted in blocking buffer at 4\u00b0C overnight. Blots were washed three times (10 min each) and incubated with fluorescently labeled IRDye 680RD goat anti-mouse (1:10,000; LI-COR, 926-68070) or goat anti-rabbit (1:10,000; LI-COR, 926-68071) secondary antibodies in TBST for 60 min at room temperature. Protein was detected in the 700-nm channel using Odyssey CLx imager (LI-COR). The following primary antibodies were used: 1C2 (1:5000, mouse; Millipore, MAB1574), GFP (1:8000, rabbit; Abcam, ab183734), hemagglutinin (HA; 1:2000, mouse; Abcam, ab18181), and \u03b2-actin (1:2000, rabbit; Cell Signaling Technology, 4970).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Immunofluorescent staining", "text": "Mice were anesthetized, perfused with fresh 4% paraformaldehyde in PBS, and postfixed overnight in the same fixative solution. Fixed brains were sliced at 40-\u03bcm thickness with a vibrating blade microtome (Leica, VT1200S). The brain slices were incubated in PBS supplemented with 0.3% (v/v) Triton X-100 for 30 min in room temperature and then blocked with 10% (v/v) goat serum (ZSGB-BIO, ZLI-9021) in PBS supplemented with 0.1 or 0.3% Triton X-100 for 60 min at room temperature. Following incubation of brain slices with primary antibodies at 4\u00b0C overnight and washes with PBS supplemented with 0.1% Triton X-100, fluor-conjugated secondary antibodies (anti-mouse, anti-rabbit, or anti-sheep Alexa Fluor 488, 555, 594, or 647; 1:1000; Thermo Fisher Scientific) were added to the samples for 60 min at room temperature. Sections were cover slipped by antifade mountant with Hoechst 33342 nuclear stain (Thermo Fisher Scientific, P36985). Images were taken by Zeiss LSM 780 or LSM 880 confocal microscope (Carl Zeiss), and whole-brain section images were taken by Axio Scan Z1 (Carl Zeiss). The following primary antibodies were used: mEM48 (1:1000, mouse; Millipore, MAB5374), S830 (1:5000, sheep, provided by G. Bates from University College London), HA (1:1000, rabbit; Cell Signaling Technology, 3724), HA (1:1000, mouse; Abcam, ab18181), GFAP (1:1000, rabbit; Abcam, ab7260), and NeuN (1:1000, mouse; Abcam, ab104224). For the quantification of S830-positive mHTT in Fig. 2 (E, F, I, and J), we performed the analysis within dorsal striatum, where AAV9 coverage is high. For the quantification of mEM48-positive mHTT in Figs. 4 (B and F) and 5 (C to E) and fig. S5 (A and C), we performed the analysis specifically within regions exhibiting clear HA staining. In fig. S2 (A and B), the proportion of mEM48-negative cells within the HA-positive population was quantified.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Immunohistochemistry", "text": "Mouse brain sections were prepared by the same procedure as the immunofluorescent staining samples. Sections were stained according to instructions provided with the ABC and DAB substrate kits (Vector Labs). Briefly, sections were first processed free-floating with blocking serum and primary antibodies. After treatment with biotinylated secondary antibody, followed by avidin-biotinylated horseradish peroxidase complexes and incubation with diaminobenzidine, sections were mounted onto slides. In addition, brain sections were dehydrated using the setup as 30% alcohol, 50% alcohol, 70% alcohol, 95% alcohol, 100% alcohol-I, and 100% alcohol-II (2 min each) and xylene-I and xylene-II (10 min each). Then, sections were cover slipped by DPX mounting medium (Sigma-Aldrich, 06522) and imaged under Axio Scan Z1 microscope. The following primary antibodies were used: S830 (1:25,000, sheep, provided by G. Bates from University College London), HA (1:1000, rabbit; Cell Signaling Technology, 3724), and GFAP (1:2000, rabbit; Abcam, ab7260). For the quantification of S830-positive mHTT in Fig. 5L and fig. S8 (A and B), we performed the analysis specifically within regions exhibiting clear HA staining.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "AAV vector production", "text": "HEK293T cells were plated on 10 15-cm dishes and transfected at 90% confluence by adding 5 ml of the plasmid and linear polyethylenimine (PEI; Sigma-Aldrich, 765090) mixture to each plate. The mixture was prepared with 70 \u03bcg of AAV9 vector (AAV-SaCas9-HTTg1, AAV-SaCas9, AAV-EGFP-gCas9, or AAV-EGFP), 200 \u03bcg of Ad-Helper plasmid, 70 \u03bcg of AAV-Rep/Cap plasmid, and PEI (1 \u03bcg/\u03bcl), and the PEI-to-DNA ratio was 5:1 (v/g) in our experiment. The mixture was filled to 50 ml with DMEM and mixed well. Cells were collected with culture medium 60 hours after transfection and centrifuged at 1000 rpm for 10 min. Cell pellet was resuspended in 5 ml of cell lysis buffer [150 mM NaCl and 20 mM tris (pH 8.0)].\n\nThe cell lysate was freeze-thawed three times by changing between lipid nitrogen and 37\u00b0C water bath to release virus from the destroyed cell membrane. Cell lysate was added with MgCl2 and Benzonase (Sigma-Aldrich, E8263-25k) to a final concentration of 1 mM and 250 U/ml, respectively, followed by incubation at 37\u00b0C for 15 min to dissolve the DNA/protein aggregation. The supernatant was collected after centrifugation of the cell lysate at 4000 rpm for 30 min at 4\u00b0C.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "AAV vector production", "text": "The viral solution was loaded on the iodixanol gradient solution for purification, and the gradient was prepared in the following order (from top to bottom): 6 ml of 17% iodixanol [5 ml of 10\u00d7\u00a0PBS, 0.05 ml of 1 M MgCl2, 0.125 ml of 1 M KCl, 10 ml of 5 M NaCl, 12.5 ml of OptiPrep (Sigma-Aldrich, D1556), and up to 50 ml of H2O], 6 ml of 25% iodixanol (5 ml of 10\u00d7\u00a0PBS, 0.05 ml of 1 M MgCl2, 0.125 ml of 1 M KCl, 20 ml of OptiPrep, and H2O up to 50 ml), 5 ml of 40% iodixanol (5 ml of 10\u00d7\u00a0PBS, 0.05 ml of 1 M MgCl2, 0.125 ml of 1 M KCl, 33.3 ml of OptiPrep, and up to 50 ml of H2O), and 4 ml of 60% iodixanol (0.05 ml of 1 M MgCl2, 0.125 ml of 1 M KCl, and 50 ml of OptiPrep). The sample was centrifuged at 53,000 rpm for 160 min at 14\u00b0C, and the viral fraction was harvested in the 40% layer with a syringe.\n\nThe viral fraction was mixed with PBS solution and 10% Poloxamer 188 solution (F188) (1:10,000; Sigma-Aldrich, P5556) and then run through the Amacon 100K filter (MilliporeSigma, UFC910008) at 3500 rpm for 20 min at 4\u00b0C to remove the iodixanol and concentrate the virus. The filtrate was removed, and the viral fraction was again filled with F188-containing PBS solution. The filter was centrifuged at 3500 rpm for 20 min at 4\u00b0C. The filtrate was discarded again, and the viral fraction was mixed with PBS. The virus was concentrated by centrifugation at 3500 rpm for 20 min at 4\u00b0C until the viral volume was about 500 \u03bcl.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "AAV vector production", "text": "Virus titer was measured by quantitative PCR. For AAV-SaCas9-HTTg1 and AAV-SaCas9, common U6-forward primer (5\u2032-CGA TAC AAG GCT GTT AGA GAG-3\u2032) was used, with reverse primers were AAV-HTTg1-reverse (5\u2032-AGG CCT TCA TCA GCT TTT CCA-3\u2032) and AAV-SaCas9-reverse (5\u2032-GTG GTC TCC GGT GTT TCG TCC-3\u2032), respectively. For AAV-EGFP-gCas9 and AAV-EGFP, the primers used are AAV-EGFP-forward (5\u2032-CGA AGG CTA CGT CCA GGA GC-3\u2032) and AAV-EGFP-reverse (5\u2032-CGA TGT TGT GGC GGA TCT TG-3\u2032). SDS-PAGE followed by Coomassie blue staining was performed to check the purity of the virus vectors, in which the only proteins visible in the gel were the VP1, VP2, and VP3 proteins that make up the capsid particle, migrating at ~87, 73, and 62 kDa, respectively.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Stereotaxic injection procedure", "text": "Stereotaxic administration of AAV9 was performed on 1-, 4-, or 7-month-old mice. Mice were anesthetized with 1.5% isoflurane inhalation and stabilized in a stereotaxic instrument (RWD, 68019). Anterior-posterior (AP), medial-lateral (ML), and dorsal-ventral (DV) injection coordinates at different ages were determined empirically and calculated relative to the bregma. The injection coordinates for striatum were as follows: AP\u00a0=\u00a00.8 mm, ML\u00a0=\u00a0\u00b12.1 mm, and DV\u00a0=\u00a0\u22123.1 mm (1-month-old mice); AP\u00a0=\u00a00.7 mm, ML\u00a0=\u00a0\u00b12.0 mm, and DV\u00a0=\u00a0\u22123.1 mm (4-month-old mice); and AP\u00a0=\u00a00.6 mm, ML\u00a0=\u00a0\u00b12.1 mm, and DV\u00a0=\u00a0\u22123.1 mm (7-month-old mice). The injection coordinates for cortex were AP\u00a0=\u00a01.5 mm, ML\u00a0=\u00a0\u00b11.5 mm, and DV\u00a0=\u00a0\u22121.0 mm (mice of all ages). AAV9 was bilaterally injected into the striatum (2.5 \u03bcl) and primary motor cortex (0.5 \u03bcl) of mice. In conventional CRISP-Cas9 system, a total dose of 1.2\u00a0\u00d7\u00a01011 viral genomes (vg) of AAV-SaCas9-HTTg1 or AAV-SaCas9 was injected into each mouse brain (virus titer: 2\u00a0\u00d7\u00a01013 vg/ml). In self-inactivating system, AAV-EGFP-gCas9 or AAV-EGFP were coinjected with AAV-SaCas9-HTTg1 or AAV-SaCas9 at 1:1 molar ratio, and a total dose of 1.2\u00a0\u00d7\u00a01010 vg of virus was administrated for each mouse (virus titer: 2\u00a0\u00d7\u00a01012 vg/ml). Injection was performed through a microinjection pump (RWD, 788130)\u2013connected Hamilton syringe at an injection rate of 0.3 \u03bcl/min in striatum and 0.1 \u03bcl/min in cortex. A 1701 Hamilton microsyringe (Hamilton, 7853-01) with 33-gauge needle (Hamilton, 7803-05) was used to deliver the virus. The needle was left in place for 15 min after each injection to minimize upward flow of viral solution after raising the needle. After surgery, the mice were placed on a heated blanket to recover from the anesthetic.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Mouse behavioral analysis", "text": "All behavioral tests were performed during the dark phase of the light cycle and at the same time of day. Mice were handled under the same conditions by one investigator who was blind to the genotype and treatment of the mice.\n\nRotarod tests were operated on a rotarod treadmill (Med Associates Inc., ENV-574 M). Mice were trained at a constant speed of 10 rpm/min for 1 min per trial for three trials per day for 3 consecutive days. During the training trials, mice that fell were gently returned to the rotarod treadmill. Animals were tested for three trials per day for 3 consecutive days in every week, allowing at least 15 min of rest between each trial. Mice were tested with the rotarod accelerating from 5 to 40 rpm with a maximum period of 300 s. The latency to fall was recorded, and the mean of all trials was used for statistical analysis. The latency to fall was defined by either the animal falling from the rotarod or when the animal holds the rod for more than 3 cycles.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Mouse behavioral analysis", "text": "The open field consisted of a clear glass box with evenly distributed photobeams (27.3 cm by 27.3 cm by 20.3 cm; Med Associates Inc., ENV-510). Mice were placed in the open field over a 10-min period and kept track of locomotion. For each time points, mice were tested for three trials per day for 3 consecutive days. Average ambulatory distance or total travel distance was recorded, and the mean of all trials was used for statistical analysis. Representative figures of heatmaps showing the time and location to stay in the open field box were drawn by R (www.r-project.org/), referring to the average of nine sessions for one mouse.\n\nGait of mice was analyzed with CatWalk XT (Noldus Information Technology, Wageningen, The Netherlands). The CatWalk system consists of an enclosed walkway on a glass plate that is traversed by a mouse from one side of the walkway to the other. A completely internally reflected green light can emit only at areas where the paws make contact with the glass plate. It scatters and illuminates the paw prints. Paw prints are captured by a high-speed video camera that is positioned underneath the walkway. The video images with paw prints are used in the footprint classification and further analysis.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Mouse behavioral analysis", "text": "Mice were trained and adapted to walk across the 70-cm glass walkway in an unforced manner without any interruption or hesitation at least four times a day for at least 4 days before measurement. After training, mice footprints were measured three times per day on 3 consecutive days at each time point. Runs with any wall climbing, grooming, and staying on the walkway were not analyzed. Mice that failed the catwalk training were excluded from the study. An average number of three to six compliant trials made by each mouse were used for analysis. Analysis was performed using the CatWalk XT 10.6 Software. In catwalk performance, there are six-step sequence patterns of walking for mice: AA (RF-RH-LF-LH), AB (LF-RH-RF-LH), CA (RF-LF-RH-LH), CB (LF-RF-LH-RH), RA (RF-LF-LH-RH), and RB (LF-RF-RH-LH). Step sequence regularity index is a measurement of walking coordination and regular step patterns (74).\n\nThe body weight of mice was measured in grams every week after surgery.\n\nThe mice were kept in mix genotype with three to five mice per cage. After surgery, mice were monitored weekly for survival for at least 13 months.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "\nRotarod test\n", "text": "Rotarod tests were operated on a rotarod treadmill (Med Associates Inc., ENV-574 M). Mice were trained at a constant speed of 10 rpm/min for 1 min per trial for three trials per day for 3 consecutive days. During the training trials, mice that fell were gently returned to the rotarod treadmill. Animals were tested for three trials per day for 3 consecutive days in every week, allowing at least 15 min of rest between each trial. Mice were tested with the rotarod accelerating from 5 to 40 rpm with a maximum period of 300 s. The latency to fall was recorded, and the mean of all trials was used for statistical analysis. The latency to fall was defined by either the animal falling from the rotarod or when the animal holds the rod for more than 3 cycles.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "\nOpen field\n", "text": "The open field consisted of a clear glass box with evenly distributed photobeams (27.3 cm by 27.3 cm by 20.3 cm; Med Associates Inc., ENV-510). Mice were placed in the open field over a 10-min period and kept track of locomotion. For each time points, mice were tested for three trials per day for 3 consecutive days. Average ambulatory distance or total travel distance was recorded, and the mean of all trials was used for statistical analysis. Representative figures of heatmaps showing the time and location to stay in the open field box were drawn by R (www.r-project.org/), referring to the average of nine sessions for one mouse.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "\nCatwalk\n", "text": "Gait of mice was analyzed with CatWalk XT (Noldus Information Technology, Wageningen, The Netherlands). The CatWalk system consists of an enclosed walkway on a glass plate that is traversed by a mouse from one side of the walkway to the other. A completely internally reflected green light can emit only at areas where the paws make contact with the glass plate. It scatters and illuminates the paw prints. Paw prints are captured by a high-speed video camera that is positioned underneath the walkway. The video images with paw prints are used in the footprint classification and further analysis.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "\nCatwalk\n", "text": "Mice were trained and adapted to walk across the 70-cm glass walkway in an unforced manner without any interruption or hesitation at least four times a day for at least 4 days before measurement. After training, mice footprints were measured three times per day on 3 consecutive days at each time point. Runs with any wall climbing, grooming, and staying on the walkway were not analyzed. Mice that failed the catwalk training were excluded from the study. An average number of three to six compliant trials made by each mouse were used for analysis. Analysis was performed using the CatWalk XT 10.6 Software. In catwalk performance, there are six-step sequence patterns of walking for mice: AA (RF-RH-LF-LH), AB (LF-RH-RF-LH), CA (RF-LF-RH-LH), CB (LF-RF-LH-RH), RA (RF-LF-LH-RH), and RB (LF-RF-RH-LH). Step sequence regularity index is a measurement of walking coordination and regular step patterns (74).", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "\nBody weight\n", "text": "The body weight of mice was measured in grams every week after surgery.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "\nSurvival\n", "text": "The mice were kept in mix genotype with three to five mice per cage. After surgery, mice were monitored weekly for survival for at least 13 months.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Primer extension\u2013mediated sequencing", "text": "Genomic DNA of mouse brain and cell line samples were extracted by a FastPure DNA isolation kit (Vazyme, DC102-01) following manufacturer\u2019s instructions and were sonicated to 300 to 500 bp. In addition, biotinylated 5\u2032-CTC AGG TTC TGC TTT TAC CTG CG-3\u2032 was used for primer extension followed by bridge adapter ligation. 5\u2032-CCG AGG CCT CCG GGG ACT GC-3\u2032 was used for nested PCR followed by tagged PCR with primers compatible for Illumina HiSeq sequencing. The raw data processing was described before (56). Editing efficiency was the percentage of indels and translocations to total events, and the off-target translocation was the percentage of junctions at off-target site to the total editing events.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Sanger sequencing", "text": "The sequence of disrupted human HTT gene in BAC226Q mouse was characterized by Sanger sequencing. Mouse brain genomic DNA was extracted as previously described. Genomic DNA (100 ng) was amplified with human HTT-specific primer (hsHTT forward: 5\u2032-CTG CCG GGA CGG GTC CAA GAT-3\u2032; and hsHTT reverse: 5\u2032-TGC AGC GGC TCC TCA GCC AC-3\u2032) with the KAPA 2G Robust HotStart PCR kit (Roche, KE5507). The resulting PCR products of human HTT exon 1 fragments were purified with the QIAquick PCR purification kit (QIAGEN, 28104) according to the manufacturer\u2019s instructions and sequenced by Sanger sequencing.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "RNA-seq and data analysis", "text": "The total stratal mRNA from 4-month-old non\u2013Tg-SaCas9, BAC226Q-SaCas9, BAC226Q-SaCas9-HTTg1, and BAC226Q-SaCas9-HTTg1/gCas9 mice were isolated by FreeZol Reagent (Vazyme, R711-01). The RNAs were sent to Beijing Novogene Technology for transcriptome sequencing and analysis. Total RNA (10 \u03bcg) for each qualified sample was used in the NEBNext Ultra RNA library prep kit (E7530L, New England Biolabs) to build the library. After cluster generation, the library preparations were sequenced on an Illumina NovaSeq platform, and 150-bp paired-end reads were generated. Index of the reference genome was built using Hisat2 v2.0.5, and paired-end clean reads were aligned to the reference genome using Hisat2 v2.0.5. Feature Counts v1.5.0-p3 was used to count the number of reads for each gene, and, then, the FPKM of each gene was calculated according to the length of the gene and reads count mapped to this gene. For differential gene analysis, the number of reads was normalized using DESeq software. The Benjamini and Hochberg\u2019s method was used for adjusting the resulting P values, and genes with absolute fold change > 1 and P < 0.05 were considered DEGs.", "title": "" }, { "pmid": 41849610, "pmcid": "PMC12998519", "section": "Statistical analysis", "text": "Unless otherwise mentioned, one-way analysis of variance (ANOVA) with Tukey\u2019s multiple comparisons test or two-way ANOVA with Tukey\u2019s multiple comparisons test was used for statistical comparison. Survival data were analyzed by log-rank (Mantel Cox) testing. Results were expressed as a means\u00a0\u00b1\u00a0SEM. Statistical tests were performed and graphs were generated using GraphPad Prism (GraphPad Software Inc.). Statistical significance was defined as P\u00a0<\u00a00.05. Statistical significance level was set as follows: *P\u00a0<\u00a00.05; **P\u00a0<\u00a00.01; ***P\u00a0<\u00a00.001.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Introduction", "text": "Non-coding RNAs (ncRNAs) are a diverse group of RNA molecules that play important roles in the central nervous system (CNS). Among these, circular RNAs (circRNAs) and enhancer RNAs (eRNAs) have become important in neurodegeneration. CircRNAs are more stable than linear ncRNAs due to their closed-loop structure. eRNAs, which come from enhancer regions, serve as regulators of gene expression in both space and time. Both types are found in high amounts in the brain, where they influence synaptic plasticity, neuroinflammation, and protein balance, linking them to neurodegenerative diseases [1, 2].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Introduction", "text": "CircRNAs and eRNAs have gained attention as key regulatory molecules among non-coding RNAs. CircRNAs have a unique closed-loop structure that makes them very stable and abundant in the mammalian brain, particularly in synapses. Their specific expression patterns suggest that they are crucial for neurodevelopment and neuronal function [3\u20136]. In contrast, eRNAs, produced from enhancer regions, play a vital role in gene expression, shaping chromatin structure and enhancing interactions between enhancers and promoters [7]. This review aims to gather existing evidence on how circRNAs and eRNAs work in the development of Alzheimer\u2019s disease, Parkinson\u2019s disease, Huntington\u2019s disease, and Multiple Sclerosis. Besides examining the mechanisms, we will explore their potential as disease-specific biomarkers that can be found in biofluids and as targets for new RNA-based therapies. By providing a clear view of their functions, this review hopes to link molecular insights with clinical uses, opening up new possibilities for diagnosis and treatment in neurology. Although this review discusses two distinct classes of noncoding RNAs\u2014circRNAs and eRNAs\u2014their inclusion within a single framework is justified by their complementary roles as master regulators of the integrity of gene expression in the nervous system. CircRNAs function primarily at the posttranscriptional level, using their stability to fine-tune gene expression through miRNA spongification, protein silencing, and translational regulation [1]. In contrast, eRNAs function at the transcriptional level, directly facilitating enhancer-promoter interaction and chromatin remodeling to initiate gene expression programs [7]. This multilevel regulatory axis is particularly important in postmitotic neurons, where precise gene control is essential for maintaining synaptic plasticity, neuronal identity, and cellular homeostasis. The simultaneous dysregulation of these regulatory layers may create destructive feeding loops in neurodegeneration, where transcriptional abnormalities (mediated by eRNAs) and posttranscriptional dysregulation (mediated by circRNAs) synergistically drive disease progression. Although the current literature on eRNAs in neurodegeneration is less extensive than on circRNAs, a difference that we acknowledge, emerging evidence strongly suggests their parallel importance in shaping the molecular landscape of neurodegenerative disorders. This narrative review synthesizes the available evidence on circRNAs and eRNAs in four major neuropathologies (AD, PD, HD, MS). We focus on identifying converging themes, such as their roles in protein aggregation, neuroinflammation, and synaptic integrity, and critically evaluate their potential as a novel class of integrative diagnostic and therapeutic targets.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Circular RNAs (circRNAs): biogenesis and multimodal functions", "text": "Circular RNAs are a large class of single-stranded, covalent RNA molecules produced by a process called \u201creverse splicing,\u201d in which a 5\u2018downstream splice site is ligated to a 3\u2019 upstream splice site of a pre-mRNA [8, 9]. This unique structure, which lacks free 5\u2018and 3\u2019 ends, confers exceptional stability to RNA exonucleases and gives circRNAs a much longer half-life than their linear counterparts [10]. Their biogenesis is regulated by cis-acting elements (e.g., reverse complementarity motifs such as Alu repeats in flanking introns) and trans-acting factors (e.g., RNA-binding proteins such as Quaking (QKI) and Muscleblind (MBL)) [11]. Once formed, circRNAs are predominantly located in the cytoplasm, although a subset can localize to the nucleus, and their function is largely determined by their subcellular location. The functional repertoire of circRNAs extends far beyond their initial description and includes several key mechanistic roles:", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Circular RNAs (circRNAs): biogenesis and multimodal functions", "text": "The most widely studied function involves circRNAs that act as competing endogenous RNAs (ceRNAs) or \u201cmiRNA sponges\u201d [12]. By possessing multiple binding sites for specific microRNAs, circRNAs can sequester them and prevent miRNAs from interacting with their target mRNAs. This sponging activity leads to the re-repression of miRNA target genes and effectively increases their expression post-transcriptionally. A classic example is ciRS-7 (also known as CDR1as), which contains more than 70 conserved binding sites for miR-7 and acts as a potent supersponge in neural tissues [13].\n\ncircRNAs can interact with RNA-binding proteins (RBPs) to act as bait or protein scaffolds. As a decoy, a circRNA can bind to and cleave a specific RBP, inhibiting its normal function. Conversely, as a scaffold, a circRNA can bring multiple proteins into proximity to facilitate the formation of functional complexes [14]. For example, circFoxo3 interacts with the senescence-associated proteins ID-1 and E2F1 and the transcription factor HIF1\u03b1, promoting cardiac aging [15].\n\ncircRNAs remaining in the nucleus can influence gene expression at the transcriptional level. Some circRNAs, such as circEIF3J and circPAIP2, can interact with the U1 small nuclear ribonucleoprotein (snRNP) and RNA polymerase II at the promoter of their parental genes, thereby enhancing their transcription [16].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Circular RNAs (circRNAs): biogenesis and multimodal functions", "text": "Although most circRNAs are non-coding, a subset containing an internal ribosome entry site (IRES) and an open reading frame (ORF) can be translated into functional peptides or proteins, often in response to cellular stress [1]. These circRNA-derived peptides can exert dominant-negative effects or have unique biological activities.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "miRNA sponging", "text": "The most widely studied function involves circRNAs that act as competing endogenous RNAs (ceRNAs) or \u201cmiRNA sponges\u201d [12]. By possessing multiple binding sites for specific microRNAs, circRNAs can sequester them and prevent miRNAs from interacting with their target mRNAs. This sponging activity leads to the re-repression of miRNA target genes and effectively increases their expression post-transcriptionally. A classic example is ciRS-7 (also known as CDR1as), which contains more than 70 conserved binding sites for miR-7 and acts as a potent supersponge in neural tissues [13].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Sponging and protein scaffolding", "text": "circRNAs can interact with RNA-binding proteins (RBPs) to act as bait or protein scaffolds. As a decoy, a circRNA can bind to and cleave a specific RBP, inhibiting its normal function. Conversely, as a scaffold, a circRNA can bring multiple proteins into proximity to facilitate the formation of functional complexes [14]. For example, circFoxo3 interacts with the senescence-associated proteins ID-1 and E2F1 and the transcription factor HIF1\u03b1, promoting cardiac aging [15].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Transcriptional regulation and splicing", "text": "circRNAs remaining in the nucleus can influence gene expression at the transcriptional level. Some circRNAs, such as circEIF3J and circPAIP2, can interact with the U1 small nuclear ribonucleoprotein (snRNP) and RNA polymerase II at the promoter of their parental genes, thereby enhancing their transcription [16].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Cap-independent translation", "text": "Although most circRNAs are non-coding, a subset containing an internal ribosome entry site (IRES) and an open reading frame (ORF) can be translated into functional peptides or proteins, often in response to cellular stress [1]. These circRNA-derived peptides can exert dominant-negative effects or have unique biological activities.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Enhancer RNAs (eRNAs): biogenesis and transcriptional roles", "text": "Enhancer RNAs are a class of relatively unstable, non-coding RNAs that are transcribed bidirectionally from active enhancer regions [17]. Enhancers are cis-terminal regulatory elements of DNA that control the spatial and temporal precision of gene expression. Transcription at these sites, which produces eRNAs, is a characteristic of functionally active enhancers. The production of eRNAs is closely linked to enhancer activation, which involves the recruitment of transcription factors, co-activators (e.g., p300/CBP), and the Mediator complex that leads to the recruitment of RNA polymerase II [18]. The main functions of eRNAs are intrinsically linked to their role in facilitating enhancer-promoter interaction and target gene transcriptional activation:\n\nA leading model postulates that eRNAs play a role in the formation and stabilization of chromatin loops that place enhancers in close physical proximity to their target gene promoters [19]. This looping is essential for the transport of the transcriptional machinery from enhancer to promoter. For example, eRNAs transcribed from the enhancer of the *Dlx5/6* locus (Evf2) help recruit the helper CBP to stabilize the enhancer-promoter complex [20].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Enhancer RNAs (eRNAs): biogenesis and transcriptional roles", "text": "eRNAs can directly interact with key helper complexes and recruit them to enhancer sites. Studies have shown that eRNAs can bind to the MED12 subunit of the mediator complex and the co-activator CBP, thereby promoting the assembly of a more potent transcriptional machinery and stimulating histone acetylation, which in turn opens chromatin and facilitates transcription [21].\n\nThe transcriptional action of eRNAs themselves or their products can affect the local chromatin state. This can include the relocation of nucleosomes or the recruitment of chromatin remodelers, which make the enhancer and its target promoter more accessible to the transcriptional machinery [22].\n\nSome eRNAs are involved in the release of stalled RNA polymerase II to induce productive elongation at target gene promoters. They can achieve this by interacting with and dissociating the negative elongation factor (NELF) complex that allows the polymerase to advance along the gene body [23].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Facilitation of enhancer-promoter looping", "text": "A leading model postulates that eRNAs play a role in the formation and stabilization of chromatin loops that place enhancers in close physical proximity to their target gene promoters [19]. This looping is essential for the transport of the transcriptional machinery from enhancer to promoter. For example, eRNAs transcribed from the enhancer of the *Dlx5/6* locus (Evf2) help recruit the helper CBP to stabilize the enhancer-promoter complex [20].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Recruitment of transcriptional co-activators", "text": "eRNAs can directly interact with key helper complexes and recruit them to enhancer sites. Studies have shown that eRNAs can bind to the MED12 subunit of the mediator complex and the co-activator CBP, thereby promoting the assembly of a more potent transcriptional machinery and stimulating histone acetylation, which in turn opens chromatin and facilitates transcription [21].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Modulation of chromatin accessibility", "text": "The transcriptional action of eRNAs themselves or their products can affect the local chromatin state. This can include the relocation of nucleosomes or the recruitment of chromatin remodelers, which make the enhancer and its target promoter more accessible to the transcriptional machinery [22].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Release of paused RNA polymerase II", "text": "Some eRNAs are involved in the release of stalled RNA polymerase II to induce productive elongation at target gene promoters. They can achieve this by interacting with and dissociating the negative elongation factor (NELF) complex that allows the polymerase to advance along the gene body [23].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Involvement of circRNAs and eRNAs in the brain", "text": "Growing evidence points out the role of circRNAs in brain development and disease. The mammalian brain has the highest concentration of circRNAs among tissues. Their presence in synapses suggests they play roles in how neurons function [4\u20136]. Their expression changes over time and varies by brain region and sex, linking them to neuronal activity and mental health issues [6, 24\u201327]. During neurodevelopment, circRNA levels peak in the second trimester, which aligns with synaptogenesis and plasticity. This peak correlates with axon guidance, dendrite formation, and neuron growth [4, 28]. CircRNA expression drops after development but stabilizes in mature neurons, where it maintains a strong association with synaptic plasticity [4, 29]. Neural circRNAs increase with age, often showing an inverse relationship with the linear transcript levels of their host genes [25, 26, 30\u201332]. This age-related rise in synaptosomes suggests a role in synaptic decline [33, 34]. For example, the primate-conserved circGRIA1, which comes from an AMPA receptor gene, builds up in the aging prefrontal cortex and hippocampus. This accumulation disrupts synaptic plasticity and calcium regulation by cis-regulating Gria1 mRNA [34]. Additionally, aging neurons exhibit unique circRNAs, primarily found in synaptosomes, which are not present in linear forms [33].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Involvement of circRNAs and eRNAs in the brain", "text": "eRNAs, first discovered in neurons during activity-dependent enhancer transcription [19], are vital for gene regulation in response to stimuli. For instance, Fos-associated eRNAs in rat neurons are crucial for the activity-dependent induction of Fos and the patterns of neuronal firing [35]. In cases of ischemic stroke, 77 eRNAs were found to be upregulated. Knocking down eRNA_06347 and eRNA_093384 increased the size of infarcts, suggesting that they play a protective role [36]. Simon et al. explored the function of eRNAs derived from enhancers that regulate activity-responsive immediate early genes associated with cognitive functions. The authors demonstrated that neuronal activity dynamically influences eRNA expression from these loci. These findings imply that RNAs transcribed from neuronal enhancers are significant contributors to epigenetic landscapes and gene regulatory mechanisms. There is a connection between enhancer activity and brain function [37]. This finding functionally links eRNA transcription directly to neuronal signaling, positioning them as dynamic mediators that translate electrical activity into sustained changes in gene expression. This mechanism is critical for synaptic plasticity and memory formation, and its dysregulation could provide a molecular bridge between synaptic dysfunction and the transcriptional changes observed in neurological diseases such as Alzheimer\u2019s disease.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Involvement of circRNAs and eRNAs in the brain", "text": "In a recent study, J. Betti et al. created in silico models to analyze the genetically regulated expression of enhancer RNAs across 49 different cell and tissue types, assessing the extent of their genetic control. Using the estimated genetically regulated expression for eRNAs and canonical genes from a large-scale DNA biobank alongside high-resolution Hi-C contact data, the authors developed a deep learning-based model to predict pairwise three-dimensional chromatin contact frequency for both enhancer-enhancer and enhancer-gene pairs in the cerebellum and whole blood. Importantly, the incorporation of genetically regulated expression of enhancer RNAs offers significant tissue-specific predictive capabilities, which points to the potential role of these transcripts in influencing spatial chromatin organization. The eRNA model, which includes a total of 14,471 unique enhancers expressed across 49 cell and tissue types, identified 64 eRNAs located in the brain, representing 34 distinct eRNAs [22]. This computational evidence significantly advances our understanding by demonstrating that eRNAs are not merely transcriptional byproducts, but may actively contribute to the three-dimensional genomic architecture of the human brain. Thus, disruption of these eRNA-mediated chromatin interactions could be a novel mechanism underlying the dysregulation of gene expression networks in neurodegenerative diseases.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "Neurodegenerative disorders are a group of progressive, irreversible diseases characterized by the slow degeneration and loss of neurons in specific regions of the brain and spinal cord. These conditions are associated with disruptions in essential cellular processes, including autophagy, apoptosis, oxidative stress, neuroinflammation, protein misfolding, mitochondrial dysfunction, and NAs. These pathological alterations ultimately result in functional, cognitive, and motor impairments. The abnormal expression of genes involved in these biological processes, particularly circRNAs and eRNAs, plays important roles in neurodegenerative disorders such as AD, PD, HD, and MS [38\u201340].\n\nWhen assessing results from animal models (such as mice, rats, and C. elegans), it is important to evaluate the level of evolutionary conservation of the related ncRNAs and their target genes. A lot of circRNAs and regulatory enhancer sequences revealed important sequence and functional homology between rodents and humans, confirming the practical importance of such models [35, 41]. However, species-specific variations in ncRNA expression, splicing, and regulatory networks need exact confirmation in human cells or tissues to completely determine their pathophysiological significance in human disorders.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "Alzheimer\u2019s disease (AD) is a neurodegenerative disrder characterized by an uncertain onset and a steady progression that results in cognitive impairments in older individuals. This condition is associated with the exacerbation of amyloid \u03b2 (A\u03b2) and the hyperphosphorylation of tau, which leads to the formation of neurofibrillary tangles (NFTs) and the failure of neurons [42\u201345]. A\u03b2 is the primary component in AD. Excessive accumulation of A\u03b2 has been implicated in the induction of neuroinflammation, oxidative stress, and ionic calcium overload, which are believed to be underlying factors contributing to neuronal malfunction and subsequent cognitive disorders [46, 47].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "Circular RNAs exhibit an accumulation phenomenon in the brains of older individuals. However, it is important to note that the changes observed in circRNAs due to aging are distinct from those linked to the progression of AD [48, 48\u201353]. They also exhibit abnormal expression patterns in the brain, cerebrospinal fluid, and serum of individuals affected by AD. It has been identified that a large number of circRNAs are involved in A\u03b2 clearance, neurotrophic signaling, neuroinflammation, and oxidative stress pathways [54\u201356]. The excessive deposition of A\u03b2 in the cerebral cortex expedites the progression of AD [57]. The process of cleavage by \u03b2-secretase and \u03b3-secretase in the APP (amyloid precursor protein) produces A\u03b2 protein, which is primarily eliminated through autophagy in brain cells [54\u201356]. We summarize the circRNAs that can be effective in AD in Table 1, Supplementary Table S1, and Fig. 1. GO enrichment analysis suggested that a significant number of circRNAs differentially expressed in AD are associated with pathways involved in A\u03b2 turnover. For example, ciRS-7 represents a candidate circRNA that shows high expression levels in healthy human brain but is downregulated in brains of individuals with AD, potentially implicating it in A\u03b2-related pathways [72]. In a study conducted by Shi et al., it was found that expression of ciRS-7 inhibited the translation of NF-\u03baB and promoted its localization in the cytoplasm, thereby allowing for the derepression of UCHL1 expression, which facilitates the degradation of APP and BACE1. Furthermore, APP decreased the levels of ciRS-7, indicating a reciprocal regulatory relationship between ciRS-7 and APP. These findings elucidated a molecular mechanism that associates reduced ciRS-7 expression with AD, implying that ciRS-7 could serve as a valuable target in the formulation of therapeutic approaches for AD and suggesting a possible neuroprotective function of ciRS-7 [73]. Zhao et al. reported that a deficiency in the sponging activity of ciRS-7 leads to an excess of ambient miRNA-7, which appears to drive the selective down-regulation of miRNA-7-sensitive mRNA targets, including the one that encodes the ubiquitin conjugating enzyme E2A (UBE2A). UBE2A typically functions as a key effector in the ubiquitin-26S proteasome system, facilitating the clearance of amyloid peptides through proteolysis. ciRS-7 is reduced in the brains of individuals with sporadic Alzheimer\u2019s disease (AD), thereby contributing to the accumulation of amyloid and the development of senile plaque deposits [74]. Similarly, CircCwc27, a newly identified circRNA derived from exons, is highly conserved and expressed in abundance within neurons. It diminished the recruitment of Pur-\u03b1 (purine-rich element-binding protein A or PURA) to the promoters of genes associated with Alzheimer\u2019s disease (AD) by binding to Pur-\u03b1. This interaction led to a reduction in A\u03b2 load, helped prevent cognitive decline in AD, and resulted in the downregulation of several genes linked to AD, including APP, Mme, dopamine receptor D1 (Drd1), protein phosphatase 1 regulatory inhibitor subunit 1B (Ppp1r1b), neurotrophic tyrosine kinase receptor type 1 (Ntrk1), and LIM homeobox 8 (Lhx8) [60]. Furthermore, it is worth noting that there is another circRNA, circA\u03b2-a, that acted as a template for the biosynthesis of an A\u03b2-related protein (A\u03b2175) in HEK293 cells. Moreover, the distinctive C-terminal peptide of A\u03b2175 was detected in human brain samples. Additionally, A\u03b2175 underwent processing into A\u03b2 peptides, indicating a novel pathway for A\u03b2 generation that may provide new insights into the molecular mechanisms contributing to the development of Alzheimer\u2019s disease. This mechanism significantly differs from the traditional A\u03b2 accumulation resulting from the pathophysiological dysregulation of full-length APP proteolytic processing [75].Table 1Key circRNAs and eRNAs in Alzheimer\u2019s disease and their applications as diagnostic/prognostic biomarkersncRNAExpression ChangeCell line studyAnimal studyClinical sampleTarget/PathwayFunctionClinical PotentialReferencesciRS-7Down--Brain and retinal tissue from sporadic Alzheimer\u2019s disease patients and healthy controlsmiR-7ceRNA sponge miR-7; translational repression of UBE2A and EGFR; impaired amyloid clearance and cell signalingA reduced biomarker for AD diagnosis and a therapeutic target to restore UBE2A-dependent amyloid clearance and restore EGFR signaling to reduce \u03b2-amyloid accumulation.[58]circHDAC9DownN2a cellsAPP/PS1 model miceSerum from individuals with MCI and ADmiR-138Reduced circHDAC9 expression allows miR-138 to repress Sirt1. This enhances amyloidogenic processing of APP and increases A\u03b2 production.Therapeutic target for regulating the circHDAC9/miR-138/Sirt1 signaling pathway.[59]circCwc27UpSH-SY5Y and HEK293T cellsAPP/PS1 transgenic miceTemporal cortex (13 AD vs 8 controls); plasma (6 AD vs 6 controls)Pur-\u03b1 protein; Mme genesIncreased expression of this circRNA exacerbates A\u03b2 pathology by increasing APP expression and decreasing the expression of the A\u03b2-degrading enzyme (Neprilysin).Silencing circCwc27 is a potential therapeutic strategy.[60]circAPP (hsa_circ_0003038)DownNPCs\u2009+\u2009oligomeric tau-MSBB BM44 cortex (195 AD vs 76 Ctrl) & BU-ADRC hippocampus (79 pure AD vs 21 Ctrl); CDR 0\u20135miR-15-5p/IGF1 & SNAP25Sponges miR-15-5p to derepress IGF1 and SNAP25, supporting synaptic maintenanceEarly diagnostic biomarker; modulating circAPP\u2013miR-15-5p axis may restore synaptic health in Alzheimer\u2019s disease[61]circHOMER1Down--Knight ADRC parietal cortex (83 AD vs 13 controls); replication: MSBB BM44 cortex (40 controls vs 155 AD)-This circRNA is involved in synaptic plasticity, learning, and memory. Its reduction is associated with synaptic dysfunction, which is a core and early feature of AD.Potential biomarker for assessing synaptic deficits in AD.[48]circPSEN1Up--parietal cortex from DIAN ADAD (N\u2009=\u200917) vs sporadic AD (N\u2009=\u200959) vs controls (N\u2009=\u200910)miR-137It acts as a regulator of A\u03b2 and tau pathways. Its increased expression may contribute to pathology by sequestering protective miRNAs and is sensitive to tau-induced neuronal stress.Potential biomarker for differentiating AD subtypes and monitoring disease-related neural stress.[62]circ_0000950UpPC12 cells; Primary rat cerebral cortex neurons; HEK293T cells--miR-103, PTGS2The ceRNA sponge miR-103 promotes neuronal apoptosis, suppresses dendritic growth, and increases IL-1\u03b2, IL-6, and TNF-\u03b1 in AD cell models.A biomarker and potential therapeutic target for regulating the miR-103/PTGS2 axis to reduce neuronal apoptosis and inflammation in Alzheimer\u2019s disease.[63]circ_0003611UpHuman SH-SY5Y and SK-N-SH neurons-Serum from Alzheimer\u2019s disease patientsmiR-383-5p, KIF1BKnockdown of circ_0003611 reduces KIF1B levels by releasing miR-383-5p, preventing apoptosis, inflammation, and oxidative stress in neurons, and restoring glycolysis.circ_0003611 can be proposed as a serum biomarker and innovative therapeutic target to reduce apoptosis, oxidative stress, and neuronal inflammation in Alzheimer\u2019s disease through the miR-383-5p/KIF1B axis.[64]circ_0049472UpHuman SK-N-SH and CHP-212 neurons-CSF and serum of 20 Alzheimer\u2019s disease patients and 20 healthy controlsmiR-107, KIF1BMiR-107 sponge by circ_0049472 leads to increased KIF1B and enhances A\u03b2 neurotoxicity by enhancing apoptosis, induction of inflammatory cytokines, and oxidative stress; suppression of circ_0049472 has the opposite effect.circ_0049472 in AD CSF/serum is a potential non-invasive biomarker, and targeting the circ_0049472/miR-107/KIF1B axis could be a novel therapeutic strategy to reduce neurotoxicity in Alzheimer\u2019s disease.[65]circ_0005835UpSH-SY5Y neurons and BV2 microglia-Serum from 30 Alzheimer\u2019s disease patients and 30 healthy controlsmiR-576-3pcirc_0005835 as a ceRNA exacerbates neuroinflammation by recruiting miR-576-3p; its knockdown reduces BV2 inflammation and promotes NSC proliferation and neuronal differentiation.circ_0005835 is proposed as a biomarker and potential therapeutic target for regulating the circ_0005835/miR-576-3p axis to reduce neuronal inflammation and promote NSC neurogenesis.[66]circAXLUpSK-N-SH human neuronal cells-Serum exosomes: 32 Alzheimer\u2019s disease patients and 19 healthy controlsmiR-1306-5p, PDE4AcircAXL knockdown inhibits PDE4A and restores cAMP levels by releasing miR-1306-5p, preventing A\u03b21\u201342 neurotoxicity, apoptosis, inflammation, oxidative stress, and ER stress in neurons.Exosomal circAXL and miR-1306-5p are diagnostic biomarkers of AD, and modulation of the circAXL/miR-1306-5p/PDE4A/cAMP axis could be an innovative therapeutic target to reduce neuronal damage in Alzheimer\u2019s disease.[67]circRNA_017,963Down-SAMP8 and SAMR1 5 and 10\u2009month old mice; n\u2009=\u20093; hippocampal tissue-miR-1896/miR-1955-5p/miR-7030-3p/miR-7033-3p/miR-542-5pSponge ceRNA miR-1896/1955-5p/7030-3p/7033-3p/542-5p; regulation of autophagosome, exostosis, apoptosis, RNA transfer, and splicingBiomarker and therapeutic target of autophagy and vesicular cycling regulation in AD[68]mmu_circRNA_012,180, mmu_circRNA_013,636Down-3-month-old SAMP8 mice-mmu-miR-7231-3pceRNA sponging Top-5 miRNAs to regulate target mRNAs involved in metabolic control, neurotrophin signaling and neuronal development pathways in the hippocampusPotential biomarker and therapeutic target for restoring circRNA-miRNA networks and associated metabolic and neuronal signaling dysregulation in Alzheimer\u2019s disease.[69]circ_0002945UpSK-N-SH human neurons, human primary neurons (HPNs)-Serum from 26 Alzheimer\u2019s disease patients and 26 healthy controlsmiR-431-5p, TNFAIP1circ_0002945 acts as a ceRNA by recruiting miR-431-5p, releasing TNFAIP1 and aggravating A\u03b2-induced ER stress and apoptosis in neurons.circ_0002945 has potential as a serum biomarker for the diagnosis and prediction of AD and a therapeutic target for inhibiting the circ_0002945/miR-431-5p/TNFAIP1 axis to reduce ER stress and apoptosis.[70]AANCR (eRNA)UpAstrocytes--APOEThis eRNA, which is transcribed from an enhancer upstream of the APOE gene, increases APOE expression and induces an inflammatory phenotype in astrocytes, which contributes to the pathogenesis of AD.Potential therapeutic target for modulating APOE-mediated inflammatory responses.[71]Fig. 1The regulatory roles of circRNAs in Alzheimer\u2019s disease. This shematic highlights some key circRNAs and how they function in the context of AD. For instance, ciRS-7 plays a role in regulating A\u03b2 production and microglial activation through the miR-7/UBE2A pathway. Then there\u2019s circ-AXL, which modulates neuroinflammation and apoptosis via miR-328/BACE1. Additionally, circHOMER1 and circCORO1C are known to influence \u03b3-secretase activity and app expression. Other circRNAs, like circRIMS2, are also implicated in synaptic dysfunction, signaling pathways, and neuronal death. Together, these circRNAs form a crucial network that contributes to the pathogenesis of Alzheimer\u2019s disease", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "Key circRNAs and eRNAs in Alzheimer\u2019s disease and their applications as diagnostic/prognostic biomarkers\n\nThe regulatory roles of circRNAs in Alzheimer\u2019s disease. This shematic highlights some key circRNAs and how they function in the context of AD. For instance, ciRS-7 plays a role in regulating A\u03b2 production and microglial activation through the miR-7/UBE2A pathway. Then there\u2019s circ-AXL, which modulates neuroinflammation and apoptosis via miR-328/BACE1. Additionally, circHOMER1 and circCORO1C are known to influence \u03b3-secretase activity and app expression. Other circRNAs, like circRIMS2, are also implicated in synaptic dysfunction, signaling pathways, and neuronal death. Together, these circRNAs form a crucial network that contributes to the pathogenesis of Alzheimer\u2019s disease", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "A study conducted by Wu et al. [76] identified a novel circRNA named circAPP, which was found to be upregulated in A\u03b2-treated microglia and in the hippocampus of APP/PS1 mice. This research demonstrated the regulatory role of circAPP in microglial polarization in AD through the miR-1906/CLIC1 axis. Specifically, circAPP interacts with miR-1906, modulating its expression and consequently increasing the expression and activity of the ion channel CLIC1. In contrast, overexpression of miR-1906 was shown to modulate microglial polarization, improve AD pathology, and enhance cognitive function both in vivo and in vitro, while also reducing the expression and activity of CLIC1. Moreover, the study revealed that silencing circAPP led to reduced hippocampal levels of A\u03b2 and phosphorylated tau (p-tau), improved hippocampal synaptic structure, and ameliorated cognitive deficits resulting from neuroinflammation caused by abnormal microglial polarization in APP/PS1 mice [76].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "Neuronal loss occurs in AD because of neuroinflammation and oxidative stress. CircRNAs control these processes. The expression of circ_0002945 was shown to be increased in the serum of individuals with Alzheimer\u2019s disease, as well as in A\u03b2-induced SK-N-SH cells and human primary neurons. The suppression of circ_0002945 reduced A\u03b225\u201335-induced cell apoptosis and endoplasmic reticulum stress. Additionally, circ_0002945 was shown to regulate miR-431-5p. Furthermore, TNFAIP1 was identified as a target of miR-431-5p, and circ_0002945 acted as a competing endogenous RNA (ceRNA) to modulate the expression of TNFAIP1 through competition with miR-431-5p. Moreover, the suppression of TNFAIP1 mediated by miR-431-5p contributed to the alleviation of A\u03b225\u201335-induced cell apoptosis and endoplasmic reticulum (ER) stress [70]. ER stress is all about the accumulation of proteins that haven\u2019t folded properly, and this can happen due to various physiological and pathological reasons. Factors like increased demand for proteins, exposure to toxins, inflammatory cytokines, and the presence of mutant proteins can all throw the endoplasmic reticulum out of balance [77]. Recent research has shown that this ER stress might actually play a significant role in both the onset and progression of AD [78].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "In experimental models, overexpression of circ-AXL was associated with increased apoptosis rates and reduced neurite outgrowth, alongside an elevation of inflammatory cytokines in cellular models of Alzheimer\u2019s disease. Furthermore, circ-AXL directly interacted with miR-328, negatively regulating it while positively modulating BACE1. Additionally, miR-328 was found to directly bind to BACE1, exerting a negative regulatory effect on it. Importantly, miR-328 knockdown diminished the impact of circ-AXL knockdown on cellular models of Alzheimer\u2019s disease. Moreover, BACE1 overexpression exacerbated neuronal injury and inflammation, while also reducing the influence of miR-328 overexpression on these functions within cellular models of Alzheimer\u2019s disease [79].\n\nThe activation of microglial cells by A\u03b2 and NFTs leads to the continuous production of inflammation-related cytokines and chemokines, ultimately resulting in the death of neighboring neurons [80\u201382]. Yang et al. have shown that circ_0000950 is associated with decreased expression of miR-103 and increased expression of Ptgs2 (prostaglandin-endoperoxide synthase 2) in AD patients. In cellular models, manipulation of circ_0000950 was associated with increased neuronal death, inhibition of neural processes, and increased expression of inflammatory cytokines, including IL-1\u03b2, IL-6, and TNF-\u03b1 levels via directly sponging miR-103 [63].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "A significant downregulation of circHOMER1 has been observed in patients with AD, and it shows a direct correlation with disease severity. Moreover, the reduced expression of circHOMER1, which normally acts as an inhibitor of miR-651, leads to decreased expression of PSEN1 and PSEN2 as key AD-related genes and predicted targets of this miRNA. This, in turn, may result in reduced activity of the \u03b3-secretase complex, which is essential for amyloid beta processing [48, 48, 83, 84]. In comparison to circHOMER1, circCORO1C is a circular RNA with increased expression in AD patients, and its upregulation is directly associated with clinical traits and AD-related genes, including APP. By acting as a sponge for miR-105, circCORO1C increases APP and SNCA42 expression, which subsequently exacerbates amyloid plaque formation and synaptic dysfunction in AD subjects. In addition, circCORO1C may mitigate the pathological process of the disease by decreasing the accumulation of A\u03b2 and SNCA in neurons, thereby contributing to symptom alleviation [48].\n\nCircPSEN1 is another circular RNA with elevated expression levels in AD patients. It is involved in the regulation of genes associated with the TGF-\u03b21 and Notch signaling pathways [85], which are key contributors to AD progression [86, 87]. Furthermore, dysregulation of circPSEN1 isoforms has been reported in autosomal dominant forms of AD [62].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "In an AD mouse model, overexpression of circRIMS2 in the hippocampus reduces dendritic spine density and impairs memory [88]. It has been shown that amyloid beta enhances the stability of circRIMS2 and its ability to sponge miR-3968, leading to increased expression of UBE2K (ubiquitin-conjugating enzyme E2 K). Elevated UBE2K levels, in turn, promote increased ubiquitination and degradation of the GRIN2B protein by UBE2K, resulting in synaptic dysfunction [88\u201390]. In a recent study, Zhao et al. developed circMeta2, a computational tool specifically designed to conduct differential expression analysis focused on circRNA clusters. The authors further have demonstrated how circRNA clusters associated with AD function in a coordinated manner both within and across various brain regions, contributing to the etiology of AD by influencing miRNA-mediated gene expression. Additionally, they have identified significant circRNA clusters that correlate with the progression and severity of AD [91]. In another study, Wang et al. have reported circGigyf2 is notably conserved and exhibited the most significant net reduction specifically in the cortex of 7-month-old 5xFAD mice. The level of circGigyf2 in the cortices of AD patients was found to have a negative correlation with the severity of dementia. The authors have identified several splicing factors affected by AD that are crucial for the biogenesis of circGigyf2. They have discovered and experimentally confirmed the conserved roles of circGigyf2 in sponging miRNAs relevant to AD and RNA-binding proteins (RBPs) associated with AD, including the cleavage and polyadenylation factor 6 (CPSF6). Additionally, the downregulation of circGigyf2 in AD facilitated the silencing activities of its sponged miRNAs (miR-17-5p/93-5p) and improved the processing efficiency of polyadenylation sites for CPSF6 targets. Furthermore, the depletion of circGigyf2 in a mouse neuronal cell line led to the dysregulation of the circGigyf2-miRNA and circGigyf2-CPSF6 pathways, which intensified apoptotic responses to various insults, thereby providing strong evidence for the causative roles of circGigyf2 deficiency in neurodegeneration associated with AD [92]. Significantly, circGigyf2 is widely conserved between mouse and human, and its underexpression in the AD mouse cortex is linked with its decrease in human AD brains, indicating the clinical importance of this approach for studying circRNA failure in neurodegeneration [92].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "In another study, Zhang et al. have clarified the patterns of the circRNA-ceRNA network in the hippocampus of APP/PS1 and WT mice through deep RNA-seq analysis. Their results further enhance the existing understanding of the biology of circRNA-ceRNA, including the signaling pathways they are involved in, such as the dysregulated circRNAs in nucleotide metabolism, the cGMP-PKG signaling pathway, the cAMP signaling pathway, platelet activation, and the Hippo signaling pathway, along with their regulatory roles in the pathogenesis of Alzheimer\u2019s disease (AD). Furthermore, their findings preliminarily confirmed that the novel_circ_0003012/mmu-miR-298-3p/Smoc2 signaling axis might influence the cGMP-PKG signaling pathway. The cGMP-PKG signaling pathway plays a crucial role in preventing cell death and promoting the survival of neural cells. In mice with AD, researchers found that the levels of PKG in the hippocampus were significantly lower, indicating that the cGMP-PKG pathway could be essential in the disease\u2019s underlying processes [93]. In addition, mmu_circ_0001125 and mmu_circ_0000672 have been identified as competing endogenous RNAs (ceRNAs) for mmu-miR-351-5p, which specifically targets Cst7. Cst7, which is also referred to as cystatin F, plays a crucial role by encoding a cathepsin inhibitor. This inhibitor works within the endosomal and lysosomal compartments, helping to regulate the activity of cathepsins in the lysosomal pathway [93]. The expression of Cst7 plays a crucial role in microglia, as it helps to diminish the phagocytic capacity of activated microglia. The amount of CST7 is higher in microglia from APP/PS1 mice and in the brains of people with Alzheimer\u2019s disease. This increase in Cst7 expression may inhibit the clearance of A\u03b2 species via microglia and macrophages [94] (Fig. 1).", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "Recent findings suggest that dysregulated eRNAs may play a role in Alzheimer\u2019s disease (AD) by affecting synaptic plasticity, neuroinflammation, and amyloid beta mechanisms. eRNA research could uncover novel molecular mechanisms and therapeutic targets for AD [95] (Table 1). Transcripted eRNAs from two enhancers named Bdnf-Enh g1 and Bdnf-Enh g2 interact with brain-derived neurotrophic factor (BDNF) in the intergenic region. After BDNF was turned on, the expression of these eRNAs increased, especially in neurons that had completed their mitotic phase. In addition, they modulate the expression of BDNF in relation to the formation of dendrites [96]. BDNF is known as the gene responsible for the production of a neurotrophic factor. BDNF protein plays various significant roles in the development of the brain, including promoting the survival and differentiation of neurons as well as facilitating long-term potentiation and synaptic plasticity in the mature brain [97, 98]. The increased expression of BDNF has been observed to have neuroprotective effects in response to several stimuli, including environmental enrichment [99, 100] and the administration of antidepressant medications [97, 101]. The dysregulation of BDNF signal transduction has been shown in the brains of individuals with AD. However, the overexpression of BDNF has been shown to enhance learning and memory abilities. Aberrant expression of this protein is believed to elevate the risk of neurodegenerative illnesses [96, 102\u2013104].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "Watts et al. have identified a noncoding enhancer RNA associated with apolipoprotein E (APOE), which was designated as APOE-activating noncoding RNA, or AANCR. The pausing of RNA Polymerase II dynamically regulates the expression and functionality of this RNA. When this noncoding enhancer RNA is in its full-length form, it promotes the expression of APOE. Conversely, if this RNA is only partially transcribed, it becomes nonfunctional, resulting in the absence of APOE expression. In certain cells, the noncoding AANCR RNA does not undergo complete transcription due to the pausing of RNA Polymerase II elongation, which is caused by R-loops that are stabilized by RNA A basic sites. Upon experiencing hypertonic stress, the R-loops that inhibit transcription are resolved, allowing AANCR RNA to be transcribed into a full-length enhancer that activates APOE expression. Through genetic analysis, it was demonstrated that sequence variants within the enhancer region influence APOE expression across various cell types, including those in the hippocampus, and are linked to Alzheimer\u2019s disease [105]. The discovery of AANCR provides a mechanistic explanation for how non-coding genetic variants influence Alzheimer\u2019s disease risk, independent of the APOE \u03b54 allele. The discovery highlights enhancer RNAs as critical functional elements in the genome, where sequence changes can modulate the expression of key disease genes such as APOE, thereby opening new avenues for therapeutic interventions that target regulatory elements rather than the protein-coding gene itself.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "Moreover, the sequence variants located in the AANCR region elucidate the individual differences in APOE expression and the levels of APOE in plasma. The variants present in the AANCR region that influence APOE levels also increase the risk of developing Alzheimer\u2019s disease. Considering the degree of linkage disequilibrium between the variants in AANCR and APOE, the impact of polymorphisms in AANCR operates independently of the risk associated with the APOE \u03f54 allele. Through the regulation of APOE expression, AANCR functions as a modifier of Alzheimer\u2019s disease [105]. Furthermore, another study examining expression profiles revealed that the interaction between DSCAML1 enhancers and their target BACE1 promoter results in the excessive production of A\u03b2 peptides, which ultimately contribute to plaque formation. In turn, A\u03b2 peptides promote the hypomethylation of enhancers that influence neurogenesis and cell cycle genes, which are already prepared for activation due to the typical loss of CpH (non-CpG sites) methylation marks associated with aging. The reactivation of cell cycle genes promotes tau hyperphosphorylation, and together with A\u03b2 plaques, this leads to the development and dissemination of tangle pathology, culminating in neuronal death and cognitive symptoms associated with Alzheimer\u2019s disease [95] (Fig. 5).", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "In a recent investigation, the existence of circRNAs in the cerebrospinal fluid of individuals diagnosed with AD was assessed. The study revealed the up-regulation of 112 circRNAs and the down-regulation of 51 circRNAs [56]. In people with AD, circ-LPAR1, circ-AXL, and circGPHN were upregulated, while circ-PCCA, circ-HAUS4, circKIF18B, and circTTC39C were downregulated [56]. Also, circRNA expression in PBMCs can be utilized to discriminate between individuals with AD and non-AD individuals [55]. Researchers found that people with AD exhibited higher levels of hsa_circRNA_405619 and hsa_circRNA_000843 in their PBMCs than those without the condition. On the other hand, the PBMCs of the same AD patients showed lower levels of hsa_circRNA_100861 and hsa_circRNA_102448 [55].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "Moreover, circRNAs demonstrate significant therapeutic potential, especially during the preclinical phases of Alzheimer\u2019s disease (AD). Importantly, circRNAs may be regarded as some of the earliest viable pharmacological targets in the context of AD. Several studies have already shown that utilizing specific miRNAs as drug targets can help alleviate or treat AD in mouse models. Consequently, circRNAs are seen as promising candidates in the realm of AD treatment. Due to their direct influence on protein expression, synthetic circRNAs could be engineered to inhibit the production of proteins associated with AD, employing a therapeutic strategy similar to that used with miRNAs. Additionally, circRNAs, functioning as miRNA sponges, represent another fascinating category of molecules that warrant investigation for AD therapy. Their ability to modify the expression of miRNAs that suppress target mRNAs presents exciting opportunities for novel therapeutic approaches in Alzheimer\u2019s disease [74, 106]. For example, this upregulation is associated with cognitive decline in these models, suggesting a potential regulatory Knockdown of circCwc27 affects the nucleocytoplasmic translocation of Pur-\u03b1 and is associated with reduced APP expression and A\u03b2 production [60]. Therefore, circCwc27 appears to exert a direct regulatory influence on pathological proteins linked to AD and may represent a viable therapeutic target for AD [107]. The circular RNA associated with histone deacetylase (circHDAC9), which is expressed in the mouse hippocampus, contains multiple binding sites for miR-138. By acting as a sponge for this microRNA, circHDAC9 increases the expression of the target gene Sirtuin 1 (Sirt1), leading to elevated levels of A\u03b2 and triggering neuronal dysfunction, characterized by synaptic deficits and impairments in cognitive functions such as learning and memory [59]. Studies have shown a decreased expression level of circHDAC9 in the serum of AD patients and individuals with mild cognitive impairment [59]. Several differentially expressed circRNAs involved in the Hippo, cGMP-PKG, and cAMP signaling pathways have been identified in the hippocampus of the AD model APP/PS1 mice. These circRNAs influence processes such as axon guidance and platelet activation in neuronal synaptic systems [93, 108] (Table 1 and Supplementary Table S1).", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "The interplay between transcriptional and post-transcriptional regulation is particularly evident in Alzheimer\u2019s disease. While the AANCR eRNA drives APOE expression at the transcriptional level [105], multiple circRNAs finely regulate amyloidogenic processing post-transcriptionally. This multilayered regulation creates both vulnerability and opportunity: dysregulation at either level can initiate pathological cascades, but coordinated targeting of eRNAs and circRNAs may offer synergistic therapeutic benefits. For example, simultaneous modulation of AANCR to reduce APOE expression and targeting of circCwc27 to enhance A\u03b2 clearance could potentially provide more comprehensive disease modification than either approach alone.\n\nParkinson\u2019s disease (PD) is the second most prevalent neurodegenerative disorder among adults in the middle-aged and elderly population [109]. PD is observed in a sporadic manner in around 90 to 95% of cases, whereas the remaining instances are familial and linked to certain autosomal genes such as SNCA [110]. PD is additionally associated with mitochondrial dysfunction, increased oxidative stress, impaired apoptosis, and disorders in the ubiquitin-proteasome system [111].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "Recent research has demonstrated that circRNAs play the key roles in the development of PD. Table 2 and Supplementary Table S2 has provided a detailed summary of the key circRNAs dysregulated across PD. To find the circRNAs in the substantia nigra (SN), middle temporal gyrus (MTG), and amygdala, Hanan et al. used RNA sequencing (RNA-seq) to compare samples from PD sufferers and healthy people. Based on available evidence, circSLC8A1 exhibits higher expression levels in the SN of individuals affected by Parkinson\u2019s disease. Furthermore, subjection of SH-SY5Y brain cells to an excessive amount of paraquat (PQ) as an indicator of oxidative stress overexpressed circSLC8A in a dose-dependent manner. Administering the neuroprotective drug simvastatin along with the LRRK2 inhibitor PF-06447475 reversed the effect. Moreover, circSLC8A1 has seven binding sites for miR-128 and regulates the levels and/or activity of it. Additionally, miR-128 target genes, including SIRT1, BMI1, and AXIN1, were upregulated in the brains of the Parkinson\u2019s patients and in cultured neurons treated with PQ [112]. One of the miR-128 targets is SIRT1, which has a role in the aging process and exhibits anti-inflammatory and neuroprotective benefits in neuroinflammatory illnesses such as PD [120]. Another target of miR-128 is BMI1. Neuronal BMI1 is a negative regulator of \u03b1-synuclein. In Parkinson\u2019s disease, misfolded and aggregated \u03b1-synuclein gives rise to Lewy bodies, which are characteristic of the condition. Studies indicate that BMI1 might play a role in regulating \u03b1-synuclein levels, with certain findings suggesting that a deficiency in BMI1 could result in heightened aggregation of \u03b1-synuclein [121]. Finally, Axin1 exhibited increased expression in the hippocampal tissues and cells of the MPTP-lesioned mouse model of Parkinson\u2019s disease (PD), whereas Hif1a and miR-128-3p showed decreased expression. The rise in HIF-1\u03b1/miR-128-3p levels led to the inhibition of apoptosis in hippocampal neurons through the activation of the Wnt/\u03b2-catenin signaling pathway, which was attributed to the suppression of Axin1 in the context of PD [122]. Similarly, circSLC8A1 exhibits binding affinity towards miR-132. This particular miRNA was modified in Parkinson\u2019s Disease (PD) and plays a crucial role in regulating key mechanisms related to PD, including the maturation of dopaminergic neurons [123].Table 2Key circRNAs and eRNAs in Parkinson\u2019s disease and their applications as diagnostic/prognostic biomarkersncRNAExpression ChangeCell line studyAnimal studyClinical sampleTarget/PathwayFunctionClinical PotentialReferencescircSLC8A1UpSH-SY5Y cells-Postmortem human SN (RNA-seq: 42 PD vs 27 controls; qRT-PCR: 24 PD vs 18 controls)miR-128/Ago2 effector complexSponges miR-128 via Ago2 binding; modulates oxidative-stress-induced changes in miR-128 target mRNAs (BMI1, SIRT1, AXIN1)Potential biomarker for PD diagnosis; targetable by oxidative-stress modulators (statins, LRRK2 inhibitors); therapeutic modulation of miR-128 network via circSLC8A1[112]circDLGAP4DownSH-SY5Y & MN9D\u2009+\u2009MPP+; viabilityC57BL/6 mice\u2009+\u2009MPTP (25\u2009mg/kg\u2009\u00d7\u20097d); expression-miR-134-5p/CREBcircDLGAP4 recruits miR-134-5p, inhibiting it and facilitating the release of the transcription factor CREB. Activated CREB increases the expression of BDNF and anti-apoptotic genes, reduces mitochondrial ROS production, and enhances the flow of autophagy to clear damaged organelles.It can be measured as a non-invasive biomarker in body fluids (serum or exosomes). Also, restoring circDLGAP4 levels or inhibiting miR-134-5p is considered as a novel therapeutic strategy to activate the CREB pathway and slow the progression of neurodegeneration in Parkinson\u2019s.[113]circSAMD4AUpSH-SY5Y, WBC57BL/6J\u2009+\u2009MPTP-miR-29c-3p/AMPK mTORSponges miR-29c-3p to activate AMPK and inhibit mTOR, thereby promoting dopaminergic neuron apoptosis and autophagyServes as a potential biomarker and therapeutic target, modulating the circSAMD4A/miR-29c-3p/AMPK/mTOR axis may slow PD progression[114]circSNCAUp in MPP+ PD model; Down after PPX treatmentSH-SY5Y, WB--miR-7/SNCA axisSponges miR-7 to upregulate SNCA and pro-apoptotic proteins (CASP3, BAX, PTEN, P53) while inhibiting autophagy (\u2193 LC3B-II); PPX-mediated downregulation reduces apoptosis and restores autophagyPotential biomarker of PPX response and therapeutic target, modulating the circSNCA/miR-7/SNCA axis may decrease neuron apoptosis and enhance autophagy in PD[115]Has-circ_0070441UpSH-SY5Y--miR-626\u2009/\u2009IRS2 axiscirc_0070441 sponges miR-626, relieving its inhibition on IRS2; elevated IRS2 amplifies MPP+ -induced neuronal apoptosis and pro-inflammatory cytokine productionTargeting circ_0070441 (knockdown) or restoring miR-626 to downregulate IRS2 offers a potential neuroprotective strategy in PD models and may guide future therapeutic development[116]circTLK1Updopaminergic cellsC57BL/6 mice-miR-26a-5p/DAPK1circTLK1 binds and sequesters miR-26a-5p, preventing it from repressing DAPK1. As a result, DAPK1 levels rise, triggering apoptotic signaling, increasing cytotoxicity, and driving dopaminergic neuron injury both in vitro and in vivo.Therapeutic silencing of circTLK1 to free miR-26a-5p and downregulate DAPK1 may protect dopaminergic neurons, ameliorate motor deficits, and represent a novel disease-modifying approach in Parkinson\u2019s disease.[117]circzip-2Detected in C. elegans PD model; differential expression not quantified-Transgenic C. elegans-miR-60/zip-2/DAF-16 axiscirczip-2 sponges miR-60, relieving its repression on zip-2; increased zip-2 promotes DAF-16\u2013mediated stress response, reduces \u03b1-synuclein aggregation and extends worm lifespan in the PD modelReveals a circRNA\u2013miRNA\u2013FOXO (DAF-16) regulatory module; targeting this axis may inspire miRNA- or circRNA-based therapies to limit \u03b1-syn toxicity and bolster neuronal resilience in Parkinson\u2019s disease[111]PD-associated eRNAsUpLUHMES---Genetic variants associated with PD are enriched in active enhancer regions in differentiating neurons, suggesting that dysregulation of these elements may contribute to neuronal vulnerability in PD.Identifying the gene targets of these enhancers could reveal new therapeutic avenues.[118]SNCA risk enhancer (intron 2)CRISPR/Cas9 deletion of the 439-bp enhancer in iPSC-derived microglia reduces SNCA and MMRN1 expression at Day 40iPSC-derived microglia--SNCA & MMRN1Two PD risk SNPs are located in a microglial enhancer in the SNCA gene. Deletion of this enhancer results in reduced expression of SNCA and MMRN1 and alterations in glucose metabolism genes.This finding provides a mechanistic link between genetic variants, microglial function, and PD pathogenesis.[119]", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "Key circRNAs and eRNAs in Parkinson\u2019s disease and their applications as diagnostic/prognostic biomarkers\n\nThe SNCA gene, in addition to producing SNCA mRNA, can also generate circular RNAs. One such circRNA, known as circSNCA, originates from the proximal 3\u2032 untranslated region (3\u2032UTR) of SNCA mRNA, which contains a high-affinity binding site for miR-7. Also, when SH-SH5Y cells were subjected to pramipexole, as a dopamine agonist, the circSNCA levels were reduced, resulting in elevated miR-7 and decreased SNCA protein [115]. Elevated levels of intracellular \u03b1-synuclein, which is primarily encoded by the SNCA gene in the nucleus, have been associated with the etiology of PD and play a significant role in neurodegenerative disorders [124, 125]. Moreover, there is a correlation between the expression of this gene and the levels of miR-7 in neurons [126]. Additionally, it has been observed that miR-7 expression is decreased in the SN of individuals with PD. The reduction in miR-7 levels is also associated with the loss of dopaminergic neurons in the substantia nigra and a consequent decrease in striatal dopamine content [127].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "Similar to circSNCA, circPANK1 upregulation has been reported in mouse and cellular models of Parkinson\u2019s disease. This circRNA also acts specifically as a sponge for miR-7. Therefore, its increased expression reduces the availability of miR-7, leading to elevated expression of alpha-synuclein (SNCA) protein and, ultimately, enhanced neurodegeneration of dopaminergic neurons [128].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "Hsa_circ_0070441 is a circRNA molecule that is derived from the coding area of the SNCA gene, and its expression is elevated in SHSY5Y cells under the influence of 1-methyl-4-phenylpyridinium ions (MPP+), a neural metabolite that leads to impaired cellular respiration and dopaminergic neuron death [116]. Moreover, Hsa_circ_0070441 acts as a sponge for a microRNA named miR-626, and its knockdown has led to a significant reduction in inflammation and apoptosis following MPP+ exposure in these cells. Furthermore, it has been observed that miR-626 modulates the expression of insulin receptor substrate 2. The miR-626/insulin receptor substrate 2 axis has been proposed as a potential diagnostic biomarker for Parkinson\u2019s disease [116]. The circular RNA known as circzip-2, derived from the zip-2 gene (Zinc-regulated transporters and iron-regulated transporter-like protein-2), shows significantly reduced expression in the PD model of C. elegans. It has also been demonstrated that circzip-2 established a competitive inhibitory interaction at the cleavage site with its parent gene, zip-2. The protein encoded by zip-2 directly affects the expression level of the SNCA gene and reactive oxygen species (ROS) concentrations [111]. Moreover, circzip-2 can bind to miR-60, thereby inhibiting the progression of PD. This neuroprotective property of circzip-2 may make it a promising candidate for the development of targeted therapeutics and precise diagnostic tools for neurodegenerative diseases such as PD [111]. Despite C. elegans missing direct orthologs of several human Parkinson\u2019s disorder genes, the common miRNA pathways (like the miR-60 family) and stress response processes make it a model for showing important principles of circRNA-based neuroprotection that may increase investigation in higher organisms.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "The circular RNA derived from the SAMD4A gene, known as circSAMD4A, plays an active role in promoting both apoptosis and autophagy. Moreover, it has been shown that circSAMD4A contributes to dopaminergic neuronal apoptosis and autophagy via the AMPK/mTOR signaling pathway in human neuroblastoma cells (SH-SY5Y) treated with the neurotoxins MPTP and MPP+ [114, 129]. Additionally, the binding of microRNA-29c-3p to circSAMD4A was associated with reduced neurotoxin-induced cytotoxicity in neuronal cells. Therefore, circSAMD4A may serve as a valuable biomarker and a potential therapeutic target for PD [107].\n\nThe circular RNA associated with DLG protein (circDLGAP4), which has been reported to be downregulated in MPTP-induced mouse models of PD and in PD cellular models treated with MPP+, may contribute to the onset and progression of PD by affecting the viability of neuronal cells (SH-SY5Y and MN9D), promoting apoptosis, impairing mitochondrial function, and enhancing autophagy [113].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "The circular RNA circTLK1, which is derived from the TLK1 (Tousled-like kinase 1) gene, is a highly upregulated circRNA in MPTP-induced PD mouse models, and its knockdown in these models is associated with reduced apoptosis and cellular toxicity, along with increased cell viability. Moreover, by sequestering miR-26a-5p, circTLK1 regulates the expression of its target gene DAPK1 (Death-associated protein kinase 1), which is involved in apoptotic processes. Through this mechanism, circTLK1 contributes to the improvement of neurological deficits caused by middle cerebral artery occlusion and reperfusion in in vivo models [117]. Therefore, the circTLK1/miR-26a-5p/DAPK1 regulatory axis is considered a prominent mechanism in PD pathogenesis and may provide a foundation for the development of innovative and targeted therapeutic approaches for this disease [107].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "In a study involving 300 PD cases and 100 healthy controls, it was found that the circRNAs circARID1B and circTCONS_l2_00002816 exhibit high sensitivity for the early diagnosis of PD. Whereas, circFAM83H, circHUWE1, circARID1B, and circTCONS_l2_00002816 were identified as potential biomarkers capable of distinguishing between early and advanced stages of PD [130]. CircEps15 has also been identified in both human and animal studies [131]. Its reduced expression in the plasma of PD patients is associated with disease progression. circEps15 functions by sponging miR-24, thereby maintaining PINK1 gene expression and promoting the PINK1-PRKN\u2013dependent mitophagy pathway, which is essential for the clearance of damaged mitochondria and maintaining mitochondrial homeostasis in neurons [131]. Top circRNAs in PD development is indicated in Fig. 2.Fig. 2Illustration depicting the roles of circRNA in Parkinson\u2019s disease. It highlights how circRNAs by sponging various miRNAs and interacting with and \u03b1\u2013synuclein for regulating PD\n\nIllustration depicting the roles of circRNA in Parkinson\u2019s disease. It highlights how circRNAs by sponging various miRNAs and interacting with and \u03b1\u2013synuclein for regulating PD", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "Dysregulation of eRNAs in PD can disrupt dopaminergic neuron function and be a cause of neurodegeneration. The study of eRNAs may reveal new insights into the mechanisms of PD and its therapeutic targets (Table 2). The production of an eRNA from intron 2 of the KANSL1 gene has been linked to Polymorphisms on chromosome 17q21 in dopamine neurons that are associated with PD. The dysregulation of KANSL1 hinders the process of autophagy, leading to the promotion of memory impairment and neurodegeneration [132, 133]. Studies have shown that the intronic enhancers of the SNCA gene form clusters of enhancers in the human brain, where single-nucleotide polymorphisms (SNPs) highly associated with PD risk are enriched. These enhancer cluster regions have the potential to liberate RNA polymerase II from the promoter region of the SNCA gene, thereby exerting regulatory control over its expression [134]. One proposed idea posits that the eRNAs generated from these enhancers are capable of promoting transcription elongation by dissociating the negative elongation factor (NELF) complex. On the other hand, deletion of the intronic enhancer clusters of the SNCA gene has been shown to significantly downregulate SNCA expression in the brain by inhibiting its transcription elongation [135]. This finding highlights the potential of targeting these enhancer regions as a novel therapeutic strategy for treating neurodegenerative disorders associated with SNCA accumulation in the brain (Fig. 5).", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "Zhong et al. [130] investigated the prognostic and diagnostic applicability of circRNAs during the early and progressive stages of PD. The analysis conducted on the plasmatic circRNAs of three individuals diagnosed with PD, compared to a control group, used microarray technology. The results of this analysis revealed that two specific circRNAs, namely circ_0004381 and circ_0017204 (referred to as circARID1B and circTCONS-l200002816, respectively), exhibited the potential to accurately predict PD at an early stage of diagnosis. This predictive ability was achieved by employing risk score formulas and conducting receiver operating characteristic (ROC) curve analysis, which demonstrated relatively high sensitivity and specificity. In addition, the circRNA molecules circ_0085869, circ_0004381, circ_0017204, and circ_0090668, which are respectively referred to as circFAM83H, circARID1B, circTCONS-l2-00002816, and circHUWE1, have potential discriminatory capabilities between late-stage and early-stage PD.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "In conclusion, these findings suggest that a cell-free circRNA panel holds promise as a non-invasive biomarker for the early detection or continuous monitoring of PD progression [130]. Ravanidis et al. conducted a study to further explore the potential of circRNAs as biomarkers in a larger group of individuals with PD. The outcome of this study identified six circRNAs (MAPK9_circ_0001566, HOMER1_circ_0006916, SLAIN1_circ_0000497, DOP1B_circ_0001187, RESP1_circ_0004368, and PSEN1_circ_0003848) that are enriched in the brain and are significantly downregulated in PBMCs of PD subjects compared to healthy ones [136]. A different study suggested that circzip-2 can bind with miR-60, consequently hindering the advancement of PD. This neuroprotective characteristic of circzip-2 could position it as a potential candidate for creating targeted therapies and accurate diagnostic instruments for neurodegenerative disorders like PD [111] (Table 2 and Supplementary Table S2).\n\nHuntington\u2019s Disease (HD) is an atypical autosomal dominant neurodegenerative condition characterized by the progressive loss of cortical and striatal neurons. Individuals with Huntington\u2019s disease (HD) have expressed motor skills, memory function, and overall cognitive abilities [137]. HD is the most prevalent trinucleotide repeat disorder. The expansion of CAG repeats within exon 1 of the Huntingtin (HTT) gene has been identified as the underlying etiology of HD [138].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "As a result of CAG expansion within the first exon, a polyglutamine mutant HTT (mHTT) protein is produced, which is prone to misfolding and aggregation and has a higher tendency to interact with proteins in ways that aren\u2019t normal. The role of HTT encompasses various cellular processes, including axonal trafficking, gene regulation, disruption of the ubiquitin-proteasome system, and autophagy [139]. Dysregulation of autophagy has been implicated in the early cognitive changes observed in Huntington\u2019s disease. Therefore, it undergoes alterations and qualifies as a potential candidate for Huntington\u2019s disease treatment [140].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "circRNAs have been shown that differentially expressed in HD. Table 3 summarizes the expression patterns of key circRNAs in HD. In addition, Several circRNAs have been shown to play a role in regulating autophagy in HD; examples include circLRP1B, circDLGAP4, and circHIPK2, which inhibit miR-27a-3p, miR-134-5p, and miR-124, respectively. Emerging evidence suggests that circRNAs may influence the pathogenesis of HD through potential modulation of the autophagy process [144]. Nonetheless, to validate this hypothesis, a rigorous investigation of the role of autophagy-related circRNAs in the progression of HD is necessary. In addition, when circ-016719 is overexpressed in neurons, it acts as a sponge for miR-29c, which starts the process of autophagy. Circ-016719 stimulates the expression of MAP2K6 and BECN1 and, as a consequence, increases the rate of autophagy within the neuronal cells. The function of circ-016719 was confirmed using knockdown experiments, which effectively inhibited the process of autophagy [145].Table 3Key circRNAs and eRNAs in Huntington\u2019s disease and their applications as diagnostic/prognostic biomarkersncRNAExpression ChangeCell line studyAnimal studyClinical sampleTarget/PathwayFunctionClinical PotentialReferencescircHTT(2\u20136) (exons 2\u20136)UpSH-SY5Y, HEK293-7 HD patients, 4 controls (frontal cortex tissue)HTTBy modulating HTT nuclear localization and influencing nuclear morphology and proliferation, circHTT(2\u20136) recapitulates HD-like cellular phenotypes without altering repeat lengthUpregulation correlates with CAG burden and nuclear pathologies, suggesting utility as a biomarker of disease severity and a target to restore HTT localization[141]circHTT(2,3,4,5,6)UpST Hdh Q111/Q111 striatal cells, HEK293T and SH-SY5YzQ175 and Htt Q111/20 knock-in miceiPSC-derived terminally differentiated neurons (hTDNs) from HD patients (CAG 43\u2013180) vs controls (\u2009<\u200936)40S ribosomal subunitAlthough this circRNA is not translated in the adult mouse brain, its binding to ribosomes suggests that it may regulate the translation of other proteins. This could contribute to the imbalance of proteostasis in HD.Potential molecular modulator in HD pathogenesis.[142]Overall circRNA populationdysregulationPC12 neuron-like cellsmESCs and mNPCs derived from Htt CAG-expansion knock-in mice-Dopaminergic synapse; MAPK; long-term depressionInferred involvement in synaptic and signaling pathways; specific mechanisms unvalidatedNot yet assessed in patient samples; stability and blood detectability suggest potential as HD biomarkers[143]", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "Key circRNAs and eRNAs in Huntington\u2019s disease and their applications as diagnostic/prognostic biomarkers\n\nThe association between miR-142 and HD has been established through the circHectd1/miR-142/autophagy axis. In astrocytes, upregulated circHectd1 has been shown to enhance TIPARP expression level via the miR-142-TIPARP axis. Subsequently, elevated TIPARP levels trigger the production of LC3-II, hence promoting the activation of autophagy [146].\n\nBased on evidence, the circHTT (2\u20136), which encompasses exons 2, 3, 4, 5, and 6 of the HTT gene, has been upregulated in the brains of HD individuals. The upregulation of circHTT (2\u20136) in HEK293 and SH-SY5Y cells demonstrated no modification to the CAG repeat region, which is considered the primary cause of HD. However, this upregulation lead to a decrease in cell proliferation, nuclear size, and nuclear localization of the HTT protein. These effects were consistent with the pathophysiological abnormalities observed in HD [141].\n\nIt was demonstrated in a study that circAKAP7 specifically interacted with miR-155-5p, hence activating ATG12 as an autophagy-related gene and consequently stimulating autophagy [147]. Growing evidence have indicated the importance of autophagy in maintenance of the synapse in HD [140].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "In a study conducted by Ayyildiz and colleagues using HD mouse models and RNA sequencing (RNA-seq), it was demonstrated that although neural progenitor cells (NPCs) predominantly express more circRNAs than embryonic stem cells (ESCs), cells harboring the mutant HTT gene exhibit an aberrant set of linear and back-splicing-generated circular RNA isoforms. Notably, circRNA expression levels in NPCs were inversely correlated with the length of the CAG repeat expansion in the HTT gene, indicating HTT mutation impairs RNA circularization. Furthermore, approximately 32.7% of the differentially expressed circRNAs showed N6-methyladenosine (m6A) modifications. This suggest that mutant huntingtin may disrupt circRNA biogenesis via m6A-mediated regulatory mechanisms, particularly in NPCs [148] (Fig. 3).Fig. 3The schematic illustration demonstrates the roles of different circRNAs in Huntington\u2019s disease. circRNAs by sponging some miRNAs and influencing various genes, modulating HD\n\nThe schematic illustration demonstrates the roles of different circRNAs in Huntington\u2019s disease. circRNAs by sponging some miRNAs and influencing various genes, modulating HD", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "In HD, eRNA dysregulation can abolish transcriptional networks to lead to mutant HTT-induced neurodegeneration. eRNA dysregulation could offer new insights into HD pathogenesis and therapeutic targeting [149]. A brain-specific ultraconservative eRNA called Evf2 can help DLX and MECP2 control the homeodomain transcription factor Dlx5/6 in their intergenic enhancer regions. Evf2 mutations during the early postnatal period decrease the number of GABAergic interneurons in the hippocampus and dentate gyrus. Additionally, these mutations have been found to diminish synaptic blockage in the brains of adult individuals [150].\n\nGABA-regulated circuit dysfunction, which disrupts the balance between excitatory and inhibitory signaling, has been strongly implicated in a lot of neurological conditions, including neurodegenerative disorders such as Huntington\u2019s disease and epilepsy, neurodevelopmental disorders like autism and Tourette\u2019s syndrome, as well as other disorders, including dystonia and hepatic encephalopathy [151\u2013153] (Table 3 and Fig. 5).", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "The study of circRNAs and eRNAs in Huntington\u2019s disease, while still in its early stages, reveals unique diagnostic and therapeutic opportunities shaped by the monogenic nature of the disease. In contrast to sporadic neurological disorders, the clear genetic etiology of HD provides a clear context for interpreting ncRNA dysregulation. The inverse correlation between CAG repeat length and expression of specific circRNAs in neural progenitor cells suggests that circRNA signatures can serve as sensitive molecular readouts of \u201cgenetic burden,\u201d potentially reflecting disease susceptibility and time to progression even before the onset of symptoms [148]. Furthermore, the accumulation of circHTT variants, which mimic cellular pathologies such as aberrant nuclear localization of HTT without themselves containing expanded CAG repeats, positions them as novel biomarkers that report cellular stress responses downstream of mutant HTT protein [142].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "Regulatory networks formed by circRNAs, such as the circHectd1/miR-142/TIPARP and circAKAP7/miR-155-5p/ATG12 axes, highlight autophagy as a central pathway that is susceptible to targeted circRNA intervention [147, 154]. Modulation of these circuits could enhance clearance of mutant huntingtin protein. From a transcriptional perspective, dysfunction of GABAergic circuits, which are regulated in part by eRNAs such as Evf2, suggests a compelling strategy [150]. Several studies have conducted additional research to explore the clinical relevance of eRNAs in neurodegenerative disorders. For instance, evidence indicates that eRNA affects neuronal genes related to HD within the striatum. This suggests targeting striatal enhancers to boost eRNA expressions and improving the neural gene repression observed in HD patients [149]. In the context of neurodegenerative disorders, most of the research on how eRNAs can be used in the clinic, such as for diagnosis or targeted therapy, is still in its early stages [155] (Tables 3). Targeted epigenetic therapies or engineered transcription factors designed to stabilize the expression of key eRNAs could restore inhibitory balance and counteract the excitotoxicity observed in HD, offering a novel approach to correct the underlying transcriptional defects caused by mutant HTT.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "While MS is classically considered a demyelinating autoimmune disease, progressive neurodegeneration is recognized as the main substrate for irreversible neurological disability in its chronic stages. The pathological continuum of MS includes inflammatory demyelination and subsequent axonal scission, neuronal loss, and synaptic damage, hallmarks shared with conventional neurodegenerative disorders. This neurodegenerative component is particularly prevalent in progressive forms of MS, where clinical deterioration continues independent of overt inflammatory activity [156]. Therefore, examining circRNAs and eRNAs in the context of MS offers a unique opportunity to investigate how ncRNA networks associated with neuroinflammatory stimuli are linked to downstream neurodegenerative processes. Specifically, these regulatory RNAs may modulate: (1) glial activation states and neuroinflammatory cascades; (2) mitochondrial dysfunction and oxidative stress in neurons; (3) remyelination failure and oligodendrocyte pathology; and (4) direct mechanisms of axonal degeneration. Therefore, the investigation of circRNAs and eRNAs in MS provides important insights into the regulatory mechanisms that drive neurodegeneration in both autoimmune and proteinopathies [157].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "The clinical subgroups encompass clinically isolated syndrome (CIS), relapsing-remitting multiple sclerosis (RRMS), primary progressive multiple sclerosis (PPMS), and secondary progressive multiple sclerosis (SPMS) [158]. The predominant manifestation of the disease is RRMS, which represents approximately 80% of cases and is characterized by recurring episodes of disease exacerbation. PPMS manifests in around 5\u201315% of cases, displaying a distinct clinical profile characterized by a gradual deterioration of neurological function without the occurrence of relapses. Following a period of 10\u201315\u2009years, RRMS undergoes a transition to a secondary progressive form characterized by a slow advancement of the disease [159].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "CircRNAs have been implicated in the development and progression of autoimmune diseases, such as MS. A large number of circular RNAs in peripheral blood exhibited differential expression in the MS cohort [160, 161]. Table 4 and Supplementary Table S3 list the key CircRNAs in MS.Table 4Key circRNAs and eRNAs in multiple sclerosis and their applications as diagnostic/prognostic biomarkersncRNAExpression ChangeCell line studyAnimal studyClinical sampleTarget/PathwayFunctionClinical PotentialReferenceshsa_circ_0007990 (PGAP3)Down--PBMCs from discovery cohort (10 RR-MS vs 10 HC); replication cohorts: US (18 RR-MS vs 9 HC), Italian (19 RR-MS vs 20 HC)PGAP3 gene; GPI-anchor remodeling enzymeLikely influences immune responses via its host gene product; downregulation may dysregulate T-cell activation and innate immunity in MSValidated as a blood-based biomarker across independent cohorts; stability in PBMCs and replication of downregulation support its diagnostic/prognostic utility in MS[162]hsa_circRNA_101348; hsa_circRNA_102611; hsa_circRNA_104361Up--Discovery microarray in PBMCs: 20 RR-MS (10 relapse, 10 remission)\u2009+\u200910 HC; Validation qRT-PCR: 47 RR-MS (19 relapse, 28 remission)\u2009+\u200927 HCPredicted sponging of 15 miRNAs across the three circRNAsModulates a circRNA\u2013miRNA network that derepresses B-cell genes AK2, IKZF3, CBX5Biomarkers of relapse activity; correlates with MRI inflammatory lesions[163]circINPP4BUpNaive CD4+ T cells polarized to Th17EAE model miceCD4+ T cells from 18 RR-MS patients in relapse, remission, and 20 healthy controlsmiR-30aSponges miR-30a to derepress Th17 differentiation program, promoting IL-17A production and EAE pathologyPotential biomarker for MS relapse phase; therapeutic targeting of circINPP4B may reduce Th17-driven neuroinflammation[164]hsa_circRNA_101145; hsa_circRNA_001896Down--PBMC microarray (20 RR-MS: 10 relapse, 10 remission; 10 HC); Validation: qRT-PCR (45 RR-MS: 18 relapse, 27 remission; 27 HC)NOS1, NOPCHAP1, DSE genesThese circRNAs create a circRNA-miRNA network that mainly affects post-transcriptional regulation. The target genes are involved in BBB permeability, microglial activity, and tissue regeneration.They may provide insight into the microenvironmental changes that control the development of MS and serve as potential differential biomarkers.[165]circ_0005402Down--PBL from RR-MS patients and controls (Discovery:4 vs 4; Validation1:20 vs 18; Paired: 21; CIS: 10 vs 10)miRNA sponging (miR-1248; miR-766; 15 others); RBP binding (AGO2; EIF4A3; LIN28A; IGF2BP1; PTB)Bioinformatic prediction: miRNA sponge; RBP interaction; potential ANXA2 regulationAUC\u2009=\u20090.899 (95% CI 0.747\u20130.976); Sens\u2009=\u200994.4%; Spec\u2009=\u200975.0%[166]hsa_circ_0106803Up--PBMCs from 30 RR-MS patients vs. 30 healthy controlsPredicted binding sites for six miRNAs (incl. miR-1275, miR-149); no pathwayPredicted miRNA sponge; potential miRNA network modulatorProposed blood-based biomarker due to high stability and differential expression in MS[167]hsa_circ_0141241 (ATP8B4)Up\u2013\u2013Leukocyte RNA from MS vs HC (Discovery: 30\u2009MS\u2009vs\u200920\u2009HC; Validation: 70\u2009MS\u2009vs\u200946\u2009HC; Paired relapse/remission: 20\u2009MS)\u2013\u2013Candidate biomarker; AUC\u2009=\u20090.722; discriminates MS from HC[168]eRNAs (including HERV-encoded sequences)UpPrimary human monocytesCbx5\u2212/\u2212 mouse model18\u2009MS patients, 7 controlsIntegrator complex (INTcom); RNAPII elongationReduced activity of the Integrator complex leads to the accumulation of eRNAs. This can lead to aberrant expression of genes, including latent viral genes (HERVs) and the inflammatory gene leptin (LEP).The Integrator complex is emerging as a novel player in MS pathology and a potential therapeutic target.[169]", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "Key circRNAs and eRNAs in multiple sclerosis and their applications as diagnostic/prognostic biomarkers\n\nIt has been observed that hsa_circ_0106803, originating from the GSDMB gene, exhibited a 2.8-fold increase in expression levels among individuals diagnosed with RRMS [167]. Multiple miRNAs were identified as having multiple target sites in hsa_circ_0106803, notably including miR-1275 and miR-149, both of which exhibit differential expression in the blood of individuals with MS [170].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "miR-149 has been observed to interact with the protein ASIC1a, resulting in ASIC1a downregulation [171]. The ASIC1a gene is responsible for encoding a specific subunit of the acid-sensing ion channel. This particular subunit is found to be highly expressed in acute MS lesions, suggesting its potential involvement in neurological disease mechanisms underlying the development of this illness [112, 120, 123, 172]. Based on the available evidence, researchers have put forth the concept of a competing endogenous RNA (ceRNA) network, suggesting that hsa_circ_0106803 may have a role in influencing the development of MS via controlling the expression of ASIC1a mRNA through miR-149 [173]. miR-1275 is a microRNA coding gene that is functional and has been directly associated with multiple sclerosis. This microRNA is involved in the TGF-beta signaling pathway and has been confirmed to exhibit varying methylation statuses along with distinct expression patterns across different IDH expression glioma subtypes [174].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "Studies have shown a significant association between the expression level of the circular RNA circINPP4B and the clinical prognosis of MS. In the experimental autoimmune encephalomyelitis (EAE) mice, circINPP4B expression was markedly elevated, and this increase was directly correlated with the severity of clinical symptoms. Furthermore, circINPP4B acts as a sponge for miR-30a, a microRNA whose overexpression inhibits excessive Th17 cell differentiation, a key pathogenic factor in MS and EAE. By sequestering miR-30a, circINPP4B suppresses its anti-inflammatory effects, leading to enhanced Th17 differentiation and progression of EAE [175].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "Recent evidence indicates that the circular RNA HECW2 (circ_HECW2) plays a pivotal role in the pathogenesis of MS. Elevated expression of circ_HECW2 has been observed in both in vitro and in vivo MS models and is associated with the induction of endothelial-to-mesenchymal transition (EndoMT), a process recognized as a critical contributor to blood-brain barrier (BBB) impairment and increased permeability. Yang et al. reported that circ_HECW2 functions as a sponge for microRNA-30D (miR-30D), thereby resulting in upregulation of ATG5, a core component of the autophagy pathway, activating the NOTCH signaling pathway, and ultimately promoting LPS-induced EndoMT [176]. In addition, Dong and his team revealed that circ_HECW2 also interacts with miR-30e-5p and, through modulation of neural growth regulator 1 (NGR1) levels, inhibits endothelial cell proliferation, increases apoptosis, and enhances LPS-induced EndoMT [177].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "For the first time, a study based on transcriptomic data classified MS into three distinct molecular subtypes: MS-FCRL1 (C1), MS-BTG1 (C2), and MS-RPL38 (C3). Each of these subtypes is linked to various immune- and infection-related signaling pathways. In particular, the investigation of circRNA/miRNA regulatory networks in this study identified 18 differentially expressed circRNAs and 22 differentially expressed miRNAs implicated in subtype C3. Furthermore, EEF1D and TUBA1A were identified as central hub genes involved in neural and immune functions within this subtype, and finally, the circ_0045537/miR-196a-5p/TUBA1A axis was proposed as a potential regulatory pathway modulating disease severity in the C3 subtype [178]. Also, Lodde et al. reported a significant decrease in the expression level of hsa_circ_0018905 in patients with higher levels of disability and in progressive forms of multiple sclerosis (MS), suggesting its potential as a biomarker for MS [179] (Fig. 4).Fig. 4The schematic diagram represents the functional importance of circRNAs in multiple sclerosis and their related roles as a inhibitor, activator, and sponge by binding to miRNA\n\nThe schematic diagram represents the functional importance of circRNAs in multiple sclerosis and their related roles as a inhibitor, activator, and sponge by binding to miRNA", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "In another study, circ_0005402 and circ_0035560 from the ANXA2 gene exhibit reduced expression levels in patients with multiple sclerosis, indicating their potential as biomarkers for the disease. They discovered that circ_0005402 contains binding sites for several RNA-binding proteins, including AGO2, EIF4A3, LIN28A, IGF2BP1, and PTB. Additionally, it features two binding sites for miR-1248 and miR-766. This suggests the possibility of a collaborative regulation involving a circRNA-miRNA-mRNA axis in the pathophysiology of MS [166]. In a study by Mohammed, it was reported that ANXA2 is recognized for its involvement in the passage through the blood-brain barrier (BBB) and for enhancing miR-155 expression through the phosphorylation of STAT3 (Signal Transducer and Activator of Transcription 3). Notably, miR-155 serves as a biomarker for multiple sclerosis (MS), detectable in the bodily fluids of patients prior to the disease\u2019s onset, and has been reported to rise in expression during relapse phases [157].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "Bioinformatics research of circRNAs derived from non-coding regions of the genome linked to MS showed that these circRNAs might make people more likely to get MS [180]. Another study revealed an association between immunoglobulin (IgG) levels in the brain fluid and the presence of exosomal circRNAs in MS [181]. In addition, Zurawska et al. have recently provided a study on circRNAs that 3 circRNAs, hsa_circRNA_101348, hsa_circRNA_102611, and hsa_circRNA104361, are differentially overexpressed in individuals with RRMS in relapse and have the ability to influence B-cell activity [160]. Furthermore, these three circRNAs exhibited a significant correlation with MRI Gd-enhancing lesions, thereby suggesting their potential involvement in the activity of inflammatory diseases. Given the critical role of B cells in the pathogenesis of multiple sclerosis (MS), they have uncovered two transcripts related to B cells, namely AK2 and IKZF3 [160]. These observations were corroborated by the direct evidence of an upregulation of AK2 and IKZF3 in the peripheral blood mononuclear cells (PBMCs) of patients experiencing a relapse of MS. The AK2 gene is recognized for encoding phosphotransferase adenylate kinase 2, fulfilling the cell-specific requirements necessary for mitochondrial functions, especially in the activation of B cells and the production of antibodies. The significance of AK2 has been particularly highlighted in B cells infected with the Epstein-Barr virus, which has been extensively discussed in relation to MS as a viral infection associated with the disease. IKZF3 regulates the development of lymphoid and myeloid cells as well as immune homeostasis [160].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "While the bulk of circRNAs that were differentially expressed exhibited upregulation, it is worth noting that a subset of circRNAs showed downregulation in those affected by MS.\n\nThe results of the global hybridization experiment, which was then confirmed by PCR in a separate group of people, showed that there were two circRNAs, hsa_circRNA_101145 and hsa_circRNA_001896, with expression levels that were significantly lower in MS patients during remission and significantly associated with alleviation in disease-induced disability [165]. In people with RRMS who are in remission, reducing the amount of hsa_circRNA_101145 and hsa_circRNA_001896 could increase the function of all miRNAs that are vulnerable to being targeted by these two circRNAs. One example of the mentioned miRNAs is miR-181c, whose increased level in the cerebrospinal fluid has been linked to a higher chance that clinically isolated syndrome (CIS) will turn into confirmed MS [182].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "There is new evidence suggesting that eRNA dysregulation may be involved in MS pathogenesis via alterations of immune response and neuronal function. Understanding of eRNAs has the potential to provide new insights into MS mechanisms and therapeutic targets [183]. Several study investigated the role of super-enhancers (SEs) and enhancer RNAs (eRNAs) in autoimmune diseases. Their findings indicated that increased activity or dysfunction of individual SEs could lead to the development of various autoimmune disorders [184, 185, 185\u2013191]. However, despite growing interest in the regulatory roles of eRNAs in autoimmune diseases and neurodegenerative disorders, there is currently no direct evidence supporting the involvement of eRNAs in the pathogenesis of MS as a disease considered both autoimmune and neurodegenerative. Nevertheless, one study demonstrated that MS-associated single-nucleotide polymorphisms (SNPs) are located within vitamin D receptor (VDR) binding sites and in activated super-enhancers (SEs), especially in VDR-associated SEs (VSEs), in CD4+ T cells and monocytes [184, 188] (Table 4 and Supplementary Table S3). These disease-linked SNPs are believed to influence the regulatory function of these SEs. This spatial correspondence between MS risk SNPs and active enhancer markers strongly suggests that dysregulation of SEs and their associated eRNAs is a key component of MS pathophysiology. This hypothesis proposes that genetic risk is exerted in part by altering the transcriptional output of immune cells via enhancer elements, and provides a new perspective linking genetic susceptibility to functional dysregulation in MS, and identifies eRNAs generated from these SEs as potential biomarkers or therapeutic targets. However, it remains unclear how the presence of these SNPs affects the expression of these SE-derived eRNAs and what impact these eRNAs might have on MS pathogenesis. Thus, this area of research remains largely unexplored and requires further investigation in the future (Fig. 5).Fig. 5Mechanistic roles of eRNAs in neurodegenerative diseases", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Functional roles of circRNAs and eRNAs in neurodegenerative disorders", "text": "Mechanistic roles of eRNAs in neurodegenerative diseases", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Alzheimer's disease (AD)", "text": "Alzheimer\u2019s disease (AD) is a neurodegenerative disrder characterized by an uncertain onset and a steady progression that results in cognitive impairments in older individuals. This condition is associated with the exacerbation of amyloid \u03b2 (A\u03b2) and the hyperphosphorylation of tau, which leads to the formation of neurofibrillary tangles (NFTs) and the failure of neurons [42\u201345]. A\u03b2 is the primary component in AD. Excessive accumulation of A\u03b2 has been implicated in the induction of neuroinflammation, oxidative stress, and ionic calcium overload, which are believed to be underlying factors contributing to neuronal malfunction and subsequent cognitive disorders [46, 47].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Alzheimer's disease (AD)", "text": "Circular RNAs exhibit an accumulation phenomenon in the brains of older individuals. However, it is important to note that the changes observed in circRNAs due to aging are distinct from those linked to the progression of AD [48, 48\u201353]. They also exhibit abnormal expression patterns in the brain, cerebrospinal fluid, and serum of individuals affected by AD. It has been identified that a large number of circRNAs are involved in A\u03b2 clearance, neurotrophic signaling, neuroinflammation, and oxidative stress pathways [54\u201356]. The excessive deposition of A\u03b2 in the cerebral cortex expedites the progression of AD [57]. The process of cleavage by \u03b2-secretase and \u03b3-secretase in the APP (amyloid precursor protein) produces A\u03b2 protein, which is primarily eliminated through autophagy in brain cells [54\u201356]. We summarize the circRNAs that can be effective in AD in Table 1, Supplementary Table S1, and Fig. 1. GO enrichment analysis suggested that a significant number of circRNAs differentially expressed in AD are associated with pathways involved in A\u03b2 turnover. For example, ciRS-7 represents a candidate circRNA that shows high expression levels in healthy human brain but is downregulated in brains of individuals with AD, potentially implicating it in A\u03b2-related pathways [72]. In a study conducted by Shi et al., it was found that expression of ciRS-7 inhibited the translation of NF-\u03baB and promoted its localization in the cytoplasm, thereby allowing for the derepression of UCHL1 expression, which facilitates the degradation of APP and BACE1. Furthermore, APP decreased the levels of ciRS-7, indicating a reciprocal regulatory relationship between ciRS-7 and APP. These findings elucidated a molecular mechanism that associates reduced ciRS-7 expression with AD, implying that ciRS-7 could serve as a valuable target in the formulation of therapeutic approaches for AD and suggesting a possible neuroprotective function of ciRS-7 [73]. Zhao et al. reported that a deficiency in the sponging activity of ciRS-7 leads to an excess of ambient miRNA-7, which appears to drive the selective down-regulation of miRNA-7-sensitive mRNA targets, including the one that encodes the ubiquitin conjugating enzyme E2A (UBE2A). UBE2A typically functions as a key effector in the ubiquitin-26S proteasome system, facilitating the clearance of amyloid peptides through proteolysis. ciRS-7 is reduced in the brains of individuals with sporadic Alzheimer\u2019s disease (AD), thereby contributing to the accumulation of amyloid and the development of senile plaque deposits [74]. Similarly, CircCwc27, a newly identified circRNA derived from exons, is highly conserved and expressed in abundance within neurons. It diminished the recruitment of Pur-\u03b1 (purine-rich element-binding protein A or PURA) to the promoters of genes associated with Alzheimer\u2019s disease (AD) by binding to Pur-\u03b1. This interaction led to a reduction in A\u03b2 load, helped prevent cognitive decline in AD, and resulted in the downregulation of several genes linked to AD, including APP, Mme, dopamine receptor D1 (Drd1), protein phosphatase 1 regulatory inhibitor subunit 1B (Ppp1r1b), neurotrophic tyrosine kinase receptor type 1 (Ntrk1), and LIM homeobox 8 (Lhx8) [60]. Furthermore, it is worth noting that there is another circRNA, circA\u03b2-a, that acted as a template for the biosynthesis of an A\u03b2-related protein (A\u03b2175) in HEK293 cells. Moreover, the distinctive C-terminal peptide of A\u03b2175 was detected in human brain samples. Additionally, A\u03b2175 underwent processing into A\u03b2 peptides, indicating a novel pathway for A\u03b2 generation that may provide new insights into the molecular mechanisms contributing to the development of Alzheimer\u2019s disease. This mechanism significantly differs from the traditional A\u03b2 accumulation resulting from the pathophysiological dysregulation of full-length APP proteolytic processing [75].Table 1Key circRNAs and eRNAs in Alzheimer\u2019s disease and their applications as diagnostic/prognostic biomarkersncRNAExpression ChangeCell line studyAnimal studyClinical sampleTarget/PathwayFunctionClinical PotentialReferencesciRS-7Down--Brain and retinal tissue from sporadic Alzheimer\u2019s disease patients and healthy controlsmiR-7ceRNA sponge miR-7; translational repression of UBE2A and EGFR; impaired amyloid clearance and cell signalingA reduced biomarker for AD diagnosis and a therapeutic target to restore UBE2A-dependent amyloid clearance and restore EGFR signaling to reduce \u03b2-amyloid accumulation.[58]circHDAC9DownN2a cellsAPP/PS1 model miceSerum from individuals with MCI and ADmiR-138Reduced circHDAC9 expression allows miR-138 to repress Sirt1. This enhances amyloidogenic processing of APP and increases A\u03b2 production.Therapeutic target for regulating the circHDAC9/miR-138/Sirt1 signaling pathway.[59]circCwc27UpSH-SY5Y and HEK293T cellsAPP/PS1 transgenic miceTemporal cortex (13 AD vs 8 controls); plasma (6 AD vs 6 controls)Pur-\u03b1 protein; Mme genesIncreased expression of this circRNA exacerbates A\u03b2 pathology by increasing APP expression and decreasing the expression of the A\u03b2-degrading enzyme (Neprilysin).Silencing circCwc27 is a potential therapeutic strategy.[60]circAPP (hsa_circ_0003038)DownNPCs\u2009+\u2009oligomeric tau-MSBB BM44 cortex (195 AD vs 76 Ctrl) & BU-ADRC hippocampus (79 pure AD vs 21 Ctrl); CDR 0\u20135miR-15-5p/IGF1 & SNAP25Sponges miR-15-5p to derepress IGF1 and SNAP25, supporting synaptic maintenanceEarly diagnostic biomarker; modulating circAPP\u2013miR-15-5p axis may restore synaptic health in Alzheimer\u2019s disease[61]circHOMER1Down--Knight ADRC parietal cortex (83 AD vs 13 controls); replication: MSBB BM44 cortex (40 controls vs 155 AD)-This circRNA is involved in synaptic plasticity, learning, and memory. Its reduction is associated with synaptic dysfunction, which is a core and early feature of AD.Potential biomarker for assessing synaptic deficits in AD.[48]circPSEN1Up--parietal cortex from DIAN ADAD (N\u2009=\u200917) vs sporadic AD (N\u2009=\u200959) vs controls (N\u2009=\u200910)miR-137It acts as a regulator of A\u03b2 and tau pathways. Its increased expression may contribute to pathology by sequestering protective miRNAs and is sensitive to tau-induced neuronal stress.Potential biomarker for differentiating AD subtypes and monitoring disease-related neural stress.[62]circ_0000950UpPC12 cells; Primary rat cerebral cortex neurons; HEK293T cells--miR-103, PTGS2The ceRNA sponge miR-103 promotes neuronal apoptosis, suppresses dendritic growth, and increases IL-1\u03b2, IL-6, and TNF-\u03b1 in AD cell models.A biomarker and potential therapeutic target for regulating the miR-103/PTGS2 axis to reduce neuronal apoptosis and inflammation in Alzheimer\u2019s disease.[63]circ_0003611UpHuman SH-SY5Y and SK-N-SH neurons-Serum from Alzheimer\u2019s disease patientsmiR-383-5p, KIF1BKnockdown of circ_0003611 reduces KIF1B levels by releasing miR-383-5p, preventing apoptosis, inflammation, and oxidative stress in neurons, and restoring glycolysis.circ_0003611 can be proposed as a serum biomarker and innovative therapeutic target to reduce apoptosis, oxidative stress, and neuronal inflammation in Alzheimer\u2019s disease through the miR-383-5p/KIF1B axis.[64]circ_0049472UpHuman SK-N-SH and CHP-212 neurons-CSF and serum of 20 Alzheimer\u2019s disease patients and 20 healthy controlsmiR-107, KIF1BMiR-107 sponge by circ_0049472 leads to increased KIF1B and enhances A\u03b2 neurotoxicity by enhancing apoptosis, induction of inflammatory cytokines, and oxidative stress; suppression of circ_0049472 has the opposite effect.circ_0049472 in AD CSF/serum is a potential non-invasive biomarker, and targeting the circ_0049472/miR-107/KIF1B axis could be a novel therapeutic strategy to reduce neurotoxicity in Alzheimer\u2019s disease.[65]circ_0005835UpSH-SY5Y neurons and BV2 microglia-Serum from 30 Alzheimer\u2019s disease patients and 30 healthy controlsmiR-576-3pcirc_0005835 as a ceRNA exacerbates neuroinflammation by recruiting miR-576-3p; its knockdown reduces BV2 inflammation and promotes NSC proliferation and neuronal differentiation.circ_0005835 is proposed as a biomarker and potential therapeutic target for regulating the circ_0005835/miR-576-3p axis to reduce neuronal inflammation and promote NSC neurogenesis.[66]circAXLUpSK-N-SH human neuronal cells-Serum exosomes: 32 Alzheimer\u2019s disease patients and 19 healthy controlsmiR-1306-5p, PDE4AcircAXL knockdown inhibits PDE4A and restores cAMP levels by releasing miR-1306-5p, preventing A\u03b21\u201342 neurotoxicity, apoptosis, inflammation, oxidative stress, and ER stress in neurons.Exosomal circAXL and miR-1306-5p are diagnostic biomarkers of AD, and modulation of the circAXL/miR-1306-5p/PDE4A/cAMP axis could be an innovative therapeutic target to reduce neuronal damage in Alzheimer\u2019s disease.[67]circRNA_017,963Down-SAMP8 and SAMR1 5 and 10\u2009month old mice; n\u2009=\u20093; hippocampal tissue-miR-1896/miR-1955-5p/miR-7030-3p/miR-7033-3p/miR-542-5pSponge ceRNA miR-1896/1955-5p/7030-3p/7033-3p/542-5p; regulation of autophagosome, exostosis, apoptosis, RNA transfer, and splicingBiomarker and therapeutic target of autophagy and vesicular cycling regulation in AD[68]mmu_circRNA_012,180, mmu_circRNA_013,636Down-3-month-old SAMP8 mice-mmu-miR-7231-3pceRNA sponging Top-5 miRNAs to regulate target mRNAs involved in metabolic control, neurotrophin signaling and neuronal development pathways in the hippocampusPotential biomarker and therapeutic target for restoring circRNA-miRNA networks and associated metabolic and neuronal signaling dysregulation in Alzheimer\u2019s disease.[69]circ_0002945UpSK-N-SH human neurons, human primary neurons (HPNs)-Serum from 26 Alzheimer\u2019s disease patients and 26 healthy controlsmiR-431-5p, TNFAIP1circ_0002945 acts as a ceRNA by recruiting miR-431-5p, releasing TNFAIP1 and aggravating A\u03b2-induced ER stress and apoptosis in neurons.circ_0002945 has potential as a serum biomarker for the diagnosis and prediction of AD and a therapeutic target for inhibiting the circ_0002945/miR-431-5p/TNFAIP1 axis to reduce ER stress and apoptosis.[70]AANCR (eRNA)UpAstrocytes--APOEThis eRNA, which is transcribed from an enhancer upstream of the APOE gene, increases APOE expression and induces an inflammatory phenotype in astrocytes, which contributes to the pathogenesis of AD.Potential therapeutic target for modulating APOE-mediated inflammatory responses.[71]Fig. 1The regulatory roles of circRNAs in Alzheimer\u2019s disease. This shematic highlights some key circRNAs and how they function in the context of AD. For instance, ciRS-7 plays a role in regulating A\u03b2 production and microglial activation through the miR-7/UBE2A pathway. Then there\u2019s circ-AXL, which modulates neuroinflammation and apoptosis via miR-328/BACE1. Additionally, circHOMER1 and circCORO1C are known to influence \u03b3-secretase activity and app expression. Other circRNAs, like circRIMS2, are also implicated in synaptic dysfunction, signaling pathways, and neuronal death. Together, these circRNAs form a crucial network that contributes to the pathogenesis of Alzheimer\u2019s disease", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Alzheimer's disease (AD)", "text": "Key circRNAs and eRNAs in Alzheimer\u2019s disease and their applications as diagnostic/prognostic biomarkers\n\nThe regulatory roles of circRNAs in Alzheimer\u2019s disease. This shematic highlights some key circRNAs and how they function in the context of AD. For instance, ciRS-7 plays a role in regulating A\u03b2 production and microglial activation through the miR-7/UBE2A pathway. Then there\u2019s circ-AXL, which modulates neuroinflammation and apoptosis via miR-328/BACE1. Additionally, circHOMER1 and circCORO1C are known to influence \u03b3-secretase activity and app expression. Other circRNAs, like circRIMS2, are also implicated in synaptic dysfunction, signaling pathways, and neuronal death. Together, these circRNAs form a crucial network that contributes to the pathogenesis of Alzheimer\u2019s disease", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Alzheimer's disease (AD)", "text": "A study conducted by Wu et al. [76] identified a novel circRNA named circAPP, which was found to be upregulated in A\u03b2-treated microglia and in the hippocampus of APP/PS1 mice. This research demonstrated the regulatory role of circAPP in microglial polarization in AD through the miR-1906/CLIC1 axis. Specifically, circAPP interacts with miR-1906, modulating its expression and consequently increasing the expression and activity of the ion channel CLIC1. In contrast, overexpression of miR-1906 was shown to modulate microglial polarization, improve AD pathology, and enhance cognitive function both in vivo and in vitro, while also reducing the expression and activity of CLIC1. Moreover, the study revealed that silencing circAPP led to reduced hippocampal levels of A\u03b2 and phosphorylated tau (p-tau), improved hippocampal synaptic structure, and ameliorated cognitive deficits resulting from neuroinflammation caused by abnormal microglial polarization in APP/PS1 mice [76].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Alzheimer's disease (AD)", "text": "Neuronal loss occurs in AD because of neuroinflammation and oxidative stress. CircRNAs control these processes. The expression of circ_0002945 was shown to be increased in the serum of individuals with Alzheimer\u2019s disease, as well as in A\u03b2-induced SK-N-SH cells and human primary neurons. The suppression of circ_0002945 reduced A\u03b225\u201335-induced cell apoptosis and endoplasmic reticulum stress. Additionally, circ_0002945 was shown to regulate miR-431-5p. Furthermore, TNFAIP1 was identified as a target of miR-431-5p, and circ_0002945 acted as a competing endogenous RNA (ceRNA) to modulate the expression of TNFAIP1 through competition with miR-431-5p. Moreover, the suppression of TNFAIP1 mediated by miR-431-5p contributed to the alleviation of A\u03b225\u201335-induced cell apoptosis and endoplasmic reticulum (ER) stress [70]. ER stress is all about the accumulation of proteins that haven\u2019t folded properly, and this can happen due to various physiological and pathological reasons. Factors like increased demand for proteins, exposure to toxins, inflammatory cytokines, and the presence of mutant proteins can all throw the endoplasmic reticulum out of balance [77]. Recent research has shown that this ER stress might actually play a significant role in both the onset and progression of AD [78].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Alzheimer's disease (AD)", "text": "In experimental models, overexpression of circ-AXL was associated with increased apoptosis rates and reduced neurite outgrowth, alongside an elevation of inflammatory cytokines in cellular models of Alzheimer\u2019s disease. Furthermore, circ-AXL directly interacted with miR-328, negatively regulating it while positively modulating BACE1. Additionally, miR-328 was found to directly bind to BACE1, exerting a negative regulatory effect on it. Importantly, miR-328 knockdown diminished the impact of circ-AXL knockdown on cellular models of Alzheimer\u2019s disease. Moreover, BACE1 overexpression exacerbated neuronal injury and inflammation, while also reducing the influence of miR-328 overexpression on these functions within cellular models of Alzheimer\u2019s disease [79].\n\nThe activation of microglial cells by A\u03b2 and NFTs leads to the continuous production of inflammation-related cytokines and chemokines, ultimately resulting in the death of neighboring neurons [80\u201382]. Yang et al. have shown that circ_0000950 is associated with decreased expression of miR-103 and increased expression of Ptgs2 (prostaglandin-endoperoxide synthase 2) in AD patients. In cellular models, manipulation of circ_0000950 was associated with increased neuronal death, inhibition of neural processes, and increased expression of inflammatory cytokines, including IL-1\u03b2, IL-6, and TNF-\u03b1 levels via directly sponging miR-103 [63].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Alzheimer's disease (AD)", "text": "A significant downregulation of circHOMER1 has been observed in patients with AD, and it shows a direct correlation with disease severity. Moreover, the reduced expression of circHOMER1, which normally acts as an inhibitor of miR-651, leads to decreased expression of PSEN1 and PSEN2 as key AD-related genes and predicted targets of this miRNA. This, in turn, may result in reduced activity of the \u03b3-secretase complex, which is essential for amyloid beta processing [48, 48, 83, 84]. In comparison to circHOMER1, circCORO1C is a circular RNA with increased expression in AD patients, and its upregulation is directly associated with clinical traits and AD-related genes, including APP. By acting as a sponge for miR-105, circCORO1C increases APP and SNCA42 expression, which subsequently exacerbates amyloid plaque formation and synaptic dysfunction in AD subjects. In addition, circCORO1C may mitigate the pathological process of the disease by decreasing the accumulation of A\u03b2 and SNCA in neurons, thereby contributing to symptom alleviation [48].\n\nCircPSEN1 is another circular RNA with elevated expression levels in AD patients. It is involved in the regulation of genes associated with the TGF-\u03b21 and Notch signaling pathways [85], which are key contributors to AD progression [86, 87]. Furthermore, dysregulation of circPSEN1 isoforms has been reported in autosomal dominant forms of AD [62].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Alzheimer's disease (AD)", "text": "In an AD mouse model, overexpression of circRIMS2 in the hippocampus reduces dendritic spine density and impairs memory [88]. It has been shown that amyloid beta enhances the stability of circRIMS2 and its ability to sponge miR-3968, leading to increased expression of UBE2K (ubiquitin-conjugating enzyme E2 K). Elevated UBE2K levels, in turn, promote increased ubiquitination and degradation of the GRIN2B protein by UBE2K, resulting in synaptic dysfunction [88\u201390]. In a recent study, Zhao et al. developed circMeta2, a computational tool specifically designed to conduct differential expression analysis focused on circRNA clusters. The authors further have demonstrated how circRNA clusters associated with AD function in a coordinated manner both within and across various brain regions, contributing to the etiology of AD by influencing miRNA-mediated gene expression. Additionally, they have identified significant circRNA clusters that correlate with the progression and severity of AD [91]. In another study, Wang et al. have reported circGigyf2 is notably conserved and exhibited the most significant net reduction specifically in the cortex of 7-month-old 5xFAD mice. The level of circGigyf2 in the cortices of AD patients was found to have a negative correlation with the severity of dementia. The authors have identified several splicing factors affected by AD that are crucial for the biogenesis of circGigyf2. They have discovered and experimentally confirmed the conserved roles of circGigyf2 in sponging miRNAs relevant to AD and RNA-binding proteins (RBPs) associated with AD, including the cleavage and polyadenylation factor 6 (CPSF6). Additionally, the downregulation of circGigyf2 in AD facilitated the silencing activities of its sponged miRNAs (miR-17-5p/93-5p) and improved the processing efficiency of polyadenylation sites for CPSF6 targets. Furthermore, the depletion of circGigyf2 in a mouse neuronal cell line led to the dysregulation of the circGigyf2-miRNA and circGigyf2-CPSF6 pathways, which intensified apoptotic responses to various insults, thereby providing strong evidence for the causative roles of circGigyf2 deficiency in neurodegeneration associated with AD [92]. Significantly, circGigyf2 is widely conserved between mouse and human, and its underexpression in the AD mouse cortex is linked with its decrease in human AD brains, indicating the clinical importance of this approach for studying circRNA failure in neurodegeneration [92].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Alzheimer's disease (AD)", "text": "In another study, Zhang et al. have clarified the patterns of the circRNA-ceRNA network in the hippocampus of APP/PS1 and WT mice through deep RNA-seq analysis. Their results further enhance the existing understanding of the biology of circRNA-ceRNA, including the signaling pathways they are involved in, such as the dysregulated circRNAs in nucleotide metabolism, the cGMP-PKG signaling pathway, the cAMP signaling pathway, platelet activation, and the Hippo signaling pathway, along with their regulatory roles in the pathogenesis of Alzheimer\u2019s disease (AD). Furthermore, their findings preliminarily confirmed that the novel_circ_0003012/mmu-miR-298-3p/Smoc2 signaling axis might influence the cGMP-PKG signaling pathway. The cGMP-PKG signaling pathway plays a crucial role in preventing cell death and promoting the survival of neural cells. In mice with AD, researchers found that the levels of PKG in the hippocampus were significantly lower, indicating that the cGMP-PKG pathway could be essential in the disease\u2019s underlying processes [93]. In addition, mmu_circ_0001125 and mmu_circ_0000672 have been identified as competing endogenous RNAs (ceRNAs) for mmu-miR-351-5p, which specifically targets Cst7. Cst7, which is also referred to as cystatin F, plays a crucial role by encoding a cathepsin inhibitor. This inhibitor works within the endosomal and lysosomal compartments, helping to regulate the activity of cathepsins in the lysosomal pathway [93]. The expression of Cst7 plays a crucial role in microglia, as it helps to diminish the phagocytic capacity of activated microglia. The amount of CST7 is higher in microglia from APP/PS1 mice and in the brains of people with Alzheimer\u2019s disease. This increase in Cst7 expression may inhibit the clearance of A\u03b2 species via microglia and macrophages [94] (Fig. 1).", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Alzheimer's disease (AD)", "text": "Recent findings suggest that dysregulated eRNAs may play a role in Alzheimer\u2019s disease (AD) by affecting synaptic plasticity, neuroinflammation, and amyloid beta mechanisms. eRNA research could uncover novel molecular mechanisms and therapeutic targets for AD [95] (Table 1). Transcripted eRNAs from two enhancers named Bdnf-Enh g1 and Bdnf-Enh g2 interact with brain-derived neurotrophic factor (BDNF) in the intergenic region. After BDNF was turned on, the expression of these eRNAs increased, especially in neurons that had completed their mitotic phase. In addition, they modulate the expression of BDNF in relation to the formation of dendrites [96]. BDNF is known as the gene responsible for the production of a neurotrophic factor. BDNF protein plays various significant roles in the development of the brain, including promoting the survival and differentiation of neurons as well as facilitating long-term potentiation and synaptic plasticity in the mature brain [97, 98]. The increased expression of BDNF has been observed to have neuroprotective effects in response to several stimuli, including environmental enrichment [99, 100] and the administration of antidepressant medications [97, 101]. The dysregulation of BDNF signal transduction has been shown in the brains of individuals with AD. However, the overexpression of BDNF has been shown to enhance learning and memory abilities. Aberrant expression of this protein is believed to elevate the risk of neurodegenerative illnesses [96, 102\u2013104].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Alzheimer's disease (AD)", "text": "Watts et al. have identified a noncoding enhancer RNA associated with apolipoprotein E (APOE), which was designated as APOE-activating noncoding RNA, or AANCR. The pausing of RNA Polymerase II dynamically regulates the expression and functionality of this RNA. When this noncoding enhancer RNA is in its full-length form, it promotes the expression of APOE. Conversely, if this RNA is only partially transcribed, it becomes nonfunctional, resulting in the absence of APOE expression. In certain cells, the noncoding AANCR RNA does not undergo complete transcription due to the pausing of RNA Polymerase II elongation, which is caused by R-loops that are stabilized by RNA A basic sites. Upon experiencing hypertonic stress, the R-loops that inhibit transcription are resolved, allowing AANCR RNA to be transcribed into a full-length enhancer that activates APOE expression. Through genetic analysis, it was demonstrated that sequence variants within the enhancer region influence APOE expression across various cell types, including those in the hippocampus, and are linked to Alzheimer\u2019s disease [105]. The discovery of AANCR provides a mechanistic explanation for how non-coding genetic variants influence Alzheimer\u2019s disease risk, independent of the APOE \u03b54 allele. The discovery highlights enhancer RNAs as critical functional elements in the genome, where sequence changes can modulate the expression of key disease genes such as APOE, thereby opening new avenues for therapeutic interventions that target regulatory elements rather than the protein-coding gene itself.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Alzheimer's disease (AD)", "text": "Moreover, the sequence variants located in the AANCR region elucidate the individual differences in APOE expression and the levels of APOE in plasma. The variants present in the AANCR region that influence APOE levels also increase the risk of developing Alzheimer\u2019s disease. Considering the degree of linkage disequilibrium between the variants in AANCR and APOE, the impact of polymorphisms in AANCR operates independently of the risk associated with the APOE \u03f54 allele. Through the regulation of APOE expression, AANCR functions as a modifier of Alzheimer\u2019s disease [105]. Furthermore, another study examining expression profiles revealed that the interaction between DSCAML1 enhancers and their target BACE1 promoter results in the excessive production of A\u03b2 peptides, which ultimately contribute to plaque formation. In turn, A\u03b2 peptides promote the hypomethylation of enhancers that influence neurogenesis and cell cycle genes, which are already prepared for activation due to the typical loss of CpH (non-CpG sites) methylation marks associated with aging. The reactivation of cell cycle genes promotes tau hyperphosphorylation, and together with A\u03b2 plaques, this leads to the development and dissemination of tangle pathology, culminating in neuronal death and cognitive symptoms associated with Alzheimer\u2019s disease [95] (Fig. 5).", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Alzheimer's disease (AD)", "text": "In a recent investigation, the existence of circRNAs in the cerebrospinal fluid of individuals diagnosed with AD was assessed. The study revealed the up-regulation of 112 circRNAs and the down-regulation of 51 circRNAs [56]. In people with AD, circ-LPAR1, circ-AXL, and circGPHN were upregulated, while circ-PCCA, circ-HAUS4, circKIF18B, and circTTC39C were downregulated [56]. Also, circRNA expression in PBMCs can be utilized to discriminate between individuals with AD and non-AD individuals [55]. Researchers found that people with AD exhibited higher levels of hsa_circRNA_405619 and hsa_circRNA_000843 in their PBMCs than those without the condition. On the other hand, the PBMCs of the same AD patients showed lower levels of hsa_circRNA_100861 and hsa_circRNA_102448 [55].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Alzheimer's disease (AD)", "text": "Moreover, circRNAs demonstrate significant therapeutic potential, especially during the preclinical phases of Alzheimer\u2019s disease (AD). Importantly, circRNAs may be regarded as some of the earliest viable pharmacological targets in the context of AD. Several studies have already shown that utilizing specific miRNAs as drug targets can help alleviate or treat AD in mouse models. Consequently, circRNAs are seen as promising candidates in the realm of AD treatment. Due to their direct influence on protein expression, synthetic circRNAs could be engineered to inhibit the production of proteins associated with AD, employing a therapeutic strategy similar to that used with miRNAs. Additionally, circRNAs, functioning as miRNA sponges, represent another fascinating category of molecules that warrant investigation for AD therapy. Their ability to modify the expression of miRNAs that suppress target mRNAs presents exciting opportunities for novel therapeutic approaches in Alzheimer\u2019s disease [74, 106]. For example, this upregulation is associated with cognitive decline in these models, suggesting a potential regulatory Knockdown of circCwc27 affects the nucleocytoplasmic translocation of Pur-\u03b1 and is associated with reduced APP expression and A\u03b2 production [60]. Therefore, circCwc27 appears to exert a direct regulatory influence on pathological proteins linked to AD and may represent a viable therapeutic target for AD [107]. The circular RNA associated with histone deacetylase (circHDAC9), which is expressed in the mouse hippocampus, contains multiple binding sites for miR-138. By acting as a sponge for this microRNA, circHDAC9 increases the expression of the target gene Sirtuin 1 (Sirt1), leading to elevated levels of A\u03b2 and triggering neuronal dysfunction, characterized by synaptic deficits and impairments in cognitive functions such as learning and memory [59]. Studies have shown a decreased expression level of circHDAC9 in the serum of AD patients and individuals with mild cognitive impairment [59]. Several differentially expressed circRNAs involved in the Hippo, cGMP-PKG, and cAMP signaling pathways have been identified in the hippocampus of the AD model APP/PS1 mice. These circRNAs influence processes such as axon guidance and platelet activation in neuronal synaptic systems [93, 108] (Table 1 and Supplementary Table S1).", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Alzheimer's disease (AD)", "text": "The interplay between transcriptional and post-transcriptional regulation is particularly evident in Alzheimer\u2019s disease. While the AANCR eRNA drives APOE expression at the transcriptional level [105], multiple circRNAs finely regulate amyloidogenic processing post-transcriptionally. This multilayered regulation creates both vulnerability and opportunity: dysregulation at either level can initiate pathological cascades, but coordinated targeting of eRNAs and circRNAs may offer synergistic therapeutic benefits. For example, simultaneous modulation of AANCR to reduce APOE expression and targeting of circCwc27 to enhance A\u03b2 clearance could potentially provide more comprehensive disease modification than either approach alone.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in AD", "text": "Circular RNAs exhibit an accumulation phenomenon in the brains of older individuals. However, it is important to note that the changes observed in circRNAs due to aging are distinct from those linked to the progression of AD [48, 48\u201353]. They also exhibit abnormal expression patterns in the brain, cerebrospinal fluid, and serum of individuals affected by AD. It has been identified that a large number of circRNAs are involved in A\u03b2 clearance, neurotrophic signaling, neuroinflammation, and oxidative stress pathways [54\u201356]. The excessive deposition of A\u03b2 in the cerebral cortex expedites the progression of AD [57]. The process of cleavage by \u03b2-secretase and \u03b3-secretase in the APP (amyloid precursor protein) produces A\u03b2 protein, which is primarily eliminated through autophagy in brain cells [54\u201356]. We summarize the circRNAs that can be effective in AD in Table 1, Supplementary Table S1, and Fig. 1. GO enrichment analysis suggested that a significant number of circRNAs differentially expressed in AD are associated with pathways involved in A\u03b2 turnover. For example, ciRS-7 represents a candidate circRNA that shows high expression levels in healthy human brain but is downregulated in brains of individuals with AD, potentially implicating it in A\u03b2-related pathways [72]. In a study conducted by Shi et al., it was found that expression of ciRS-7 inhibited the translation of NF-\u03baB and promoted its localization in the cytoplasm, thereby allowing for the derepression of UCHL1 expression, which facilitates the degradation of APP and BACE1. Furthermore, APP decreased the levels of ciRS-7, indicating a reciprocal regulatory relationship between ciRS-7 and APP. These findings elucidated a molecular mechanism that associates reduced ciRS-7 expression with AD, implying that ciRS-7 could serve as a valuable target in the formulation of therapeutic approaches for AD and suggesting a possible neuroprotective function of ciRS-7 [73]. Zhao et al. reported that a deficiency in the sponging activity of ciRS-7 leads to an excess of ambient miRNA-7, which appears to drive the selective down-regulation of miRNA-7-sensitive mRNA targets, including the one that encodes the ubiquitin conjugating enzyme E2A (UBE2A). UBE2A typically functions as a key effector in the ubiquitin-26S proteasome system, facilitating the clearance of amyloid peptides through proteolysis. ciRS-7 is reduced in the brains of individuals with sporadic Alzheimer\u2019s disease (AD), thereby contributing to the accumulation of amyloid and the development of senile plaque deposits [74]. Similarly, CircCwc27, a newly identified circRNA derived from exons, is highly conserved and expressed in abundance within neurons. It diminished the recruitment of Pur-\u03b1 (purine-rich element-binding protein A or PURA) to the promoters of genes associated with Alzheimer\u2019s disease (AD) by binding to Pur-\u03b1. This interaction led to a reduction in A\u03b2 load, helped prevent cognitive decline in AD, and resulted in the downregulation of several genes linked to AD, including APP, Mme, dopamine receptor D1 (Drd1), protein phosphatase 1 regulatory inhibitor subunit 1B (Ppp1r1b), neurotrophic tyrosine kinase receptor type 1 (Ntrk1), and LIM homeobox 8 (Lhx8) [60]. Furthermore, it is worth noting that there is another circRNA, circA\u03b2-a, that acted as a template for the biosynthesis of an A\u03b2-related protein (A\u03b2175) in HEK293 cells. Moreover, the distinctive C-terminal peptide of A\u03b2175 was detected in human brain samples. Additionally, A\u03b2175 underwent processing into A\u03b2 peptides, indicating a novel pathway for A\u03b2 generation that may provide new insights into the molecular mechanisms contributing to the development of Alzheimer\u2019s disease. This mechanism significantly differs from the traditional A\u03b2 accumulation resulting from the pathophysiological dysregulation of full-length APP proteolytic processing [75].Table 1Key circRNAs and eRNAs in Alzheimer\u2019s disease and their applications as diagnostic/prognostic biomarkersncRNAExpression ChangeCell line studyAnimal studyClinical sampleTarget/PathwayFunctionClinical PotentialReferencesciRS-7Down--Brain and retinal tissue from sporadic Alzheimer\u2019s disease patients and healthy controlsmiR-7ceRNA sponge miR-7; translational repression of UBE2A and EGFR; impaired amyloid clearance and cell signalingA reduced biomarker for AD diagnosis and a therapeutic target to restore UBE2A-dependent amyloid clearance and restore EGFR signaling to reduce \u03b2-amyloid accumulation.[58]circHDAC9DownN2a cellsAPP/PS1 model miceSerum from individuals with MCI and ADmiR-138Reduced circHDAC9 expression allows miR-138 to repress Sirt1. This enhances amyloidogenic processing of APP and increases A\u03b2 production.Therapeutic target for regulating the circHDAC9/miR-138/Sirt1 signaling pathway.[59]circCwc27UpSH-SY5Y and HEK293T cellsAPP/PS1 transgenic miceTemporal cortex (13 AD vs 8 controls); plasma (6 AD vs 6 controls)Pur-\u03b1 protein; Mme genesIncreased expression of this circRNA exacerbates A\u03b2 pathology by increasing APP expression and decreasing the expression of the A\u03b2-degrading enzyme (Neprilysin).Silencing circCwc27 is a potential therapeutic strategy.[60]circAPP (hsa_circ_0003038)DownNPCs\u2009+\u2009oligomeric tau-MSBB BM44 cortex (195 AD vs 76 Ctrl) & BU-ADRC hippocampus (79 pure AD vs 21 Ctrl); CDR 0\u20135miR-15-5p/IGF1 & SNAP25Sponges miR-15-5p to derepress IGF1 and SNAP25, supporting synaptic maintenanceEarly diagnostic biomarker; modulating circAPP\u2013miR-15-5p axis may restore synaptic health in Alzheimer\u2019s disease[61]circHOMER1Down--Knight ADRC parietal cortex (83 AD vs 13 controls); replication: MSBB BM44 cortex (40 controls vs 155 AD)-This circRNA is involved in synaptic plasticity, learning, and memory. Its reduction is associated with synaptic dysfunction, which is a core and early feature of AD.Potential biomarker for assessing synaptic deficits in AD.[48]circPSEN1Up--parietal cortex from DIAN ADAD (N\u2009=\u200917) vs sporadic AD (N\u2009=\u200959) vs controls (N\u2009=\u200910)miR-137It acts as a regulator of A\u03b2 and tau pathways. Its increased expression may contribute to pathology by sequestering protective miRNAs and is sensitive to tau-induced neuronal stress.Potential biomarker for differentiating AD subtypes and monitoring disease-related neural stress.[62]circ_0000950UpPC12 cells; Primary rat cerebral cortex neurons; HEK293T cells--miR-103, PTGS2The ceRNA sponge miR-103 promotes neuronal apoptosis, suppresses dendritic growth, and increases IL-1\u03b2, IL-6, and TNF-\u03b1 in AD cell models.A biomarker and potential therapeutic target for regulating the miR-103/PTGS2 axis to reduce neuronal apoptosis and inflammation in Alzheimer\u2019s disease.[63]circ_0003611UpHuman SH-SY5Y and SK-N-SH neurons-Serum from Alzheimer\u2019s disease patientsmiR-383-5p, KIF1BKnockdown of circ_0003611 reduces KIF1B levels by releasing miR-383-5p, preventing apoptosis, inflammation, and oxidative stress in neurons, and restoring glycolysis.circ_0003611 can be proposed as a serum biomarker and innovative therapeutic target to reduce apoptosis, oxidative stress, and neuronal inflammation in Alzheimer\u2019s disease through the miR-383-5p/KIF1B axis.[64]circ_0049472UpHuman SK-N-SH and CHP-212 neurons-CSF and serum of 20 Alzheimer\u2019s disease patients and 20 healthy controlsmiR-107, KIF1BMiR-107 sponge by circ_0049472 leads to increased KIF1B and enhances A\u03b2 neurotoxicity by enhancing apoptosis, induction of inflammatory cytokines, and oxidative stress; suppression of circ_0049472 has the opposite effect.circ_0049472 in AD CSF/serum is a potential non-invasive biomarker, and targeting the circ_0049472/miR-107/KIF1B axis could be a novel therapeutic strategy to reduce neurotoxicity in Alzheimer\u2019s disease.[65]circ_0005835UpSH-SY5Y neurons and BV2 microglia-Serum from 30 Alzheimer\u2019s disease patients and 30 healthy controlsmiR-576-3pcirc_0005835 as a ceRNA exacerbates neuroinflammation by recruiting miR-576-3p; its knockdown reduces BV2 inflammation and promotes NSC proliferation and neuronal differentiation.circ_0005835 is proposed as a biomarker and potential therapeutic target for regulating the circ_0005835/miR-576-3p axis to reduce neuronal inflammation and promote NSC neurogenesis.[66]circAXLUpSK-N-SH human neuronal cells-Serum exosomes: 32 Alzheimer\u2019s disease patients and 19 healthy controlsmiR-1306-5p, PDE4AcircAXL knockdown inhibits PDE4A and restores cAMP levels by releasing miR-1306-5p, preventing A\u03b21\u201342 neurotoxicity, apoptosis, inflammation, oxidative stress, and ER stress in neurons.Exosomal circAXL and miR-1306-5p are diagnostic biomarkers of AD, and modulation of the circAXL/miR-1306-5p/PDE4A/cAMP axis could be an innovative therapeutic target to reduce neuronal damage in Alzheimer\u2019s disease.[67]circRNA_017,963Down-SAMP8 and SAMR1 5 and 10\u2009month old mice; n\u2009=\u20093; hippocampal tissue-miR-1896/miR-1955-5p/miR-7030-3p/miR-7033-3p/miR-542-5pSponge ceRNA miR-1896/1955-5p/7030-3p/7033-3p/542-5p; regulation of autophagosome, exostosis, apoptosis, RNA transfer, and splicingBiomarker and therapeutic target of autophagy and vesicular cycling regulation in AD[68]mmu_circRNA_012,180, mmu_circRNA_013,636Down-3-month-old SAMP8 mice-mmu-miR-7231-3pceRNA sponging Top-5 miRNAs to regulate target mRNAs involved in metabolic control, neurotrophin signaling and neuronal development pathways in the hippocampusPotential biomarker and therapeutic target for restoring circRNA-miRNA networks and associated metabolic and neuronal signaling dysregulation in Alzheimer\u2019s disease.[69]circ_0002945UpSK-N-SH human neurons, human primary neurons (HPNs)-Serum from 26 Alzheimer\u2019s disease patients and 26 healthy controlsmiR-431-5p, TNFAIP1circ_0002945 acts as a ceRNA by recruiting miR-431-5p, releasing TNFAIP1 and aggravating A\u03b2-induced ER stress and apoptosis in neurons.circ_0002945 has potential as a serum biomarker for the diagnosis and prediction of AD and a therapeutic target for inhibiting the circ_0002945/miR-431-5p/TNFAIP1 axis to reduce ER stress and apoptosis.[70]AANCR (eRNA)UpAstrocytes--APOEThis eRNA, which is transcribed from an enhancer upstream of the APOE gene, increases APOE expression and induces an inflammatory phenotype in astrocytes, which contributes to the pathogenesis of AD.Potential therapeutic target for modulating APOE-mediated inflammatory responses.[71]Fig. 1The regulatory roles of circRNAs in Alzheimer\u2019s disease. This shematic highlights some key circRNAs and how they function in the context of AD. For instance, ciRS-7 plays a role in regulating A\u03b2 production and microglial activation through the miR-7/UBE2A pathway. Then there\u2019s circ-AXL, which modulates neuroinflammation and apoptosis via miR-328/BACE1. Additionally, circHOMER1 and circCORO1C are known to influence \u03b3-secretase activity and app expression. Other circRNAs, like circRIMS2, are also implicated in synaptic dysfunction, signaling pathways, and neuronal death. Together, these circRNAs form a crucial network that contributes to the pathogenesis of Alzheimer\u2019s disease", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in AD", "text": "Key circRNAs and eRNAs in Alzheimer\u2019s disease and their applications as diagnostic/prognostic biomarkers\n\nThe regulatory roles of circRNAs in Alzheimer\u2019s disease. This shematic highlights some key circRNAs and how they function in the context of AD. For instance, ciRS-7 plays a role in regulating A\u03b2 production and microglial activation through the miR-7/UBE2A pathway. Then there\u2019s circ-AXL, which modulates neuroinflammation and apoptosis via miR-328/BACE1. Additionally, circHOMER1 and circCORO1C are known to influence \u03b3-secretase activity and app expression. Other circRNAs, like circRIMS2, are also implicated in synaptic dysfunction, signaling pathways, and neuronal death. Together, these circRNAs form a crucial network that contributes to the pathogenesis of Alzheimer\u2019s disease", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in AD", "text": "A study conducted by Wu et al. [76] identified a novel circRNA named circAPP, which was found to be upregulated in A\u03b2-treated microglia and in the hippocampus of APP/PS1 mice. This research demonstrated the regulatory role of circAPP in microglial polarization in AD through the miR-1906/CLIC1 axis. Specifically, circAPP interacts with miR-1906, modulating its expression and consequently increasing the expression and activity of the ion channel CLIC1. In contrast, overexpression of miR-1906 was shown to modulate microglial polarization, improve AD pathology, and enhance cognitive function both in vivo and in vitro, while also reducing the expression and activity of CLIC1. Moreover, the study revealed that silencing circAPP led to reduced hippocampal levels of A\u03b2 and phosphorylated tau (p-tau), improved hippocampal synaptic structure, and ameliorated cognitive deficits resulting from neuroinflammation caused by abnormal microglial polarization in APP/PS1 mice [76].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in AD", "text": "Neuronal loss occurs in AD because of neuroinflammation and oxidative stress. CircRNAs control these processes. The expression of circ_0002945 was shown to be increased in the serum of individuals with Alzheimer\u2019s disease, as well as in A\u03b2-induced SK-N-SH cells and human primary neurons. The suppression of circ_0002945 reduced A\u03b225\u201335-induced cell apoptosis and endoplasmic reticulum stress. Additionally, circ_0002945 was shown to regulate miR-431-5p. Furthermore, TNFAIP1 was identified as a target of miR-431-5p, and circ_0002945 acted as a competing endogenous RNA (ceRNA) to modulate the expression of TNFAIP1 through competition with miR-431-5p. Moreover, the suppression of TNFAIP1 mediated by miR-431-5p contributed to the alleviation of A\u03b225\u201335-induced cell apoptosis and endoplasmic reticulum (ER) stress [70]. ER stress is all about the accumulation of proteins that haven\u2019t folded properly, and this can happen due to various physiological and pathological reasons. Factors like increased demand for proteins, exposure to toxins, inflammatory cytokines, and the presence of mutant proteins can all throw the endoplasmic reticulum out of balance [77]. Recent research has shown that this ER stress might actually play a significant role in both the onset and progression of AD [78].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in AD", "text": "In experimental models, overexpression of circ-AXL was associated with increased apoptosis rates and reduced neurite outgrowth, alongside an elevation of inflammatory cytokines in cellular models of Alzheimer\u2019s disease. Furthermore, circ-AXL directly interacted with miR-328, negatively regulating it while positively modulating BACE1. Additionally, miR-328 was found to directly bind to BACE1, exerting a negative regulatory effect on it. Importantly, miR-328 knockdown diminished the impact of circ-AXL knockdown on cellular models of Alzheimer\u2019s disease. Moreover, BACE1 overexpression exacerbated neuronal injury and inflammation, while also reducing the influence of miR-328 overexpression on these functions within cellular models of Alzheimer\u2019s disease [79].\n\nThe activation of microglial cells by A\u03b2 and NFTs leads to the continuous production of inflammation-related cytokines and chemokines, ultimately resulting in the death of neighboring neurons [80\u201382]. Yang et al. have shown that circ_0000950 is associated with decreased expression of miR-103 and increased expression of Ptgs2 (prostaglandin-endoperoxide synthase 2) in AD patients. In cellular models, manipulation of circ_0000950 was associated with increased neuronal death, inhibition of neural processes, and increased expression of inflammatory cytokines, including IL-1\u03b2, IL-6, and TNF-\u03b1 levels via directly sponging miR-103 [63].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in AD", "text": "A significant downregulation of circHOMER1 has been observed in patients with AD, and it shows a direct correlation with disease severity. Moreover, the reduced expression of circHOMER1, which normally acts as an inhibitor of miR-651, leads to decreased expression of PSEN1 and PSEN2 as key AD-related genes and predicted targets of this miRNA. This, in turn, may result in reduced activity of the \u03b3-secretase complex, which is essential for amyloid beta processing [48, 48, 83, 84]. In comparison to circHOMER1, circCORO1C is a circular RNA with increased expression in AD patients, and its upregulation is directly associated with clinical traits and AD-related genes, including APP. By acting as a sponge for miR-105, circCORO1C increases APP and SNCA42 expression, which subsequently exacerbates amyloid plaque formation and synaptic dysfunction in AD subjects. In addition, circCORO1C may mitigate the pathological process of the disease by decreasing the accumulation of A\u03b2 and SNCA in neurons, thereby contributing to symptom alleviation [48].\n\nCircPSEN1 is another circular RNA with elevated expression levels in AD patients. It is involved in the regulation of genes associated with the TGF-\u03b21 and Notch signaling pathways [85], which are key contributors to AD progression [86, 87]. Furthermore, dysregulation of circPSEN1 isoforms has been reported in autosomal dominant forms of AD [62].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in AD", "text": "In an AD mouse model, overexpression of circRIMS2 in the hippocampus reduces dendritic spine density and impairs memory [88]. It has been shown that amyloid beta enhances the stability of circRIMS2 and its ability to sponge miR-3968, leading to increased expression of UBE2K (ubiquitin-conjugating enzyme E2 K). Elevated UBE2K levels, in turn, promote increased ubiquitination and degradation of the GRIN2B protein by UBE2K, resulting in synaptic dysfunction [88\u201390]. In a recent study, Zhao et al. developed circMeta2, a computational tool specifically designed to conduct differential expression analysis focused on circRNA clusters. The authors further have demonstrated how circRNA clusters associated with AD function in a coordinated manner both within and across various brain regions, contributing to the etiology of AD by influencing miRNA-mediated gene expression. Additionally, they have identified significant circRNA clusters that correlate with the progression and severity of AD [91]. In another study, Wang et al. have reported circGigyf2 is notably conserved and exhibited the most significant net reduction specifically in the cortex of 7-month-old 5xFAD mice. The level of circGigyf2 in the cortices of AD patients was found to have a negative correlation with the severity of dementia. The authors have identified several splicing factors affected by AD that are crucial for the biogenesis of circGigyf2. They have discovered and experimentally confirmed the conserved roles of circGigyf2 in sponging miRNAs relevant to AD and RNA-binding proteins (RBPs) associated with AD, including the cleavage and polyadenylation factor 6 (CPSF6). Additionally, the downregulation of circGigyf2 in AD facilitated the silencing activities of its sponged miRNAs (miR-17-5p/93-5p) and improved the processing efficiency of polyadenylation sites for CPSF6 targets. Furthermore, the depletion of circGigyf2 in a mouse neuronal cell line led to the dysregulation of the circGigyf2-miRNA and circGigyf2-CPSF6 pathways, which intensified apoptotic responses to various insults, thereby providing strong evidence for the causative roles of circGigyf2 deficiency in neurodegeneration associated with AD [92]. Significantly, circGigyf2 is widely conserved between mouse and human, and its underexpression in the AD mouse cortex is linked with its decrease in human AD brains, indicating the clinical importance of this approach for studying circRNA failure in neurodegeneration [92].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in AD", "text": "In another study, Zhang et al. have clarified the patterns of the circRNA-ceRNA network in the hippocampus of APP/PS1 and WT mice through deep RNA-seq analysis. Their results further enhance the existing understanding of the biology of circRNA-ceRNA, including the signaling pathways they are involved in, such as the dysregulated circRNAs in nucleotide metabolism, the cGMP-PKG signaling pathway, the cAMP signaling pathway, platelet activation, and the Hippo signaling pathway, along with their regulatory roles in the pathogenesis of Alzheimer\u2019s disease (AD). Furthermore, their findings preliminarily confirmed that the novel_circ_0003012/mmu-miR-298-3p/Smoc2 signaling axis might influence the cGMP-PKG signaling pathway. The cGMP-PKG signaling pathway plays a crucial role in preventing cell death and promoting the survival of neural cells. In mice with AD, researchers found that the levels of PKG in the hippocampus were significantly lower, indicating that the cGMP-PKG pathway could be essential in the disease\u2019s underlying processes [93]. In addition, mmu_circ_0001125 and mmu_circ_0000672 have been identified as competing endogenous RNAs (ceRNAs) for mmu-miR-351-5p, which specifically targets Cst7. Cst7, which is also referred to as cystatin F, plays a crucial role by encoding a cathepsin inhibitor. This inhibitor works within the endosomal and lysosomal compartments, helping to regulate the activity of cathepsins in the lysosomal pathway [93]. The expression of Cst7 plays a crucial role in microglia, as it helps to diminish the phagocytic capacity of activated microglia. The amount of CST7 is higher in microglia from APP/PS1 mice and in the brains of people with Alzheimer\u2019s disease. This increase in Cst7 expression may inhibit the clearance of A\u03b2 species via microglia and macrophages [94] (Fig. 1).", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "eRNAs in AD", "text": "Recent findings suggest that dysregulated eRNAs may play a role in Alzheimer\u2019s disease (AD) by affecting synaptic plasticity, neuroinflammation, and amyloid beta mechanisms. eRNA research could uncover novel molecular mechanisms and therapeutic targets for AD [95] (Table 1). Transcripted eRNAs from two enhancers named Bdnf-Enh g1 and Bdnf-Enh g2 interact with brain-derived neurotrophic factor (BDNF) in the intergenic region. After BDNF was turned on, the expression of these eRNAs increased, especially in neurons that had completed their mitotic phase. In addition, they modulate the expression of BDNF in relation to the formation of dendrites [96]. BDNF is known as the gene responsible for the production of a neurotrophic factor. BDNF protein plays various significant roles in the development of the brain, including promoting the survival and differentiation of neurons as well as facilitating long-term potentiation and synaptic plasticity in the mature brain [97, 98]. The increased expression of BDNF has been observed to have neuroprotective effects in response to several stimuli, including environmental enrichment [99, 100] and the administration of antidepressant medications [97, 101]. The dysregulation of BDNF signal transduction has been shown in the brains of individuals with AD. However, the overexpression of BDNF has been shown to enhance learning and memory abilities. Aberrant expression of this protein is believed to elevate the risk of neurodegenerative illnesses [96, 102\u2013104].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "eRNAs in AD", "text": "Watts et al. have identified a noncoding enhancer RNA associated with apolipoprotein E (APOE), which was designated as APOE-activating noncoding RNA, or AANCR. The pausing of RNA Polymerase II dynamically regulates the expression and functionality of this RNA. When this noncoding enhancer RNA is in its full-length form, it promotes the expression of APOE. Conversely, if this RNA is only partially transcribed, it becomes nonfunctional, resulting in the absence of APOE expression. In certain cells, the noncoding AANCR RNA does not undergo complete transcription due to the pausing of RNA Polymerase II elongation, which is caused by R-loops that are stabilized by RNA A basic sites. Upon experiencing hypertonic stress, the R-loops that inhibit transcription are resolved, allowing AANCR RNA to be transcribed into a full-length enhancer that activates APOE expression. Through genetic analysis, it was demonstrated that sequence variants within the enhancer region influence APOE expression across various cell types, including those in the hippocampus, and are linked to Alzheimer\u2019s disease [105]. The discovery of AANCR provides a mechanistic explanation for how non-coding genetic variants influence Alzheimer\u2019s disease risk, independent of the APOE \u03b54 allele. The discovery highlights enhancer RNAs as critical functional elements in the genome, where sequence changes can modulate the expression of key disease genes such as APOE, thereby opening new avenues for therapeutic interventions that target regulatory elements rather than the protein-coding gene itself.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "eRNAs in AD", "text": "Moreover, the sequence variants located in the AANCR region elucidate the individual differences in APOE expression and the levels of APOE in plasma. The variants present in the AANCR region that influence APOE levels also increase the risk of developing Alzheimer\u2019s disease. Considering the degree of linkage disequilibrium between the variants in AANCR and APOE, the impact of polymorphisms in AANCR operates independently of the risk associated with the APOE \u03f54 allele. Through the regulation of APOE expression, AANCR functions as a modifier of Alzheimer\u2019s disease [105]. Furthermore, another study examining expression profiles revealed that the interaction between DSCAML1 enhancers and their target BACE1 promoter results in the excessive production of A\u03b2 peptides, which ultimately contribute to plaque formation. In turn, A\u03b2 peptides promote the hypomethylation of enhancers that influence neurogenesis and cell cycle genes, which are already prepared for activation due to the typical loss of CpH (non-CpG sites) methylation marks associated with aging. The reactivation of cell cycle genes promotes tau hyperphosphorylation, and together with A\u03b2 plaques, this leads to the development and dissemination of tangle pathology, culminating in neuronal death and cognitive symptoms associated with Alzheimer\u2019s disease [95] (Fig. 5).", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Clinical implications of CircRNAs and eRNAS in AD", "text": "In a recent investigation, the existence of circRNAs in the cerebrospinal fluid of individuals diagnosed with AD was assessed. The study revealed the up-regulation of 112 circRNAs and the down-regulation of 51 circRNAs [56]. In people with AD, circ-LPAR1, circ-AXL, and circGPHN were upregulated, while circ-PCCA, circ-HAUS4, circKIF18B, and circTTC39C were downregulated [56]. Also, circRNA expression in PBMCs can be utilized to discriminate between individuals with AD and non-AD individuals [55]. Researchers found that people with AD exhibited higher levels of hsa_circRNA_405619 and hsa_circRNA_000843 in their PBMCs than those without the condition. On the other hand, the PBMCs of the same AD patients showed lower levels of hsa_circRNA_100861 and hsa_circRNA_102448 [55].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Clinical implications of CircRNAs and eRNAS in AD", "text": "Moreover, circRNAs demonstrate significant therapeutic potential, especially during the preclinical phases of Alzheimer\u2019s disease (AD). Importantly, circRNAs may be regarded as some of the earliest viable pharmacological targets in the context of AD. Several studies have already shown that utilizing specific miRNAs as drug targets can help alleviate or treat AD in mouse models. Consequently, circRNAs are seen as promising candidates in the realm of AD treatment. Due to their direct influence on protein expression, synthetic circRNAs could be engineered to inhibit the production of proteins associated with AD, employing a therapeutic strategy similar to that used with miRNAs. Additionally, circRNAs, functioning as miRNA sponges, represent another fascinating category of molecules that warrant investigation for AD therapy. Their ability to modify the expression of miRNAs that suppress target mRNAs presents exciting opportunities for novel therapeutic approaches in Alzheimer\u2019s disease [74, 106]. For example, this upregulation is associated with cognitive decline in these models, suggesting a potential regulatory Knockdown of circCwc27 affects the nucleocytoplasmic translocation of Pur-\u03b1 and is associated with reduced APP expression and A\u03b2 production [60]. Therefore, circCwc27 appears to exert a direct regulatory influence on pathological proteins linked to AD and may represent a viable therapeutic target for AD [107]. The circular RNA associated with histone deacetylase (circHDAC9), which is expressed in the mouse hippocampus, contains multiple binding sites for miR-138. By acting as a sponge for this microRNA, circHDAC9 increases the expression of the target gene Sirtuin 1 (Sirt1), leading to elevated levels of A\u03b2 and triggering neuronal dysfunction, characterized by synaptic deficits and impairments in cognitive functions such as learning and memory [59]. Studies have shown a decreased expression level of circHDAC9 in the serum of AD patients and individuals with mild cognitive impairment [59]. Several differentially expressed circRNAs involved in the Hippo, cGMP-PKG, and cAMP signaling pathways have been identified in the hippocampus of the AD model APP/PS1 mice. These circRNAs influence processes such as axon guidance and platelet activation in neuronal synaptic systems [93, 108] (Table 1 and Supplementary Table S1).", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Clinical implications of CircRNAs and eRNAS in AD", "text": "The interplay between transcriptional and post-transcriptional regulation is particularly evident in Alzheimer\u2019s disease. While the AANCR eRNA drives APOE expression at the transcriptional level [105], multiple circRNAs finely regulate amyloidogenic processing post-transcriptionally. This multilayered regulation creates both vulnerability and opportunity: dysregulation at either level can initiate pathological cascades, but coordinated targeting of eRNAs and circRNAs may offer synergistic therapeutic benefits. For example, simultaneous modulation of AANCR to reduce APOE expression and targeting of circCwc27 to enhance A\u03b2 clearance could potentially provide more comprehensive disease modification than either approach alone.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Parkinson\u2019s disease (PD)", "text": "Parkinson\u2019s disease (PD) is the second most prevalent neurodegenerative disorder among adults in the middle-aged and elderly population [109]. PD is observed in a sporadic manner in around 90 to 95% of cases, whereas the remaining instances are familial and linked to certain autosomal genes such as SNCA [110]. PD is additionally associated with mitochondrial dysfunction, increased oxidative stress, impaired apoptosis, and disorders in the ubiquitin-proteasome system [111].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Parkinson\u2019s disease (PD)", "text": "Recent research has demonstrated that circRNAs play the key roles in the development of PD. Table 2 and Supplementary Table S2 has provided a detailed summary of the key circRNAs dysregulated across PD. To find the circRNAs in the substantia nigra (SN), middle temporal gyrus (MTG), and amygdala, Hanan et al. used RNA sequencing (RNA-seq) to compare samples from PD sufferers and healthy people. Based on available evidence, circSLC8A1 exhibits higher expression levels in the SN of individuals affected by Parkinson\u2019s disease. Furthermore, subjection of SH-SY5Y brain cells to an excessive amount of paraquat (PQ) as an indicator of oxidative stress overexpressed circSLC8A in a dose-dependent manner. Administering the neuroprotective drug simvastatin along with the LRRK2 inhibitor PF-06447475 reversed the effect. Moreover, circSLC8A1 has seven binding sites for miR-128 and regulates the levels and/or activity of it. Additionally, miR-128 target genes, including SIRT1, BMI1, and AXIN1, were upregulated in the brains of the Parkinson\u2019s patients and in cultured neurons treated with PQ [112]. One of the miR-128 targets is SIRT1, which has a role in the aging process and exhibits anti-inflammatory and neuroprotective benefits in neuroinflammatory illnesses such as PD [120]. Another target of miR-128 is BMI1. Neuronal BMI1 is a negative regulator of \u03b1-synuclein. In Parkinson\u2019s disease, misfolded and aggregated \u03b1-synuclein gives rise to Lewy bodies, which are characteristic of the condition. Studies indicate that BMI1 might play a role in regulating \u03b1-synuclein levels, with certain findings suggesting that a deficiency in BMI1 could result in heightened aggregation of \u03b1-synuclein [121]. Finally, Axin1 exhibited increased expression in the hippocampal tissues and cells of the MPTP-lesioned mouse model of Parkinson\u2019s disease (PD), whereas Hif1a and miR-128-3p showed decreased expression. The rise in HIF-1\u03b1/miR-128-3p levels led to the inhibition of apoptosis in hippocampal neurons through the activation of the Wnt/\u03b2-catenin signaling pathway, which was attributed to the suppression of Axin1 in the context of PD [122]. Similarly, circSLC8A1 exhibits binding affinity towards miR-132. This particular miRNA was modified in Parkinson\u2019s Disease (PD) and plays a crucial role in regulating key mechanisms related to PD, including the maturation of dopaminergic neurons [123].Table 2Key circRNAs and eRNAs in Parkinson\u2019s disease and their applications as diagnostic/prognostic biomarkersncRNAExpression ChangeCell line studyAnimal studyClinical sampleTarget/PathwayFunctionClinical PotentialReferencescircSLC8A1UpSH-SY5Y cells-Postmortem human SN (RNA-seq: 42 PD vs 27 controls; qRT-PCR: 24 PD vs 18 controls)miR-128/Ago2 effector complexSponges miR-128 via Ago2 binding; modulates oxidative-stress-induced changes in miR-128 target mRNAs (BMI1, SIRT1, AXIN1)Potential biomarker for PD diagnosis; targetable by oxidative-stress modulators (statins, LRRK2 inhibitors); therapeutic modulation of miR-128 network via circSLC8A1[112]circDLGAP4DownSH-SY5Y & MN9D\u2009+\u2009MPP+; viabilityC57BL/6 mice\u2009+\u2009MPTP (25\u2009mg/kg\u2009\u00d7\u20097d); expression-miR-134-5p/CREBcircDLGAP4 recruits miR-134-5p, inhibiting it and facilitating the release of the transcription factor CREB. Activated CREB increases the expression of BDNF and anti-apoptotic genes, reduces mitochondrial ROS production, and enhances the flow of autophagy to clear damaged organelles.It can be measured as a non-invasive biomarker in body fluids (serum or exosomes). Also, restoring circDLGAP4 levels or inhibiting miR-134-5p is considered as a novel therapeutic strategy to activate the CREB pathway and slow the progression of neurodegeneration in Parkinson\u2019s.[113]circSAMD4AUpSH-SY5Y, WBC57BL/6J\u2009+\u2009MPTP-miR-29c-3p/AMPK mTORSponges miR-29c-3p to activate AMPK and inhibit mTOR, thereby promoting dopaminergic neuron apoptosis and autophagyServes as a potential biomarker and therapeutic target, modulating the circSAMD4A/miR-29c-3p/AMPK/mTOR axis may slow PD progression[114]circSNCAUp in MPP+ PD model; Down after PPX treatmentSH-SY5Y, WB--miR-7/SNCA axisSponges miR-7 to upregulate SNCA and pro-apoptotic proteins (CASP3, BAX, PTEN, P53) while inhibiting autophagy (\u2193 LC3B-II); PPX-mediated downregulation reduces apoptosis and restores autophagyPotential biomarker of PPX response and therapeutic target, modulating the circSNCA/miR-7/SNCA axis may decrease neuron apoptosis and enhance autophagy in PD[115]Has-circ_0070441UpSH-SY5Y--miR-626\u2009/\u2009IRS2 axiscirc_0070441 sponges miR-626, relieving its inhibition on IRS2; elevated IRS2 amplifies MPP+ -induced neuronal apoptosis and pro-inflammatory cytokine productionTargeting circ_0070441 (knockdown) or restoring miR-626 to downregulate IRS2 offers a potential neuroprotective strategy in PD models and may guide future therapeutic development[116]circTLK1Updopaminergic cellsC57BL/6 mice-miR-26a-5p/DAPK1circTLK1 binds and sequesters miR-26a-5p, preventing it from repressing DAPK1. As a result, DAPK1 levels rise, triggering apoptotic signaling, increasing cytotoxicity, and driving dopaminergic neuron injury both in vitro and in vivo.Therapeutic silencing of circTLK1 to free miR-26a-5p and downregulate DAPK1 may protect dopaminergic neurons, ameliorate motor deficits, and represent a novel disease-modifying approach in Parkinson\u2019s disease.[117]circzip-2Detected in C. elegans PD model; differential expression not quantified-Transgenic C. elegans-miR-60/zip-2/DAF-16 axiscirczip-2 sponges miR-60, relieving its repression on zip-2; increased zip-2 promotes DAF-16\u2013mediated stress response, reduces \u03b1-synuclein aggregation and extends worm lifespan in the PD modelReveals a circRNA\u2013miRNA\u2013FOXO (DAF-16) regulatory module; targeting this axis may inspire miRNA- or circRNA-based therapies to limit \u03b1-syn toxicity and bolster neuronal resilience in Parkinson\u2019s disease[111]PD-associated eRNAsUpLUHMES---Genetic variants associated with PD are enriched in active enhancer regions in differentiating neurons, suggesting that dysregulation of these elements may contribute to neuronal vulnerability in PD.Identifying the gene targets of these enhancers could reveal new therapeutic avenues.[118]SNCA risk enhancer (intron 2)CRISPR/Cas9 deletion of the 439-bp enhancer in iPSC-derived microglia reduces SNCA and MMRN1 expression at Day 40iPSC-derived microglia--SNCA & MMRN1Two PD risk SNPs are located in a microglial enhancer in the SNCA gene. Deletion of this enhancer results in reduced expression of SNCA and MMRN1 and alterations in glucose metabolism genes.This finding provides a mechanistic link between genetic variants, microglial function, and PD pathogenesis.[119]", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Parkinson\u2019s disease (PD)", "text": "Key circRNAs and eRNAs in Parkinson\u2019s disease and their applications as diagnostic/prognostic biomarkers\n\nThe SNCA gene, in addition to producing SNCA mRNA, can also generate circular RNAs. One such circRNA, known as circSNCA, originates from the proximal 3\u2032 untranslated region (3\u2032UTR) of SNCA mRNA, which contains a high-affinity binding site for miR-7. Also, when SH-SH5Y cells were subjected to pramipexole, as a dopamine agonist, the circSNCA levels were reduced, resulting in elevated miR-7 and decreased SNCA protein [115]. Elevated levels of intracellular \u03b1-synuclein, which is primarily encoded by the SNCA gene in the nucleus, have been associated with the etiology of PD and play a significant role in neurodegenerative disorders [124, 125]. Moreover, there is a correlation between the expression of this gene and the levels of miR-7 in neurons [126]. Additionally, it has been observed that miR-7 expression is decreased in the SN of individuals with PD. The reduction in miR-7 levels is also associated with the loss of dopaminergic neurons in the substantia nigra and a consequent decrease in striatal dopamine content [127].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Parkinson\u2019s disease (PD)", "text": "Similar to circSNCA, circPANK1 upregulation has been reported in mouse and cellular models of Parkinson\u2019s disease. This circRNA also acts specifically as a sponge for miR-7. Therefore, its increased expression reduces the availability of miR-7, leading to elevated expression of alpha-synuclein (SNCA) protein and, ultimately, enhanced neurodegeneration of dopaminergic neurons [128].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Parkinson\u2019s disease (PD)", "text": "Hsa_circ_0070441 is a circRNA molecule that is derived from the coding area of the SNCA gene, and its expression is elevated in SHSY5Y cells under the influence of 1-methyl-4-phenylpyridinium ions (MPP+), a neural metabolite that leads to impaired cellular respiration and dopaminergic neuron death [116]. Moreover, Hsa_circ_0070441 acts as a sponge for a microRNA named miR-626, and its knockdown has led to a significant reduction in inflammation and apoptosis following MPP+ exposure in these cells. Furthermore, it has been observed that miR-626 modulates the expression of insulin receptor substrate 2. The miR-626/insulin receptor substrate 2 axis has been proposed as a potential diagnostic biomarker for Parkinson\u2019s disease [116]. The circular RNA known as circzip-2, derived from the zip-2 gene (Zinc-regulated transporters and iron-regulated transporter-like protein-2), shows significantly reduced expression in the PD model of C. elegans. It has also been demonstrated that circzip-2 established a competitive inhibitory interaction at the cleavage site with its parent gene, zip-2. The protein encoded by zip-2 directly affects the expression level of the SNCA gene and reactive oxygen species (ROS) concentrations [111]. Moreover, circzip-2 can bind to miR-60, thereby inhibiting the progression of PD. This neuroprotective property of circzip-2 may make it a promising candidate for the development of targeted therapeutics and precise diagnostic tools for neurodegenerative diseases such as PD [111]. Despite C. elegans missing direct orthologs of several human Parkinson\u2019s disorder genes, the common miRNA pathways (like the miR-60 family) and stress response processes make it a model for showing important principles of circRNA-based neuroprotection that may increase investigation in higher organisms.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Parkinson\u2019s disease (PD)", "text": "The circular RNA derived from the SAMD4A gene, known as circSAMD4A, plays an active role in promoting both apoptosis and autophagy. Moreover, it has been shown that circSAMD4A contributes to dopaminergic neuronal apoptosis and autophagy via the AMPK/mTOR signaling pathway in human neuroblastoma cells (SH-SY5Y) treated with the neurotoxins MPTP and MPP+ [114, 129]. Additionally, the binding of microRNA-29c-3p to circSAMD4A was associated with reduced neurotoxin-induced cytotoxicity in neuronal cells. Therefore, circSAMD4A may serve as a valuable biomarker and a potential therapeutic target for PD [107].\n\nThe circular RNA associated with DLG protein (circDLGAP4), which has been reported to be downregulated in MPTP-induced mouse models of PD and in PD cellular models treated with MPP+, may contribute to the onset and progression of PD by affecting the viability of neuronal cells (SH-SY5Y and MN9D), promoting apoptosis, impairing mitochondrial function, and enhancing autophagy [113].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Parkinson\u2019s disease (PD)", "text": "The circular RNA circTLK1, which is derived from the TLK1 (Tousled-like kinase 1) gene, is a highly upregulated circRNA in MPTP-induced PD mouse models, and its knockdown in these models is associated with reduced apoptosis and cellular toxicity, along with increased cell viability. Moreover, by sequestering miR-26a-5p, circTLK1 regulates the expression of its target gene DAPK1 (Death-associated protein kinase 1), which is involved in apoptotic processes. Through this mechanism, circTLK1 contributes to the improvement of neurological deficits caused by middle cerebral artery occlusion and reperfusion in in vivo models [117]. Therefore, the circTLK1/miR-26a-5p/DAPK1 regulatory axis is considered a prominent mechanism in PD pathogenesis and may provide a foundation for the development of innovative and targeted therapeutic approaches for this disease [107].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Parkinson\u2019s disease (PD)", "text": "In a study involving 300 PD cases and 100 healthy controls, it was found that the circRNAs circARID1B and circTCONS_l2_00002816 exhibit high sensitivity for the early diagnosis of PD. Whereas, circFAM83H, circHUWE1, circARID1B, and circTCONS_l2_00002816 were identified as potential biomarkers capable of distinguishing between early and advanced stages of PD [130]. CircEps15 has also been identified in both human and animal studies [131]. Its reduced expression in the plasma of PD patients is associated with disease progression. circEps15 functions by sponging miR-24, thereby maintaining PINK1 gene expression and promoting the PINK1-PRKN\u2013dependent mitophagy pathway, which is essential for the clearance of damaged mitochondria and maintaining mitochondrial homeostasis in neurons [131]. Top circRNAs in PD development is indicated in Fig. 2.Fig. 2Illustration depicting the roles of circRNA in Parkinson\u2019s disease. It highlights how circRNAs by sponging various miRNAs and interacting with and \u03b1\u2013synuclein for regulating PD\n\nIllustration depicting the roles of circRNA in Parkinson\u2019s disease. It highlights how circRNAs by sponging various miRNAs and interacting with and \u03b1\u2013synuclein for regulating PD", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Parkinson\u2019s disease (PD)", "text": "Dysregulation of eRNAs in PD can disrupt dopaminergic neuron function and be a cause of neurodegeneration. The study of eRNAs may reveal new insights into the mechanisms of PD and its therapeutic targets (Table 2). The production of an eRNA from intron 2 of the KANSL1 gene has been linked to Polymorphisms on chromosome 17q21 in dopamine neurons that are associated with PD. The dysregulation of KANSL1 hinders the process of autophagy, leading to the promotion of memory impairment and neurodegeneration [132, 133]. Studies have shown that the intronic enhancers of the SNCA gene form clusters of enhancers in the human brain, where single-nucleotide polymorphisms (SNPs) highly associated with PD risk are enriched. These enhancer cluster regions have the potential to liberate RNA polymerase II from the promoter region of the SNCA gene, thereby exerting regulatory control over its expression [134]. One proposed idea posits that the eRNAs generated from these enhancers are capable of promoting transcription elongation by dissociating the negative elongation factor (NELF) complex. On the other hand, deletion of the intronic enhancer clusters of the SNCA gene has been shown to significantly downregulate SNCA expression in the brain by inhibiting its transcription elongation [135]. This finding highlights the potential of targeting these enhancer regions as a novel therapeutic strategy for treating neurodegenerative disorders associated with SNCA accumulation in the brain (Fig. 5).", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Parkinson\u2019s disease (PD)", "text": "Zhong et al. [130] investigated the prognostic and diagnostic applicability of circRNAs during the early and progressive stages of PD. The analysis conducted on the plasmatic circRNAs of three individuals diagnosed with PD, compared to a control group, used microarray technology. The results of this analysis revealed that two specific circRNAs, namely circ_0004381 and circ_0017204 (referred to as circARID1B and circTCONS-l200002816, respectively), exhibited the potential to accurately predict PD at an early stage of diagnosis. This predictive ability was achieved by employing risk score formulas and conducting receiver operating characteristic (ROC) curve analysis, which demonstrated relatively high sensitivity and specificity. In addition, the circRNA molecules circ_0085869, circ_0004381, circ_0017204, and circ_0090668, which are respectively referred to as circFAM83H, circARID1B, circTCONS-l2-00002816, and circHUWE1, have potential discriminatory capabilities between late-stage and early-stage PD.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Parkinson\u2019s disease (PD)", "text": "In conclusion, these findings suggest that a cell-free circRNA panel holds promise as a non-invasive biomarker for the early detection or continuous monitoring of PD progression [130]. Ravanidis et al. conducted a study to further explore the potential of circRNAs as biomarkers in a larger group of individuals with PD. The outcome of this study identified six circRNAs (MAPK9_circ_0001566, HOMER1_circ_0006916, SLAIN1_circ_0000497, DOP1B_circ_0001187, RESP1_circ_0004368, and PSEN1_circ_0003848) that are enriched in the brain and are significantly downregulated in PBMCs of PD subjects compared to healthy ones [136]. A different study suggested that circzip-2 can bind with miR-60, consequently hindering the advancement of PD. This neuroprotective characteristic of circzip-2 could position it as a potential candidate for creating targeted therapies and accurate diagnostic instruments for neurodegenerative disorders like PD [111] (Table 2 and Supplementary Table S2).", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in PD", "text": "Recent research has demonstrated that circRNAs play the key roles in the development of PD. Table 2 and Supplementary Table S2 has provided a detailed summary of the key circRNAs dysregulated across PD. To find the circRNAs in the substantia nigra (SN), middle temporal gyrus (MTG), and amygdala, Hanan et al. used RNA sequencing (RNA-seq) to compare samples from PD sufferers and healthy people. Based on available evidence, circSLC8A1 exhibits higher expression levels in the SN of individuals affected by Parkinson\u2019s disease. Furthermore, subjection of SH-SY5Y brain cells to an excessive amount of paraquat (PQ) as an indicator of oxidative stress overexpressed circSLC8A in a dose-dependent manner. Administering the neuroprotective drug simvastatin along with the LRRK2 inhibitor PF-06447475 reversed the effect. Moreover, circSLC8A1 has seven binding sites for miR-128 and regulates the levels and/or activity of it. Additionally, miR-128 target genes, including SIRT1, BMI1, and AXIN1, were upregulated in the brains of the Parkinson\u2019s patients and in cultured neurons treated with PQ [112]. One of the miR-128 targets is SIRT1, which has a role in the aging process and exhibits anti-inflammatory and neuroprotective benefits in neuroinflammatory illnesses such as PD [120]. Another target of miR-128 is BMI1. Neuronal BMI1 is a negative regulator of \u03b1-synuclein. In Parkinson\u2019s disease, misfolded and aggregated \u03b1-synuclein gives rise to Lewy bodies, which are characteristic of the condition. Studies indicate that BMI1 might play a role in regulating \u03b1-synuclein levels, with certain findings suggesting that a deficiency in BMI1 could result in heightened aggregation of \u03b1-synuclein [121]. Finally, Axin1 exhibited increased expression in the hippocampal tissues and cells of the MPTP-lesioned mouse model of Parkinson\u2019s disease (PD), whereas Hif1a and miR-128-3p showed decreased expression. The rise in HIF-1\u03b1/miR-128-3p levels led to the inhibition of apoptosis in hippocampal neurons through the activation of the Wnt/\u03b2-catenin signaling pathway, which was attributed to the suppression of Axin1 in the context of PD [122]. Similarly, circSLC8A1 exhibits binding affinity towards miR-132. This particular miRNA was modified in Parkinson\u2019s Disease (PD) and plays a crucial role in regulating key mechanisms related to PD, including the maturation of dopaminergic neurons [123].Table 2Key circRNAs and eRNAs in Parkinson\u2019s disease and their applications as diagnostic/prognostic biomarkersncRNAExpression ChangeCell line studyAnimal studyClinical sampleTarget/PathwayFunctionClinical PotentialReferencescircSLC8A1UpSH-SY5Y cells-Postmortem human SN (RNA-seq: 42 PD vs 27 controls; qRT-PCR: 24 PD vs 18 controls)miR-128/Ago2 effector complexSponges miR-128 via Ago2 binding; modulates oxidative-stress-induced changes in miR-128 target mRNAs (BMI1, SIRT1, AXIN1)Potential biomarker for PD diagnosis; targetable by oxidative-stress modulators (statins, LRRK2 inhibitors); therapeutic modulation of miR-128 network via circSLC8A1[112]circDLGAP4DownSH-SY5Y & MN9D\u2009+\u2009MPP+; viabilityC57BL/6 mice\u2009+\u2009MPTP (25\u2009mg/kg\u2009\u00d7\u20097d); expression-miR-134-5p/CREBcircDLGAP4 recruits miR-134-5p, inhibiting it and facilitating the release of the transcription factor CREB. Activated CREB increases the expression of BDNF and anti-apoptotic genes, reduces mitochondrial ROS production, and enhances the flow of autophagy to clear damaged organelles.It can be measured as a non-invasive biomarker in body fluids (serum or exosomes). Also, restoring circDLGAP4 levels or inhibiting miR-134-5p is considered as a novel therapeutic strategy to activate the CREB pathway and slow the progression of neurodegeneration in Parkinson\u2019s.[113]circSAMD4AUpSH-SY5Y, WBC57BL/6J\u2009+\u2009MPTP-miR-29c-3p/AMPK mTORSponges miR-29c-3p to activate AMPK and inhibit mTOR, thereby promoting dopaminergic neuron apoptosis and autophagyServes as a potential biomarker and therapeutic target, modulating the circSAMD4A/miR-29c-3p/AMPK/mTOR axis may slow PD progression[114]circSNCAUp in MPP+ PD model; Down after PPX treatmentSH-SY5Y, WB--miR-7/SNCA axisSponges miR-7 to upregulate SNCA and pro-apoptotic proteins (CASP3, BAX, PTEN, P53) while inhibiting autophagy (\u2193 LC3B-II); PPX-mediated downregulation reduces apoptosis and restores autophagyPotential biomarker of PPX response and therapeutic target, modulating the circSNCA/miR-7/SNCA axis may decrease neuron apoptosis and enhance autophagy in PD[115]Has-circ_0070441UpSH-SY5Y--miR-626\u2009/\u2009IRS2 axiscirc_0070441 sponges miR-626, relieving its inhibition on IRS2; elevated IRS2 amplifies MPP+ -induced neuronal apoptosis and pro-inflammatory cytokine productionTargeting circ_0070441 (knockdown) or restoring miR-626 to downregulate IRS2 offers a potential neuroprotective strategy in PD models and may guide future therapeutic development[116]circTLK1Updopaminergic cellsC57BL/6 mice-miR-26a-5p/DAPK1circTLK1 binds and sequesters miR-26a-5p, preventing it from repressing DAPK1. As a result, DAPK1 levels rise, triggering apoptotic signaling, increasing cytotoxicity, and driving dopaminergic neuron injury both in vitro and in vivo.Therapeutic silencing of circTLK1 to free miR-26a-5p and downregulate DAPK1 may protect dopaminergic neurons, ameliorate motor deficits, and represent a novel disease-modifying approach in Parkinson\u2019s disease.[117]circzip-2Detected in C. elegans PD model; differential expression not quantified-Transgenic C. elegans-miR-60/zip-2/DAF-16 axiscirczip-2 sponges miR-60, relieving its repression on zip-2; increased zip-2 promotes DAF-16\u2013mediated stress response, reduces \u03b1-synuclein aggregation and extends worm lifespan in the PD modelReveals a circRNA\u2013miRNA\u2013FOXO (DAF-16) regulatory module; targeting this axis may inspire miRNA- or circRNA-based therapies to limit \u03b1-syn toxicity and bolster neuronal resilience in Parkinson\u2019s disease[111]PD-associated eRNAsUpLUHMES---Genetic variants associated with PD are enriched in active enhancer regions in differentiating neurons, suggesting that dysregulation of these elements may contribute to neuronal vulnerability in PD.Identifying the gene targets of these enhancers could reveal new therapeutic avenues.[118]SNCA risk enhancer (intron 2)CRISPR/Cas9 deletion of the 439-bp enhancer in iPSC-derived microglia reduces SNCA and MMRN1 expression at Day 40iPSC-derived microglia--SNCA & MMRN1Two PD risk SNPs are located in a microglial enhancer in the SNCA gene. Deletion of this enhancer results in reduced expression of SNCA and MMRN1 and alterations in glucose metabolism genes.This finding provides a mechanistic link between genetic variants, microglial function, and PD pathogenesis.[119]", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in PD", "text": "Key circRNAs and eRNAs in Parkinson\u2019s disease and their applications as diagnostic/prognostic biomarkers\n\nThe SNCA gene, in addition to producing SNCA mRNA, can also generate circular RNAs. One such circRNA, known as circSNCA, originates from the proximal 3\u2032 untranslated region (3\u2032UTR) of SNCA mRNA, which contains a high-affinity binding site for miR-7. Also, when SH-SH5Y cells were subjected to pramipexole, as a dopamine agonist, the circSNCA levels were reduced, resulting in elevated miR-7 and decreased SNCA protein [115]. Elevated levels of intracellular \u03b1-synuclein, which is primarily encoded by the SNCA gene in the nucleus, have been associated with the etiology of PD and play a significant role in neurodegenerative disorders [124, 125]. Moreover, there is a correlation between the expression of this gene and the levels of miR-7 in neurons [126]. Additionally, it has been observed that miR-7 expression is decreased in the SN of individuals with PD. The reduction in miR-7 levels is also associated with the loss of dopaminergic neurons in the substantia nigra and a consequent decrease in striatal dopamine content [127].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in PD", "text": "Similar to circSNCA, circPANK1 upregulation has been reported in mouse and cellular models of Parkinson\u2019s disease. This circRNA also acts specifically as a sponge for miR-7. Therefore, its increased expression reduces the availability of miR-7, leading to elevated expression of alpha-synuclein (SNCA) protein and, ultimately, enhanced neurodegeneration of dopaminergic neurons [128].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in PD", "text": "Hsa_circ_0070441 is a circRNA molecule that is derived from the coding area of the SNCA gene, and its expression is elevated in SHSY5Y cells under the influence of 1-methyl-4-phenylpyridinium ions (MPP+), a neural metabolite that leads to impaired cellular respiration and dopaminergic neuron death [116]. Moreover, Hsa_circ_0070441 acts as a sponge for a microRNA named miR-626, and its knockdown has led to a significant reduction in inflammation and apoptosis following MPP+ exposure in these cells. Furthermore, it has been observed that miR-626 modulates the expression of insulin receptor substrate 2. The miR-626/insulin receptor substrate 2 axis has been proposed as a potential diagnostic biomarker for Parkinson\u2019s disease [116]. The circular RNA known as circzip-2, derived from the zip-2 gene (Zinc-regulated transporters and iron-regulated transporter-like protein-2), shows significantly reduced expression in the PD model of C. elegans. It has also been demonstrated that circzip-2 established a competitive inhibitory interaction at the cleavage site with its parent gene, zip-2. The protein encoded by zip-2 directly affects the expression level of the SNCA gene and reactive oxygen species (ROS) concentrations [111]. Moreover, circzip-2 can bind to miR-60, thereby inhibiting the progression of PD. This neuroprotective property of circzip-2 may make it a promising candidate for the development of targeted therapeutics and precise diagnostic tools for neurodegenerative diseases such as PD [111]. Despite C. elegans missing direct orthologs of several human Parkinson\u2019s disorder genes, the common miRNA pathways (like the miR-60 family) and stress response processes make it a model for showing important principles of circRNA-based neuroprotection that may increase investigation in higher organisms.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in PD", "text": "The circular RNA derived from the SAMD4A gene, known as circSAMD4A, plays an active role in promoting both apoptosis and autophagy. Moreover, it has been shown that circSAMD4A contributes to dopaminergic neuronal apoptosis and autophagy via the AMPK/mTOR signaling pathway in human neuroblastoma cells (SH-SY5Y) treated with the neurotoxins MPTP and MPP+ [114, 129]. Additionally, the binding of microRNA-29c-3p to circSAMD4A was associated with reduced neurotoxin-induced cytotoxicity in neuronal cells. Therefore, circSAMD4A may serve as a valuable biomarker and a potential therapeutic target for PD [107].\n\nThe circular RNA associated with DLG protein (circDLGAP4), which has been reported to be downregulated in MPTP-induced mouse models of PD and in PD cellular models treated with MPP+, may contribute to the onset and progression of PD by affecting the viability of neuronal cells (SH-SY5Y and MN9D), promoting apoptosis, impairing mitochondrial function, and enhancing autophagy [113].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in PD", "text": "The circular RNA circTLK1, which is derived from the TLK1 (Tousled-like kinase 1) gene, is a highly upregulated circRNA in MPTP-induced PD mouse models, and its knockdown in these models is associated with reduced apoptosis and cellular toxicity, along with increased cell viability. Moreover, by sequestering miR-26a-5p, circTLK1 regulates the expression of its target gene DAPK1 (Death-associated protein kinase 1), which is involved in apoptotic processes. Through this mechanism, circTLK1 contributes to the improvement of neurological deficits caused by middle cerebral artery occlusion and reperfusion in in vivo models [117]. Therefore, the circTLK1/miR-26a-5p/DAPK1 regulatory axis is considered a prominent mechanism in PD pathogenesis and may provide a foundation for the development of innovative and targeted therapeutic approaches for this disease [107].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in PD", "text": "In a study involving 300 PD cases and 100 healthy controls, it was found that the circRNAs circARID1B and circTCONS_l2_00002816 exhibit high sensitivity for the early diagnosis of PD. Whereas, circFAM83H, circHUWE1, circARID1B, and circTCONS_l2_00002816 were identified as potential biomarkers capable of distinguishing between early and advanced stages of PD [130]. CircEps15 has also been identified in both human and animal studies [131]. Its reduced expression in the plasma of PD patients is associated with disease progression. circEps15 functions by sponging miR-24, thereby maintaining PINK1 gene expression and promoting the PINK1-PRKN\u2013dependent mitophagy pathway, which is essential for the clearance of damaged mitochondria and maintaining mitochondrial homeostasis in neurons [131]. Top circRNAs in PD development is indicated in Fig. 2.Fig. 2Illustration depicting the roles of circRNA in Parkinson\u2019s disease. It highlights how circRNAs by sponging various miRNAs and interacting with and \u03b1\u2013synuclein for regulating PD\n\nIllustration depicting the roles of circRNA in Parkinson\u2019s disease. It highlights how circRNAs by sponging various miRNAs and interacting with and \u03b1\u2013synuclein for regulating PD", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "eRNAs in PD", "text": "Dysregulation of eRNAs in PD can disrupt dopaminergic neuron function and be a cause of neurodegeneration. The study of eRNAs may reveal new insights into the mechanisms of PD and its therapeutic targets (Table 2). The production of an eRNA from intron 2 of the KANSL1 gene has been linked to Polymorphisms on chromosome 17q21 in dopamine neurons that are associated with PD. The dysregulation of KANSL1 hinders the process of autophagy, leading to the promotion of memory impairment and neurodegeneration [132, 133]. Studies have shown that the intronic enhancers of the SNCA gene form clusters of enhancers in the human brain, where single-nucleotide polymorphisms (SNPs) highly associated with PD risk are enriched. These enhancer cluster regions have the potential to liberate RNA polymerase II from the promoter region of the SNCA gene, thereby exerting regulatory control over its expression [134]. One proposed idea posits that the eRNAs generated from these enhancers are capable of promoting transcription elongation by dissociating the negative elongation factor (NELF) complex. On the other hand, deletion of the intronic enhancer clusters of the SNCA gene has been shown to significantly downregulate SNCA expression in the brain by inhibiting its transcription elongation [135]. This finding highlights the potential of targeting these enhancer regions as a novel therapeutic strategy for treating neurodegenerative disorders associated with SNCA accumulation in the brain (Fig. 5).", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Clinical implications circRNAs and eRNAs in PD", "text": "Zhong et al. [130] investigated the prognostic and diagnostic applicability of circRNAs during the early and progressive stages of PD. The analysis conducted on the plasmatic circRNAs of three individuals diagnosed with PD, compared to a control group, used microarray technology. The results of this analysis revealed that two specific circRNAs, namely circ_0004381 and circ_0017204 (referred to as circARID1B and circTCONS-l200002816, respectively), exhibited the potential to accurately predict PD at an early stage of diagnosis. This predictive ability was achieved by employing risk score formulas and conducting receiver operating characteristic (ROC) curve analysis, which demonstrated relatively high sensitivity and specificity. In addition, the circRNA molecules circ_0085869, circ_0004381, circ_0017204, and circ_0090668, which are respectively referred to as circFAM83H, circARID1B, circTCONS-l2-00002816, and circHUWE1, have potential discriminatory capabilities between late-stage and early-stage PD.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Clinical implications circRNAs and eRNAs in PD", "text": "In conclusion, these findings suggest that a cell-free circRNA panel holds promise as a non-invasive biomarker for the early detection or continuous monitoring of PD progression [130]. Ravanidis et al. conducted a study to further explore the potential of circRNAs as biomarkers in a larger group of individuals with PD. The outcome of this study identified six circRNAs (MAPK9_circ_0001566, HOMER1_circ_0006916, SLAIN1_circ_0000497, DOP1B_circ_0001187, RESP1_circ_0004368, and PSEN1_circ_0003848) that are enriched in the brain and are significantly downregulated in PBMCs of PD subjects compared to healthy ones [136]. A different study suggested that circzip-2 can bind with miR-60, consequently hindering the advancement of PD. This neuroprotective characteristic of circzip-2 could position it as a potential candidate for creating targeted therapies and accurate diagnostic instruments for neurodegenerative disorders like PD [111] (Table 2 and Supplementary Table S2).", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Huntington\u2019s disease (HD)", "text": "Huntington\u2019s Disease (HD) is an atypical autosomal dominant neurodegenerative condition characterized by the progressive loss of cortical and striatal neurons. Individuals with Huntington\u2019s disease (HD) have expressed motor skills, memory function, and overall cognitive abilities [137]. HD is the most prevalent trinucleotide repeat disorder. The expansion of CAG repeats within exon 1 of the Huntingtin (HTT) gene has been identified as the underlying etiology of HD [138].\n\nAs a result of CAG expansion within the first exon, a polyglutamine mutant HTT (mHTT) protein is produced, which is prone to misfolding and aggregation and has a higher tendency to interact with proteins in ways that aren\u2019t normal. The role of HTT encompasses various cellular processes, including axonal trafficking, gene regulation, disruption of the ubiquitin-proteasome system, and autophagy [139]. Dysregulation of autophagy has been implicated in the early cognitive changes observed in Huntington\u2019s disease. Therefore, it undergoes alterations and qualifies as a potential candidate for Huntington\u2019s disease treatment [140].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Huntington\u2019s disease (HD)", "text": "circRNAs have been shown that differentially expressed in HD. Table 3 summarizes the expression patterns of key circRNAs in HD. In addition, Several circRNAs have been shown to play a role in regulating autophagy in HD; examples include circLRP1B, circDLGAP4, and circHIPK2, which inhibit miR-27a-3p, miR-134-5p, and miR-124, respectively. Emerging evidence suggests that circRNAs may influence the pathogenesis of HD through potential modulation of the autophagy process [144]. Nonetheless, to validate this hypothesis, a rigorous investigation of the role of autophagy-related circRNAs in the progression of HD is necessary. In addition, when circ-016719 is overexpressed in neurons, it acts as a sponge for miR-29c, which starts the process of autophagy. Circ-016719 stimulates the expression of MAP2K6 and BECN1 and, as a consequence, increases the rate of autophagy within the neuronal cells. The function of circ-016719 was confirmed using knockdown experiments, which effectively inhibited the process of autophagy [145].Table 3Key circRNAs and eRNAs in Huntington\u2019s disease and their applications as diagnostic/prognostic biomarkersncRNAExpression ChangeCell line studyAnimal studyClinical sampleTarget/PathwayFunctionClinical PotentialReferencescircHTT(2\u20136) (exons 2\u20136)UpSH-SY5Y, HEK293-7 HD patients, 4 controls (frontal cortex tissue)HTTBy modulating HTT nuclear localization and influencing nuclear morphology and proliferation, circHTT(2\u20136) recapitulates HD-like cellular phenotypes without altering repeat lengthUpregulation correlates with CAG burden and nuclear pathologies, suggesting utility as a biomarker of disease severity and a target to restore HTT localization[141]circHTT(2,3,4,5,6)UpST Hdh Q111/Q111 striatal cells, HEK293T and SH-SY5YzQ175 and Htt Q111/20 knock-in miceiPSC-derived terminally differentiated neurons (hTDNs) from HD patients (CAG 43\u2013180) vs controls (\u2009<\u200936)40S ribosomal subunitAlthough this circRNA is not translated in the adult mouse brain, its binding to ribosomes suggests that it may regulate the translation of other proteins. This could contribute to the imbalance of proteostasis in HD.Potential molecular modulator in HD pathogenesis.[142]Overall circRNA populationdysregulationPC12 neuron-like cellsmESCs and mNPCs derived from Htt CAG-expansion knock-in mice-Dopaminergic synapse; MAPK; long-term depressionInferred involvement in synaptic and signaling pathways; specific mechanisms unvalidatedNot yet assessed in patient samples; stability and blood detectability suggest potential as HD biomarkers[143]", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Huntington\u2019s disease (HD)", "text": "Key circRNAs and eRNAs in Huntington\u2019s disease and their applications as diagnostic/prognostic biomarkers\n\nThe association between miR-142 and HD has been established through the circHectd1/miR-142/autophagy axis. In astrocytes, upregulated circHectd1 has been shown to enhance TIPARP expression level via the miR-142-TIPARP axis. Subsequently, elevated TIPARP levels trigger the production of LC3-II, hence promoting the activation of autophagy [146].\n\nBased on evidence, the circHTT (2\u20136), which encompasses exons 2, 3, 4, 5, and 6 of the HTT gene, has been upregulated in the brains of HD individuals. The upregulation of circHTT (2\u20136) in HEK293 and SH-SY5Y cells demonstrated no modification to the CAG repeat region, which is considered the primary cause of HD. However, this upregulation lead to a decrease in cell proliferation, nuclear size, and nuclear localization of the HTT protein. These effects were consistent with the pathophysiological abnormalities observed in HD [141].\n\nIt was demonstrated in a study that circAKAP7 specifically interacted with miR-155-5p, hence activating ATG12 as an autophagy-related gene and consequently stimulating autophagy [147]. Growing evidence have indicated the importance of autophagy in maintenance of the synapse in HD [140].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Huntington\u2019s disease (HD)", "text": "In a study conducted by Ayyildiz and colleagues using HD mouse models and RNA sequencing (RNA-seq), it was demonstrated that although neural progenitor cells (NPCs) predominantly express more circRNAs than embryonic stem cells (ESCs), cells harboring the mutant HTT gene exhibit an aberrant set of linear and back-splicing-generated circular RNA isoforms. Notably, circRNA expression levels in NPCs were inversely correlated with the length of the CAG repeat expansion in the HTT gene, indicating HTT mutation impairs RNA circularization. Furthermore, approximately 32.7% of the differentially expressed circRNAs showed N6-methyladenosine (m6A) modifications. This suggest that mutant huntingtin may disrupt circRNA biogenesis via m6A-mediated regulatory mechanisms, particularly in NPCs [148] (Fig. 3).Fig. 3The schematic illustration demonstrates the roles of different circRNAs in Huntington\u2019s disease. circRNAs by sponging some miRNAs and influencing various genes, modulating HD\n\nThe schematic illustration demonstrates the roles of different circRNAs in Huntington\u2019s disease. circRNAs by sponging some miRNAs and influencing various genes, modulating HD", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Huntington\u2019s disease (HD)", "text": "In HD, eRNA dysregulation can abolish transcriptional networks to lead to mutant HTT-induced neurodegeneration. eRNA dysregulation could offer new insights into HD pathogenesis and therapeutic targeting [149]. A brain-specific ultraconservative eRNA called Evf2 can help DLX and MECP2 control the homeodomain transcription factor Dlx5/6 in their intergenic enhancer regions. Evf2 mutations during the early postnatal period decrease the number of GABAergic interneurons in the hippocampus and dentate gyrus. Additionally, these mutations have been found to diminish synaptic blockage in the brains of adult individuals [150].\n\nGABA-regulated circuit dysfunction, which disrupts the balance between excitatory and inhibitory signaling, has been strongly implicated in a lot of neurological conditions, including neurodegenerative disorders such as Huntington\u2019s disease and epilepsy, neurodevelopmental disorders like autism and Tourette\u2019s syndrome, as well as other disorders, including dystonia and hepatic encephalopathy [151\u2013153] (Table 3 and Fig. 5).", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Huntington\u2019s disease (HD)", "text": "The study of circRNAs and eRNAs in Huntington\u2019s disease, while still in its early stages, reveals unique diagnostic and therapeutic opportunities shaped by the monogenic nature of the disease. In contrast to sporadic neurological disorders, the clear genetic etiology of HD provides a clear context for interpreting ncRNA dysregulation. The inverse correlation between CAG repeat length and expression of specific circRNAs in neural progenitor cells suggests that circRNA signatures can serve as sensitive molecular readouts of \u201cgenetic burden,\u201d potentially reflecting disease susceptibility and time to progression even before the onset of symptoms [148]. Furthermore, the accumulation of circHTT variants, which mimic cellular pathologies such as aberrant nuclear localization of HTT without themselves containing expanded CAG repeats, positions them as novel biomarkers that report cellular stress responses downstream of mutant HTT protein [142].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Huntington\u2019s disease (HD)", "text": "Regulatory networks formed by circRNAs, such as the circHectd1/miR-142/TIPARP and circAKAP7/miR-155-5p/ATG12 axes, highlight autophagy as a central pathway that is susceptible to targeted circRNA intervention [147, 154]. Modulation of these circuits could enhance clearance of mutant huntingtin protein. From a transcriptional perspective, dysfunction of GABAergic circuits, which are regulated in part by eRNAs such as Evf2, suggests a compelling strategy [150]. Several studies have conducted additional research to explore the clinical relevance of eRNAs in neurodegenerative disorders. For instance, evidence indicates that eRNA affects neuronal genes related to HD within the striatum. This suggests targeting striatal enhancers to boost eRNA expressions and improving the neural gene repression observed in HD patients [149]. In the context of neurodegenerative disorders, most of the research on how eRNAs can be used in the clinic, such as for diagnosis or targeted therapy, is still in its early stages [155] (Tables 3). Targeted epigenetic therapies or engineered transcription factors designed to stabilize the expression of key eRNAs could restore inhibitory balance and counteract the excitotoxicity observed in HD, offering a novel approach to correct the underlying transcriptional defects caused by mutant HTT.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "circRNAs in HD", "text": "circRNAs have been shown that differentially expressed in HD. Table 3 summarizes the expression patterns of key circRNAs in HD. In addition, Several circRNAs have been shown to play a role in regulating autophagy in HD; examples include circLRP1B, circDLGAP4, and circHIPK2, which inhibit miR-27a-3p, miR-134-5p, and miR-124, respectively. Emerging evidence suggests that circRNAs may influence the pathogenesis of HD through potential modulation of the autophagy process [144]. Nonetheless, to validate this hypothesis, a rigorous investigation of the role of autophagy-related circRNAs in the progression of HD is necessary. In addition, when circ-016719 is overexpressed in neurons, it acts as a sponge for miR-29c, which starts the process of autophagy. Circ-016719 stimulates the expression of MAP2K6 and BECN1 and, as a consequence, increases the rate of autophagy within the neuronal cells. The function of circ-016719 was confirmed using knockdown experiments, which effectively inhibited the process of autophagy [145].Table 3Key circRNAs and eRNAs in Huntington\u2019s disease and their applications as diagnostic/prognostic biomarkersncRNAExpression ChangeCell line studyAnimal studyClinical sampleTarget/PathwayFunctionClinical PotentialReferencescircHTT(2\u20136) (exons 2\u20136)UpSH-SY5Y, HEK293-7 HD patients, 4 controls (frontal cortex tissue)HTTBy modulating HTT nuclear localization and influencing nuclear morphology and proliferation, circHTT(2\u20136) recapitulates HD-like cellular phenotypes without altering repeat lengthUpregulation correlates with CAG burden and nuclear pathologies, suggesting utility as a biomarker of disease severity and a target to restore HTT localization[141]circHTT(2,3,4,5,6)UpST Hdh Q111/Q111 striatal cells, HEK293T and SH-SY5YzQ175 and Htt Q111/20 knock-in miceiPSC-derived terminally differentiated neurons (hTDNs) from HD patients (CAG 43\u2013180) vs controls (\u2009<\u200936)40S ribosomal subunitAlthough this circRNA is not translated in the adult mouse brain, its binding to ribosomes suggests that it may regulate the translation of other proteins. This could contribute to the imbalance of proteostasis in HD.Potential molecular modulator in HD pathogenesis.[142]Overall circRNA populationdysregulationPC12 neuron-like cellsmESCs and mNPCs derived from Htt CAG-expansion knock-in mice-Dopaminergic synapse; MAPK; long-term depressionInferred involvement in synaptic and signaling pathways; specific mechanisms unvalidatedNot yet assessed in patient samples; stability and blood detectability suggest potential as HD biomarkers[143]", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "circRNAs in HD", "text": "Key circRNAs and eRNAs in Huntington\u2019s disease and their applications as diagnostic/prognostic biomarkers\n\nThe association between miR-142 and HD has been established through the circHectd1/miR-142/autophagy axis. In astrocytes, upregulated circHectd1 has been shown to enhance TIPARP expression level via the miR-142-TIPARP axis. Subsequently, elevated TIPARP levels trigger the production of LC3-II, hence promoting the activation of autophagy [146].\n\nBased on evidence, the circHTT (2\u20136), which encompasses exons 2, 3, 4, 5, and 6 of the HTT gene, has been upregulated in the brains of HD individuals. The upregulation of circHTT (2\u20136) in HEK293 and SH-SY5Y cells demonstrated no modification to the CAG repeat region, which is considered the primary cause of HD. However, this upregulation lead to a decrease in cell proliferation, nuclear size, and nuclear localization of the HTT protein. These effects were consistent with the pathophysiological abnormalities observed in HD [141].\n\nIt was demonstrated in a study that circAKAP7 specifically interacted with miR-155-5p, hence activating ATG12 as an autophagy-related gene and consequently stimulating autophagy [147]. Growing evidence have indicated the importance of autophagy in maintenance of the synapse in HD [140].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "circRNAs in HD", "text": "In a study conducted by Ayyildiz and colleagues using HD mouse models and RNA sequencing (RNA-seq), it was demonstrated that although neural progenitor cells (NPCs) predominantly express more circRNAs than embryonic stem cells (ESCs), cells harboring the mutant HTT gene exhibit an aberrant set of linear and back-splicing-generated circular RNA isoforms. Notably, circRNA expression levels in NPCs were inversely correlated with the length of the CAG repeat expansion in the HTT gene, indicating HTT mutation impairs RNA circularization. Furthermore, approximately 32.7% of the differentially expressed circRNAs showed N6-methyladenosine (m6A) modifications. This suggest that mutant huntingtin may disrupt circRNA biogenesis via m6A-mediated regulatory mechanisms, particularly in NPCs [148] (Fig. 3).Fig. 3The schematic illustration demonstrates the roles of different circRNAs in Huntington\u2019s disease. circRNAs by sponging some miRNAs and influencing various genes, modulating HD\n\nThe schematic illustration demonstrates the roles of different circRNAs in Huntington\u2019s disease. circRNAs by sponging some miRNAs and influencing various genes, modulating HD", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "eRNAs in HD", "text": "In HD, eRNA dysregulation can abolish transcriptional networks to lead to mutant HTT-induced neurodegeneration. eRNA dysregulation could offer new insights into HD pathogenesis and therapeutic targeting [149]. A brain-specific ultraconservative eRNA called Evf2 can help DLX and MECP2 control the homeodomain transcription factor Dlx5/6 in their intergenic enhancer regions. Evf2 mutations during the early postnatal period decrease the number of GABAergic interneurons in the hippocampus and dentate gyrus. Additionally, these mutations have been found to diminish synaptic blockage in the brains of adult individuals [150].\n\nGABA-regulated circuit dysfunction, which disrupts the balance between excitatory and inhibitory signaling, has been strongly implicated in a lot of neurological conditions, including neurodegenerative disorders such as Huntington\u2019s disease and epilepsy, neurodevelopmental disorders like autism and Tourette\u2019s syndrome, as well as other disorders, including dystonia and hepatic encephalopathy [151\u2013153] (Table 3 and Fig. 5).", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Clinical implications of circRNAs and eRNAs in HD", "text": "The study of circRNAs and eRNAs in Huntington\u2019s disease, while still in its early stages, reveals unique diagnostic and therapeutic opportunities shaped by the monogenic nature of the disease. In contrast to sporadic neurological disorders, the clear genetic etiology of HD provides a clear context for interpreting ncRNA dysregulation. The inverse correlation between CAG repeat length and expression of specific circRNAs in neural progenitor cells suggests that circRNA signatures can serve as sensitive molecular readouts of \u201cgenetic burden,\u201d potentially reflecting disease susceptibility and time to progression even before the onset of symptoms [148]. Furthermore, the accumulation of circHTT variants, which mimic cellular pathologies such as aberrant nuclear localization of HTT without themselves containing expanded CAG repeats, positions them as novel biomarkers that report cellular stress responses downstream of mutant HTT protein [142].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Clinical implications of circRNAs and eRNAs in HD", "text": "Regulatory networks formed by circRNAs, such as the circHectd1/miR-142/TIPARP and circAKAP7/miR-155-5p/ATG12 axes, highlight autophagy as a central pathway that is susceptible to targeted circRNA intervention [147, 154]. Modulation of these circuits could enhance clearance of mutant huntingtin protein. From a transcriptional perspective, dysfunction of GABAergic circuits, which are regulated in part by eRNAs such as Evf2, suggests a compelling strategy [150]. Several studies have conducted additional research to explore the clinical relevance of eRNAs in neurodegenerative disorders. For instance, evidence indicates that eRNA affects neuronal genes related to HD within the striatum. This suggests targeting striatal enhancers to boost eRNA expressions and improving the neural gene repression observed in HD patients [149]. In the context of neurodegenerative disorders, most of the research on how eRNAs can be used in the clinic, such as for diagnosis or targeted therapy, is still in its early stages [155] (Tables 3). Targeted epigenetic therapies or engineered transcription factors designed to stabilize the expression of key eRNAs could restore inhibitory balance and counteract the excitotoxicity observed in HD, offering a novel approach to correct the underlying transcriptional defects caused by mutant HTT.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Multiple sclerosis (MS)", "text": "While MS is classically considered a demyelinating autoimmune disease, progressive neurodegeneration is recognized as the main substrate for irreversible neurological disability in its chronic stages. The pathological continuum of MS includes inflammatory demyelination and subsequent axonal scission, neuronal loss, and synaptic damage, hallmarks shared with conventional neurodegenerative disorders. This neurodegenerative component is particularly prevalent in progressive forms of MS, where clinical deterioration continues independent of overt inflammatory activity [156]. Therefore, examining circRNAs and eRNAs in the context of MS offers a unique opportunity to investigate how ncRNA networks associated with neuroinflammatory stimuli are linked to downstream neurodegenerative processes. Specifically, these regulatory RNAs may modulate: (1) glial activation states and neuroinflammatory cascades; (2) mitochondrial dysfunction and oxidative stress in neurons; (3) remyelination failure and oligodendrocyte pathology; and (4) direct mechanisms of axonal degeneration. Therefore, the investigation of circRNAs and eRNAs in MS provides important insights into the regulatory mechanisms that drive neurodegeneration in both autoimmune and proteinopathies [157].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Multiple sclerosis (MS)", "text": "The clinical subgroups encompass clinically isolated syndrome (CIS), relapsing-remitting multiple sclerosis (RRMS), primary progressive multiple sclerosis (PPMS), and secondary progressive multiple sclerosis (SPMS) [158]. The predominant manifestation of the disease is RRMS, which represents approximately 80% of cases and is characterized by recurring episodes of disease exacerbation. PPMS manifests in around 5\u201315% of cases, displaying a distinct clinical profile characterized by a gradual deterioration of neurological function without the occurrence of relapses. Following a period of 10\u201315\u2009years, RRMS undergoes a transition to a secondary progressive form characterized by a slow advancement of the disease [159].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Multiple sclerosis (MS)", "text": "CircRNAs have been implicated in the development and progression of autoimmune diseases, such as MS. A large number of circular RNAs in peripheral blood exhibited differential expression in the MS cohort [160, 161]. Table 4 and Supplementary Table S3 list the key CircRNAs in MS.Table 4Key circRNAs and eRNAs in multiple sclerosis and their applications as diagnostic/prognostic biomarkersncRNAExpression ChangeCell line studyAnimal studyClinical sampleTarget/PathwayFunctionClinical PotentialReferenceshsa_circ_0007990 (PGAP3)Down--PBMCs from discovery cohort (10 RR-MS vs 10 HC); replication cohorts: US (18 RR-MS vs 9 HC), Italian (19 RR-MS vs 20 HC)PGAP3 gene; GPI-anchor remodeling enzymeLikely influences immune responses via its host gene product; downregulation may dysregulate T-cell activation and innate immunity in MSValidated as a blood-based biomarker across independent cohorts; stability in PBMCs and replication of downregulation support its diagnostic/prognostic utility in MS[162]hsa_circRNA_101348; hsa_circRNA_102611; hsa_circRNA_104361Up--Discovery microarray in PBMCs: 20 RR-MS (10 relapse, 10 remission)\u2009+\u200910 HC; Validation qRT-PCR: 47 RR-MS (19 relapse, 28 remission)\u2009+\u200927 HCPredicted sponging of 15 miRNAs across the three circRNAsModulates a circRNA\u2013miRNA network that derepresses B-cell genes AK2, IKZF3, CBX5Biomarkers of relapse activity; correlates with MRI inflammatory lesions[163]circINPP4BUpNaive CD4+ T cells polarized to Th17EAE model miceCD4+ T cells from 18 RR-MS patients in relapse, remission, and 20 healthy controlsmiR-30aSponges miR-30a to derepress Th17 differentiation program, promoting IL-17A production and EAE pathologyPotential biomarker for MS relapse phase; therapeutic targeting of circINPP4B may reduce Th17-driven neuroinflammation[164]hsa_circRNA_101145; hsa_circRNA_001896Down--PBMC microarray (20 RR-MS: 10 relapse, 10 remission; 10 HC); Validation: qRT-PCR (45 RR-MS: 18 relapse, 27 remission; 27 HC)NOS1, NOPCHAP1, DSE genesThese circRNAs create a circRNA-miRNA network that mainly affects post-transcriptional regulation. The target genes are involved in BBB permeability, microglial activity, and tissue regeneration.They may provide insight into the microenvironmental changes that control the development of MS and serve as potential differential biomarkers.[165]circ_0005402Down--PBL from RR-MS patients and controls (Discovery:4 vs 4; Validation1:20 vs 18; Paired: 21; CIS: 10 vs 10)miRNA sponging (miR-1248; miR-766; 15 others); RBP binding (AGO2; EIF4A3; LIN28A; IGF2BP1; PTB)Bioinformatic prediction: miRNA sponge; RBP interaction; potential ANXA2 regulationAUC\u2009=\u20090.899 (95% CI 0.747\u20130.976); Sens\u2009=\u200994.4%; Spec\u2009=\u200975.0%[166]hsa_circ_0106803Up--PBMCs from 30 RR-MS patients vs. 30 healthy controlsPredicted binding sites for six miRNAs (incl. miR-1275, miR-149); no pathwayPredicted miRNA sponge; potential miRNA network modulatorProposed blood-based biomarker due to high stability and differential expression in MS[167]hsa_circ_0141241 (ATP8B4)Up\u2013\u2013Leukocyte RNA from MS vs HC (Discovery: 30\u2009MS\u2009vs\u200920\u2009HC; Validation: 70\u2009MS\u2009vs\u200946\u2009HC; Paired relapse/remission: 20\u2009MS)\u2013\u2013Candidate biomarker; AUC\u2009=\u20090.722; discriminates MS from HC[168]eRNAs (including HERV-encoded sequences)UpPrimary human monocytesCbx5\u2212/\u2212 mouse model18\u2009MS patients, 7 controlsIntegrator complex (INTcom); RNAPII elongationReduced activity of the Integrator complex leads to the accumulation of eRNAs. This can lead to aberrant expression of genes, including latent viral genes (HERVs) and the inflammatory gene leptin (LEP).The Integrator complex is emerging as a novel player in MS pathology and a potential therapeutic target.[169]", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Multiple sclerosis (MS)", "text": "Key circRNAs and eRNAs in multiple sclerosis and their applications as diagnostic/prognostic biomarkers\n\nIt has been observed that hsa_circ_0106803, originating from the GSDMB gene, exhibited a 2.8-fold increase in expression levels among individuals diagnosed with RRMS [167]. Multiple miRNAs were identified as having multiple target sites in hsa_circ_0106803, notably including miR-1275 and miR-149, both of which exhibit differential expression in the blood of individuals with MS [170].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Multiple sclerosis (MS)", "text": "miR-149 has been observed to interact with the protein ASIC1a, resulting in ASIC1a downregulation [171]. The ASIC1a gene is responsible for encoding a specific subunit of the acid-sensing ion channel. This particular subunit is found to be highly expressed in acute MS lesions, suggesting its potential involvement in neurological disease mechanisms underlying the development of this illness [112, 120, 123, 172]. Based on the available evidence, researchers have put forth the concept of a competing endogenous RNA (ceRNA) network, suggesting that hsa_circ_0106803 may have a role in influencing the development of MS via controlling the expression of ASIC1a mRNA through miR-149 [173]. miR-1275 is a microRNA coding gene that is functional and has been directly associated with multiple sclerosis. This microRNA is involved in the TGF-beta signaling pathway and has been confirmed to exhibit varying methylation statuses along with distinct expression patterns across different IDH expression glioma subtypes [174].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Multiple sclerosis (MS)", "text": "Studies have shown a significant association between the expression level of the circular RNA circINPP4B and the clinical prognosis of MS. In the experimental autoimmune encephalomyelitis (EAE) mice, circINPP4B expression was markedly elevated, and this increase was directly correlated with the severity of clinical symptoms. Furthermore, circINPP4B acts as a sponge for miR-30a, a microRNA whose overexpression inhibits excessive Th17 cell differentiation, a key pathogenic factor in MS and EAE. By sequestering miR-30a, circINPP4B suppresses its anti-inflammatory effects, leading to enhanced Th17 differentiation and progression of EAE [175].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Multiple sclerosis (MS)", "text": "Recent evidence indicates that the circular RNA HECW2 (circ_HECW2) plays a pivotal role in the pathogenesis of MS. Elevated expression of circ_HECW2 has been observed in both in vitro and in vivo MS models and is associated with the induction of endothelial-to-mesenchymal transition (EndoMT), a process recognized as a critical contributor to blood-brain barrier (BBB) impairment and increased permeability. Yang et al. reported that circ_HECW2 functions as a sponge for microRNA-30D (miR-30D), thereby resulting in upregulation of ATG5, a core component of the autophagy pathway, activating the NOTCH signaling pathway, and ultimately promoting LPS-induced EndoMT [176]. In addition, Dong and his team revealed that circ_HECW2 also interacts with miR-30e-5p and, through modulation of neural growth regulator 1 (NGR1) levels, inhibits endothelial cell proliferation, increases apoptosis, and enhances LPS-induced EndoMT [177].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Multiple sclerosis (MS)", "text": "For the first time, a study based on transcriptomic data classified MS into three distinct molecular subtypes: MS-FCRL1 (C1), MS-BTG1 (C2), and MS-RPL38 (C3). Each of these subtypes is linked to various immune- and infection-related signaling pathways. In particular, the investigation of circRNA/miRNA regulatory networks in this study identified 18 differentially expressed circRNAs and 22 differentially expressed miRNAs implicated in subtype C3. Furthermore, EEF1D and TUBA1A were identified as central hub genes involved in neural and immune functions within this subtype, and finally, the circ_0045537/miR-196a-5p/TUBA1A axis was proposed as a potential regulatory pathway modulating disease severity in the C3 subtype [178]. Also, Lodde et al. reported a significant decrease in the expression level of hsa_circ_0018905 in patients with higher levels of disability and in progressive forms of multiple sclerosis (MS), suggesting its potential as a biomarker for MS [179] (Fig. 4).Fig. 4The schematic diagram represents the functional importance of circRNAs in multiple sclerosis and their related roles as a inhibitor, activator, and sponge by binding to miRNA\n\nThe schematic diagram represents the functional importance of circRNAs in multiple sclerosis and their related roles as a inhibitor, activator, and sponge by binding to miRNA", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Multiple sclerosis (MS)", "text": "In another study, circ_0005402 and circ_0035560 from the ANXA2 gene exhibit reduced expression levels in patients with multiple sclerosis, indicating their potential as biomarkers for the disease. They discovered that circ_0005402 contains binding sites for several RNA-binding proteins, including AGO2, EIF4A3, LIN28A, IGF2BP1, and PTB. Additionally, it features two binding sites for miR-1248 and miR-766. This suggests the possibility of a collaborative regulation involving a circRNA-miRNA-mRNA axis in the pathophysiology of MS [166]. In a study by Mohammed, it was reported that ANXA2 is recognized for its involvement in the passage through the blood-brain barrier (BBB) and for enhancing miR-155 expression through the phosphorylation of STAT3 (Signal Transducer and Activator of Transcription 3). Notably, miR-155 serves as a biomarker for multiple sclerosis (MS), detectable in the bodily fluids of patients prior to the disease\u2019s onset, and has been reported to rise in expression during relapse phases [157].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Multiple sclerosis (MS)", "text": "Bioinformatics research of circRNAs derived from non-coding regions of the genome linked to MS showed that these circRNAs might make people more likely to get MS [180]. Another study revealed an association between immunoglobulin (IgG) levels in the brain fluid and the presence of exosomal circRNAs in MS [181]. In addition, Zurawska et al. have recently provided a study on circRNAs that 3 circRNAs, hsa_circRNA_101348, hsa_circRNA_102611, and hsa_circRNA104361, are differentially overexpressed in individuals with RRMS in relapse and have the ability to influence B-cell activity [160]. Furthermore, these three circRNAs exhibited a significant correlation with MRI Gd-enhancing lesions, thereby suggesting their potential involvement in the activity of inflammatory diseases. Given the critical role of B cells in the pathogenesis of multiple sclerosis (MS), they have uncovered two transcripts related to B cells, namely AK2 and IKZF3 [160]. These observations were corroborated by the direct evidence of an upregulation of AK2 and IKZF3 in the peripheral blood mononuclear cells (PBMCs) of patients experiencing a relapse of MS. The AK2 gene is recognized for encoding phosphotransferase adenylate kinase 2, fulfilling the cell-specific requirements necessary for mitochondrial functions, especially in the activation of B cells and the production of antibodies. The significance of AK2 has been particularly highlighted in B cells infected with the Epstein-Barr virus, which has been extensively discussed in relation to MS as a viral infection associated with the disease. IKZF3 regulates the development of lymphoid and myeloid cells as well as immune homeostasis [160].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Multiple sclerosis (MS)", "text": "While the bulk of circRNAs that were differentially expressed exhibited upregulation, it is worth noting that a subset of circRNAs showed downregulation in those affected by MS.\n\nThe results of the global hybridization experiment, which was then confirmed by PCR in a separate group of people, showed that there were two circRNAs, hsa_circRNA_101145 and hsa_circRNA_001896, with expression levels that were significantly lower in MS patients during remission and significantly associated with alleviation in disease-induced disability [165]. In people with RRMS who are in remission, reducing the amount of hsa_circRNA_101145 and hsa_circRNA_001896 could increase the function of all miRNAs that are vulnerable to being targeted by these two circRNAs. One example of the mentioned miRNAs is miR-181c, whose increased level in the cerebrospinal fluid has been linked to a higher chance that clinically isolated syndrome (CIS) will turn into confirmed MS [182].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Multiple sclerosis (MS)", "text": "There is new evidence suggesting that eRNA dysregulation may be involved in MS pathogenesis via alterations of immune response and neuronal function. Understanding of eRNAs has the potential to provide new insights into MS mechanisms and therapeutic targets [183]. Several study investigated the role of super-enhancers (SEs) and enhancer RNAs (eRNAs) in autoimmune diseases. Their findings indicated that increased activity or dysfunction of individual SEs could lead to the development of various autoimmune disorders [184, 185, 185\u2013191]. However, despite growing interest in the regulatory roles of eRNAs in autoimmune diseases and neurodegenerative disorders, there is currently no direct evidence supporting the involvement of eRNAs in the pathogenesis of MS as a disease considered both autoimmune and neurodegenerative. Nevertheless, one study demonstrated that MS-associated single-nucleotide polymorphisms (SNPs) are located within vitamin D receptor (VDR) binding sites and in activated super-enhancers (SEs), especially in VDR-associated SEs (VSEs), in CD4+ T cells and monocytes [184, 188] (Table 4 and Supplementary Table S3). These disease-linked SNPs are believed to influence the regulatory function of these SEs. This spatial correspondence between MS risk SNPs and active enhancer markers strongly suggests that dysregulation of SEs and their associated eRNAs is a key component of MS pathophysiology. This hypothesis proposes that genetic risk is exerted in part by altering the transcriptional output of immune cells via enhancer elements, and provides a new perspective linking genetic susceptibility to functional dysregulation in MS, and identifies eRNAs generated from these SEs as potential biomarkers or therapeutic targets. However, it remains unclear how the presence of these SNPs affects the expression of these SE-derived eRNAs and what impact these eRNAs might have on MS pathogenesis. Thus, this area of research remains largely unexplored and requires further investigation in the future (Fig. 5).Fig. 5Mechanistic roles of eRNAs in neurodegenerative diseases", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Multiple sclerosis (MS)", "text": "Mechanistic roles of eRNAs in neurodegenerative diseases", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in MS", "text": "CircRNAs have been implicated in the development and progression of autoimmune diseases, such as MS. A large number of circular RNAs in peripheral blood exhibited differential expression in the MS cohort [160, 161]. Table 4 and Supplementary Table S3 list the key CircRNAs in MS.Table 4Key circRNAs and eRNAs in multiple sclerosis and their applications as diagnostic/prognostic biomarkersncRNAExpression ChangeCell line studyAnimal studyClinical sampleTarget/PathwayFunctionClinical PotentialReferenceshsa_circ_0007990 (PGAP3)Down--PBMCs from discovery cohort (10 RR-MS vs 10 HC); replication cohorts: US (18 RR-MS vs 9 HC), Italian (19 RR-MS vs 20 HC)PGAP3 gene; GPI-anchor remodeling enzymeLikely influences immune responses via its host gene product; downregulation may dysregulate T-cell activation and innate immunity in MSValidated as a blood-based biomarker across independent cohorts; stability in PBMCs and replication of downregulation support its diagnostic/prognostic utility in MS[162]hsa_circRNA_101348; hsa_circRNA_102611; hsa_circRNA_104361Up--Discovery microarray in PBMCs: 20 RR-MS (10 relapse, 10 remission)\u2009+\u200910 HC; Validation qRT-PCR: 47 RR-MS (19 relapse, 28 remission)\u2009+\u200927 HCPredicted sponging of 15 miRNAs across the three circRNAsModulates a circRNA\u2013miRNA network that derepresses B-cell genes AK2, IKZF3, CBX5Biomarkers of relapse activity; correlates with MRI inflammatory lesions[163]circINPP4BUpNaive CD4+ T cells polarized to Th17EAE model miceCD4+ T cells from 18 RR-MS patients in relapse, remission, and 20 healthy controlsmiR-30aSponges miR-30a to derepress Th17 differentiation program, promoting IL-17A production and EAE pathologyPotential biomarker for MS relapse phase; therapeutic targeting of circINPP4B may reduce Th17-driven neuroinflammation[164]hsa_circRNA_101145; hsa_circRNA_001896Down--PBMC microarray (20 RR-MS: 10 relapse, 10 remission; 10 HC); Validation: qRT-PCR (45 RR-MS: 18 relapse, 27 remission; 27 HC)NOS1, NOPCHAP1, DSE genesThese circRNAs create a circRNA-miRNA network that mainly affects post-transcriptional regulation. The target genes are involved in BBB permeability, microglial activity, and tissue regeneration.They may provide insight into the microenvironmental changes that control the development of MS and serve as potential differential biomarkers.[165]circ_0005402Down--PBL from RR-MS patients and controls (Discovery:4 vs 4; Validation1:20 vs 18; Paired: 21; CIS: 10 vs 10)miRNA sponging (miR-1248; miR-766; 15 others); RBP binding (AGO2; EIF4A3; LIN28A; IGF2BP1; PTB)Bioinformatic prediction: miRNA sponge; RBP interaction; potential ANXA2 regulationAUC\u2009=\u20090.899 (95% CI 0.747\u20130.976); Sens\u2009=\u200994.4%; Spec\u2009=\u200975.0%[166]hsa_circ_0106803Up--PBMCs from 30 RR-MS patients vs. 30 healthy controlsPredicted binding sites for six miRNAs (incl. miR-1275, miR-149); no pathwayPredicted miRNA sponge; potential miRNA network modulatorProposed blood-based biomarker due to high stability and differential expression in MS[167]hsa_circ_0141241 (ATP8B4)Up\u2013\u2013Leukocyte RNA from MS vs HC (Discovery: 30\u2009MS\u2009vs\u200920\u2009HC; Validation: 70\u2009MS\u2009vs\u200946\u2009HC; Paired relapse/remission: 20\u2009MS)\u2013\u2013Candidate biomarker; AUC\u2009=\u20090.722; discriminates MS from HC[168]eRNAs (including HERV-encoded sequences)UpPrimary human monocytesCbx5\u2212/\u2212 mouse model18\u2009MS patients, 7 controlsIntegrator complex (INTcom); RNAPII elongationReduced activity of the Integrator complex leads to the accumulation of eRNAs. This can lead to aberrant expression of genes, including latent viral genes (HERVs) and the inflammatory gene leptin (LEP).The Integrator complex is emerging as a novel player in MS pathology and a potential therapeutic target.[169]", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in MS", "text": "Key circRNAs and eRNAs in multiple sclerosis and their applications as diagnostic/prognostic biomarkers\n\nIt has been observed that hsa_circ_0106803, originating from the GSDMB gene, exhibited a 2.8-fold increase in expression levels among individuals diagnosed with RRMS [167]. Multiple miRNAs were identified as having multiple target sites in hsa_circ_0106803, notably including miR-1275 and miR-149, both of which exhibit differential expression in the blood of individuals with MS [170].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in MS", "text": "miR-149 has been observed to interact with the protein ASIC1a, resulting in ASIC1a downregulation [171]. The ASIC1a gene is responsible for encoding a specific subunit of the acid-sensing ion channel. This particular subunit is found to be highly expressed in acute MS lesions, suggesting its potential involvement in neurological disease mechanisms underlying the development of this illness [112, 120, 123, 172]. Based on the available evidence, researchers have put forth the concept of a competing endogenous RNA (ceRNA) network, suggesting that hsa_circ_0106803 may have a role in influencing the development of MS via controlling the expression of ASIC1a mRNA through miR-149 [173]. miR-1275 is a microRNA coding gene that is functional and has been directly associated with multiple sclerosis. This microRNA is involved in the TGF-beta signaling pathway and has been confirmed to exhibit varying methylation statuses along with distinct expression patterns across different IDH expression glioma subtypes [174].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in MS", "text": "Studies have shown a significant association between the expression level of the circular RNA circINPP4B and the clinical prognosis of MS. In the experimental autoimmune encephalomyelitis (EAE) mice, circINPP4B expression was markedly elevated, and this increase was directly correlated with the severity of clinical symptoms. Furthermore, circINPP4B acts as a sponge for miR-30a, a microRNA whose overexpression inhibits excessive Th17 cell differentiation, a key pathogenic factor in MS and EAE. By sequestering miR-30a, circINPP4B suppresses its anti-inflammatory effects, leading to enhanced Th17 differentiation and progression of EAE [175].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in MS", "text": "Recent evidence indicates that the circular RNA HECW2 (circ_HECW2) plays a pivotal role in the pathogenesis of MS. Elevated expression of circ_HECW2 has been observed in both in vitro and in vivo MS models and is associated with the induction of endothelial-to-mesenchymal transition (EndoMT), a process recognized as a critical contributor to blood-brain barrier (BBB) impairment and increased permeability. Yang et al. reported that circ_HECW2 functions as a sponge for microRNA-30D (miR-30D), thereby resulting in upregulation of ATG5, a core component of the autophagy pathway, activating the NOTCH signaling pathway, and ultimately promoting LPS-induced EndoMT [176]. In addition, Dong and his team revealed that circ_HECW2 also interacts with miR-30e-5p and, through modulation of neural growth regulator 1 (NGR1) levels, inhibits endothelial cell proliferation, increases apoptosis, and enhances LPS-induced EndoMT [177].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in MS", "text": "For the first time, a study based on transcriptomic data classified MS into three distinct molecular subtypes: MS-FCRL1 (C1), MS-BTG1 (C2), and MS-RPL38 (C3). Each of these subtypes is linked to various immune- and infection-related signaling pathways. In particular, the investigation of circRNA/miRNA regulatory networks in this study identified 18 differentially expressed circRNAs and 22 differentially expressed miRNAs implicated in subtype C3. Furthermore, EEF1D and TUBA1A were identified as central hub genes involved in neural and immune functions within this subtype, and finally, the circ_0045537/miR-196a-5p/TUBA1A axis was proposed as a potential regulatory pathway modulating disease severity in the C3 subtype [178]. Also, Lodde et al. reported a significant decrease in the expression level of hsa_circ_0018905 in patients with higher levels of disability and in progressive forms of multiple sclerosis (MS), suggesting its potential as a biomarker for MS [179] (Fig. 4).Fig. 4The schematic diagram represents the functional importance of circRNAs in multiple sclerosis and their related roles as a inhibitor, activator, and sponge by binding to miRNA\n\nThe schematic diagram represents the functional importance of circRNAs in multiple sclerosis and their related roles as a inhibitor, activator, and sponge by binding to miRNA", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in MS", "text": "In another study, circ_0005402 and circ_0035560 from the ANXA2 gene exhibit reduced expression levels in patients with multiple sclerosis, indicating their potential as biomarkers for the disease. They discovered that circ_0005402 contains binding sites for several RNA-binding proteins, including AGO2, EIF4A3, LIN28A, IGF2BP1, and PTB. Additionally, it features two binding sites for miR-1248 and miR-766. This suggests the possibility of a collaborative regulation involving a circRNA-miRNA-mRNA axis in the pathophysiology of MS [166]. In a study by Mohammed, it was reported that ANXA2 is recognized for its involvement in the passage through the blood-brain barrier (BBB) and for enhancing miR-155 expression through the phosphorylation of STAT3 (Signal Transducer and Activator of Transcription 3). Notably, miR-155 serves as a biomarker for multiple sclerosis (MS), detectable in the bodily fluids of patients prior to the disease\u2019s onset, and has been reported to rise in expression during relapse phases [157].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in MS", "text": "Bioinformatics research of circRNAs derived from non-coding regions of the genome linked to MS showed that these circRNAs might make people more likely to get MS [180]. Another study revealed an association between immunoglobulin (IgG) levels in the brain fluid and the presence of exosomal circRNAs in MS [181]. In addition, Zurawska et al. have recently provided a study on circRNAs that 3 circRNAs, hsa_circRNA_101348, hsa_circRNA_102611, and hsa_circRNA104361, are differentially overexpressed in individuals with RRMS in relapse and have the ability to influence B-cell activity [160]. Furthermore, these three circRNAs exhibited a significant correlation with MRI Gd-enhancing lesions, thereby suggesting their potential involvement in the activity of inflammatory diseases. Given the critical role of B cells in the pathogenesis of multiple sclerosis (MS), they have uncovered two transcripts related to B cells, namely AK2 and IKZF3 [160]. These observations were corroborated by the direct evidence of an upregulation of AK2 and IKZF3 in the peripheral blood mononuclear cells (PBMCs) of patients experiencing a relapse of MS. The AK2 gene is recognized for encoding phosphotransferase adenylate kinase 2, fulfilling the cell-specific requirements necessary for mitochondrial functions, especially in the activation of B cells and the production of antibodies. The significance of AK2 has been particularly highlighted in B cells infected with the Epstein-Barr virus, which has been extensively discussed in relation to MS as a viral infection associated with the disease. IKZF3 regulates the development of lymphoid and myeloid cells as well as immune homeostasis [160].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNAs in MS", "text": "While the bulk of circRNAs that were differentially expressed exhibited upregulation, it is worth noting that a subset of circRNAs showed downregulation in those affected by MS.\n\nThe results of the global hybridization experiment, which was then confirmed by PCR in a separate group of people, showed that there were two circRNAs, hsa_circRNA_101145 and hsa_circRNA_001896, with expression levels that were significantly lower in MS patients during remission and significantly associated with alleviation in disease-induced disability [165]. In people with RRMS who are in remission, reducing the amount of hsa_circRNA_101145 and hsa_circRNA_001896 could increase the function of all miRNAs that are vulnerable to being targeted by these two circRNAs. One example of the mentioned miRNAs is miR-181c, whose increased level in the cerebrospinal fluid has been linked to a higher chance that clinically isolated syndrome (CIS) will turn into confirmed MS [182].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "eRNAs in MS", "text": "There is new evidence suggesting that eRNA dysregulation may be involved in MS pathogenesis via alterations of immune response and neuronal function. Understanding of eRNAs has the potential to provide new insights into MS mechanisms and therapeutic targets [183]. Several study investigated the role of super-enhancers (SEs) and enhancer RNAs (eRNAs) in autoimmune diseases. Their findings indicated that increased activity or dysfunction of individual SEs could lead to the development of various autoimmune disorders [184, 185, 185\u2013191]. However, despite growing interest in the regulatory roles of eRNAs in autoimmune diseases and neurodegenerative disorders, there is currently no direct evidence supporting the involvement of eRNAs in the pathogenesis of MS as a disease considered both autoimmune and neurodegenerative. Nevertheless, one study demonstrated that MS-associated single-nucleotide polymorphisms (SNPs) are located within vitamin D receptor (VDR) binding sites and in activated super-enhancers (SEs), especially in VDR-associated SEs (VSEs), in CD4+ T cells and monocytes [184, 188] (Table 4 and Supplementary Table S3). These disease-linked SNPs are believed to influence the regulatory function of these SEs. This spatial correspondence between MS risk SNPs and active enhancer markers strongly suggests that dysregulation of SEs and their associated eRNAs is a key component of MS pathophysiology. This hypothesis proposes that genetic risk is exerted in part by altering the transcriptional output of immune cells via enhancer elements, and provides a new perspective linking genetic susceptibility to functional dysregulation in MS, and identifies eRNAs generated from these SEs as potential biomarkers or therapeutic targets. However, it remains unclear how the presence of these SNPs affects the expression of these SE-derived eRNAs and what impact these eRNAs might have on MS pathogenesis. Thus, this area of research remains largely unexplored and requires further investigation in the future (Fig. 5).Fig. 5Mechanistic roles of eRNAs in neurodegenerative diseases", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "eRNAs in MS", "text": "Mechanistic roles of eRNAs in neurodegenerative diseases", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Clinical implications of circRNAs and eRNAs in MS", "text": "Iparraguirre et al. identified six sex-dependent circRNAs as potential biomarkers for MS. Their study was conducted on a cohort consisting of 30\u2009MS patients and 20 healthy controls and subsequently validated in an independent validation group of 70 additional MS patients. The identified circRNAs -hsa_circ_0141241, hsa_circ_0058514, circPADI4, hsa_circ_0001947, hsa_circ_0001707, and hsa_circ_0001459\u2014exhibited significant upregulation in female MS subjects, whereas male subjects showed no comparable increase in their expression [168]. In addition, another study reported 166 differentially expressed circRNAs in individuals with MS, of which 125 showed decreased expression. Among these, hsa_circ_0007990 (derived from the PGAP3 gene) was notably downregulated. This reduction in expression was also confirmed in two independent replication cohorts, suggesting circPGAP3 as a promising biomarker for multiple sclerosis [162].\n\nNumerous studies have highlighted the significance of circRNA expression levels, including circINPP4B, circ_0005402, circ_0035560, and hsa_circ_0018905, in predicting the clinical trajectory of MS patients. The expression levels of these transcripts may assist clinicians in selecting more effective treatment options tailored to the specific circRNA expression patterns observed in each patient, thereby indicating a possible advantage as biomarkers for the detection of MS and facilitating personalized therapeutic approaches.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Clinical implications of circRNAs and eRNAs in MS", "text": "The association between dysregulation of eRNAs and a range of human disorders suggests the potential of these molecules to be used as novel therapeutic tools in various diseases. The attractiveness of eRNA therapy lies in its ability to regulate transcription, in contrast to other non-coding RNAs such as miRNAs, which mostly exert their effects post-transcriptionally. Moreover, it should be noted that each eRNA-producing domain exerts a limited impact on a localized set of adjacent genes in a cis-acting manner, thereby ensuring a high degree of target specificity and reducing the likelihood of unintended consequences. The expression pattern of eRNAs in specific tissues and cell types is considered a valuable characteristic contributing to the mitigation of potential off-target effects [155].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Converging subjects and integrated regulatory networks", "text": "Our comprehensive review reveals significant convergent patterns of circRNA and eRNA dysregulation across major neurological disorders, suggesting common molecular mechanisms underlying apparently distinct pathological processes. Three core regulatory subjects consistently emerge:\n\nSeveral synapse-enriched circRNAs (e.g., circHOMER1, circGRIA1) show consistent downregulation in AD, PD, and HD, suggesting a common disruption in synaptic maintenance programs [192\u2013194].\n\nActivity-dependent eRNAs that regulate immediate early genes (e.g., Fos, Arc) show impaired induction in disease models, suggesting compromised activity-transcription coupling [195, 196].\n\nThe circHOMER1/miR-651/PSEN1-2 axis in AD [197] parallels the circSNCA/miR-7/SNCA axis in PD [198], revealing a conserved pattern of synaptic circRNAs that regulate key disease proteins through miRNA spongification.\n\nProinflammatory circRNAs (e.g., circ_0000950, circ_0005835) are upregulated in multiple disorders and drive microglial activation and cytokine production through shared miRNA pathways (miR-103, miR-576-3p) [63, 66].\n\nEnhancer RNAs that regulate immune response genes (e.g., AANCR-APOE in AD, SNCA enhancer RNAs in PD) demonstrate disease-specific modulation of neuroinflammatory pathways [71, 119].\n\nConsistent involvement of NF-\u03baB and cytokine signaling pathways implicates neuroinflammation as a global enhancer of neurodegeneration [199].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Converging subjects and integrated regulatory networks", "text": "Several circRNAs (circHectd1, circAKAP7, circLRP1B) converge in regulating autophagy in HD, PD, and AD [144, 146, 147], suggesting that maintaining proteostasis is a critical vulnerability node.\n\ncircRNA-mediated regulation of ubiquitin enzymes (e.g., ciRS-7/UBE2A in AD, circRIMS2/UBE2K in synaptic dysfunction) reveals common protein degradation defects [74, 88].\n\nEnhancer RNAs that modulate proteostasis genes exhibit disease-specific expression patterns but target common biological processes.\n\nEarly studies indicate potential direct and indirect interactions between circRNA and eRNA networks. For example, eRNAs that regulate BDNF expression may influence the biogenesis of activity-dependent circRNAs, while specific circRNAs could potentially feed back to modulate chromatin accessibility through RBP interactions. This multi-layered regulatory crosstalk represents a promising area for future research.\n\nA recurring pattern emerges throughout this review: eRNAs and circRNAs often converge on the same pathogenic pathways, but act at distinct regulatory levels. For example, in Alzheimer\u2019s disease, the eRNA AANCR transcriptionally activates APOE expression [105], whereas several circRNAs (e.g., circCORO1C, circHOMER1) posttranscriptionally regulate APP processing and \u03b3-secretase activity [48]. This multilevel regulation may provide robust homeostatic control in healthy states but amplify pathological cascades in disease.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Converging subjects and integrated regulatory networks", "text": "The potential for eRNA-circRNA interference\u2014in which enhancer-derived transcripts influence circRNA biogenesis or vice versa\u2014demonstrates an intriguing, albeit still speculative, layer of complexity. For instance, neuronal activity-dependent eRNAs could potentially modulate the expression of RNA-binding proteins (e.g., QKI, MBL) that control circRNA biogenesis, thereby creating a feed-forward loop between transcriptional and post-transcriptional regulation [200, 201].\n\nWhile miRNA sponging represents a well-known mechanism, our synthesis provides significant evidence for alternative circRNA functions in neurodegeneration:\n\ncircCwc27 modulates the nucleocytoplasmic trafficking of Pur-\u03b1, affecting its DNA binding capacity and transcriptional regulation of AD-related genes [60].\n\ncircFoxo3 acts as a protein scaffold and enhances interactions between senescence-related proteins in age-related cognitive decline [15].\n\ncircGigyf2 forms complexes containing CPSF6 to regulate mRNA processing and nuclear polyadenylation site selection [92].\n\ncircA\u03b2-a potentially encodes pathogenic peptides that contribute to amyloid pathology through non-canonical translational mechanisms [75].\n\nNuclear circRNAs (e.g., circEIF3J, circPAIP2) interact with U1 snRNP and RNA Pol II to enhance transcription of the parental gene [16].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Converging subjects and integrated regulatory networks", "text": "Emerging evidence suggests that circRNAs may participate in biomolecular condensations and potentially influence the subcellular organization of RNA-protein complexes in neurons. High local concentrations of RBPs on circRNA scaffolds could nucleate phase separation events associated with pathological aggregation.\n\nSimilarly, eRNAs exhibit mechanisms beyond the enhancer-promoter loop, including:\n\nDissociation of the NELF complex to release stalled RNA polymerase II [23].\n\nModulation of chromatin accessibility through recruitment of remodeling complexes [22].\n\nFacilitation of three-dimensional genome organization as predicted by computational models [22].\n\nThis functional diversity suggests that these ncRNAs act as multifaceted regulators rather than unidimensional sponges, opening new avenues for therapeutic targeting beyond miRNA inhibition.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Cross-disease patterns of circRNA/eRNA dysregulation", "text": "Our comprehensive review reveals significant convergent patterns of circRNA and eRNA dysregulation across major neurological disorders, suggesting common molecular mechanisms underlying apparently distinct pathological processes. Three core regulatory subjects consistently emerge:\n\nSeveral synapse-enriched circRNAs (e.g., circHOMER1, circGRIA1) show consistent downregulation in AD, PD, and HD, suggesting a common disruption in synaptic maintenance programs [192\u2013194].\n\nActivity-dependent eRNAs that regulate immediate early genes (e.g., Fos, Arc) show impaired induction in disease models, suggesting compromised activity-transcription coupling [195, 196].\n\nThe circHOMER1/miR-651/PSEN1-2 axis in AD [197] parallels the circSNCA/miR-7/SNCA axis in PD [198], revealing a conserved pattern of synaptic circRNAs that regulate key disease proteins through miRNA spongification.\n\nProinflammatory circRNAs (e.g., circ_0000950, circ_0005835) are upregulated in multiple disorders and drive microglial activation and cytokine production through shared miRNA pathways (miR-103, miR-576-3p) [63, 66].\n\nEnhancer RNAs that regulate immune response genes (e.g., AANCR-APOE in AD, SNCA enhancer RNAs in PD) demonstrate disease-specific modulation of neuroinflammatory pathways [71, 119].\n\nConsistent involvement of NF-\u03baB and cytokine signaling pathways implicates neuroinflammation as a global enhancer of neurodegeneration [199].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Cross-disease patterns of circRNA/eRNA dysregulation", "text": "Several circRNAs (circHectd1, circAKAP7, circLRP1B) converge in regulating autophagy in HD, PD, and AD [144, 146, 147], suggesting that maintaining proteostasis is a critical vulnerability node.\n\ncircRNA-mediated regulation of ubiquitin enzymes (e.g., ciRS-7/UBE2A in AD, circRIMS2/UBE2K in synaptic dysfunction) reveals common protein degradation defects [74, 88].\n\nEnhancer RNAs that modulate proteostasis genes exhibit disease-specific expression patterns but target common biological processes.\n\nEarly studies indicate potential direct and indirect interactions between circRNA and eRNA networks. For example, eRNAs that regulate BDNF expression may influence the biogenesis of activity-dependent circRNAs, while specific circRNAs could potentially feed back to modulate chromatin accessibility through RBP interactions. This multi-layered regulatory crosstalk represents a promising area for future research.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Synaptic plasticity and neuronal homeostasis", "text": "Several synapse-enriched circRNAs (e.g., circHOMER1, circGRIA1) show consistent downregulation in AD, PD, and HD, suggesting a common disruption in synaptic maintenance programs [192\u2013194].\n\nActivity-dependent eRNAs that regulate immediate early genes (e.g., Fos, Arc) show impaired induction in disease models, suggesting compromised activity-transcription coupling [195, 196].\n\nThe circHOMER1/miR-651/PSEN1-2 axis in AD [197] parallels the circSNCA/miR-7/SNCA axis in PD [198], revealing a conserved pattern of synaptic circRNAs that regulate key disease proteins through miRNA spongification.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "circRNAs:", "text": "Several synapse-enriched circRNAs (e.g., circHOMER1, circGRIA1) show consistent downregulation in AD, PD, and HD, suggesting a common disruption in synaptic maintenance programs [192\u2013194].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "eRNAs:", "text": "Activity-dependent eRNAs that regulate immediate early genes (e.g., Fos, Arc) show impaired induction in disease models, suggesting compromised activity-transcription coupling [195, 196].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Convergence:", "text": "The circHOMER1/miR-651/PSEN1-2 axis in AD [197] parallels the circSNCA/miR-7/SNCA axis in PD [198], revealing a conserved pattern of synaptic circRNAs that regulate key disease proteins through miRNA spongification.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Neuroinflammatory cascades", "text": "Proinflammatory circRNAs (e.g., circ_0000950, circ_0005835) are upregulated in multiple disorders and drive microglial activation and cytokine production through shared miRNA pathways (miR-103, miR-576-3p) [63, 66].\n\nEnhancer RNAs that regulate immune response genes (e.g., AANCR-APOE in AD, SNCA enhancer RNAs in PD) demonstrate disease-specific modulation of neuroinflammatory pathways [71, 119].\n\nConsistent involvement of NF-\u03baB and cytokine signaling pathways implicates neuroinflammation as a global enhancer of neurodegeneration [199].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "circRNAs:", "text": "Proinflammatory circRNAs (e.g., circ_0000950, circ_0005835) are upregulated in multiple disorders and drive microglial activation and cytokine production through shared miRNA pathways (miR-103, miR-576-3p) [63, 66].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "eRNAs:", "text": "Enhancer RNAs that regulate immune response genes (e.g., AANCR-APOE in AD, SNCA enhancer RNAs in PD) demonstrate disease-specific modulation of neuroinflammatory pathways [71, 119].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Cross-disease paradigm:", "text": "Consistent involvement of NF-\u03baB and cytokine signaling pathways implicates neuroinflammation as a global enhancer of neurodegeneration [199].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Protein quality control systems", "text": "Several circRNAs (circHectd1, circAKAP7, circLRP1B) converge in regulating autophagy in HD, PD, and AD [144, 146, 147], suggesting that maintaining proteostasis is a critical vulnerability node.\n\ncircRNA-mediated regulation of ubiquitin enzymes (e.g., ciRS-7/UBE2A in AD, circRIMS2/UBE2K in synaptic dysfunction) reveals common protein degradation defects [74, 88].\n\nEnhancer RNAs that modulate proteostasis genes exhibit disease-specific expression patterns but target common biological processes.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Autophagy-lysosomal pathways:", "text": "Several circRNAs (circHectd1, circAKAP7, circLRP1B) converge in regulating autophagy in HD, PD, and AD [144, 146, 147], suggesting that maintaining proteostasis is a critical vulnerability node.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Ubiquitin-proteasome system:", "text": "circRNA-mediated regulation of ubiquitin enzymes (e.g., ciRS-7/UBE2A in AD, circRIMS2/UBE2K in synaptic dysfunction) reveals common protein degradation defects [74, 88].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "eRNA contributions:", "text": "Enhancer RNAs that modulate proteostasis genes exhibit disease-specific expression patterns but target common biological processes.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Emerging evidence of crosstalk:", "text": "Early studies indicate potential direct and indirect interactions between circRNA and eRNA networks. For example, eRNAs that regulate BDNF expression may influence the biogenesis of activity-dependent circRNAs, while specific circRNAs could potentially feed back to modulate chromatin accessibility through RBP interactions. This multi-layered regulatory crosstalk represents a promising area for future research.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Transcriptional-posttranscriptional interface in neurodegeneration", "text": "A recurring pattern emerges throughout this review: eRNAs and circRNAs often converge on the same pathogenic pathways, but act at distinct regulatory levels. For example, in Alzheimer\u2019s disease, the eRNA AANCR transcriptionally activates APOE expression [105], whereas several circRNAs (e.g., circCORO1C, circHOMER1) posttranscriptionally regulate APP processing and \u03b3-secretase activity [48]. This multilevel regulation may provide robust homeostatic control in healthy states but amplify pathological cascades in disease.\n\nThe potential for eRNA-circRNA interference\u2014in which enhancer-derived transcripts influence circRNA biogenesis or vice versa\u2014demonstrates an intriguing, albeit still speculative, layer of complexity. For instance, neuronal activity-dependent eRNAs could potentially modulate the expression of RNA-binding proteins (e.g., QKI, MBL) that control circRNA biogenesis, thereby creating a feed-forward loop between transcriptional and post-transcriptional regulation [200, 201].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Beyond miRNA sponging: expanding the functional paradigm", "text": "While miRNA sponging represents a well-known mechanism, our synthesis provides significant evidence for alternative circRNA functions in neurodegeneration:", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Protein splicing and scaffolding", "text": "circCwc27 modulates the nucleocytoplasmic trafficking of Pur-\u03b1, affecting its DNA binding capacity and transcriptional regulation of AD-related genes [60].\n\ncircFoxo3 acts as a protein scaffold and enhances interactions between senescence-related proteins in age-related cognitive decline [15].\n\ncircGigyf2 forms complexes containing CPSF6 to regulate mRNA processing and nuclear polyadenylation site selection [92].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Translation regulation", "text": "circA\u03b2-a potentially encodes pathogenic peptides that contribute to amyloid pathology through non-canonical translational mechanisms [75].\n\nNuclear circRNAs (e.g., circEIF3J, circPAIP2) interact with U1 snRNP and RNA Pol II to enhance transcription of the parental gene [16].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "RBP coordination and phase separation", "text": "Emerging evidence suggests that circRNAs may participate in biomolecular condensations and potentially influence the subcellular organization of RNA-protein complexes in neurons. High local concentrations of RBPs on circRNA scaffolds could nucleate phase separation events associated with pathological aggregation.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "eRNA functional diversity", "text": "Similarly, eRNAs exhibit mechanisms beyond the enhancer-promoter loop, including:\n\nDissociation of the NELF complex to release stalled RNA polymerase II [23].\n\nModulation of chromatin accessibility through recruitment of remodeling complexes [22].\n\nFacilitation of three-dimensional genome organization as predicted by computational models [22].\n\nThis functional diversity suggests that these ncRNAs act as multifaceted regulators rather than unidimensional sponges, opening new avenues for therapeutic targeting beyond miRNA inhibition.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Challenges and future directions", "text": "Despite the growing interest in circRNAs and eRNAs as key regulators of neurodegeneration, we face several challenges before these findings can be applied in clinical settings:\n\ncircRNAs and eRNAs are often present in low concentrations in biofluids like blood, plasma, serum, or exosomes [202]. To identify disease-specific signatures amidst the background noise, we need highly sensitive techniques such as single-molecule sequencing [203] or digital PCR [204]. Many eRNAs are fleeting and unstable, which complicates functional testing [205]. We need better RNA-stabilization methods and nascent RNA capture techniques, like GRO-seq [206].\n\nBulk tissue studies often mix the circRNA/eRNA profiles from both neuronal and glial cells. By using single-cell or nucleus RNA sequencing on samples from patients, we can start to understand what these molecules do in specific cell types [207, 208].\n\nWhile many studies link circRNA/eRNA dysregulation to various diseases, they don\u2019t always show that one causes the other. To establish these connections, we need to use conditional knockout or overexpression models, like AAV-driven circRNA sponges [209].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Challenges and future directions", "text": "In terms of specificity and standardization, some candidate circRNAs linked to Alzheimer\u2019s disease also show changes in conditions like cancer or inflammation [210]. This means we need to validate these findings across different diseases. Moreover, the lack of standardized methods for isolating circRNAs from biofluids\u2014whether they are exosomal or free\u2014limits our ability to reproduce results.\n\nMost current biomarker research is primarily cross-sectional. We really need large-scale longitudinal studies that include pre-symptomatic patients with Alzheimer\u2019s or Parkinson\u2019s disease to track circRNA and eRNA changes as the diseases progress.\n\nDelivering RNA-based therapies across the blood-brain barrier remains difficult. Researchers are exploring nanoparticle and exosome-based delivery systems that specifically target brain-related circRNAs and eRNAs [211, 212]. Also, the use of artificial intelligence in bioinformatics is improving how we model antigen screening and delivery systems [213]. We need to understand the off-target effects of circRNA modulation, like unintended miRNA sponging, which requires thorough in vivo safety tests.\n\nSince eRNAs operate in cis, it\u2019s tricky to target them without affecting the nearby chromatin structure. Techniques such as CircRNA-based AntimiR therapy [214], CRISPR-dCas9 [215], or antisense oligonucleotides could enable precise modulation of enhancers.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Challenges and future directions", "text": "Combining circRNA/eRNA data with epigenomic studies (like ATAC-seq and ChIP-seq) and proteomic data will reveal the regulatory networks involved in neurodegeneration.\n\nBesides miRNA sponging, circRNAs can encode peptides or serve as scaffolds for protein complexes, like Tau aggregation inhibitors [1]. We need a detailed analysis of their protein-binding landscapes.\n\nCurrent animal models often fail to accurately reflect human-specific circRNA/eRNA expression. Investigating cerebral organoids or xenotransplant models might help close this gap.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Technical and mechanistic challenges", "text": "circRNAs and eRNAs are often present in low concentrations in biofluids like blood, plasma, serum, or exosomes [202]. To identify disease-specific signatures amidst the background noise, we need highly sensitive techniques such as single-molecule sequencing [203] or digital PCR [204]. Many eRNAs are fleeting and unstable, which complicates functional testing [205]. We need better RNA-stabilization methods and nascent RNA capture techniques, like GRO-seq [206].\n\nBulk tissue studies often mix the circRNA/eRNA profiles from both neuronal and glial cells. By using single-cell or nucleus RNA sequencing on samples from patients, we can start to understand what these molecules do in specific cell types [207, 208].\n\nWhile many studies link circRNA/eRNA dysregulation to various diseases, they don\u2019t always show that one causes the other. To establish these connections, we need to use conditional knockout or overexpression models, like AAV-driven circRNA sponges [209].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Detection and quantification:", "text": "circRNAs and eRNAs are often present in low concentrations in biofluids like blood, plasma, serum, or exosomes [202]. To identify disease-specific signatures amidst the background noise, we need highly sensitive techniques such as single-molecule sequencing [203] or digital PCR [204]. Many eRNAs are fleeting and unstable, which complicates functional testing [205]. We need better RNA-stabilization methods and nascent RNA capture techniques, like GRO-seq [206].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Cell-type specificity:", "text": "Bulk tissue studies often mix the circRNA/eRNA profiles from both neuronal and glial cells. By using single-cell or nucleus RNA sequencing on samples from patients, we can start to understand what these molecules do in specific cell types [207, 208].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Cause vs. consequence:", "text": "While many studies link circRNA/eRNA dysregulation to various diseases, they don\u2019t always show that one causes the other. To establish these connections, we need to use conditional knockout or overexpression models, like AAV-driven circRNA sponges [209].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Biomarker development", "text": "In terms of specificity and standardization, some candidate circRNAs linked to Alzheimer\u2019s disease also show changes in conditions like cancer or inflammation [210]. This means we need to validate these findings across different diseases. Moreover, the lack of standardized methods for isolating circRNAs from biofluids\u2014whether they are exosomal or free\u2014limits our ability to reproduce results.\n\nMost current biomarker research is primarily cross-sectional. We really need large-scale longitudinal studies that include pre-symptomatic patients with Alzheimer\u2019s or Parkinson\u2019s disease to track circRNA and eRNA changes as the diseases progress.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Therapeutic potential", "text": "Delivering RNA-based therapies across the blood-brain barrier remains difficult. Researchers are exploring nanoparticle and exosome-based delivery systems that specifically target brain-related circRNAs and eRNAs [211, 212]. Also, the use of artificial intelligence in bioinformatics is improving how we model antigen screening and delivery systems [213]. We need to understand the off-target effects of circRNA modulation, like unintended miRNA sponging, which requires thorough in vivo safety tests.\n\nSince eRNAs operate in cis, it\u2019s tricky to target them without affecting the nearby chromatin structure. Techniques such as CircRNA-based AntimiR therapy [214], CRISPR-dCas9 [215], or antisense oligonucleotides could enable precise modulation of enhancers.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Delivery challenges:", "text": "Delivering RNA-based therapies across the blood-brain barrier remains difficult. Researchers are exploring nanoparticle and exosome-based delivery systems that specifically target brain-related circRNAs and eRNAs [211, 212]. Also, the use of artificial intelligence in bioinformatics is improving how we model antigen screening and delivery systems [213]. We need to understand the off-target effects of circRNA modulation, like unintended miRNA sponging, which requires thorough in vivo safety tests.\n\nSince eRNAs operate in cis, it\u2019s tricky to target them without affecting the nearby chromatin structure. Techniques such as CircRNA-based AntimiR therapy [214], CRISPR-dCas9 [215], or antisense oligonucleotides could enable precise modulation of enhancers.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Precision targeting:", "text": "Since eRNAs operate in cis, it\u2019s tricky to target them without affecting the nearby chromatin structure. Techniques such as CircRNA-based AntimiR therapy [214], CRISPR-dCas9 [215], or antisense oligonucleotides could enable precise modulation of enhancers.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Emerging opportunities", "text": "Combining circRNA/eRNA data with epigenomic studies (like ATAC-seq and ChIP-seq) and proteomic data will reveal the regulatory networks involved in neurodegeneration.\n\nBesides miRNA sponging, circRNAs can encode peptides or serve as scaffolds for protein complexes, like Tau aggregation inhibitors [1]. We need a detailed analysis of their protein-binding landscapes.\n\nCurrent animal models often fail to accurately reflect human-specific circRNA/eRNA expression. Investigating cerebral organoids or xenotransplant models might help close this gap.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Multi-omics integration:", "text": "Combining circRNA/eRNA data with epigenomic studies (like ATAC-seq and ChIP-seq) and proteomic data will reveal the regulatory networks involved in neurodegeneration.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "CircRNA/protein interactions:", "text": "Besides miRNA sponging, circRNAs can encode peptides or serve as scaffolds for protein complexes, like Tau aggregation inhibitors [1]. We need a detailed analysis of their protein-binding landscapes.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Humanized models:", "text": "Current animal models often fail to accurately reflect human-specific circRNA/eRNA expression. Investigating cerebral organoids or xenotransplant models might help close this gap.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Conclusion and perspective", "text": "Our understanding of circRNA biogenesis and function in the brain is still in its early stages, requiring more rigorous investigation. Implementing sophisticated research methods and strategies to modulate circRNA expression and function while maintaining the integrity of parental gene transcription would greatly enhance our understanding of circRNA biology in the brain [41]. The complex regulatory potential of circRNAs is evident in various human diseases, where they operate within networks comprising circRNAs, RNA-binding proteins, and microRNAs [216].\n\nEven with the progress in identifying and characterizing circRNAs, many questions remain unanswered. Healthy cells intricately regulate circRNA biogenesis through both cis sequences and trans-acting factors. However, changes in the activity of these factors under pathological conditions can either increase or decrease circRNA production. We still don\u2019t fully understand how function-related circRNAs are generated, distributed, or what role tissue-specific expression plays in creating different regulatory patterns in various biological processes. Additionally, the mechanisms governing tissue-specific expression of circRNAs and the link between their specificity and abundance in particular tissues and developmental stages are still unclear [217].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Conclusion and perspective", "text": "As we pay more attention to RNA biomarkers in neurodegenerative diseases, circRNAs stand out as potential reliable and cost-effective candidates. They have a unique circular structure that makes them more resistant to RNase degradation and gives them distinct structural forms compared to linear RNAs. Notably, these molecules can be detected in the bloodstream, boosting their potential not just as indicators of disease but also as targets for molecular interventions. They also tend to accumulate in the brain, especially with age. This characteristic increases their appeal for research into neurodegenerative biomarkers. It has been found that the circRNAs differentially expressed in the brain also overlap with those found in the plasma of patients with neurodegenerative diseases. This opens up the possibility of using circRNAs as peripheral biomarkers, which could lead to non-invasive diagnostic methods in the future [217].\n\nMoreover, active enhancers produce eRNAs, which play a crucial role in regulating gene expression by affecting the transcription of target genes and exhibit unique expression profiles. eRNAs come in various types with different structures, lengths, and post-transcriptional modifications. New evidence indicates that eRNAs are significant in various cellular processes and are involved in gene regulation through enhancer-promoter chromatin looping, modulating chromatin state, and interacting with transcriptional machinery [218].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Conclusion and perspective", "text": "Despite the recent surge in research into eRNAs, considerable gaps remain in our understanding of their biological functions, necessitating more investigation and clarity. Since most eRNAs are unstable, we must determine if they can function like other non-coding RNAs.\n\nWe also need to develop better frameworks for modulating eRNA expression to ensure successful functional eRNA research [19]. Research has recently clarified the role of eRNAs in neurodegenerative diseases using a mix of bioinformatic and experimental approaches. Bioinformatics has shown that neurodegenerative diseases have a broader range of eRNA expression patterns. This finding enhances our understanding of the regulatory components linked to neurodegenerative disorders [40]. Recent studies have connected eRNA dysregulation to neurodegenerative disorders like AD, PD, and HD. This association offers the potential to deepen our understanding of the underlying disease mechanisms and explore possible clinical applications of eRNAs [219\u2013221].", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Conclusion and perspective", "text": "In summary, this review goes beyond cataloging individual ncRNAs to provide an integrated model of neurodegenerative pathogenesis. We demonstrate that circRNAs and eRNAs, although acting at different regulatory levels, often converge on critical pathways governing proteostasis, neuroinflammation, and synaptic function. Their potential interactions\u2014whether synergistic, compensatory, or sequential\u2014create complex regulatory networks that both maintain neuronal health and, when disrupted, drive disease progression. Future research should prioritize mapping these inter-RNA networks, developing multitarget strategies, and using computational approaches to identify key nodal points for therapeutic intervention. By moving beyond the conventional silo approach to ncRNA research, we can unleash their full potential as biomarkers and targets in the fight against neurodegenerative disorders. Ultimately, we understand that this research on eRNAs in neurodegeneration is less comprehensive than that on circRNAs, a variation observed in the depth of discussion between parts of this review. More studies focused on describing the enhancer topography and effective eRNAs in the human brain are fundamental to obtaining a more balanced and integrated comprehensive understanding of ncRNA-based regulatory networks in health and condition.", "title": "" }, { "pmid": 41618467, "pmcid": "PMC12895692", "section": "Electronic supplementary material", "text": "Below is the link to the electronic supplementary material.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Introduction", "text": "Neuroinflammation is a common characteristic feature in neurodegenerative diseases (NDDs) and aging brains, which plays a major role in cognitive decline and neurological problems. NDDs such as Alzheimer\u2019s disease (AD), Parkinson\u2019s disease (PD), Huntington\u2019s disease (HD), and amyotrophic lateral sclerosis (ALS) have all been associated with neuroinflammation, which further speeds up disease progression.\n\nMicroglia, the brain resident immune cells, play an important role in the regulation of neuroinflammation and synaptic degeneration. Microglial activation leads to both beneficial and harmful effects [1, 2]. In a healthy brain, microglia maintain a dynamic balance between protective, anti-inflammatory functions and detrimental, pro-inflammatory signaling [1, 3, 4]. However, in NDDs, this microglia balance is disrupted, and the activation of pro-inflammatory microglia contributes to neuronal damage and synaptic loss [5\u20137]. The inflammatory response amplifies through the release of cytokines, chemokines, and reactive oxygen species (ROS), creating a cycle of chronic inflammation that accelerates neurodegeneration and results in cognitive and motor impairments [8, 9].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Introduction", "text": "Mitochondria, \u201cpowerhouses\u201d of cells, are critical for neuronal survival due to their central role in energy production, calcium buffering, redox balance, and regulation of apoptotic pathways. Mitochondria contain their own DNA (mitochondrial DNA, mtDNA) in multiple copies and encode for 13 proteins required for oxidative phosphorylation (OXPHOS). mtDNA is vulnerable to oxidative damage due to its proximity to ROS production sites and the relatively lower capacity of DNA repair pathways. Unlike nuclear DNA, mtDNA lacks histones and is therefore more susceptible to oxidative damage, deletions, and point mutations. Accumulation of mtDNA damage could impair the OXPHOS system, leading to defects in electron transport chain (ETC) activity, increased ROS generation, and progressive bioenergetic failure, which contribute to the pathogenesis and progression of NDDs [10]. Further, mitochondrial structural damage compromises membrane integrity, which allows the release of mitochondrial contents such as mtDNA, cardiolipin, and mitochondrial proteins into the cytosol. The released mitochondrial content in cytosol or extracellular spaces acts as danger-associated molecular patterns (DAMP) and triggers innate immune responses. Cytosolic or extracellular mtDNA induces an innate immune response, which triggers neuroinflammation via immune sensors like the cyclic GMP\u2013AMP synthase (cGAS)-stimulator of interferon genes (STING) pathway and Toll-like receptors (TLR) [11, 12]. Therefore, understanding the mtDNA-mediated neuroinflammation is crucial to develop targeted therapies for NDDs and the aging brain.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Introduction", "text": "In recent years, substantial progress has been made to understand the role of neuroinflammation in the aging brain and NDDs; however, many questions remain unanswered. Specifically, the mechanisms that activate and regulate neuroinflammation and interplay between mtDNA, immune signaling, and neuroinflammatory pathways are not fully understood. This review aims to explore the emerging role of mtDNA as a neuroinflammatory switch. Here, we review the mechanisms of mtDNA release that triggers immune responses in the brain, and we highlight its potential as a therapeutic target in aging brain and NDDs. Understanding how mitochondrial dysfunction contributes to mtDNA release and subsequent neuroinflammation will open new avenues for therapeutic interventions aimed at mitigating neuroinflammation and its detrimental effects on brain health.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "mtDNA structure, maintenance, and vulnerability", "text": "In humans, mtDNA is a circular genome of 16,500 base pairs, which is located in many copies in the mitochondrial matrix [13]. mtDNA encodes a total of 37 genes, of which 13 code for proteins involved in OXPHOS, 22 code for tRNA, and 2 code for rRNA [14]. This distinct gene set on mtDNA enables mitochondria to have their own independent protein synthesis system, allowing them to carry out ATP synthesis efficiently. However, the structural, functional, and maintenance work of mitochondria are dependent on thousands of nuclear-encoded proteins. Neurons are among the most energy-demanding cells in the body, and their health and proper functioning depend on mitochondria to keep up with this constant demand. mtDNA is an important factor for OXPHOS and ATP production; therefore, it supports neuronal activity, synaptic transmission, and the maintenance of bioenergetic homeostasis\u00a0[15]. Mitochondria synthesize ATP through OXPHOS, which involves the transfer of electrons through the ETC. However, during ETC, excessive ROS is generated, including superoxide anions and hydrogen peroxide. mtDNA proximity to these ROS production sites makes them more vulnerable to damage, mutations, deletions, and genetic instability [16]. This constant mtDNA exposure to ROS increases the oxidative damage probability, which accumulates over time, and leads to mitochondrial dysfunction [17, 18]. Mitochondria possess some DNA repair enzymes, such as mitochondrial polymerase gamma (POLG) and a version of base excision repair; however, these systems are not as efficient as those in the nucleus [19]. The combination of high exposure to ROS and inefficient repair systems results in a higher mutation rate in mtDNA and a decline of mitochondrial ability to produce energy [20]. Some mtDNA regions are more prone to mutations, particularly in genes encoding respiratory complex components. One of the most frequently observed mtDNA mutations is the large-scale 4977-bp deletion, also known as the \u201ccommon deletion\u201d, which spans several OXPHOS genes and is often flanked by direct repeats that predispose these regions to slipped-strand mispairing during replication [21, 22]. These deletions accumulate with age and have been observed at high levels in NDD [23\u201325]. mtDNA copy number is identified as a dynamic parameter that represents mitochondrial biogenesis and turnover. In aging brains and NDDs, mtDNA copy number alterations have been documented, which further compromise OXPHOS and respiratory capacity [26\u201330]. The functional consequences of accumulated mtDNA damage are profound, as oxidization, mutation, or deletion of mtDNA molecules can impair the assembly and function of respiratory chain complexes, leading to deficient ATP synthesis and increased ROS production [19]. Beyond the bioenergetic defects, damaged mtDNA may leak from mitochondria into the cytoplasm or extracellular space [31]. Such extranuclear mtDNA acts as a potent DAMP, capable of activating innate immune sensors including cGAS, TLR9, and the NOD, LRR and pyrin domain-containing protein 3 (NLRP3) inflammasome, thus triggering neuroinflammatory cascades [32\u201334]. Neurons, defined by their high metabolic demand, long lifespan, and limited regenerative capacity, are especially vulnerable to mitochondrial dysfunction. Accumulation of mtDNA damage is particularly detrimental and has been implicated in the pathogenesis of NDDs such as AD, PD, HD, and ALS. Over time, progressive mtDNA mutations compromise mitochondrial integrity, leading to increased neuroinflammation, synaptic dysfunction, and neuronal loss. This vicious cycle of mtDNA damage accumulation and mitochondrial dysfunction further accelerates neurodegeneration and disease progression.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Mechanisms of mtDNA release", "text": "The mtDNA is normally stored in the mitochondrial matrix, enclosed by the inner and outer mitochondrial membranes (OMM), protecting it from recognition by cellular immune sensors. Under physiological conditions, mitochondrial outer and inner membranes ensure that mtDNA primarily functions for mitochondrial gene expression and energy metabolism. However, during mitochondrial stress (physical, chemical, and biological) and damage, mtDNA could be abnormally released into the cytosol or extracellular space. This cytosolic or extracellular mtDNA functions as a potential DAMP, triggering innate immune responses and neuroinflammation. Diverse physical, chemical, and biological insults can trigger mtDNA release from mitochondria by different mechanisms (Fig.\u00a01).Fig.\u00a01Schematic illustration depicting major cellular pathways that promote mtDNA escape from mitochondria into the cytosol and extracellular space under neurodegenerative conditions. Neurodegenerative conditions induce mitochondrial membrane permeabilization through BAX/BAK macropores, mitochondrial permeability transition pore (mPTP) opening, and Gasdermin (GSDMD) pores. Defective mitophagy, apoptosis, and necrosis further exacerbate mtDNA accumulation", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Mechanisms of mtDNA release", "text": "Schematic illustration depicting major cellular pathways that promote mtDNA escape from mitochondria into the cytosol and extracellular space under neurodegenerative conditions. Neurodegenerative conditions induce mitochondrial membrane permeabilization through BAX/BAK macropores, mitochondrial permeability transition pore (mPTP) opening, and Gasdermin (GSDMD) pores. Defective mitophagy, apoptosis, and necrosis further exacerbate mtDNA accumulation", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Mechanisms of mtDNA release", "text": "mPTP is a multi-protein complex that forms a non-selective channel across the inner mitochondrial membrane [35]. When mPTP is open, it allows the passage of solutes\u2009\u2264\u20091.5\u00a0kDa, which leads to mitochondrial depolarization, swelling, rupture of the outer membrane, and release of material including mtDNA and cytochrome c [36]. mPTP opening can be caused by several factors, including higher mitochondrial Ca2\u207a level, increased oxidative stress, inorganic phosphate, and depletion of adenine nucleotide [35]. Opening of the mPTP plays a crucial role in cell death pathways, particularly in ischemia\u2013reperfusion injury and different NDDs [37, 222]. Under pathological conditions such as calcium overload, higher ROS, ATP depletion, or ischemic injury, the mPTP can open transiently or persistently, leading to dissipation of the mitochondrial membrane potential, matrix swelling, and rupture of OMM [38\u201341]. This rupture of OMM permits the release of mitochondrial contents, which includes mtDNA fragments, into the cytosol, where they act as a potent DAMP [42, 43]. Although the precise molecular composition of the mPTP remains under investigation, cyclophilin D (CypD) is recognized as a key regulatory component [44, 45]. Pharmacological inhibition of CypD by cyclosporin A or genetic deletion reduces mPTP opening, mtDNA release, and downstream inflammation in various disease models [46\u201348]. In neurons, mPTP-mediated mtDNA release has been implicated in excitotoxicity and neuroinflammation, linking mitochondrial dysfunction to neurodegenerative pathology [12, 49].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Mechanisms of mtDNA release", "text": "In NDDs, persistent mPTP opening is increasingly recognized as a common mechanism driving neuronal injury [50]. In AD, A\u03b2 accumulates in mitochondria and interacts with CypD, and promotes oxidative stress, collectively lowering the calcium threshold for mPTP opening. CypD deletion or inhibition rescued mitochondrial function and cognitive performance in murine models [51]. In PD, mitochondrial toxins like MPTP (1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine) and rotenone induce oxidative stress and calcium overload, while \u03b1-synuclein (\u03b1-syn) accumulates at mitochondria and induces mPTP opening [52]. In HD, mutant Huntingtin (mHTT) disrupts calcium levels and increases mitochondrial ROS, lowering the threshold for calcium-induced mPTP opening in mitochondria [53, 54]. Together, these studies establish a common pathological cascade in NDDs. Disease-specific proteins (A\u03b2, \u03b1-syn, mHTT) induce mitochondrial calcium overload and oxidative stress, which sensitize the mPTP, leading to mitochondrial failure, mtDNA release, and neuronal degeneration.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Mechanisms of mtDNA release", "text": "Mitochondria maintain their functional integrity through a coordinated quality control network, which includes mitochondrial fission and fusion, proteolytic degradation of oxidized/damaged proteins, and the removal and/or recycle of dysfunctional mitochondria via mitophagy [55, 56]. Mitophagy is crucial for removing depolarized, dysfunctional, or oxidized mitochondria that contain oxidatively damaged mtDNA [57, 58]. The canonical PINK1\u2013Parkin-mediated mitophagy enables the recognition and ubiquitination of damaged mitochondria, facilitating their autophagic clearance, while the receptor-mediated pathways involving BNIP3, NIX, and FUNDC1 provide additional layers of regulation, particularly under stress conditions relevant to aging and NDDs [59\u201362]. Compromised mitophagy, which is increasingly observed with aging and in NDDs, allows damaged mitochondria to accumulate, increasing the possibility of mtDNA instability and release [60, 63]. In microglia, age-associated decline in mitophagic efficiency exacerbates mitochondrial dysfunction, promoting excessive ROS production and calcium dysregulation. These conditions sensitize mitochondria to prolonged mPTP opening, leading to matrix swelling, rupture of the OMM, and subsequent release of mtDNA fragments into the cytosol [42, 45, 48]. The cytosolic mtDNA can then act as a potent DAMP, activating innate immune pathways and sustaining chronic neuroinflammatory signaling in disorders such as AD, PD and HD.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Mechanisms of mtDNA release", "text": "In NDDs, impaired mitophagy causes accumulation of defective mitochondria containing oxidatively damaged mtDNA, which, upon release, can activate innate immune sensors, like cGAS, and promote neuroinflammation [63, 64]. Notably, the presence of A\u03b2 exacerbates mitochondrial ROS production and impairs mitophagy, further enhancing mitochondrial dysfunction [65, 66]. In PD, mutations in the PINK1 and Parkin genes, which encode key regulatory proteins of mitophagy, prevent the efficient clearance of damaged mitochondria [60]. As a result, neurons accumulate damaged or dysfunctional mitochondria,\u00a0leading to excessive ROS production, increased sensitivity to mPTP opening, and cytosolic release of mtDNA, which activates innate immune signaling in microglia and accelerates neurodegeneration [67]. Similarly, in HD, mHTT not only impairs autophagy initiation and mitochondrial trafficking, but also directly inhibits the TIM23 protein import complex, reducing the import of nuclear-encoded mitochondrial proteins and causing early mitochondrial dysfunction and oxidative stress [68, 69]. This combination of defective mitophagy and impaired protein import exacerbates the proteostatic stress, promotes aggregate formation, and further compromises mitochondrial dynamics. The released mtDNA can activate cGAS\u2013STING signaling, creating a self-reinforcing loop where inflammation further impairs mitochondrial quality control, promotes additional mtDNA release, and accelerates neuronal degeneration. Collectively, these observations highlight a convergent pathogenic mechanism in NDDs, in which mitochondrial quality control failure, through impaired mitophagy, defective protein import, and ROS-driven dysfunction, links mitochondrial damage, mtDNA-mediated inflammation, and protein aggregation [68\u201373].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Mechanisms of mtDNA release", "text": "Mitochondria are known to be a center for neuronal survival and death, which integrates metabolic signals and stress responses to decide and activate apoptosis. In the intrinsic apoptotic pathway, the BCL-2 family proteins play a critical regulatory role. Pro-apoptotic effectors BAX and BAK oligomerize upon activation, inducing mitochondrial outer membrane permeabilization (MOMP), a crucial event during apoptosis [73, 74]. Traditionally, MOMP is recognized for cytosolic release of cytochrome c, Smac/DIABLO, and other intermembrane space proteins, thereby activating caspase cascades and executing apoptosis [75]. However, recent work has demonstrated that MOMP facilitates the mtDNA release into the cytosol [76, 77]. Mechanistically, BAX and BAK assemble into large, lipidic pores within the OMM that can dynamically enlarge over time [78]. During apoptosis, the BAX/BAK pores permit protrusion or \u201cherniation\u201d of the inner mitochondrial membrane, carrying mitochondrial matrix content, including mtDNA, into the cytosol [76]. This process effectively breaks the mitochondrial compartmentalization that normally keeps mtDNA away from innate immune sensors. Unlike transient mPTP observed during mitophagy, apoptotic MOMP often results in continuous membrane disruption, greatly increasing the possibility of mtDNA release. Importantly, even in the absence of full caspase activation, BAX/BAK-mediated membrane permeabilization can still cause mtDNA leakage and trigger inflammatory responses [79]. This phenomenon is particularly relevant in post-mitotic neurons, where caspase activation is often tightly controlled, yet sub-lethal mitochondrial permeabilization can still promote chronic inflammation.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Mechanisms of mtDNA release", "text": "In the context of NDDs, this mechanism has profound implications. In AD, A\u03b2 accumulation within neurons increases mitochondrial ROS production and sensitizes mitochondria to apoptosis by promoting BAX activation [80]. The activated BAX forms large enough pores to release mtDNA into the cytosol, which then amplifies neuroinflammation through the cGAS\u2013STING signaling. Similarly, in PD, \u03b1-syn aggregates localize to mitochondria, impair mitochondrial dynamics, and increase oxidative stress, which together trigger BAX/BAK-mediated MOMP [81]. In HD, mHTT impairs mitochondrial calcium buffering and promotes oxidative stress, facilitating BAX recruitment to mitochondria [82]. BAX/BAK pore formation in HD models can similarly enable mtDNA release, feeding into a cycle of inflammation and neurodegeneration. These insights reveal a dual pathogenic role of BAX/BAK-mediated MOMP in neurological diseases beyond executing apoptosis: it can directly promote mtDNA release, which can activate neuroinflammation and drive disease progression. This convergence of mitochondrial dysfunction, apoptosis, and innate immune activation highlights therapeutic opportunities. Strategies that stabilize mitochondrial membranes, modulate BAX/BAK pore dynamics [83], or block downstream mtDNA sensing pathways [84] could inhibit inflammation without fully blocking apoptosis.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Mechanisms of mtDNA release", "text": "Unlike the controlled, typically non-inflammatory process of apoptosis, necrotic and necroptotic cell death are characterized by plasma membrane integrity loss, resulting in the extracellular release of intracellular components, including intact mitochondria or mitochondrial fragments that harbor mtDNA [85, 86]. Similarly, pyroptosis, a caspase-1-dependent form of cell death, induces gasdermin-D pore formation in the plasma membrane, culminating in cellular swelling and rupture [87, 88]. These lytic cell death pathways promote the release of DAMPs, notably mtDNA, into the extracellular milieu [33, 89]. Extracellular mtDNA could engage pattern recognition receptors (PRR) on microglia, astrocytes, and infiltrating immune cells, serving as a potent pro-inflammatory signal that drives neuroinflammation in acute brain injury, ischemia, and chronic NDDs.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Mechanisms of mtDNA release", "text": "Recent studies have revealed that cells actively secrete EVs, such as exosomes and microvesicles [90]. These EVs can encapsulate whole mitochondria, mitochondrial fragments, or naked mtDNA, shielding them from extracellular degradation and enabling intercellular communication [90]. Under conditions of cellular stress, neurons, astrocytes, and microglia have all been shown to release mtDNA-containing EVs [67, 91]. This EV-mediated transfer of mtDNA may amplify neuroinflammatory responses by activating PRR in recipient cells, thereby facilitating the spread of pathological signals across neural networks. Mechanistically, EV biogenesis depends on the endosomal sorting complexes required for transport machinery and is modulated by stress-responsive pathways, although the precise molecular mechanisms that govern selective packaging of mtDNA into EV remain incompletely understood [92].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Mitochondrial permeability transition pore (mPTP) opening and membrane rupture", "text": "mPTP is a multi-protein complex that forms a non-selective channel across the inner mitochondrial membrane [35]. When mPTP is open, it allows the passage of solutes\u2009\u2264\u20091.5\u00a0kDa, which leads to mitochondrial depolarization, swelling, rupture of the outer membrane, and release of material including mtDNA and cytochrome c [36]. mPTP opening can be caused by several factors, including higher mitochondrial Ca2\u207a level, increased oxidative stress, inorganic phosphate, and depletion of adenine nucleotide [35]. Opening of the mPTP plays a crucial role in cell death pathways, particularly in ischemia\u2013reperfusion injury and different NDDs [37, 222]. Under pathological conditions such as calcium overload, higher ROS, ATP depletion, or ischemic injury, the mPTP can open transiently or persistently, leading to dissipation of the mitochondrial membrane potential, matrix swelling, and rupture of OMM [38\u201341]. This rupture of OMM permits the release of mitochondrial contents, which includes mtDNA fragments, into the cytosol, where they act as a potent DAMP [42, 43]. Although the precise molecular composition of the mPTP remains under investigation, cyclophilin D (CypD) is recognized as a key regulatory component [44, 45]. Pharmacological inhibition of CypD by cyclosporin A or genetic deletion reduces mPTP opening, mtDNA release, and downstream inflammation in various disease models [46\u201348]. In neurons, mPTP-mediated mtDNA release has been implicated in excitotoxicity and neuroinflammation, linking mitochondrial dysfunction to neurodegenerative pathology [12, 49].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Mitochondrial permeability transition pore (mPTP) opening and membrane rupture", "text": "In NDDs, persistent mPTP opening is increasingly recognized as a common mechanism driving neuronal injury [50]. In AD, A\u03b2 accumulates in mitochondria and interacts with CypD, and promotes oxidative stress, collectively lowering the calcium threshold for mPTP opening. CypD deletion or inhibition rescued mitochondrial function and cognitive performance in murine models [51]. In PD, mitochondrial toxins like MPTP (1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine) and rotenone induce oxidative stress and calcium overload, while \u03b1-synuclein (\u03b1-syn) accumulates at mitochondria and induces mPTP opening [52]. In HD, mutant Huntingtin (mHTT) disrupts calcium levels and increases mitochondrial ROS, lowering the threshold for calcium-induced mPTP opening in mitochondria [53, 54]. Together, these studies establish a common pathological cascade in NDDs. Disease-specific proteins (A\u03b2, \u03b1-syn, mHTT) induce mitochondrial calcium overload and oxidative stress, which sensitize the mPTP, leading to mitochondrial failure, mtDNA release, and neuronal degeneration.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Dysfunctional mitophagy and impaired mitochondrial quality control", "text": "Mitochondria maintain their functional integrity through a coordinated quality control network, which includes mitochondrial fission and fusion, proteolytic degradation of oxidized/damaged proteins, and the removal and/or recycle of dysfunctional mitochondria via mitophagy [55, 56]. Mitophagy is crucial for removing depolarized, dysfunctional, or oxidized mitochondria that contain oxidatively damaged mtDNA [57, 58]. The canonical PINK1\u2013Parkin-mediated mitophagy enables the recognition and ubiquitination of damaged mitochondria, facilitating their autophagic clearance, while the receptor-mediated pathways involving BNIP3, NIX, and FUNDC1 provide additional layers of regulation, particularly under stress conditions relevant to aging and NDDs [59\u201362]. Compromised mitophagy, which is increasingly observed with aging and in NDDs, allows damaged mitochondria to accumulate, increasing the possibility of mtDNA instability and release [60, 63]. In microglia, age-associated decline in mitophagic efficiency exacerbates mitochondrial dysfunction, promoting excessive ROS production and calcium dysregulation. These conditions sensitize mitochondria to prolonged mPTP opening, leading to matrix swelling, rupture of the OMM, and subsequent release of mtDNA fragments into the cytosol [42, 45, 48]. The cytosolic mtDNA can then act as a potent DAMP, activating innate immune pathways and sustaining chronic neuroinflammatory signaling in disorders such as AD, PD and HD.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Dysfunctional mitophagy and impaired mitochondrial quality control", "text": "In NDDs, impaired mitophagy causes accumulation of defective mitochondria containing oxidatively damaged mtDNA, which, upon release, can activate innate immune sensors, like cGAS, and promote neuroinflammation [63, 64]. Notably, the presence of A\u03b2 exacerbates mitochondrial ROS production and impairs mitophagy, further enhancing mitochondrial dysfunction [65, 66]. In PD, mutations in the PINK1 and Parkin genes, which encode key regulatory proteins of mitophagy, prevent the efficient clearance of damaged mitochondria [60]. As a result, neurons accumulate damaged or dysfunctional mitochondria,\u00a0leading to excessive ROS production, increased sensitivity to mPTP opening, and cytosolic release of mtDNA, which activates innate immune signaling in microglia and accelerates neurodegeneration [67]. Similarly, in HD, mHTT not only impairs autophagy initiation and mitochondrial trafficking, but also directly inhibits the TIM23 protein import complex, reducing the import of nuclear-encoded mitochondrial proteins and causing early mitochondrial dysfunction and oxidative stress [68, 69]. This combination of defective mitophagy and impaired protein import exacerbates the proteostatic stress, promotes aggregate formation, and further compromises mitochondrial dynamics. The released mtDNA can activate cGAS\u2013STING signaling, creating a self-reinforcing loop where inflammation further impairs mitochondrial quality control, promotes additional mtDNA release, and accelerates neuronal degeneration. Collectively, these observations highlight a convergent pathogenic mechanism in NDDs, in which mitochondrial quality control failure, through impaired mitophagy, defective protein import, and ROS-driven dysfunction, links mitochondrial damage, mtDNA-mediated inflammation, and protein aggregation [68\u201373].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Apoptosis and BAX/BAK-mediated outer membrane permeabilization", "text": "Mitochondria are known to be a center for neuronal survival and death, which integrates metabolic signals and stress responses to decide and activate apoptosis. In the intrinsic apoptotic pathway, the BCL-2 family proteins play a critical regulatory role. Pro-apoptotic effectors BAX and BAK oligomerize upon activation, inducing mitochondrial outer membrane permeabilization (MOMP), a crucial event during apoptosis [73, 74]. Traditionally, MOMP is recognized for cytosolic release of cytochrome c, Smac/DIABLO, and other intermembrane space proteins, thereby activating caspase cascades and executing apoptosis [75]. However, recent work has demonstrated that MOMP facilitates the mtDNA release into the cytosol [76, 77]. Mechanistically, BAX and BAK assemble into large, lipidic pores within the OMM that can dynamically enlarge over time [78]. During apoptosis, the BAX/BAK pores permit protrusion or \u201cherniation\u201d of the inner mitochondrial membrane, carrying mitochondrial matrix content, including mtDNA, into the cytosol [76]. This process effectively breaks the mitochondrial compartmentalization that normally keeps mtDNA away from innate immune sensors. Unlike transient mPTP observed during mitophagy, apoptotic MOMP often results in continuous membrane disruption, greatly increasing the possibility of mtDNA release. Importantly, even in the absence of full caspase activation, BAX/BAK-mediated membrane permeabilization can still cause mtDNA leakage and trigger inflammatory responses [79]. This phenomenon is particularly relevant in post-mitotic neurons, where caspase activation is often tightly controlled, yet sub-lethal mitochondrial permeabilization can still promote chronic inflammation.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Apoptosis and BAX/BAK-mediated outer membrane permeabilization", "text": "In the context of NDDs, this mechanism has profound implications. In AD, A\u03b2 accumulation within neurons increases mitochondrial ROS production and sensitizes mitochondria to apoptosis by promoting BAX activation [80]. The activated BAX forms large enough pores to release mtDNA into the cytosol, which then amplifies neuroinflammation through the cGAS\u2013STING signaling. Similarly, in PD, \u03b1-syn aggregates localize to mitochondria, impair mitochondrial dynamics, and increase oxidative stress, which together trigger BAX/BAK-mediated MOMP [81]. In HD, mHTT impairs mitochondrial calcium buffering and promotes oxidative stress, facilitating BAX recruitment to mitochondria [82]. BAX/BAK pore formation in HD models can similarly enable mtDNA release, feeding into a cycle of inflammation and neurodegeneration. These insights reveal a dual pathogenic role of BAX/BAK-mediated MOMP in neurological diseases beyond executing apoptosis: it can directly promote mtDNA release, which can activate neuroinflammation and drive disease progression. This convergence of mitochondrial dysfunction, apoptosis, and innate immune activation highlights therapeutic opportunities. Strategies that stabilize mitochondrial membranes, modulate BAX/BAK pore dynamics [83], or block downstream mtDNA sensing pathways [84] could inhibit inflammation without fully blocking apoptosis.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Necrosis, necroptosis, and pyroptosis, inflammatory cell death modalities", "text": "Unlike the controlled, typically non-inflammatory process of apoptosis, necrotic and necroptotic cell death are characterized by plasma membrane integrity loss, resulting in the extracellular release of intracellular components, including intact mitochondria or mitochondrial fragments that harbor mtDNA [85, 86]. Similarly, pyroptosis, a caspase-1-dependent form of cell death, induces gasdermin-D pore formation in the plasma membrane, culminating in cellular swelling and rupture [87, 88]. These lytic cell death pathways promote the release of DAMPs, notably mtDNA, into the extracellular milieu [33, 89]. Extracellular mtDNA could engage pattern recognition receptors (PRR) on microglia, astrocytes, and infiltrating immune cells, serving as a potent pro-inflammatory signal that drives neuroinflammation in acute brain injury, ischemia, and chronic NDDs.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Extracellular vesicle (EV)-mediated secretion of mtDNA", "text": "Recent studies have revealed that cells actively secrete EVs, such as exosomes and microvesicles [90]. These EVs can encapsulate whole mitochondria, mitochondrial fragments, or naked mtDNA, shielding them from extracellular degradation and enabling intercellular communication [90]. Under conditions of cellular stress, neurons, astrocytes, and microglia have all been shown to release mtDNA-containing EVs [67, 91]. This EV-mediated transfer of mtDNA may amplify neuroinflammatory responses by activating PRR in recipient cells, thereby facilitating the spread of pathological signals across neural networks. Mechanistically, EV biogenesis depends on the endosomal sorting complexes required for transport machinery and is modulated by stress-responsive pathways, although the precise molecular mechanisms that govern selective packaging of mtDNA into EV remain incompletely understood [92].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "mtDNA: sensing and immune activation", "text": "Under physiological conditions, mtDNA is securely sequestered within the mitochondrial matrix. However, physical injury, oxidative stress, mPTP opening, or BAX/BAK-mediated membrane permeabilization can trigger its release into the cytosol or extracellular space [32, 76]. Once liberated, mtDNA is perceived as a foreign-like molecular signature, activating an evolutionarily conserved network of nucleic acid sensors such as cGAS\u2013STING, NLRP3 inflammasome, and TLR9 [33]. This signaling cascade effectively links mitochondrial dysfunction to sterile inflammation.\n\nIn the central nervous system (CNS), these processes are especially consequential. Neurons and microglia can detect and sense cytosolic mtDNA, driving neuroinflammatory responses characterized by cytokine production, inflammasome activation, and recruitment of peripheral immune cells [12]. Dysregulated mtDNA-mediated signaling has emerged as a major contributor to NDD progression [12, 93\u201395]. Thus, mtDNA release acts as a molecular bridge coupling cellular stress to chronic neuroinflammation. The following sections examine the major mtDNA sensors and their downstream pathways that sculpt immune responses in the aging and diseased brains.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "mtDNA: sensing and immune activation", "text": "The discovery of the cGAS\u2013STING pathway has fundamentally reshaped our understanding of how cells detect cytosolic DNA as a danger signal, establishing a direct molecular link between mitochondrial injury and innate immunity [96\u201398]. Historically, innate DNA sensing was studied primarily in the context of viral infection; however, subsequent research demonstrated that endogenous nucleic acids, particularly mtDNA released under cellular stress, can also activate this pathway and trigger sterile inflammation [43]. Mechanistically, cGAS is a cytosolic sensor that binds dsDNA independent of sequence [96]. Upon dsDNA binding, cGAS catalyzes the synthesis of the second messenger cyclic GMP\u2013AMP (cGAMP), which uniquely contains a mixed 2\u2032\u20135\u2032 and 3\u2032\u20135\u2032 phosphodiester linkage conferring high stability and potent signaling capacity [97]. cGAMP binds and activates the adaptor protein STING, which is localized on the endoplasmic reticulum (ER) membrane. Ligand binding induces STING to oligomerize and translocate from the ER to the Golgi apparatus, where it recruits and activates TANK-binding kinase 1 (TBK1) [99, 100]. TBK1 phosphorylates IRF3 (interferon regulatory factor 3), which then dimerizes and translocates to the nucleus to induce transcription of type I interferons (IFN-\u03b1 and IFN-\u03b2) and pro-inflammatory cytokines (Fig.\u00a02) [98, 101].Fig.\u00a02Schematic illustrating activation of the cGAS\u2013STING pathway following the release of mitochondrial DNA (mtDNA) into the cytosol during cellular stress or mitochondrial dysfunction. Cytosolic mtDNA is sensed by cyclic cGAS, leading to the production of the second messenger cGAMP. cGAMP binds to and activates STING on the endoplasmic reticulum, triggering its translocation to the Golgi apparatus and subsequent recruitment of TBK1. This signaling cascade results in phosphorylation and nuclear translocation of IRF3, culminating in the induction of type I interferons and proinflammatory cytokines", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "mtDNA: sensing and immune activation", "text": "Schematic illustrating activation of the cGAS\u2013STING pathway following the release of mitochondrial DNA (mtDNA) into the cytosol during cellular stress or mitochondrial dysfunction. Cytosolic mtDNA is sensed by cyclic cGAS, leading to the production of the second messenger cGAMP. cGAMP binds to and activates STING on the endoplasmic reticulum, triggering its translocation to the Golgi apparatus and subsequent recruitment of TBK1. This signaling cascade results in phosphorylation and nuclear translocation of IRF3, culminating in the induction of type I interferons and proinflammatory cytokines", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "mtDNA: sensing and immune activation", "text": "In the CNS, both neurons and glial cells express cGAS and STING, and their activation by mtDNA has been directly implicated in neuroinflammation and NDDs [12, 93]. For instance, in AD\u00a0mice, mitochondrial damage increases mtDNA leakage into the cytosol, where it engages cGAS and activates STING, driving microglial secretion of type I interferons and pro-inflammatory cytokines that contribute to neuronal dysfunction [93]. In PD, impaired mitophagy due to PINK1 or Parkin mutations results in the accumulation of damaged, ROS-producing mitochondria with mtDNA leakage, triggering cGAS\u2013STING signaling and promoting dopaminergic neuron degeneration [102, 103]. Importantly, beyond neurons, microglia are particularly sensitive to this pathway. Their activation sustains chronic inflammation, recruits peripheral immune cells, and amplifies neuronal injury. While transient activation of cGAS\u2013STING is protective against pathogens, chronic or dysregulated activation by endogenous mtDNA in aging or neurodegeneration becomes maladaptive.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "mtDNA: sensing and immune activation", "text": "In addition to cGAS\u2013STING, recent evidence indicates that Z-DNA binding protein 1 (ZBP1) can act as an innate immune sensor for cytosolic Z-DNA, a structural form that can arise from damaged mtDNA. Upon recognition of Z-DNA, ZBP1 can activate downstream signaling pathways that overlap with and amplify cGAS\u2013STING\u2013mediated inflammatory responses, including type I interferon and inflammasome activation. This parallel sensing mechanism suggests that mtDNA released from dysfunctional mitochondria can engage multiple cytosolic DNA sensors, thereby enhancing microglial activation and sustaining neuroinflammation. Acknowledging ZBP1 expands our understanding of how aberrant mtDNA contributes to innate immune signaling in NDDs and highlights additional potential targets for therapeutic modulation of sterile inflammation [104].\n\nTherapeutically, targeting this pathway is now being actively explored. Pharmacological inhibitors of cGAS and STING, as well as strategies to neutralize type I interferon signaling, have shown benefits in reducing neuroinflammation and preserving neuronal function in preclinical. Together, these findings underscore that the cGAS\u2013STING pathway is not only a molecular sentinel for pathogen detection but also a critical driver linking mitochondrial stress and mtDNA release to chronic neuroinflammation in NDDs. Understanding this mechanism has opened new opportunities to modulate inflammation and protect neurons in disorders like AD, PD, and HD.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "mtDNA: sensing and immune activation", "text": "NLRP3 inflammasome is a well-characterized multiprotein complex that serves as a critical intracellular sensor of diverse DAMPs and pathogen-associated molecular patterns (PAMPs) [105, 106]. Initially discovered in the context of autoinflammatory syndromes, the NLRP3 inflammasome has now been recognized as a central mediator of sterile inflammation in NDDs. Mechanistically, cytosolic mtDNA is a potent activator of NLRP3, providing a direct molecular link between mitochondrial dysfunction and chronic inflammation in the brain [42, 107].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "mtDNA: sensing and immune activation", "text": "Under stress conditions, such as oxidative injury, mPTP opening, or impaired mitophagy,\u00a0damaged mitochondria release oxidized mtDNA into the cytosol. This mtDNA can directly bind to and activate NLRP3, triggering its conformational change and oligomerization [223]. Activated NLRP3 recruits the adaptor protein apoptosis-associated speck-like protein containing a CARD (ASC) through pyrin domain interactions. ASC then nucleates the clustering of pro-caspase-1, forming a large supramolecular assembly known as the ASC speck [108, 109]. This close proximity facilitates autocatalytic activation of caspase-1, which is the key effector protease of the inflammasome complex. Activated caspase-1 processes the precursor cytokines pro-IL-1\u03b2 and pro-IL-18 into their mature, bioactive forms [109]. These cytokines are subsequently secreted and act as potent amplifiers of inflammation: IL-1\u03b2 promotes the expression of additional inflammatory mediators and adhesion molecules, while IL-18 synergizes to enhance IFN-\u03b3 production and immune cell recruitment (Fig.\u00a03) [110]. Beyond cytokine maturation, caspase-1 also induces pyroptosis, a lytic and inflammatory form of programmed cell death that further releases intracellular DAMP, amplifying the inflammatory cascade [87].Fig.\u00a03Diagram depicting the mechanisms by which mitochondrial dysfunction and mitochondrial DNA (mtDNA) release promote activation of the NLRP3 inflammasome. Cellular stress leads to mitochondrial damage and the release of oxidized mtDNA into the cytosol. Cytosolic mtDNA acts as a DAMP that facilitates assembly of the NLRP3 inflammasome complex, composed of NLRP3, the adaptor protein ASC, and pro-caspase-1. Inflammasome assembly results in caspase-1 activation, leading to cleavage and maturation of the proinflammatory cytokines IL-1\u03b2 and IL-18. Sustained mtDNA-driven NLRP3 activation contributes to chronic neuroinflammation and neuronal injury in neurodegenerative disorders", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "mtDNA: sensing and immune activation", "text": "Diagram depicting the mechanisms by which mitochondrial dysfunction and mitochondrial DNA (mtDNA) release promote activation of the NLRP3 inflammasome. Cellular stress leads to mitochondrial damage and the release of oxidized mtDNA into the cytosol. Cytosolic mtDNA acts as a DAMP that facilitates assembly of the NLRP3 inflammasome complex, composed of NLRP3, the adaptor protein ASC, and pro-caspase-1. Inflammasome assembly results in caspase-1 activation, leading to cleavage and maturation of the proinflammatory cytokines IL-1\u03b2 and IL-18. Sustained mtDNA-driven NLRP3 activation contributes to chronic neuroinflammation and neuronal injury in neurodegenerative disorders", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "mtDNA: sensing and immune activation", "text": "In the brain, the NLRP3 inflammasome is primarily expressed in microglia. Activation of NLRP3 by mtDNA released from stressed or damaged neurons leads to microglial secretion of IL-1\u03b2 and IL-18, establishing a feed-forward loop of neuroinflammation [111]. This process has been directly implicated in the pathogenesis of several NDDs. For example, in AD, A\u03b2 aggregates induce mitochondrial dysfunction and ROS production, leading to mtDNA leakage and NLRP3 activation in microglia [111, 112]. The resulting inflammatory environment contributes to synaptic loss and cognitive decline. In PD, \u03b1-syn aggregates also stimulate NLRP3 activation through mitochondrial stress and mtDNA release, exacerbating dopaminergic neuron degeneration [113, 114]. Similar mechanisms are proposed in HD, where mHTT induces mitochondrial fragmentation and oxidative damage, promoting NLRP3-driven inflammation [115].\n\nHistorically, these discoveries have reframed the NLRP3 inflammasome not merely as a pathogen sensor, but as a central mediator of sterile, mtDNA-driven neuroinflammation. Therapeutically, pharmacological inhibitors targeting NLRP3 (e.g., MCC950), caspase-1, or IL-1\u03b2 signaling have shown promise in preclinical neurodegenerative models by reducing neuroinflammation and preserving neuronal function [116\u2013122].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "mtDNA: sensing and immune activation", "text": "In summary, the NLRP3 inflammasome detects danger signals, including cytosolic mtDNA released from damaged mitochondria, and orchestrates a powerful inflammatory response through caspase-1 activation, cytokine maturation, and pyroptosis. In NDDs, chronic mtDNA-driven NLRP3 activation in microglia fuels sustained neuroinflammation, linking mitochondrial dysfunction directly to neuronal injury and disease progression.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "mtDNA: sensing and immune activation", "text": "TLR9 is a specialized PRR that evolved to detect unmethylated CpG motifs, which are abundant in bacterial DNA\u00a0and mtDNA but relatively rare and typically methylated in nuclear DNA [123, 124]. Once released, mtDNA can be internalized into endolysosomal compartments through processes like phagocytosis, autophagy, or uptake of EVs [125, 126]). Within the acidified endosomes, TLR9 directly binds these CpG-rich mtDNA fragments, initiating an innate immune response. Mechanistically, ligand engagement causes TLR9 to recruit the adaptor protein myeloid differentiation primary response 88, via its Toll/IL-1 receptor domain [127]. This interaction triggers the assembly of a signaling complex that includes interleukin-1 receptor-associated kinases and TNF receptor-associated factor 6, leading to the activation of nuclear factor kappa B (NF-\u03baB) and mitogen-activated protein kinases (MAPKs) such as p38 and JNK [128]. These transcription factors translocate to the nucleus, driving the expression of pro-inflammatory cytokines, including TNF-\u03b1, IL-6, and IL-1\u03b2 (Fig.\u00a04) [128].Fig.\u00a04Schematic illustrating how mtDNA released from damaged mitochondria activates TLR-dependent inflammatory signaling. Extracellular or endolysosomal mtDNA is recognized primarily by TLR9 within endosomes of microglia, astrocytes, and infiltrating immune cells. TLR9 engagement triggers MyD88-dependent signaling cascades, leading to the activation of NF-\u03baB and induction of proinflammatory cytokines and chemokines. This mtDNA\u2013TLR axis amplifies innate immune responses, promotes glial activation, and contributes to sustained neuroinflammation and neuronal dysfunction in neurodegenerative and neuroinflammatory conditions", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "mtDNA: sensing and immune activation", "text": "Schematic illustrating how mtDNA released from damaged mitochondria activates TLR-dependent inflammatory signaling. Extracellular or endolysosomal mtDNA is recognized primarily by TLR9 within endosomes of microglia, astrocytes, and infiltrating immune cells. TLR9 engagement triggers MyD88-dependent signaling cascades, leading to the activation of NF-\u03baB and induction of proinflammatory cytokines and chemokines. This mtDNA\u2013TLR axis amplifies innate immune responses, promotes glial activation, and contributes to sustained neuroinflammation and neuronal dysfunction in neurodegenerative and neuroinflammatory conditions", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "mtDNA: sensing and immune activation", "text": "TLR9 is functionally expressed in microglia and astrocytes, which serve as the brain\u2019s primary immune effector and supporting glial cells, respectively [129]. Activation of TLR9 in these cells by mtDNA released from injured neurons amplifies local inflammation. Microglial TLR9 stimulation increases production of pro-inflammatory cytokines and chemokines, while astrocytic activation contributes to gliosis and release of neurotoxic mediators. In PD and HD, similar processes have been proposed, where mitochondrial dysfunction and defective mitophagy increase mtDNA availability, fueling chronic TLR9-driven inflammation that exacerbates neuronal degeneration. Historically, TLR9 was first recognized for its antiviral and antibacterial roles, sensing pathogen DNA within endosomes [123]. Its role in sterile inflammation, driven by self-DNA such as mtDNA, was later established through studies showing that oxidatively damaged mtDNA can activate TLR9 and trigger inflammation even in the absence of infection [124]. This discovery highlighted how mtDNA, because of its bacterial ancestry, can inadvertently activate ancient immune pathways designed to detect pathogens. Therapeutically, targeting TLR9 or its downstream effectors holds potential to dampen chronic inflammation in NDDs. Preclinical studies using TLR9 antagonists or genetic deletion have shown attenuation of neuroinflammation and improved neuronal survival in models of CNS injury and disease [130].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "mtDNA: sensing and immune activation", "text": "Beyond the well-characterized cGAS\u2013STING pathway, accumulating evidence indicates that additional cytosolic DNA sensors may recognize mtDNA and contribute to innate immune activation, although their specific roles within the CNS remain incompletely understood. One notable sensor is absent in melanoma 2 (AIM2), which belongs to the AIM2-like receptor (ALR) family [131]. AIM2 directly binds cytosolic dsDNA through its HIN-200 domain, independent of sequence, and assembles an inflammasome complex via its pyrin domain that recruits ASC and activates caspase-1. This activation leads to the maturation of IL-1\u03b2 and IL-18 and can trigger pyroptosis[131, 132]. Although most studies have focused on AIM2 sensing of nuclear or pathogen-derived DNA, recent work suggests that mtDNA released during apoptosis or mitochondrial damage can also activate the AIM2 inflammasome, amplifying inflammation [42]. In the CNS, AIM2 expression has been detected in microglia and astrocytes, and AIM2 activation has been linked to neuroinflammatory responses and cognitive deficits in experimental models of neurodegeneration and traumatic brain injury.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "mtDNA: sensing and immune activation", "text": "In addition, mtDNA can interact indirectly with the mitochondrial antiviral-signaling protein (MAVS), which is anchored on the OMM and classically activated by viral RNA sensors such as RIG-I and MDA5 [133]. Recent work indicates that oxidative stress or mitochondrial damage can promote MAVS signaling, enhancing the expression of type I interferons and pro-inflammatory cytokines [134, 135]. Although traditionally associated with antiviral responses, MAVS signaling is increasingly recognized as a modulator of sterile inflammation, including in the brain [136].\n\nCollectively, these emerging pathways suggest that mtDNA serves as a versatile DAMP capable of engaging multiple cytosolic sensors beyond cGAS. The activation of AIM2 and MAVS by mtDNA could act in parallel or synergistically with cGAS\u2013STING and NLRP3 inflammasome pathways, amplifying innate immune responses and neuroinflammation. Future studies are needed to dissect the cell-type specificity, temporal dynamics, and pathological consequences of these alternative DNA sensors in NDDs.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "cGAS\u2013STING pathway\u2014cytosolic DNA sensing", "text": "The discovery of the cGAS\u2013STING pathway has fundamentally reshaped our understanding of how cells detect cytosolic DNA as a danger signal, establishing a direct molecular link between mitochondrial injury and innate immunity [96\u201398]. Historically, innate DNA sensing was studied primarily in the context of viral infection; however, subsequent research demonstrated that endogenous nucleic acids, particularly mtDNA released under cellular stress, can also activate this pathway and trigger sterile inflammation [43]. Mechanistically, cGAS is a cytosolic sensor that binds dsDNA independent of sequence [96]. Upon dsDNA binding, cGAS catalyzes the synthesis of the second messenger cyclic GMP\u2013AMP (cGAMP), which uniquely contains a mixed 2\u2032\u20135\u2032 and 3\u2032\u20135\u2032 phosphodiester linkage conferring high stability and potent signaling capacity [97]. cGAMP binds and activates the adaptor protein STING, which is localized on the endoplasmic reticulum (ER) membrane. Ligand binding induces STING to oligomerize and translocate from the ER to the Golgi apparatus, where it recruits and activates TANK-binding kinase 1 (TBK1) [99, 100]. TBK1 phosphorylates IRF3 (interferon regulatory factor 3), which then dimerizes and translocates to the nucleus to induce transcription of type I interferons (IFN-\u03b1 and IFN-\u03b2) and pro-inflammatory cytokines (Fig.\u00a02) [98, 101].Fig.\u00a02Schematic illustrating activation of the cGAS\u2013STING pathway following the release of mitochondrial DNA (mtDNA) into the cytosol during cellular stress or mitochondrial dysfunction. Cytosolic mtDNA is sensed by cyclic cGAS, leading to the production of the second messenger cGAMP. cGAMP binds to and activates STING on the endoplasmic reticulum, triggering its translocation to the Golgi apparatus and subsequent recruitment of TBK1. This signaling cascade results in phosphorylation and nuclear translocation of IRF3, culminating in the induction of type I interferons and proinflammatory cytokines", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "cGAS\u2013STING pathway\u2014cytosolic DNA sensing", "text": "Schematic illustrating activation of the cGAS\u2013STING pathway following the release of mitochondrial DNA (mtDNA) into the cytosol during cellular stress or mitochondrial dysfunction. Cytosolic mtDNA is sensed by cyclic cGAS, leading to the production of the second messenger cGAMP. cGAMP binds to and activates STING on the endoplasmic reticulum, triggering its translocation to the Golgi apparatus and subsequent recruitment of TBK1. This signaling cascade results in phosphorylation and nuclear translocation of IRF3, culminating in the induction of type I interferons and proinflammatory cytokines", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "cGAS\u2013STING pathway\u2014cytosolic DNA sensing", "text": "In the CNS, both neurons and glial cells express cGAS and STING, and their activation by mtDNA has been directly implicated in neuroinflammation and NDDs [12, 93]. For instance, in AD\u00a0mice, mitochondrial damage increases mtDNA leakage into the cytosol, where it engages cGAS and activates STING, driving microglial secretion of type I interferons and pro-inflammatory cytokines that contribute to neuronal dysfunction [93]. In PD, impaired mitophagy due to PINK1 or Parkin mutations results in the accumulation of damaged, ROS-producing mitochondria with mtDNA leakage, triggering cGAS\u2013STING signaling and promoting dopaminergic neuron degeneration [102, 103]. Importantly, beyond neurons, microglia are particularly sensitive to this pathway. Their activation sustains chronic inflammation, recruits peripheral immune cells, and amplifies neuronal injury. While transient activation of cGAS\u2013STING is protective against pathogens, chronic or dysregulated activation by endogenous mtDNA in aging or neurodegeneration becomes maladaptive.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "cGAS\u2013STING pathway\u2014cytosolic DNA sensing", "text": "In addition to cGAS\u2013STING, recent evidence indicates that Z-DNA binding protein 1 (ZBP1) can act as an innate immune sensor for cytosolic Z-DNA, a structural form that can arise from damaged mtDNA. Upon recognition of Z-DNA, ZBP1 can activate downstream signaling pathways that overlap with and amplify cGAS\u2013STING\u2013mediated inflammatory responses, including type I interferon and inflammasome activation. This parallel sensing mechanism suggests that mtDNA released from dysfunctional mitochondria can engage multiple cytosolic DNA sensors, thereby enhancing microglial activation and sustaining neuroinflammation. Acknowledging ZBP1 expands our understanding of how aberrant mtDNA contributes to innate immune signaling in NDDs and highlights additional potential targets for therapeutic modulation of sterile inflammation [104].\n\nTherapeutically, targeting this pathway is now being actively explored. Pharmacological inhibitors of cGAS and STING, as well as strategies to neutralize type I interferon signaling, have shown benefits in reducing neuroinflammation and preserving neuronal function in preclinical. Together, these findings underscore that the cGAS\u2013STING pathway is not only a molecular sentinel for pathogen detection but also a critical driver linking mitochondrial stress and mtDNA release to chronic neuroinflammation in NDDs. Understanding this mechanism has opened new opportunities to modulate inflammation and protect neurons in disorders like AD, PD, and HD.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "NLRP3 inflammasome activation\u2014cytosolic danger sensing", "text": "NLRP3 inflammasome is a well-characterized multiprotein complex that serves as a critical intracellular sensor of diverse DAMPs and pathogen-associated molecular patterns (PAMPs) [105, 106]. Initially discovered in the context of autoinflammatory syndromes, the NLRP3 inflammasome has now been recognized as a central mediator of sterile inflammation in NDDs. Mechanistically, cytosolic mtDNA is a potent activator of NLRP3, providing a direct molecular link between mitochondrial dysfunction and chronic inflammation in the brain [42, 107].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "NLRP3 inflammasome activation\u2014cytosolic danger sensing", "text": "Under stress conditions, such as oxidative injury, mPTP opening, or impaired mitophagy,\u00a0damaged mitochondria release oxidized mtDNA into the cytosol. This mtDNA can directly bind to and activate NLRP3, triggering its conformational change and oligomerization [223]. Activated NLRP3 recruits the adaptor protein apoptosis-associated speck-like protein containing a CARD (ASC) through pyrin domain interactions. ASC then nucleates the clustering of pro-caspase-1, forming a large supramolecular assembly known as the ASC speck [108, 109]. This close proximity facilitates autocatalytic activation of caspase-1, which is the key effector protease of the inflammasome complex. Activated caspase-1 processes the precursor cytokines pro-IL-1\u03b2 and pro-IL-18 into their mature, bioactive forms [109]. These cytokines are subsequently secreted and act as potent amplifiers of inflammation: IL-1\u03b2 promotes the expression of additional inflammatory mediators and adhesion molecules, while IL-18 synergizes to enhance IFN-\u03b3 production and immune cell recruitment (Fig.\u00a03) [110]. Beyond cytokine maturation, caspase-1 also induces pyroptosis, a lytic and inflammatory form of programmed cell death that further releases intracellular DAMP, amplifying the inflammatory cascade [87].Fig.\u00a03Diagram depicting the mechanisms by which mitochondrial dysfunction and mitochondrial DNA (mtDNA) release promote activation of the NLRP3 inflammasome. Cellular stress leads to mitochondrial damage and the release of oxidized mtDNA into the cytosol. Cytosolic mtDNA acts as a DAMP that facilitates assembly of the NLRP3 inflammasome complex, composed of NLRP3, the adaptor protein ASC, and pro-caspase-1. Inflammasome assembly results in caspase-1 activation, leading to cleavage and maturation of the proinflammatory cytokines IL-1\u03b2 and IL-18. Sustained mtDNA-driven NLRP3 activation contributes to chronic neuroinflammation and neuronal injury in neurodegenerative disorders", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "NLRP3 inflammasome activation\u2014cytosolic danger sensing", "text": "Diagram depicting the mechanisms by which mitochondrial dysfunction and mitochondrial DNA (mtDNA) release promote activation of the NLRP3 inflammasome. Cellular stress leads to mitochondrial damage and the release of oxidized mtDNA into the cytosol. Cytosolic mtDNA acts as a DAMP that facilitates assembly of the NLRP3 inflammasome complex, composed of NLRP3, the adaptor protein ASC, and pro-caspase-1. Inflammasome assembly results in caspase-1 activation, leading to cleavage and maturation of the proinflammatory cytokines IL-1\u03b2 and IL-18. Sustained mtDNA-driven NLRP3 activation contributes to chronic neuroinflammation and neuronal injury in neurodegenerative disorders", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "NLRP3 inflammasome activation\u2014cytosolic danger sensing", "text": "In the brain, the NLRP3 inflammasome is primarily expressed in microglia. Activation of NLRP3 by mtDNA released from stressed or damaged neurons leads to microglial secretion of IL-1\u03b2 and IL-18, establishing a feed-forward loop of neuroinflammation [111]. This process has been directly implicated in the pathogenesis of several NDDs. For example, in AD, A\u03b2 aggregates induce mitochondrial dysfunction and ROS production, leading to mtDNA leakage and NLRP3 activation in microglia [111, 112]. The resulting inflammatory environment contributes to synaptic loss and cognitive decline. In PD, \u03b1-syn aggregates also stimulate NLRP3 activation through mitochondrial stress and mtDNA release, exacerbating dopaminergic neuron degeneration [113, 114]. Similar mechanisms are proposed in HD, where mHTT induces mitochondrial fragmentation and oxidative damage, promoting NLRP3-driven inflammation [115].\n\nHistorically, these discoveries have reframed the NLRP3 inflammasome not merely as a pathogen sensor, but as a central mediator of sterile, mtDNA-driven neuroinflammation. Therapeutically, pharmacological inhibitors targeting NLRP3 (e.g., MCC950), caspase-1, or IL-1\u03b2 signaling have shown promise in preclinical neurodegenerative models by reducing neuroinflammation and preserving neuronal function [116\u2013122].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "NLRP3 inflammasome activation\u2014cytosolic danger sensing", "text": "In summary, the NLRP3 inflammasome detects danger signals, including cytosolic mtDNA released from damaged mitochondria, and orchestrates a powerful inflammatory response through caspase-1 activation, cytokine maturation, and pyroptosis. In NDDs, chronic mtDNA-driven NLRP3 activation in microglia fuels sustained neuroinflammation, linking mitochondrial dysfunction directly to neuronal injury and disease progression.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "TLR9\u2014endosomal sensing of CpG DNA", "text": "TLR9 is a specialized PRR that evolved to detect unmethylated CpG motifs, which are abundant in bacterial DNA\u00a0and mtDNA but relatively rare and typically methylated in nuclear DNA [123, 124]. Once released, mtDNA can be internalized into endolysosomal compartments through processes like phagocytosis, autophagy, or uptake of EVs [125, 126]). Within the acidified endosomes, TLR9 directly binds these CpG-rich mtDNA fragments, initiating an innate immune response. Mechanistically, ligand engagement causes TLR9 to recruit the adaptor protein myeloid differentiation primary response 88, via its Toll/IL-1 receptor domain [127]. This interaction triggers the assembly of a signaling complex that includes interleukin-1 receptor-associated kinases and TNF receptor-associated factor 6, leading to the activation of nuclear factor kappa B (NF-\u03baB) and mitogen-activated protein kinases (MAPKs) such as p38 and JNK [128]. These transcription factors translocate to the nucleus, driving the expression of pro-inflammatory cytokines, including TNF-\u03b1, IL-6, and IL-1\u03b2 (Fig.\u00a04) [128].Fig.\u00a04Schematic illustrating how mtDNA released from damaged mitochondria activates TLR-dependent inflammatory signaling. Extracellular or endolysosomal mtDNA is recognized primarily by TLR9 within endosomes of microglia, astrocytes, and infiltrating immune cells. TLR9 engagement triggers MyD88-dependent signaling cascades, leading to the activation of NF-\u03baB and induction of proinflammatory cytokines and chemokines. This mtDNA\u2013TLR axis amplifies innate immune responses, promotes glial activation, and contributes to sustained neuroinflammation and neuronal dysfunction in neurodegenerative and neuroinflammatory conditions", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "TLR9\u2014endosomal sensing of CpG DNA", "text": "Schematic illustrating how mtDNA released from damaged mitochondria activates TLR-dependent inflammatory signaling. Extracellular or endolysosomal mtDNA is recognized primarily by TLR9 within endosomes of microglia, astrocytes, and infiltrating immune cells. TLR9 engagement triggers MyD88-dependent signaling cascades, leading to the activation of NF-\u03baB and induction of proinflammatory cytokines and chemokines. This mtDNA\u2013TLR axis amplifies innate immune responses, promotes glial activation, and contributes to sustained neuroinflammation and neuronal dysfunction in neurodegenerative and neuroinflammatory conditions", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "TLR9\u2014endosomal sensing of CpG DNA", "text": "TLR9 is functionally expressed in microglia and astrocytes, which serve as the brain\u2019s primary immune effector and supporting glial cells, respectively [129]. Activation of TLR9 in these cells by mtDNA released from injured neurons amplifies local inflammation. Microglial TLR9 stimulation increases production of pro-inflammatory cytokines and chemokines, while astrocytic activation contributes to gliosis and release of neurotoxic mediators. In PD and HD, similar processes have been proposed, where mitochondrial dysfunction and defective mitophagy increase mtDNA availability, fueling chronic TLR9-driven inflammation that exacerbates neuronal degeneration. Historically, TLR9 was first recognized for its antiviral and antibacterial roles, sensing pathogen DNA within endosomes [123]. Its role in sterile inflammation, driven by self-DNA such as mtDNA, was later established through studies showing that oxidatively damaged mtDNA can activate TLR9 and trigger inflammation even in the absence of infection [124]. This discovery highlighted how mtDNA, because of its bacterial ancestry, can inadvertently activate ancient immune pathways designed to detect pathogens. Therapeutically, targeting TLR9 or its downstream effectors holds potential to dampen chronic inflammation in NDDs. Preclinical studies using TLR9 antagonists or genetic deletion have shown attenuation of neuroinflammation and improved neuronal survival in models of CNS injury and disease [130].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Other potential mtDNA sensing pathways", "text": "Beyond the well-characterized cGAS\u2013STING pathway, accumulating evidence indicates that additional cytosolic DNA sensors may recognize mtDNA and contribute to innate immune activation, although their specific roles within the CNS remain incompletely understood. One notable sensor is absent in melanoma 2 (AIM2), which belongs to the AIM2-like receptor (ALR) family [131]. AIM2 directly binds cytosolic dsDNA through its HIN-200 domain, independent of sequence, and assembles an inflammasome complex via its pyrin domain that recruits ASC and activates caspase-1. This activation leads to the maturation of IL-1\u03b2 and IL-18 and can trigger pyroptosis[131, 132]. Although most studies have focused on AIM2 sensing of nuclear or pathogen-derived DNA, recent work suggests that mtDNA released during apoptosis or mitochondrial damage can also activate the AIM2 inflammasome, amplifying inflammation [42]. In the CNS, AIM2 expression has been detected in microglia and astrocytes, and AIM2 activation has been linked to neuroinflammatory responses and cognitive deficits in experimental models of neurodegeneration and traumatic brain injury.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Other potential mtDNA sensing pathways", "text": "In addition, mtDNA can interact indirectly with the mitochondrial antiviral-signaling protein (MAVS), which is anchored on the OMM and classically activated by viral RNA sensors such as RIG-I and MDA5 [133]. Recent work indicates that oxidative stress or mitochondrial damage can promote MAVS signaling, enhancing the expression of type I interferons and pro-inflammatory cytokines [134, 135]. Although traditionally associated with antiviral responses, MAVS signaling is increasingly recognized as a modulator of sterile inflammation, including in the brain [136].\n\nCollectively, these emerging pathways suggest that mtDNA serves as a versatile DAMP capable of engaging multiple cytosolic sensors beyond cGAS. The activation of AIM2 and MAVS by mtDNA could act in parallel or synergistically with cGAS\u2013STING and NLRP3 inflammasome pathways, amplifying innate immune responses and neuroinflammation. Future studies are needed to dissect the cell-type specificity, temporal dynamics, and pathological consequences of these alternative DNA sensors in NDDs.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Neuroinflammation and mtDNA", "text": "Neuroinflammation is a key process in the CNS, playing a vital role in both maintaining brain health and contributing to disease when dysregulated [137]. The main cells involved in neuroinflammation are microglia, astrocytes, and peripheral immune cells. Microglia are the resident immune cells in the CNS, performing crucial tasks in monitoring and maintaining neuronal homeostasis [138]. Microglia are involved in synaptic pruning, debris clearance, and supporting neuronal health in their resting state [138, 139]. However, when activated in response to injury, infection, or disease, microglia can change into a pro-inflammatory state, producing a wide range of inflammatory molecules such as cytokines (e.g., TNF-\u03b1, IL-1\u03b2), ROS, and proteases, all of which contribute to tissue damage and neurodegeneration [140]. Astrocytes, the most abundant glial cells in the brain, also play an essential role in neuroinflammation [141]. Under normal conditions, astrocytes help maintain the blood\u2013brain barrier, provide metabolic support to neurons, and regulate synaptic function [142]. However, astrocytes can become reactive and secrete pro-inflammatory molecules in response to inflammation or injury, amplifying the inflammatory response [141]. These reactive astrocytes may further exacerbate neuronal damage, particularly when neuroinflammation persists over time. Peripheral immune cells, such as macrophages and T lymphocytes, can infiltrate the CNS during chronic inflammation. Normally confined to the periphery, these immune cells migrate into the brain in conditions like multiple sclerosis, AD, and PD, where they contribute to the inflammatory cascade [143]. This infiltration can intensify the neuroinflammatory response, further driving neurodegeneration [143].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Neuroinflammation and mtDNA", "text": "The immune response in the CNS is highly complex and needs to be finely balanced. On one hand, microglia and astrocytes can mount protective, anti-inflammatory responses, which aid in tissue repair, clearance of damaged cells, and resolution of inflammation [144]. Anti-inflammatory microglia release factors that promote neuronal survival and tissue healing, effectively dampening the inflammatory response once the threat has been removed [144]. The shift from anti-inflammatory to pro-inflammatory microglial activation is a key factor in the transition from protective to harmful inflammation. The imbalance between these pro-inflammatory and anti-inflammatory responses is a central feature in conditions like AD, PD, and HD, where chronic neuroinflammation exacerbates neuronal injury and accelerates disease progression [145]. Furthermore, chronic neuroinflammation is self-perpetuating, creating a vicious cycle where neurodegeneration and inflammation feed into each other. The presence of inflammatory mediators can induce further activation of microglia and astrocytes, which in turn release additional inflammatory factors, leading to even greater neuronal damage. mtDNA itself can act as a transferable inflammatory signal that drives non cell-autonomous propagation of neurodegeneration. Under conditions of mitochondrial stress or impaired mitophagy, mtDNA can be released into the cytosol or the extracellular space. Once in these compartments, mtDNA may be taken up by neighboring neurons, microglia, astrocytes, and, if it enters the bloodstream, by peripheral immune cells, including T cells. mtDNA internalization, or receptor-mediated activation of immune response as DAMP, triggers sterile inflammation and amplifies cytokine and type I interferon signaling, thereby establishing feed-forward loops of mitochondrial dysfunction, immune activation, and secondary mtDNA release. This cycle perpetuates the progression of the disease, ultimately contributing to cognitive decline, motor dysfunction, and other behavioral changes commonly observed in NDDs. Understanding the complex interactions between microglia, astrocytes, and peripheral immune cells is essential for developing targeted therapies that can harness the beneficial aspects of neuroinflammation while mitigating its destructive effects in NDDs.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Neuroinflammation and mtDNA", "text": "mtDNA is increasingly recognized as a pivotal mediator linking mitochondrial dysfunction to neuroinflammation in a broad range of neurological disorders. The unique pro-inflammatory potential of mtDNA stems from its bacterial ancestry, allowing it to activate innate immune receptors upon aberrant release into the cytosol or extracellular space. Neuroinflammation, characterized by activation of microglia\u00a0and astrocytes, and recruitment of peripheral immune cells, plays a dual role in nervous system pathophysiology, where acute responses may be protective, but chronic inflammation leads to progressive neuronal damage and neurodegeneration. This section reviews the roles of mtDNA-driven neuroinflammation in major NDDs and acute CNS injuries, detailing underlying molecular mechanisms, pathological consequences, and therapeutic prospects.\n\nAD is marked by progressive cognitive decline linked to extracellular A\u03b2 plaques and intraneuronal neurofibrillary tangles composed of hyperphosphorylated tau protein [146]. In addition to these hallmark lesions, mitochondrial dysfunction appears early and consistently in AD pathology [147]. OXPHOS defects, increased production of ROS, and structural abnormalities of mitochondria are often seen in vulnerable brain areas such as the hippocampus and cortex [148]. Postmortem studies have shown higher mtDNA point mutations, large deletions, and lower mtDNA copy numbers in AD patients [149, 150].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Neuroinflammation and mtDNA", "text": "There is growing evidence supporting a central role for mtDNA release in driving neuroinflammation and pathology in AD. Multiple human, animal, and cellular studies converge on the view that mitochondrial dysfunction, oxidative stress, and impaired mitophagy lead to leakage of mtDNA into the cytosol and extracellular space, where it activates innate immune sensors and sustains chronic inflammation. Clinical and postmortem human studies show elevated levels of circulating cell-free mtDNA in the cerebrospinal fluid (CSF) and plasma of AD patients, often correlating with cognitive impairment and disease severity. Mechanistic studies in animals have highlighted the role of mtDNA in activating key innate immune pathways. For example, Xie et al. [93] demonstrated in 5\u2009\u00d7\u2009FAD mice and human AD brains that cytosolic mtDNA engages the cGAS\u2013STING pathway in microglia, driving production of type I interferons and pro-inflammatory cytokines. Similarly, Hou et al. [151] found upregulation of cGAS and STING in APP/PS1 mice; genetic deletion of cGAS reduced neuroinflammation and improved cognitive function. In parallel, Heneka et al. [111] showed that mtDNA released from damaged mitochondria activates the NLRP3 inflammasome, leading to IL-1\u03b2 production and promoting amyloid pathology in AD mice and human samples. Complementary work by Fang et al. [63] demonstrated that pharmacologically enhancing mitophagy (e.g., with urolithin A) can reduce mtDNA leakage and ameliorate inflammation in APP/PS1 mice. Evidence for TLR9 activation comes from studies like Scholtzova et al. [152], showing that a TLR9 agonist (CpG oligodeoxynucleotides) modulates immune responses and reduces amyloid burden in transgenic AD mouse models, although effects on cognition and plaque clearance appear context-dependent. Roy et al. [224] demonstrated that oxidative stress triggers mtDNA leakage, which activates the cGAS\u2013STING pathway and induces the secretion of pro-inflammatory cytokines in human iPSC-derived neurons. Finally, several reviews [147, 153] have synthesized this body of work, emphasizing how mitochondrial damage and mtDNA release create a vicious cycle: oxidative stress damages mtDNA, increasing its release, which then sustains inflammation and exacerbates amyloid and tau pathology. Additionally, sustained activation of the cGAS\u2013STING signaling impairs microglial phagocytosis of A\u03b2, promoting A\u03b2 accumulation [151]. This microglial clearance impairment may result from reprogramming microglial function from a homeostatic, phagocytic state to a pro-inflammatory phenotype, which further worsens AD pathology. Overall, these findings highlight the pathogenic role of mtDNA as a neuroinflammatory trigger and its contribution to AD progression.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Neuroinflammation and mtDNA", "text": "Pharmacological inhibition of TLR9 or the cGAS\u2013STING DNA-sensing pathway has demonstrated efficacy in mitigating neuroinflammation and cognitive deficits in AD mouse models. For instance, STING inhibition in AppNL\u2011G\u2011F/hTau knock\u2011in mice normalizes microglial function, reduces type I interferon and NLRP3 activation, decreases A\u03b2 and tau pathology, limits synaptic loss, and preserves memory performance [154, 155]. Complementarily, microglial cGAS deletion significantly curbs early plaque formation and cognitive decline in AD models, underscoring the central role of this innate immune sensor in disease progression [156, 157]. Similarly, treatment with mitochondria\u2011targeted antioxidants such as Mitoquinone (MitoQ) has shown potent protective effects in preclinical AD settings. In 3\u2009\u00d7\u2009Tg\u2011AD mice, 5\u2011month MitoQ administration prevented the A\u03b2\u2011induced oxidative stress, astrogliosis, synaptic loss, and cognitive decline [158]. Although earlier studies in C. elegans models suggested MitoQ did not reduce mtDNA oxidative lesions directly, more recent reviews confirm that MitoQ enhances mitochondrial antioxidant defenses, reduces mtROS, suppresses inflammasome activation, and promotes anti-inflammatory\u00a0microglial polarization, all contributing to reduced mtDNA damage and dampened inflammatory signaling [159]. Together, these findings emphasize the therapeutic potential of targeting mtDNA-mediated innate immune activation, either through TLR9/cGAS\u2013STING pathway inhibition or mitochondria-targeted antioxidant therapy, to attenuate neuroinflammation and cognitive impairment in AD models.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Neuroinflammation and mtDNA", "text": "PD is a progressive neurodegenerative disorder primarily characterized by the selective loss of dopaminergic neurons in the substantia nigra [160]. While the accumulation of \u03b1-syn aggregates and mitochondrial dysfunction have long been established as central features of PD pathology, increasing evidence implicates mtDNA as a potent DAMP that links mitochondrial dysfunction to innate immune activation [161]. This connection is largely mediated through the cGAS\u2013STING pathway, which plays a pivotal role in driving neuroinflammation. Genetic mutations in PINK1 and Parkin, two key regulators of mitophagy, are strongly associated with early-onset familial PD [60]. PINK1 accumulates on the OMM in response to mitochondrial damage, recruiting Parkin to ubiquitinate mitochondrial proteins and promote autophagosome-mediated clearance [162]. When this system is impaired, dysfunctional mitochondria accumulate, leading to excessive production of mitochondrial ROS and leakage of mtDNA into the cytosol [163, 164]. Cytosolic mtDNA acts as a ligand for cGAS, initiating a cascade via STING that culminates in production of type I interferons, IL-6, and other pro-inflammatory cytokines. The NLRP3 inflammasome is also activated by mtDNA and mtROS, further contributing to neuroinflammatory cascades [223]. Elevated levels of circulating mtDNA and inflammatory cytokines such as IL-6 and IFN-\u03b2 have been reported in the CSF and serum of PD patients, particularly those carrying biallelic PINK1 or Parkin mutations [165]. Postmortem analyses of PD brains have revealed increased cytosolic mtDNA and upregulation of cGAS\u2013STING pathway components, especially in dopaminergic neurons of the substantia nigra [166]. Mouse models lacking Parkin or PINK1 develop motor deficits and dopaminergic neuron loss when subjected to mitochondrial stress, particularly in combination with the mutator mouse model, which accumulates mtDNA mutations due to POlG dysfunction [60]. Emerging data indicate that \u03b1-syn aggregates not only contribute to mitochondrial damage but also directly or indirectly induce genomic DNA damage. Through oxidative and nitrosative stress, \u03b1-syn facilitates the formation of DNA double-strand breaks (DSB), which can also activate cGAS\u2013STING in both neurons and glial cells [167, 168]. This is further supported by the observation of elevated \u03b3H2AX, a marker of DSBs, in PD brains. Interestingly, microglia and astrocytes exposed to \u03b1-syn fibrils exhibit STING-dependent inflammatory responses, including upregulation of IFN-stimulated genes (ISGs) and increased secretion of IL-1\u03b2, IL-6, CXCL10, and CCL20 [169]. These responses can precede neurodegeneration and are significantly ameliorated in STING-knockout models, indicating that cGAS\u2013STING activation plays a pivotal role in \u03b1-syn-induced neuroinflammation [170, 171].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Neuroinflammation and mtDNA", "text": "The interplay between impaired mitophagy, mtDNA leakage, and cGAS\u2013STING signaling provides several promising therapeutic targets. Enhancing mitophagy (e.g., via urolithin A or nicotinamide riboside) reduces mtDNA burden and inflammation in preclinical models [151]. STING inhibitors, such as H-151 and C-176, have been shown to reduce neuroinflammation and dopaminergic loss in PD mouse models. Likewise, improving mitochondrial dynamics and reducing fission via Drp1 inhibition helps prevent mtDNA fragmentation and release [172, 173]. Additionally, strategies aimed at preserving DNA repair mechanisms, such as enhancing the expression of MRE11 or APE1, could counteract \u03b1-syn-induced DNA damage and reduce inflammation [174]. Modulating astrocytic and microglial responses to mtDNA and genomic DAMPs may offer further benefits.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Neuroinflammation and mtDNA", "text": "Altogether, mtDNA acts as a molecular switch linking mitochondrial dysfunction to innate immune activation in PD. The cGAS\u2013STING pathway is emerging as a key inflammatory driver in this context, particularly when mitophagy fails to contain mitochondrial damage. Targeting mtDNA release, enhancing its clearance, or inhibiting downstream signaling may open new avenues for disease-modifying therapies in PD. Table 1 summarizes key studies that elucidate the involvement of mtDNA in mediating neuroinflammatory responses in PD.\nTable\u00a01Studies linking mtDNA damage to inflammation in PDStudy/yearModel/systemKey findingsReferenceMatsui et al., 2021Cell models & zebrafish PD modelsCytosolic double-stranded DNA of mitochondrial origin escapes lysosomal degradation and induces cytotoxicity and neurodegeneration[175]Borsche et al., 2020Parkin/PINK1 deficiency models (mouse) & reviewParkin/PINK1 loss impairs mitophagy, leads to mtDNA release and triggers inflammatory responses[165]Tresse et al., 2023Mouse models / brain injection experimentsDamaged mtDNA in brain can be released and, when introduced, elicits PD-like behavioral and pathological changes, linking mtDNA damage to PD phenotypes[176]Hinkle et al., 2022Human tissue\u2009+\u2009modelsSTING (downstream of cytosolic DNA sensing) mediates \u03b1-syn\u2013driven neuroinflammation and neurodegeneration; implicates DNA-sensing pathways in PD[170]Standaert et al., 2022Microglia / in vitro modelsMisfolded \u03b1-syn causes DNA damage in microglia and activates cGAS\u2013STING signaling to drive neuroinflammation[177]Lowes et al., 2020CSF from PD patients (PPMI cohort)Cerebrospinal fluid cell-free mtDNA levels were altered in PD cases and associated with cognitive measures, supporting relevance of mtDNA as a biomarker[178]Qi et al., 2023Peripheral blood cells from PD patientsIncreased mtDNA damage in PBMCs from idiopathic PD and LRRK2 mutation carriers proposes mtDNA damage as a peripheral PD marker[179]", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Neuroinflammation and mtDNA", "text": "HD is a fatal, autosomal-dominant neurodegenerative disorder characterized by progressive motor dysfunction, cognitive decline, and psychiatric disturbances. The disease is caused by an expanded CAG repeat in the HTT gene, leading to mHTT expression and widespread neuronal dysfunction and loss, especially in the striatum and cortex [180]. While the genetic basis of HD is well defined, the downstream molecular mechanisms contributing to neuronal death and disease progression remain an area of active research. Among these mechanisms, mitochondrial dysfunction and neuroinflammation have emerged as critical contributors to HD pathogenesis [181, 182]. Recent evidence implicates mtDNA release and activation of cGAS-STING innate immune signaling pathways, in driving neuroinflammation in HD. In HD, mitochondrial abnormalities including impaired respiration, altered calcium handling, and increased oxidative stress have been consistently reported [183, 184]. Dysfunctional mitochondria can undergo permeabilization and fragmentation, leading to the release of mitochondrial components such as mtDNA into the cytoplasm and extracellular space [134].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Neuroinflammation and mtDNA", "text": "The release of mtDNA, a pro-inflammatory DAMP, has been shown to activate innate immune responses [33]. mPTP opening, mitochondrial fission regulated by Drp1, and impaired mitophagy are mechanisms contributing to mtDNA release in NDDs, including HD [63, 185]. In HD models, increased mitochondrial fragmentation and impaired clearance have been reported, supporting the possibility of mtDNA leakage as a neuroinflammatory trigger [186]. Notably, we have demonstrated in an HD mouse model that mitochondrial dysfunction leads to enhanced release of mtDNA into the cytosol, which primes innate immune activation via the cGAS\u2013STING pathway. We have shown that the mtDNA-induced cGAS activation in the striatum leads to a robust type I interferon response, exacerbating neuroinflammation and accelerating neurodegeneration. This study provided mechanistic insights linking mitochondrial dysfunction, mtDNA release, and cGAS-STING activation with progressive HD pathology, highlighting a feed-forward cycle of neuroinflammation [12]. cGAS and downstream signaling (STING, TBK1) are upregulated in HD striatal neurons, promoting expression of proinflammatory genes like Ccl5 and Cxcl10 [187]. Sun et al. [188] revealed that mHTT disrupts DNA repair by degrading MLH1, causing DNA hyperexcision via Exo1 and cytosolic DNA accumulation. This activates the cGAS\u2013STING-dependent apoptosis and type I interferon responses. Together, these studies link mHTT to innate immune activation through DNA damage and cytosolic DNA sensing.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Neuroinflammation and mtDNA", "text": "Chronic activation of the cGAS\u2013STING pathway by mtDNA promotes sustained type I interferon and NF-\u03baB signaling, amplifying neuroinflammation. Neuroinflammation in HD has been linked to exacerbation of neuronal injury via multiple mechanisms, including microglial toxicity, altered synaptic function, and impaired clearance of mutant proteins [189]. By releasing mtDNA and triggering cGAS\u2013STING activation, mitochondrial dysfunction acts as a key upstream driver of neuroimmune responses in HD. These findings open new avenues for therapeutic intervention, such as targeting mitochondrial quality control, mtDNA release, or cGAS-STING signaling to mitigate neuroinflammation and slow disease progression. Modulation of the cGAS-STING pathway holds therapeutic promise. Pharmacological inhibitors of cGAS or STING are under development and have shown efficacy in reducing inflammation in autoimmune and neurodegenerative models. Additionally, improving mitochondrial function, enhancing mitophagy, or preventing mitochondrial fragmentation could reduce mtDNA release and subsequent inflammatory signaling. Inhibiting cGAS\u2013STING signaling with small molecules alleviated neuroinflammation in HD mice, providing a proof-of-concept for translational development. Future studies are needed to precisely delineate the role of the mtDNA\u2013cGAS\u2013STING signaling in HD progression and to explore combinational approaches targeting both mitochondrial health and innate immunity. Advanced techniques such as single-cell RNA sequencing and in vivo imaging will aid in unraveling cell-type-specific dynamics of this pathway in HD. Table 2 summarizes key studies that elucidate the involvement of mtDNA in mediating neuroinflammatory responses in HD.\nTable\u00a02Studies linking mtDNA instability and release to inflammation in HDStudy/yearModel/systemKey findingsReferenceJauhari et al., 2020HD mice & Q7 vs Q111 striatal cellsCytosolic mtDNA activates cGAS-STING-IRF3 signaling, driving inflammatory cytokine production; effects reduced by DNase1 or melatonin[12]Neueder et al., 2024Human post-mortem skeletal muscle & cellsLifelong mutant HTT expression predisposes to mitochondrial network defects and mtDNA instability in post-mitotic tissues[190]Beatriz et al., 2022HD cellular models/tissuesDefective mitochondria-lysosome axis increases release of extracellular EVs that contain mtDNA and mitochondrial proteins in HD[191]Sharma et al., 2020Cellular HD models/mechanistic assayscGAS (DNA sensor) is a key regulator of inflammatory and autophagy responses in HD models,\u00a0linking DNA sensing to HD inflammation[187]Franco-Iborra et al., 2021Cellular models (mitophagy assays)Mutant HTT impairs mitophagy steps, promoting accumulation of damaged mitochondria (mechanistic route for potential mtDNA release)[192]Siddiqui et al., 2012STHdhQ7 vs STHdhQ111 striatal cellsMutant HTT increases oxidative mtDNA damage and reduces mitochondrial bioenergetics-early biochemical evidence linking mHTT\u00a0to\u00a0mtDNA damage[28]Hwang et al., 2015Cellular models/mechanisticMutant HTT interacts with GAPDH and impairs GAPDH-induced (micro)mitophagy, promoting persistence of damaged mitochondria[193]Sun et al., 2024Cellular/molecular studies (HD mechanisms)Recent work on DNA repair/nuclear genome instability in HD that complements mtDNA instability findings (mechanistic context)[188]This table summarizes key experimental and clinical studies demonstrating mtDNA damage, instability, or cytosolic/extracellular release in HD models and patient samples. For each study, the experimental model/system (cellular, animal, or human), and key findings are indicated, highlighting the mechanistic link between mitochondrial dysfunction and innate immune activation in HD", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Neuroinflammation and mtDNA", "text": "Studies linking mtDNA instability and release to inflammation in HD\n\nThis table summarizes key experimental and clinical studies demonstrating mtDNA damage, instability, or cytosolic/extracellular release in HD models and patient samples. For each study, the experimental model/system (cellular, animal, or human), and key findings are indicated, highlighting the mechanistic link between mitochondrial dysfunction and innate immune activation in HD", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "AD", "text": "AD is marked by progressive cognitive decline linked to extracellular A\u03b2 plaques and intraneuronal neurofibrillary tangles composed of hyperphosphorylated tau protein [146]. In addition to these hallmark lesions, mitochondrial dysfunction appears early and consistently in AD pathology [147]. OXPHOS defects, increased production of ROS, and structural abnormalities of mitochondria are often seen in vulnerable brain areas such as the hippocampus and cortex [148]. Postmortem studies have shown higher mtDNA point mutations, large deletions, and lower mtDNA copy numbers in AD patients [149, 150].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "AD", "text": "There is growing evidence supporting a central role for mtDNA release in driving neuroinflammation and pathology in AD. Multiple human, animal, and cellular studies converge on the view that mitochondrial dysfunction, oxidative stress, and impaired mitophagy lead to leakage of mtDNA into the cytosol and extracellular space, where it activates innate immune sensors and sustains chronic inflammation. Clinical and postmortem human studies show elevated levels of circulating cell-free mtDNA in the cerebrospinal fluid (CSF) and plasma of AD patients, often correlating with cognitive impairment and disease severity. Mechanistic studies in animals have highlighted the role of mtDNA in activating key innate immune pathways. For example, Xie et al. [93] demonstrated in 5\u2009\u00d7\u2009FAD mice and human AD brains that cytosolic mtDNA engages the cGAS\u2013STING pathway in microglia, driving production of type I interferons and pro-inflammatory cytokines. Similarly, Hou et al. [151] found upregulation of cGAS and STING in APP/PS1 mice; genetic deletion of cGAS reduced neuroinflammation and improved cognitive function. In parallel, Heneka et al. [111] showed that mtDNA released from damaged mitochondria activates the NLRP3 inflammasome, leading to IL-1\u03b2 production and promoting amyloid pathology in AD mice and human samples. Complementary work by Fang et al. [63] demonstrated that pharmacologically enhancing mitophagy (e.g., with urolithin A) can reduce mtDNA leakage and ameliorate inflammation in APP/PS1 mice. Evidence for TLR9 activation comes from studies like Scholtzova et al. [152], showing that a TLR9 agonist (CpG oligodeoxynucleotides) modulates immune responses and reduces amyloid burden in transgenic AD mouse models, although effects on cognition and plaque clearance appear context-dependent. Roy et al. [224] demonstrated that oxidative stress triggers mtDNA leakage, which activates the cGAS\u2013STING pathway and induces the secretion of pro-inflammatory cytokines in human iPSC-derived neurons. Finally, several reviews [147, 153] have synthesized this body of work, emphasizing how mitochondrial damage and mtDNA release create a vicious cycle: oxidative stress damages mtDNA, increasing its release, which then sustains inflammation and exacerbates amyloid and tau pathology. Additionally, sustained activation of the cGAS\u2013STING signaling impairs microglial phagocytosis of A\u03b2, promoting A\u03b2 accumulation [151]. This microglial clearance impairment may result from reprogramming microglial function from a homeostatic, phagocytic state to a pro-inflammatory phenotype, which further worsens AD pathology. Overall, these findings highlight the pathogenic role of mtDNA as a neuroinflammatory trigger and its contribution to AD progression.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "AD", "text": "Pharmacological inhibition of TLR9 or the cGAS\u2013STING DNA-sensing pathway has demonstrated efficacy in mitigating neuroinflammation and cognitive deficits in AD mouse models. For instance, STING inhibition in AppNL\u2011G\u2011F/hTau knock\u2011in mice normalizes microglial function, reduces type I interferon and NLRP3 activation, decreases A\u03b2 and tau pathology, limits synaptic loss, and preserves memory performance [154, 155]. Complementarily, microglial cGAS deletion significantly curbs early plaque formation and cognitive decline in AD models, underscoring the central role of this innate immune sensor in disease progression [156, 157]. Similarly, treatment with mitochondria\u2011targeted antioxidants such as Mitoquinone (MitoQ) has shown potent protective effects in preclinical AD settings. In 3\u2009\u00d7\u2009Tg\u2011AD mice, 5\u2011month MitoQ administration prevented the A\u03b2\u2011induced oxidative stress, astrogliosis, synaptic loss, and cognitive decline [158]. Although earlier studies in C. elegans models suggested MitoQ did not reduce mtDNA oxidative lesions directly, more recent reviews confirm that MitoQ enhances mitochondrial antioxidant defenses, reduces mtROS, suppresses inflammasome activation, and promotes anti-inflammatory\u00a0microglial polarization, all contributing to reduced mtDNA damage and dampened inflammatory signaling [159]. Together, these findings emphasize the therapeutic potential of targeting mtDNA-mediated innate immune activation, either through TLR9/cGAS\u2013STING pathway inhibition or mitochondria-targeted antioxidant therapy, to attenuate neuroinflammation and cognitive impairment in AD models.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "PD", "text": "PD is a progressive neurodegenerative disorder primarily characterized by the selective loss of dopaminergic neurons in the substantia nigra [160]. While the accumulation of \u03b1-syn aggregates and mitochondrial dysfunction have long been established as central features of PD pathology, increasing evidence implicates mtDNA as a potent DAMP that links mitochondrial dysfunction to innate immune activation [161]. This connection is largely mediated through the cGAS\u2013STING pathway, which plays a pivotal role in driving neuroinflammation. Genetic mutations in PINK1 and Parkin, two key regulators of mitophagy, are strongly associated with early-onset familial PD [60]. PINK1 accumulates on the OMM in response to mitochondrial damage, recruiting Parkin to ubiquitinate mitochondrial proteins and promote autophagosome-mediated clearance [162]. When this system is impaired, dysfunctional mitochondria accumulate, leading to excessive production of mitochondrial ROS and leakage of mtDNA into the cytosol [163, 164]. Cytosolic mtDNA acts as a ligand for cGAS, initiating a cascade via STING that culminates in production of type I interferons, IL-6, and other pro-inflammatory cytokines. The NLRP3 inflammasome is also activated by mtDNA and mtROS, further contributing to neuroinflammatory cascades [223]. Elevated levels of circulating mtDNA and inflammatory cytokines such as IL-6 and IFN-\u03b2 have been reported in the CSF and serum of PD patients, particularly those carrying biallelic PINK1 or Parkin mutations [165]. Postmortem analyses of PD brains have revealed increased cytosolic mtDNA and upregulation of cGAS\u2013STING pathway components, especially in dopaminergic neurons of the substantia nigra [166]. Mouse models lacking Parkin or PINK1 develop motor deficits and dopaminergic neuron loss when subjected to mitochondrial stress, particularly in combination with the mutator mouse model, which accumulates mtDNA mutations due to POlG dysfunction [60]. Emerging data indicate that \u03b1-syn aggregates not only contribute to mitochondrial damage but also directly or indirectly induce genomic DNA damage. Through oxidative and nitrosative stress, \u03b1-syn facilitates the formation of DNA double-strand breaks (DSB), which can also activate cGAS\u2013STING in both neurons and glial cells [167, 168]. This is further supported by the observation of elevated \u03b3H2AX, a marker of DSBs, in PD brains. Interestingly, microglia and astrocytes exposed to \u03b1-syn fibrils exhibit STING-dependent inflammatory responses, including upregulation of IFN-stimulated genes (ISGs) and increased secretion of IL-1\u03b2, IL-6, CXCL10, and CCL20 [169]. These responses can precede neurodegeneration and are significantly ameliorated in STING-knockout models, indicating that cGAS\u2013STING activation plays a pivotal role in \u03b1-syn-induced neuroinflammation [170, 171].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "PD", "text": "The interplay between impaired mitophagy, mtDNA leakage, and cGAS\u2013STING signaling provides several promising therapeutic targets. Enhancing mitophagy (e.g., via urolithin A or nicotinamide riboside) reduces mtDNA burden and inflammation in preclinical models [151]. STING inhibitors, such as H-151 and C-176, have been shown to reduce neuroinflammation and dopaminergic loss in PD mouse models. Likewise, improving mitochondrial dynamics and reducing fission via Drp1 inhibition helps prevent mtDNA fragmentation and release [172, 173]. Additionally, strategies aimed at preserving DNA repair mechanisms, such as enhancing the expression of MRE11 or APE1, could counteract \u03b1-syn-induced DNA damage and reduce inflammation [174]. Modulating astrocytic and microglial responses to mtDNA and genomic DAMPs may offer further benefits.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "PD", "text": "Altogether, mtDNA acts as a molecular switch linking mitochondrial dysfunction to innate immune activation in PD. The cGAS\u2013STING pathway is emerging as a key inflammatory driver in this context, particularly when mitophagy fails to contain mitochondrial damage. Targeting mtDNA release, enhancing its clearance, or inhibiting downstream signaling may open new avenues for disease-modifying therapies in PD. Table 1 summarizes key studies that elucidate the involvement of mtDNA in mediating neuroinflammatory responses in PD.\nTable\u00a01Studies linking mtDNA damage to inflammation in PDStudy/yearModel/systemKey findingsReferenceMatsui et al., 2021Cell models & zebrafish PD modelsCytosolic double-stranded DNA of mitochondrial origin escapes lysosomal degradation and induces cytotoxicity and neurodegeneration[175]Borsche et al., 2020Parkin/PINK1 deficiency models (mouse) & reviewParkin/PINK1 loss impairs mitophagy, leads to mtDNA release and triggers inflammatory responses[165]Tresse et al., 2023Mouse models / brain injection experimentsDamaged mtDNA in brain can be released and, when introduced, elicits PD-like behavioral and pathological changes, linking mtDNA damage to PD phenotypes[176]Hinkle et al., 2022Human tissue\u2009+\u2009modelsSTING (downstream of cytosolic DNA sensing) mediates \u03b1-syn\u2013driven neuroinflammation and neurodegeneration; implicates DNA-sensing pathways in PD[170]Standaert et al., 2022Microglia / in vitro modelsMisfolded \u03b1-syn causes DNA damage in microglia and activates cGAS\u2013STING signaling to drive neuroinflammation[177]Lowes et al., 2020CSF from PD patients (PPMI cohort)Cerebrospinal fluid cell-free mtDNA levels were altered in PD cases and associated with cognitive measures, supporting relevance of mtDNA as a biomarker[178]Qi et al., 2023Peripheral blood cells from PD patientsIncreased mtDNA damage in PBMCs from idiopathic PD and LRRK2 mutation carriers proposes mtDNA damage as a peripheral PD marker[179]", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "HD", "text": "HD is a fatal, autosomal-dominant neurodegenerative disorder characterized by progressive motor dysfunction, cognitive decline, and psychiatric disturbances. The disease is caused by an expanded CAG repeat in the HTT gene, leading to mHTT expression and widespread neuronal dysfunction and loss, especially in the striatum and cortex [180]. While the genetic basis of HD is well defined, the downstream molecular mechanisms contributing to neuronal death and disease progression remain an area of active research. Among these mechanisms, mitochondrial dysfunction and neuroinflammation have emerged as critical contributors to HD pathogenesis [181, 182]. Recent evidence implicates mtDNA release and activation of cGAS-STING innate immune signaling pathways, in driving neuroinflammation in HD. In HD, mitochondrial abnormalities including impaired respiration, altered calcium handling, and increased oxidative stress have been consistently reported [183, 184]. Dysfunctional mitochondria can undergo permeabilization and fragmentation, leading to the release of mitochondrial components such as mtDNA into the cytoplasm and extracellular space [134].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "HD", "text": "The release of mtDNA, a pro-inflammatory DAMP, has been shown to activate innate immune responses [33]. mPTP opening, mitochondrial fission regulated by Drp1, and impaired mitophagy are mechanisms contributing to mtDNA release in NDDs, including HD [63, 185]. In HD models, increased mitochondrial fragmentation and impaired clearance have been reported, supporting the possibility of mtDNA leakage as a neuroinflammatory trigger [186]. Notably, we have demonstrated in an HD mouse model that mitochondrial dysfunction leads to enhanced release of mtDNA into the cytosol, which primes innate immune activation via the cGAS\u2013STING pathway. We have shown that the mtDNA-induced cGAS activation in the striatum leads to a robust type I interferon response, exacerbating neuroinflammation and accelerating neurodegeneration. This study provided mechanistic insights linking mitochondrial dysfunction, mtDNA release, and cGAS-STING activation with progressive HD pathology, highlighting a feed-forward cycle of neuroinflammation [12]. cGAS and downstream signaling (STING, TBK1) are upregulated in HD striatal neurons, promoting expression of proinflammatory genes like Ccl5 and Cxcl10 [187]. Sun et al. [188] revealed that mHTT disrupts DNA repair by degrading MLH1, causing DNA hyperexcision via Exo1 and cytosolic DNA accumulation. This activates the cGAS\u2013STING-dependent apoptosis and type I interferon responses. Together, these studies link mHTT to innate immune activation through DNA damage and cytosolic DNA sensing.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "HD", "text": "Chronic activation of the cGAS\u2013STING pathway by mtDNA promotes sustained type I interferon and NF-\u03baB signaling, amplifying neuroinflammation. Neuroinflammation in HD has been linked to exacerbation of neuronal injury via multiple mechanisms, including microglial toxicity, altered synaptic function, and impaired clearance of mutant proteins [189]. By releasing mtDNA and triggering cGAS\u2013STING activation, mitochondrial dysfunction acts as a key upstream driver of neuroimmune responses in HD. These findings open new avenues for therapeutic intervention, such as targeting mitochondrial quality control, mtDNA release, or cGAS-STING signaling to mitigate neuroinflammation and slow disease progression. Modulation of the cGAS-STING pathway holds therapeutic promise. Pharmacological inhibitors of cGAS or STING are under development and have shown efficacy in reducing inflammation in autoimmune and neurodegenerative models. Additionally, improving mitochondrial function, enhancing mitophagy, or preventing mitochondrial fragmentation could reduce mtDNA release and subsequent inflammatory signaling. Inhibiting cGAS\u2013STING signaling with small molecules alleviated neuroinflammation in HD mice, providing a proof-of-concept for translational development. Future studies are needed to precisely delineate the role of the mtDNA\u2013cGAS\u2013STING signaling in HD progression and to explore combinational approaches targeting both mitochondrial health and innate immunity. Advanced techniques such as single-cell RNA sequencing and in vivo imaging will aid in unraveling cell-type-specific dynamics of this pathway in HD. Table 2 summarizes key studies that elucidate the involvement of mtDNA in mediating neuroinflammatory responses in HD.\nTable\u00a02Studies linking mtDNA instability and release to inflammation in HDStudy/yearModel/systemKey findingsReferenceJauhari et al., 2020HD mice & Q7 vs Q111 striatal cellsCytosolic mtDNA activates cGAS-STING-IRF3 signaling, driving inflammatory cytokine production; effects reduced by DNase1 or melatonin[12]Neueder et al., 2024Human post-mortem skeletal muscle & cellsLifelong mutant HTT expression predisposes to mitochondrial network defects and mtDNA instability in post-mitotic tissues[190]Beatriz et al., 2022HD cellular models/tissuesDefective mitochondria-lysosome axis increases release of extracellular EVs that contain mtDNA and mitochondrial proteins in HD[191]Sharma et al., 2020Cellular HD models/mechanistic assayscGAS (DNA sensor) is a key regulator of inflammatory and autophagy responses in HD models,\u00a0linking DNA sensing to HD inflammation[187]Franco-Iborra et al., 2021Cellular models (mitophagy assays)Mutant HTT impairs mitophagy steps, promoting accumulation of damaged mitochondria (mechanistic route for potential mtDNA release)[192]Siddiqui et al., 2012STHdhQ7 vs STHdhQ111 striatal cellsMutant HTT increases oxidative mtDNA damage and reduces mitochondrial bioenergetics-early biochemical evidence linking mHTT\u00a0to\u00a0mtDNA damage[28]Hwang et al., 2015Cellular models/mechanisticMutant HTT interacts with GAPDH and impairs GAPDH-induced (micro)mitophagy, promoting persistence of damaged mitochondria[193]Sun et al., 2024Cellular/molecular studies (HD mechanisms)Recent work on DNA repair/nuclear genome instability in HD that complements mtDNA instability findings (mechanistic context)[188]This table summarizes key experimental and clinical studies demonstrating mtDNA damage, instability, or cytosolic/extracellular release in HD models and patient samples. For each study, the experimental model/system (cellular, animal, or human), and key findings are indicated, highlighting the mechanistic link between mitochondrial dysfunction and innate immune activation in HD", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "HD", "text": "Studies linking mtDNA instability and release to inflammation in HD\n\nThis table summarizes key experimental and clinical studies demonstrating mtDNA damage, instability, or cytosolic/extracellular release in HD models and patient samples. For each study, the experimental model/system (cellular, animal, or human), and key findings are indicated, highlighting the mechanistic link between mitochondrial dysfunction and innate immune activation in HD", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Therapeutic strategies targeting mtDNA-driven neuroinflammation", "text": "The mtDNA-driven neuroinflammation has emerged as a pivotal pathological mechanism underlying various neurodegenerative and neuroinflammatory disorders [12, 95]. Therapeutic interventions aimed at mitigating this process can be categorized into strategies that reduce mtDNA release, inhibit downstream immune-sensing pathways, promote mtDNA repair, and modulate the metabolic-immune interface (Fig.\u00a05).Fig.\u00a05Therapeutic strategies to limit mtDNA-driven neuroinflammation. The strategies include pharmacological inhibition of mtDNA-mediated cGAS\u2013STING signaling using small-molecule inhibitors or antagonists, suppression of NLRP3 inflammasome activation, and blockade of TLR9-mediated inflammatory signaling through targeted compounds. Collectively, these interventions are designed to dampen innate immune activation, reduce glial reactivity, and preserve neuronal function in neurodegenerative and neuroinflammatory disorders\n\nTherapeutic strategies to limit mtDNA-driven neuroinflammation. The strategies include pharmacological inhibition of mtDNA-mediated cGAS\u2013STING signaling using small-molecule inhibitors or antagonists, suppression of NLRP3 inflammasome activation, and blockade of TLR9-mediated inflammatory signaling through targeted compounds. Collectively, these interventions are designed to dampen innate immune activation, reduce glial reactivity, and preserve neuronal function in neurodegenerative and neuroinflammatory disorders", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Therapeutic strategies targeting mtDNA-driven neuroinflammation", "text": "The pathological release of mtDNA into the cytosol or extracellular space\u2014often due to mitochondrial membrane permeabilization or dysfunction\u2014is the initiating event driving mtDNA-induced inflammation. Therapeutic approaches focus primarily on stabilizing mitochondrial membranes and enhancing mitophagy to limit mtDNA leakage.\n\nMaintaining mitochondrial membrane integrity is crucial to prevent mtDNA escape. Inhibitors of the mPTP, such as cyclosporin A and sanglifehrin A, have demonstrated efficacy in preserving mitochondrial membrane potential and preventing mtDNA release in neurodegeneration and acute brain injury models [45, 95]. By restricting mPTP opening, these agents reduce mitochondrial swelling and rupture, thereby limiting the release of DAMPs, including mtDNA.\n\nMitophagy selectively clears damaged mitochondria, thereby preventing the accumulation of mitochondria prone to mtDNA leakage [57]. The PINK1/Parkin pathway is a key regulator of this process. Pharmacological or genetic activation of PINK1 stabilization and Parkin recruitment enhances mitochondrial turnover and reduces cytosolic mtDNA levels [60]. Urolithin A, a natural mitophagy inducer, has shown promise in improving mitochondrial quality control and attenuating neuroinflammation in preclinical models [194]. Other autophagy enhancers, such as spermidine, also indirectly diminish mtDNA release by promoting the clearance of defective mitochondria [195].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Therapeutic strategies targeting mtDNA-driven neuroinflammation", "text": "Despite a reduction in mtDNA release, residual cytosolic or extracellular mtDNA can activate innate immune receptors to sustain inflammation. Targeting these sensing pathways is therefore a critical therapeutic strategy. Small-molecule inhibitors of cGAS enzymatic activity (e.g., RU.521, G140, PF-06928215) or STING signaling (e.g., H-151, C-176) have shown promise in preclinical neuroinflammation and neurodegeneration models by suppressing excessive type I interferon and cytokine production [12, 196]. TLR9 antagonists, such as synthetic oligodeoxynucleotide antagonists (e.g., IRS-954) and chloroquine, inhibit TLR9 activation and downstream NF-\u03baB signaling, reducing pro-inflammatory cytokine release from microglia and astrocytes [197]. NLRP3 inflammasome inhibitors, such as MCC950, reduce neuroinflammatory cytokines and improve outcomes in neurodegenerative and acute brain injury models [121, 198]. Other indirect inhibitors targeting inflammasome assembly or upstream signaling cascades may also mitigate mtDNA-mediated inflammation [106].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Therapeutic strategies targeting mtDNA-driven neuroinflammation", "text": "Repairing damaged mtDNA decreases the pool of immunogenic substrates and preserves mitochondrial function, thereby mitigating neuroinflammation [199]. Enhancing mtDNA repair enzymes, POLG, which serves as the primary mtDNA polymerase, is\u00a0essential for mtDNA replication and base excision repair of oxidative lesions [200]. Therapeutic strategies aiming to upregulate POLG expression or augment its enzymatic activity may improve mtDNA integrity. These include gene therapy approaches to deliver functional POLG or small molecules that stabilize POLG complexes [201, 202]. Furthermore, boosting other mitochondrial repair enzymes such as mitochondrial uracil-DNA glycosylase (UNG1) and 8-oxoguanine DNA glycosylase (OGG1) is reported to enhance mtDNA repair capacity and reduce the mutation burden [203].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Therapeutic strategies targeting mtDNA-driven neuroinflammation", "text": "Mitochondrial metabolism is tightly coupled to innate immune signaling, which may be an additional target to attenuate mtDNA-driven neuroinflammation. Excess mitochondrial ROS causes oxidative damage to mtDNA and triggers its release, activating downstream immune pathways. Mitochondria-targeted antioxidants such as MitoQ, SkQ1, and SS-31 have been developed to selectively scavenge mitochondrial ROS, thereby protecting mtDNA from oxidative lesions and preventing inflammatory activation [204]. Clinical trials evaluating these antioxidants in NDDs have reported safety and initial efficacy, supporting their therapeutic potential. Further, metabolic drugs, metformin, a first-line treatment for type 2 diabetes, exhibits neuroprotective and anti-inflammatory effects via activation of AMP-activated protein kinase, inhibition of mitochondrial complex I, and modulation of mitochondrial metabolism. By reducing mitochondrial ROS production and enhancing bioenergetic efficiency, metformin decreases mtDNA damage and attenuates mtDNA-mediated activation of innate immune sensors such as the cGAS\u2013STING pathway and inflammasomes [205]. Emerging evidence also suggests that metformin promotes mitophagy and mitochondrial biogenesis, further limiting mtDNA release and neuroinflammation [206\u2013208]. We have summarized current mitochondria-targeted therapeutic developments in Table\u00a03.\nTable\u00a03Mitochondria-targeted therapeutic strategies for NDDsTherapeuticsMechanismIndication(s)Stage of developmentReferenceElamipretide (SS-31 / MTP-131)Mitochondrial membrane/cardiolipin stabilization; reduces oxidative stressPD, AD models; approved for Barth syndromePreclinical for NDDs; FDA-approved for Barth syndrome (Sept 2025)[209]Mitophagy enhancers (e.g., PINK1 activator ABBV-1088)Enhance mitophagy via PINK1PD, NDDsPhase 1 (NCT06414798)[210]Mitophagy enhancers (e.g., USP30 inhibitors MTX115325, FT385)Inhibit USP30 to boost mitophagyPD, ADEarly clinical/Phase I targeted[211]cGAS\u2013STING inhibitors (e.g., H-151 and others)Inhibit cytosolic DNA sensing & type I IFN inflammationNeuroinflammatory NDDsPreclinical[212]NLRP3 inflammasome inhibitorsSuppress inflammasome activation & IL-1\u03b2/IL-18PD, AD models & other inflammatory conditionsEarly clinical/preclinical[213]Selnoflast (RO7486967)NLRP3 inflammasome inhibitorEarly PD neuroinflammationPhase Ib PD safety[214]Nicotinamide Riboside (NR)NAD+\u2009precursor; supports mitochondrial metabolismPDPhase I trial (1000\u00a0mg\u2009\u00d7\u200930\u00a0days)[215]Coenzyme Q10 & derivatives (MitoQ, SkQ1)Mitochondrial antioxidants; reduce ROSAD/PD/HD modelsClinical/Preclinical[216]Melatonin/AstaxanthinmPTP inhibition; antioxidantAD/MCIClinical (Phase II/III)[217, 218]HydralazineAutophagy activation; NRF2 antioxidant regulationADPhase III[219]Glutathione (GlyNAC)Antioxidant supportAD/MCIClinical[225]PridopidineSigma-1 receptor modulator; mitochondrial supportHD, PD, AD modelsPreclinical/early clinical interest[220]Metabolic modulators (e.g., terazosin, deferiprone, sulforaphane)Mito metabolism/antioxidant pathwaysPDPhase II or early[221]This table summarizes current and emerging therapeutic approaches aimed at mitigating mitochondrial dysfunction in NDDs. For each intervention, the molecular target, proposed mechanism, disease context, and stage of development (preclinical or clinical) are indicated", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Therapeutic strategies targeting mtDNA-driven neuroinflammation", "text": "Mitochondria-targeted therapeutic strategies for NDDs\n\nThis table summarizes current and emerging therapeutic approaches aimed at mitigating mitochondrial dysfunction in NDDs. For each intervention, the molecular target, proposed mechanism, disease context, and stage of development (preclinical or clinical) are indicated", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Reducing mtDNA release", "text": "The pathological release of mtDNA into the cytosol or extracellular space\u2014often due to mitochondrial membrane permeabilization or dysfunction\u2014is the initiating event driving mtDNA-induced inflammation. Therapeutic approaches focus primarily on stabilizing mitochondrial membranes and enhancing mitophagy to limit mtDNA leakage.\n\nMaintaining mitochondrial membrane integrity is crucial to prevent mtDNA escape. Inhibitors of the mPTP, such as cyclosporin A and sanglifehrin A, have demonstrated efficacy in preserving mitochondrial membrane potential and preventing mtDNA release in neurodegeneration and acute brain injury models [45, 95]. By restricting mPTP opening, these agents reduce mitochondrial swelling and rupture, thereby limiting the release of DAMPs, including mtDNA.\n\nMitophagy selectively clears damaged mitochondria, thereby preventing the accumulation of mitochondria prone to mtDNA leakage [57]. The PINK1/Parkin pathway is a key regulator of this process. Pharmacological or genetic activation of PINK1 stabilization and Parkin recruitment enhances mitochondrial turnover and reduces cytosolic mtDNA levels [60]. Urolithin A, a natural mitophagy inducer, has shown promise in improving mitochondrial quality control and attenuating neuroinflammation in preclinical models [194]. Other autophagy enhancers, such as spermidine, also indirectly diminish mtDNA release by promoting the clearance of defective mitochondria [195].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Blocking downstream sensing pathways", "text": "Despite a reduction in mtDNA release, residual cytosolic or extracellular mtDNA can activate innate immune receptors to sustain inflammation. Targeting these sensing pathways is therefore a critical therapeutic strategy. Small-molecule inhibitors of cGAS enzymatic activity (e.g., RU.521, G140, PF-06928215) or STING signaling (e.g., H-151, C-176) have shown promise in preclinical neuroinflammation and neurodegeneration models by suppressing excessive type I interferon and cytokine production [12, 196]. TLR9 antagonists, such as synthetic oligodeoxynucleotide antagonists (e.g., IRS-954) and chloroquine, inhibit TLR9 activation and downstream NF-\u03baB signaling, reducing pro-inflammatory cytokine release from microglia and astrocytes [197]. NLRP3 inflammasome inhibitors, such as MCC950, reduce neuroinflammatory cytokines and improve outcomes in neurodegenerative and acute brain injury models [121, 198]. Other indirect inhibitors targeting inflammasome assembly or upstream signaling cascades may also mitigate mtDNA-mediated inflammation [106].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Promoting mtDNA repair", "text": "Repairing damaged mtDNA decreases the pool of immunogenic substrates and preserves mitochondrial function, thereby mitigating neuroinflammation [199]. Enhancing mtDNA repair enzymes, POLG, which serves as the primary mtDNA polymerase, is\u00a0essential for mtDNA replication and base excision repair of oxidative lesions [200]. Therapeutic strategies aiming to upregulate POLG expression or augment its enzymatic activity may improve mtDNA integrity. These include gene therapy approaches to deliver functional POLG or small molecules that stabilize POLG complexes [201, 202]. Furthermore, boosting other mitochondrial repair enzymes such as mitochondrial uracil-DNA glycosylase (UNG1) and 8-oxoguanine DNA glycosylase (OGG1) is reported to enhance mtDNA repair capacity and reduce the mutation burden [203].", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Modulating the metabolic\u2013immune interface", "text": "Mitochondrial metabolism is tightly coupled to innate immune signaling, which may be an additional target to attenuate mtDNA-driven neuroinflammation. Excess mitochondrial ROS causes oxidative damage to mtDNA and triggers its release, activating downstream immune pathways. Mitochondria-targeted antioxidants such as MitoQ, SkQ1, and SS-31 have been developed to selectively scavenge mitochondrial ROS, thereby protecting mtDNA from oxidative lesions and preventing inflammatory activation [204]. Clinical trials evaluating these antioxidants in NDDs have reported safety and initial efficacy, supporting their therapeutic potential. Further, metabolic drugs, metformin, a first-line treatment for type 2 diabetes, exhibits neuroprotective and anti-inflammatory effects via activation of AMP-activated protein kinase, inhibition of mitochondrial complex I, and modulation of mitochondrial metabolism. By reducing mitochondrial ROS production and enhancing bioenergetic efficiency, metformin decreases mtDNA damage and attenuates mtDNA-mediated activation of innate immune sensors such as the cGAS\u2013STING pathway and inflammasomes [205]. Emerging evidence also suggests that metformin promotes mitophagy and mitochondrial biogenesis, further limiting mtDNA release and neuroinflammation [206\u2013208]. We have summarized current mitochondria-targeted therapeutic developments in Table\u00a03.\nTable\u00a03Mitochondria-targeted therapeutic strategies for NDDsTherapeuticsMechanismIndication(s)Stage of developmentReferenceElamipretide (SS-31 / MTP-131)Mitochondrial membrane/cardiolipin stabilization; reduces oxidative stressPD, AD models; approved for Barth syndromePreclinical for NDDs; FDA-approved for Barth syndrome (Sept 2025)[209]Mitophagy enhancers (e.g., PINK1 activator ABBV-1088)Enhance mitophagy via PINK1PD, NDDsPhase 1 (NCT06414798)[210]Mitophagy enhancers (e.g., USP30 inhibitors MTX115325, FT385)Inhibit USP30 to boost mitophagyPD, ADEarly clinical/Phase I targeted[211]cGAS\u2013STING inhibitors (e.g., H-151 and others)Inhibit cytosolic DNA sensing & type I IFN inflammationNeuroinflammatory NDDsPreclinical[212]NLRP3 inflammasome inhibitorsSuppress inflammasome activation & IL-1\u03b2/IL-18PD, AD models & other inflammatory conditionsEarly clinical/preclinical[213]Selnoflast (RO7486967)NLRP3 inflammasome inhibitorEarly PD neuroinflammationPhase Ib PD safety[214]Nicotinamide Riboside (NR)NAD+\u2009precursor; supports mitochondrial metabolismPDPhase I trial (1000\u00a0mg\u2009\u00d7\u200930\u00a0days)[215]Coenzyme Q10 & derivatives (MitoQ, SkQ1)Mitochondrial antioxidants; reduce ROSAD/PD/HD modelsClinical/Preclinical[216]Melatonin/AstaxanthinmPTP inhibition; antioxidantAD/MCIClinical (Phase II/III)[217, 218]HydralazineAutophagy activation; NRF2 antioxidant regulationADPhase III[219]Glutathione (GlyNAC)Antioxidant supportAD/MCIClinical[225]PridopidineSigma-1 receptor modulator; mitochondrial supportHD, PD, AD modelsPreclinical/early clinical interest[220]Metabolic modulators (e.g., terazosin, deferiprone, sulforaphane)Mito metabolism/antioxidant pathwaysPDPhase II or early[221]This table summarizes current and emerging therapeutic approaches aimed at mitigating mitochondrial dysfunction in NDDs. For each intervention, the molecular target, proposed mechanism, disease context, and stage of development (preclinical or clinical) are indicated", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Modulating the metabolic\u2013immune interface", "text": "Mitochondria-targeted therapeutic strategies for NDDs\n\nThis table summarizes current and emerging therapeutic approaches aimed at mitigating mitochondrial dysfunction in NDDs. For each intervention, the molecular target, proposed mechanism, disease context, and stage of development (preclinical or clinical) are indicated", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Challenges and future perspectives", "text": "mtDNA has emerged as a critical driver of neuroinflammation, linking mitochondrial dysfunction to innate immune activation across diverse neurological disorders. However, translating this growing knowledge into reliable diagnostics and effective therapies remains challenging. Accurate detection and quantification of mtDNA release in the CNS are hindered by several factors. Cytosolic and extracellular mtDNA are often present at low levels and rapidly degraded by nucleases, complicating their measurement. Moreover, distinguishing mtDNA from nuclear mitochondrial DNA segments requires highly specific assays. While PCR-based methods are widely used, they lack sensitivity and specificity; droplet digital PCR offers improvements but lacks standardization. In biofluids like CSF and plasma, fragmented cell-free DNA further complicates interpretation, underscoring the need for enhanced sample processing and mtDNA-targeted probes. These constraints limit the development of robust mtDNA biomarkers and impede therapeutic monitoring. mtDNA-driven neuroinflammation varies widely by cell type and disease context. Microglia predominantly utilize cGAS\u2013STING and inflammasome pathways, whereas astrocytes may engage TLR9 signaling. The temporal dynamics of mtDNA release and immune responses differ between acute injuries (e.g., stroke, traumatic brain injury) and chronic NDDs. Furthermore, brain region-specific vulnerabilities and aging influence mtDNA-mediated inflammation. Current experimental models often fail to capture this heterogeneity, highlighting the need for cell-type-specific, spatially resolved, and longitudinal studies.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Challenges and future perspectives", "text": "Recent advances offer powerful tools to overcome existing challenges. Single-cell RNA sequencing combined with spatial transcriptomics enables high-resolution mapping of mtDNA sensing and inflammatory pathways within distinct cell populations in situ. Advanced imaging techniques, including super-resolution microscopy and mitochondrial-targeted fluorescent reporters, allow real-time visualization of mtDNA release and mitochondrial dynamics in vivo. Integration with proteomics, metabolomics, and epigenomics provides a comprehensive multi-omics perspective. Artificial intelligence and computational modeling can synthesize these datasets to identify novel regulatory networks and therapeutic targets.", "title": "" }, { "pmid": 41689141, "pmcid": "PMC12903234", "section": "Conclusion", "text": "mtDNA functions as a crucial regulator of neuroinflammation, bridging mitochondrial dysfunction with innate immune activation through DAMP/PRR such as cGAS\u2013STING, TLR9, and inflammasomes. This dual role of mtDNA, initiator of acute responses and perpetuator of chronic inflammation, links metabolic distress to progressive neuronal injury in aging and NDDs. Despite advances elucidating mtDNA release, sensing, and downstream signaling, critical gaps remain, particularly concerning cell-type-specific dynamics and integration with other pathological processes. Improved detection technologies and multi-omics approaches promise to resolve these complexities and accelerate biomarker discovery and therapeutic development. Therapeutic strategies targeting mitochondrial integrity, mtDNA repair, and selective inhibition of mtDNA sensing pathways hold promise to mitigate neuroinflammation and neurodegeneration. Metabolic modulators such as antioxidants, metformin, and melatonin also show potential by modulating mitochondrial and immune pathways. Realizing these advances clinically demands multidisciplinary collaboration spanning mitochondrial biology, immunology, neuroscience, and clinical research. Ultimately, harnessing mtDNA\u2019s central role could transform diagnosis, prognosis, and treatment of devastating neurological diseases.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Introduction", "text": "Neurodegenerative disorders including Alzheimer\u2019s disease (AD), Parkinson\u2019s disease (PD), Huntington\u2019s disease (HD), and Amyotrophic Lateral Sclerosis (ALS) are rising sharply with global ageing. By 2050, dementia alone is projected to impose a global cost of over US $9 trillion [1]. Despite improved insights into pathogenesis, available treatments remain only modestly effective [2, 3].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Introduction", "text": "Canonical cell death apoptosis and regulated necrosis has long framed neurodegeneration research. Apoptosis engages intrinsic/extrinsic routes to activate executioner caspases and typically ends with immunologically quiet efferocytosis [4, 5]. Regulated necrosis, once deemed passive, includes programmed forms such as necroptosis driven by RIPK1/RIPK3/MLKL, linking death to inflammatory signaling [6]. By contrast, non-canonical pathways ferroptosis, necroptosis, pyroptosis, parthanatos carry distinct biochemical/evolutionary signatures and, unlike apoptosis, frequently release DAMPs/alarmins that amplify neuroinflammation [7\u201310]. Recent multi-omic analyses reveal dense crosstalk between canonical and non-canonical regulated cell death, with shared intermediates, reciprocal checkpoints, and hierarchical decision nodes. Ferroptosis-peroxisome coupling and PANoptosome assembly exemplify coordination among apoptosis, pyroptosis, and necroptosis [11, 12]. Accordingly, perturbing one pathway can reverberate across the broader death network, arguing for therapies that anticipate these interdependencies. Pathway engagement is also stage-dependent apoptosis may prevail early, while non-canonical programs increasingly dominate chronic/advanced disease supporting precision, time-resolved interventions [13].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Introduction", "text": "Beyond canonical paradigms, recent work has increasingly implicated four regulated, non-apoptotic cell-death programmes in driving neuronal loss and neuroinflammation. These include ferroptosis, characterized by iron-driven lipid peroxidation [14\u201316], pyroptosis, mediated through Gasdermin-D-dependent inflammasome activation [17], necroptosis, executed via RIPK1-RIPK3-MLKL signalling [18], and parthanatos, initiated by PARP-1 overactivation [19]. Recognition of these pathways has opened new translational avenues, positioning them as tractable therapeutic targets to alleviate neuronal damage and improve patient outcome [20].\n\nThis review surveys recent therapeutic advances that modulate these non-canonical death programmes in neurodegenerative disease. We first outline their core machinery and disease phenotypes, then synthesize convergence and crosstalk, next discuss pathway-proximal pharmacodynamic anchors and delivery constraints, and finally propose an integrative framework that maps pathway nodes to cell-type programmes and candidate biomarkers to guide clinical translation.\n\nFig. 1Overview of molecular mechanisms involved in non-canonical cell death pathways in neurons. This schematic illustrates the core signaling cascades and key effectors that define ferroptosis, necroptosis, pyroptosis, and parthanatos. Created in https://BioRender.com/5ujmmks", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Introduction", "text": "Overview of molecular mechanisms involved in non-canonical cell death pathways in neurons. This schematic illustrates the core signaling cascades and key effectors that define ferroptosis, necroptosis, pyroptosis, and parthanatos. Created in https://BioRender.com/5ujmmks", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Involvement of non-canonical cell death mechanisms in neurodegenerative disease", "text": "Growing evidence identifies the non-canonical death programmes as compelling therapeutic targets for neurodegenerative disorders [21\u201324]. Because these pathways converge on oxidative stress, neuroinflammation and mitochondrial dysfunction, intervening in them offers a fresh angle on the multifactorial nature of diseases such as AD, PD, HD, and ALS [25]. Studies have mapped druggable nodes within each pathway and shown that their inhibition confers marked neuroprotection in pre-clinical models [26\u201329] (Fig. 1). At the same time, integrative work is clarifying where lipid peroxidation and inflammatory signalling intersect, creating new opportunities for drug design and biomarker discovery [30]. An integrated schematic of these pathways and their roles in neuronal damage is illustrated in Fig. 2. In parallel, their key mechanistic features and distinguishing biomarkers are summarized in Table 1.\n\nFig. 2Interplay of non-canonical regulated cell death mechanisms in the pathology of neurodegenerative diseases. This figure depicts how ferroptosis, necroptosis, pyroptosis, and parthanatos contribute to disease progression in AD, PD, HD, and ALS. Pathway-specific triggers and cellular outcomes such as neuroinflammation, oxidative damage, and metabolic collapse are mapped across disease types. The diagram highlights the overlapping cellular stress responses and potential intervention points for therapeutic modulation of these pathways. Created in BioRender. https://BioRender.com/pzbivta", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Involvement of non-canonical cell death mechanisms in neurodegenerative disease", "text": "Interplay of non-canonical regulated cell death mechanisms in the pathology of neurodegenerative diseases. This figure depicts how ferroptosis, necroptosis, pyroptosis, and parthanatos contribute to disease progression in AD, PD, HD, and ALS. Pathway-specific triggers and cellular outcomes such as neuroinflammation, oxidative damage, and metabolic collapse are mapped across disease types. The diagram highlights the overlapping cellular stress responses and potential intervention points for therapeutic modulation of these pathways. Created in BioRender. https://BioRender.com/pzbivta", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Involvement of non-canonical cell death mechanisms in neurodegenerative disease", "text": "First described in 2012, ferroptosis is a regulated cell-death programme distinguished by iron-dependent lipid peroxidation rather than the caspase cascades typical of apoptosis or the membrane rupture seen in necrosis [31]. Its execution hinges on the interplay of iron handling, lipid metabolism and antioxidant defences [32]. Oxidation of membrane polyunsaturated fatty acids (PUFAs) generates lipid peroxides that, if not detoxified, compromise membrane integrity and trigger cell death [33]. Iron chelators can block this process, underscoring the central role of iron homeostasis and suggesting therapeutic potential in modulating intracellular iron levels [34, 35]. Two key protective systems counterbalance ferroptotic stress. First, glutathione peroxidase-4 (GPX4) reduces lipid hydroperoxides to non-toxic alcohols, a reaction that relies on glutathione (GSH). Second, the cystine/glutamate antiporter system Xc\u2212 sustains GSH synthesis by importing cystine, its inhibition sensitises cells to ferroptosis [36\u201338]. Transcriptional activation of Nrf2 (NFE2L2) further bolsters antioxidant capacity by up-regulating detoxifying enzymes [36].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Involvement of non-canonical cell death mechanisms in neurodegenerative disease", "text": "Ferroptosis is tightly coupled to cellular iron handling and oxidative stress, positioning it as a central mediator of neurodegenerative pathology [16]. Dysregulated iron homeostasis drives Fenton chemistry and PUFA lipid peroxidation, producing lethal lipid peroxide accumulation and membrane failure [16, 39, 40]. For disease-specific mechanistic summaries in AD, PD, ALS, and HD, see Sect. 3. The resulting redox imbalance heightens neuronal vulnerability and contributes to progressive loss in disorders such as AD and PD [41\u201343] This profile also anticipates the inter-pathway interactions detailed in Sect. 5, particularly where oxidative stress and mitochondrial dysfunction signaling converge.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Involvement of non-canonical cell death mechanisms in neurodegenerative disease", "text": "Table 1Comparison of mechanisms, biomarkers, and pharmacological agentsFerroptosisNecroptosisPyroptosisParthanatosDefinitionIron-dependent, lipid peroxidation-driven cell deathCaspase-independent, inflammatory necrotic deathInflammatory cell death mediated by inflammasomesPARP1-dependent cell death triggered by DNA damageTriggersIron overload, erastin, RSL3TNF-\u03b1, TLRs, interferonsLPS, nigericin, microbial infectionROS, DNA damage, alkylating agentsKey RegulatorsGPX4, xCT/SLC7A11, ACSL4, Nrf2, Fe\u00b2\u207aRIPK1, RIPK3, MLKL, caspase-8, FADDNLRP3, Caspase-1/4/511, GSDMDPARP1, AIF, PARGPotential BiomarkersACSL4, lipid ROS, MDAp-MLKL, RIPK3GSDMD-N, IL-1\u03b2, caspase-1/4/511PAR, nuclear AIFEnergy DependenceModerate (depends on GSH and NADPH)Yes (requires ATP)Yes (inflammasome activation)Yes (ATP and NAD\u207a depletion)Morphological aspects and Membrane DisruptionCellular swelling, mitochondrial shrinkage, loss of cristae structure, and increased mitochondrial membrane density, and lipid ROS-induced lipid bilayer damage [8]Cellular swelling, intermediate levels of chromatin condensation, and MLKL-mediated pore formation [8]Cellular swelling, intermediate levels of chromatin condensation, and gasdermin D pore formation [8]Shrinkage of the cell, AIF-mediated mitochondrial dysfunction, condensed chromatin, extensive DNA fragmentation, and compromised membrane integrity [77]Therapeutic InhibitorsFerrostatin-1, Liproxstatin-1Necrostatin-1 (RIPK1 inhibitor)MCC950 (NLRP3 inhibitor), VX-765 (Caspase inhibitor)PARP inhibitors (Olaparib, Veliparib)Nomenclature Introduced2012 [7]2005 [7]2001 [7]2009 [7]", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Involvement of non-canonical cell death mechanisms in neurodegenerative disease", "text": "Comparison of mechanisms, biomarkers, and pharmacological agents\n\nPARP-1: Poly(ADP-ribose) polymerase 1, RSL3: RAS-selective lethal 3, TNF-\u03b1: Tumor necrosis factor alpha, TLRs: Toll-like receptors, LPS: Lipopolysaccharide, ROS: Reactive oxygen species, GPX4: Glutathione peroxidase 4, xCT/SLC7A11: Cystine/glutamate antiporter, ACSL4: Acyl-CoA synthetase long-chain family member 4, Nrf2: Nuclear factor erythroid 2-related factor 2, RIPK1: Receptor-interacting protein kinase 1, RIPK3: Receptor-interacting protein kinase 3, MLKL: Mixed lineage kinase domain-like pseudokinase, FADD: Fas-associated protein with death domain, NLRP3: NOD-like receptor family pyrin domain containing 3, GSDMD: Gasdermin D, AIF: Apoptosis-inducing factor, PARG: Poly(ADP-ribose) glycohydrolase, MDA: Malondialdehyde, p-MLKL: phosphorylated-Mixed lineage kinase domain-like pseudokinase, GSDMD-N: N-terminal fragment of Gasdermin D, IL-1\u03b2: Interleukin-1 beta, PAR: Poly(ADP-ribose), GSH: Glutathione, NADPH: Nicotinamide adenine dinucleotide phosphate, ATP: Adenosine triphosphate, NAD\u207a: Nicotinamide adenine dinucleotide, PARP: Poly(ADP-ribose) polymerase.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Involvement of non-canonical cell death mechanisms in neurodegenerative disease", "text": "Necroptosis is a caspase-independent, pro-inflammatory cell-death programme that becomes prominent when apoptotic signalling is blocked [44]. The loss of plasma membrane integrity and the induction of inflammatory responses link this process particularly to neurodegenerative diseases, infections, and inflammatory disorders. The RIPK1-RIPK3-MLKL axis constitutes the core signaling pathway of necroptosis [18]. Necroptosis is triggered by the activation of death receptors through the binding of death ligands such as TNF-\u03b1 to receptors like TNFR1. Following this binding, Complex I is formed, which includes proteins such as RIPK1, TRADD, TRAF2, and cIAP1/2 [45]. This complex initially transmits cell survival signals via NF-\u03baB [46]. Upon deubiquitination of RIP1 by CYLD, caspase-8, FADD, RIP1, and RIP3 assemble to form complex IIa. Within this complex, caspase-8 remains active and suppresses the necroptotic activity of RIP1 and RIP3, leading to apoptosis. However, when caspase-8 is inhibited or silenced, RIP1, RIP3, and MLKL associate to form the necrosome (complex IIb). In this context, the mutual direct or indirect phosphorylation between these components occurs [47]. Phosphorylated MLKL then oligomerises and inserts into the plasma membrane, disturbing ion homeostasis and causing swelling, membrane rupture and a burst of reactive oxygen species (ROS) [46] (Table 2).", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Involvement of non-canonical cell death mechanisms in neurodegenerative disease", "text": "Table 2Biomarkers of non-canonical cell death pathways in neurodegenerative diseasesDeath typeBiomarker typeBiomarkersRole / mechanismFerroptosisLipid peroxidation productsMDA, 4-HNEIndicators of oxidative stress and ferroptosisFerroptosisOther lipid markersF\u2082-isoprostanes, oxidized PE (PUFA-PE-OOH)Products of ACSL4/lipoxygenase activity; ferroptosis-specific lipid peroxidationFerroptosisNeuroimagingQSM, T2 MRIDetect iron deposits in brain regionsFerroptosisIron transport proteinsFerroportinRegulates iron export; dysfunction causes accumulationFerroptosisAntioxidant defensesGPX4Protects against lipid peroxidesFerroptosisTranscriptomic signaturesACSL4, SAT1, ALOX12, TrioSig, FerrDb/KEGG panelsReflect ferroptosis susceptibility and heterogeneityNecroptosisPhosphorylated proteinspRIPK1, pRIPK3, pMLKLMark sequential stages of death signalingNecroptosisNecrosome complexRIPK1-RIPK3-MLKL complexPlatform for necroptosis executionNecroptosis(non-canonical)Upstream mediatorsTRIF, ZBP1Bypass RIPK1, directly activate RIPK3-MLKL axisNecroptosis (non-canonical)Biomarker signaturepRIPK3, pMLKL, low/absent pRIPK1, TRIF/ZBP1Distinguishes non-canonical necroptosisNecroptosisDAMPsHMGB1Released during necroptosis, measurable in fluidsPyroptosis (non-canonical)Effector proteinsCleaved GSDMD (GSDMD-N)Pore formation and membrane rupturePyroptosis (non-canonical)Initiator caspasesCaspase-4/5 (human), Caspase-11 (mouse)Direct sensing of cytosolic danger signals; cleave GSDMDPyroptosisInflammatory cytokinesIL-1\u03b2, IL-18 (CSF)Released through GSDMD poresPyroptosis (alternative)Effector proteinsCleaved GSDME (GSDME-N)Caspase-3 cleavage switches apoptosis-pyroptosisPyroptosis (alternative)Initiator caspaseCaspase-8Cleaves GSDMD under inflammatory stressParthanatos (non-canonical)Nuclear eventsAIF-MIF translocationNuclear co-translocation mediates large-scale DNA fragmentationParthanatosPAR-protein complexesPAR-\u03b1-synucleinPAR interacts with \u03b1-syn aggregatesParthanatosRepair enzyme dysregulationARH3 deficiency, Ser-ADPr accumulationImpaired reversal of ADP-ribosylationParthanatosMetabolic markersSARM1 activation, NAD\u207a depletionAxonal degeneration through NAD\u207a collapseMDA: Malondialdehyde, 4-HNE: 4-Hydroxynonenal, PE: Phosphatidylethanolamine, PUFA-PE-OOH: Polyunsaturated fatty acid\u2013phosphatidylethanolamine hydroperoxide, ACSL4: Acyl-CoA synthetase long-chain family member 4, QSM: Quantitative susceptibility mapping, MRI: Magnetic resonance imaging, GPX4: Glutathione peroxidase 4, SAT1: Spermidine/spermine N1-acetyltransferase 1, ALOX12: Arachidonate 12-lipoxygenase, TrioSig: Ferroptosis-related gene signature (Trio signature), FerrDb: Ferroptosis Database, KEGG: Kyoto Encyclopedia of Genes and Genomes, pRIPK1: Phospho-receptor-interacting serine/threonine-protein kinase 1, pRIPK3: phosphorylated-Receptor-interacting protein kinase 3, pMLKL: phosphorylated-mixed lineage kinase domain-like pseudokinase, TRIF: TIR-domain-containing adapter-inducing interferon-\u03b2, ZBP1: Z-DNA-binding protein 1, HMGB1: High-mobility group box 1, DAMP: Damage-associated molecular pattern, GSDMD: Gasdermin D, GSDMD-N: N-terminal fragment of gasdermin D, IL-1\u03b2: Interleukin-1 beta, IL-18: Interleukin-18, CSF: Cerebrospinal fluid, GSDME: Gasdermin E, GSDME-N: N-terminal fragment of gasdermin E, AIF: Apoptosis-inducing factor, MIF: Macrophage migration inhibitory factor, PAR: Poly(ADP-ribose), ARH3: ADP-ribosylhydrolase 3, Ser-ADPr: Serine-ADP-ribosylation, SARM1: Sterile alpha and TIR motif-containing 1, NAD\u207a: Nicotinamide adenine dinucleotide", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Involvement of non-canonical cell death mechanisms in neurodegenerative disease", "text": "Biomarkers of non-canonical cell death pathways in neurodegenerative diseases\n\nMDA: Malondialdehyde, 4-HNE: 4-Hydroxynonenal, PE: Phosphatidylethanolamine, PUFA-PE-OOH: Polyunsaturated fatty acid\u2013phosphatidylethanolamine hydroperoxide, ACSL4: Acyl-CoA synthetase long-chain family member 4, QSM: Quantitative susceptibility mapping, MRI: Magnetic resonance imaging, GPX4: Glutathione peroxidase 4, SAT1: Spermidine/spermine N1-acetyltransferase 1, ALOX12: Arachidonate 12-lipoxygenase, TrioSig: Ferroptosis-related gene signature (Trio signature), FerrDb: Ferroptosis Database, KEGG: Kyoto Encyclopedia of Genes and Genomes, pRIPK1: Phospho-receptor-interacting serine/threonine-protein kinase 1, pRIPK3: phosphorylated-Receptor-interacting protein kinase 3, pMLKL: phosphorylated-mixed lineage kinase domain-like pseudokinase, TRIF: TIR-domain-containing adapter-inducing interferon-\u03b2, ZBP1: Z-DNA-binding protein 1, HMGB1: High-mobility group box 1, DAMP: Damage-associated molecular pattern, GSDMD: Gasdermin D, GSDMD-N: N-terminal fragment of gasdermin D, IL-1\u03b2: Interleukin-1 beta, IL-18: Interleukin-18, CSF: Cerebrospinal fluid, GSDME: Gasdermin E, GSDME-N: N-terminal fragment of gasdermin E, AIF: Apoptosis-inducing factor, MIF: Macrophage migration inhibitory factor, PAR: Poly(ADP-ribose), ARH3: ADP-ribosylhydrolase 3, Ser-ADPr: Serine-ADP-ribosylation, SARM1: Sterile alpha and TIR motif-containing 1, NAD\u207a: Nicotinamide adenine dinucleotide", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Involvement of non-canonical cell death mechanisms in neurodegenerative disease", "text": "Elevated membrane permeability and cytokine release place necroptosis at the intersection of neuroinflammation and neuronal loss [44]. Studies have guided treatments by reducing both neuronal loss and neuroinflammation.\n\nIn AD, elevated necroptotic signalling has been documented in patient brains and aligns with worsening cognition [48, 49]. Salvadores et al. emphasize in their study that amyloid- \u03b2 peptides can trigger necroptotic pathways through microglial activation, a process that further spreads neuroinflammation and accelerates neuronal deterioration [48]. Experimental blockade of the RIPK1-RIPK3-MLKL axis in AD models limits neuronal death and helps maintain cognitive performance underscoring the pathway\u2019s therapeutic appeal [50, 51]. Comparable observations have emerged in PD and vascular dementia: dopaminergic neurons and glia display necroptotic markers, and pathway inhibition affords measurable neuroprotection [52, 53]. The interplay between necroptosis and neuroinflammation is particularly critical; necroptosis in microglia promotes a chronic inflammatory phenotype [44, 54]. Necroptosis also intersects with oxidative and other cellular stress responses [55]. This suggests that multifaceted treatment strategies targeting necroptosis and associated inflammatory and oxidative cascades may be necessary for optimal benefit [56, 57].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Involvement of non-canonical cell death mechanisms in neurodegenerative disease", "text": "These insights have sparked interest in small-molecule inhibitors of RIPK1, RIPK3 and MLKL, several of which confer neuroprotection in pre-clinical models and are advancing toward clinical evaluation [58] (Table 3). Beyond therapy, necroptotic markers detected in biofluids or imaging may aid prognosis and treatment monitoring [55, 59]. A fuller understanding of how necroptosis interfaces with oxidative and inflammatory stress should therefore inform both biomarker development and multipronged therapeutic strategies for neurodegenerative disease. This profile predicts inter-pathway interactions detailed in Sect. 5, particularly where oxidative stress signaling converges.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Involvement of non-canonical cell death mechanisms in neurodegenerative disease", "text": "Pyroptosis is an inflammatory form of programmed cell death. The term combines pyro (\u201cfire,\u201d reflecting its pro-inflammatory nature) and ptosis (\u201cfalling,\u201d as with other programmed deaths) [28]. It is typically initiated by innate immune stimuli that activate inflammasomes most prominently members of the NOD-like receptor (NLR) family and is associated with the rapid release of IL-1\u03b2 and IL-18. Morphologically, cells often swell, the plasma membrane ruptures, and intracellular contents spill into the extracellular space, which can amplify local inflammation [60, 61]. Two main routes have been described. In the canonical pathway, sensors such as NLRP3 assemble an inflammasome that activates caspase-1. Caspase-1 cleaves gasdermin D (GSDMD) to liberate its N-terminal fragment (GSDMD-N), which oligomerizes in the plasma membrane to form pores and thereby drives pyroptotic death [62]. In the non-canonical pathway, human caspase-4/5 (and murine caspase-11) are directly engaged by intracellular lipopolysaccharide; molecules such as oxPAPC may compete with LPS and blunt this activation. These caspases also cleave GSDMD to generate pore-forming GSDMD-N. While they do not directly mature IL-1\u03b2 or IL-18, the subsequent ionic fluxes (particularly K\u207a efflux) could permissively activate NLRP3 and lead to cytokine release [28].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Involvement of non-canonical cell death mechanisms in neurodegenerative disease", "text": "Multiple factors appear to modulate pyroptosis. ROS and other stress signals can promote inflammasome priming or activation [63, 64], while NF-\u03baB signaling frequently serves as a transcriptional \u201cpriming\u201d step for pro-IL-1\u03b2 and inflammasome components. Pathogens, toxins, and cytokine milieus further tune the response, suggesting substantial context dependence across tissues and disease states [65\u201367].\n\nIn the nervous system, pyroptosis seems to play a dual role. As part of host defense it may help contain infection; however, dysregulated or chronic activation can contribute to sterile inflammation and tissue injury [67\u201369]. Emerging literature links inflammasome activity and pyroptotic signaling to neurodegenerative settings for example, amyloid-\u03b2 or tau in AD, dopaminergic vulnerability in PD, and related mechanisms in HD and multiple sclerosis [70\u201374]. In such contexts, sustained production of inflammatory mediators may create conditions that favor synaptic dysfunction, neuronal loss, and circuit impairment [75].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Involvement of non-canonical cell death mechanisms in neurodegenerative disease", "text": "These insights have encouraged exploration of pharmacological modulators of inflammasome pyroptosis signaling. Inhibitors that target NLRP3, (e.g., MCC950, OLT1177/dapansutrile) have shown promise in experimental model systems and continue to be evaluated. Representative compounds and their reported effects are summarized in Table\u00a03. This profile also predicts intra-pathway interactions, specifically where inflammatory signaling converges, which are detailed in Sect.\u00a05.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Involvement of non-canonical cell death mechanisms in neurodegenerative disease", "text": "Parthanatos, derived from \u201cpoly(ADP-ribose) (PAR)\u201d and the Greek word \u201cThanatos\u201d (death), refers to a form of cell death caused by the accumulation of PAR polymers [76]. Parthanatos can be induced by various stressors, including ROS, DNA-damaging agents, nitric oxide (NO)-derived peroxynitrite, DNA alkylating compounds such as N-methyl-N\u2019-nitro-N-nitrosoguanidine (MNNG), ultraviolet radiation, and ionizing irradiation [20]. Calcium overload triggered by glutamate receptor activation, along with ROS produced under oxidative stress, leads to both single- and double-stranded DNA breaks. These lesions result in the excessive activation of Poly(ADP-ribose) polymerase 1 (PARP-1). Poly(ADP-ribose) glycohydrolase (PARG), the main enzyme responsible for degrading poly(ADP-ribose) (PAR) chains, cleaves and releases small PAR oligomers from PARP1 and associated proteins. These PAR fragments subsequently migrate to the cytoplasm, where they interact with mitochondria, promoting the release of a cleaved form of apoptosis-inducing factor (AIF) from mitochondrial membranes. AIF then translocates to the nucleus partly guided by its nuclear localization sequence where it facilitates large-scale DNA fragmentation by activating endonucleases. A downstream bioenergetic crisis often accompanies this pathway, with NAD\u207a depletion, ATP loss, and broader metabolic impairment frequently described as hallmarks [77]. Neurons may be particularly vulnerable given their high energetic demand and limited regenerative capacity, which could help explain why parthanatos-like signatures are discussed in neurodegenerative contexts [78]. In AD, for instance, amyloid-\u03b2 accumulation has been associated with oxidative stress, heightened PARP-1 activity, mitochondrial dysfunction, and indices of NAD\u207a/ATP depletion findings that together suggest a potential parthanatos component and motivate interest in targeting PARP-1 [77, 79].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Involvement of non-canonical cell death mechanisms in neurodegenerative disease", "text": "Experimental studies increasingly indicate that attenuating this pathway can be neuroprotective in vitro and in vivo models, although the magnitude and generalizability of benefit likely depend on context [80, 81]. Persistent PARP-1 activation under chronic oxidative stress and cumulative DNA damage has been linked to greater neuronal susceptibility, and dysregulated PARP-1 activity is sometimes correlated with disease severity, raising the possibility that it acts both as a marker and a mediator. These observations have encouraged exploration of PARP inhibitors in neurological settings [81\u201383]. Timing, dosing, and brain penetration appear critical variables, and several agents including olaparib and nicotinamide have been evaluated for effects consistent with dampening parthanatos-related injury (see Table 3 for examples in AD and PD models) [81, 84].\n\nLooking ahead, closing the translational gap will likely require finer mapping of cell-type specific vulnerabilities, clarification of context-dependent triggers, and a better understanding of how parthanatos intersects with other regulated death programs such as ferroptosis and necroptosis. Parallel advances in more selective, CNS-penetrant PARP inhibitors and in biomarker strategies that enable real-time pathway monitoring could improve therapeutic precision. If successful, parthanatos modulation may eventually align with multi-target strategies aimed at slowing progressive neuronal loss.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Untitled", "text": "First described in 2012, ferroptosis is a regulated cell-death programme distinguished by iron-dependent lipid peroxidation rather than the caspase cascades typical of apoptosis or the membrane rupture seen in necrosis [31]. Its execution hinges on the interplay of iron handling, lipid metabolism and antioxidant defences [32]. Oxidation of membrane polyunsaturated fatty acids (PUFAs) generates lipid peroxides that, if not detoxified, compromise membrane integrity and trigger cell death [33]. Iron chelators can block this process, underscoring the central role of iron homeostasis and suggesting therapeutic potential in modulating intracellular iron levels [34, 35]. Two key protective systems counterbalance ferroptotic stress. First, glutathione peroxidase-4 (GPX4) reduces lipid hydroperoxides to non-toxic alcohols, a reaction that relies on glutathione (GSH). Second, the cystine/glutamate antiporter system Xc\u2212 sustains GSH synthesis by importing cystine, its inhibition sensitises cells to ferroptosis [36\u201338]. Transcriptional activation of Nrf2 (NFE2L2) further bolsters antioxidant capacity by up-regulating detoxifying enzymes [36].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Untitled", "text": "Ferroptosis is tightly coupled to cellular iron handling and oxidative stress, positioning it as a central mediator of neurodegenerative pathology [16]. Dysregulated iron homeostasis drives Fenton chemistry and PUFA lipid peroxidation, producing lethal lipid peroxide accumulation and membrane failure [16, 39, 40]. For disease-specific mechanistic summaries in AD, PD, ALS, and HD, see Sect. 3. The resulting redox imbalance heightens neuronal vulnerability and contributes to progressive loss in disorders such as AD and PD [41\u201343] This profile also anticipates the inter-pathway interactions detailed in Sect. 5, particularly where oxidative stress and mitochondrial dysfunction signaling converge.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Untitled", "text": "Table 1Comparison of mechanisms, biomarkers, and pharmacological agentsFerroptosisNecroptosisPyroptosisParthanatosDefinitionIron-dependent, lipid peroxidation-driven cell deathCaspase-independent, inflammatory necrotic deathInflammatory cell death mediated by inflammasomesPARP1-dependent cell death triggered by DNA damageTriggersIron overload, erastin, RSL3TNF-\u03b1, TLRs, interferonsLPS, nigericin, microbial infectionROS, DNA damage, alkylating agentsKey RegulatorsGPX4, xCT/SLC7A11, ACSL4, Nrf2, Fe\u00b2\u207aRIPK1, RIPK3, MLKL, caspase-8, FADDNLRP3, Caspase-1/4/511, GSDMDPARP1, AIF, PARGPotential BiomarkersACSL4, lipid ROS, MDAp-MLKL, RIPK3GSDMD-N, IL-1\u03b2, caspase-1/4/511PAR, nuclear AIFEnergy DependenceModerate (depends on GSH and NADPH)Yes (requires ATP)Yes (inflammasome activation)Yes (ATP and NAD\u207a depletion)Morphological aspects and Membrane DisruptionCellular swelling, mitochondrial shrinkage, loss of cristae structure, and increased mitochondrial membrane density, and lipid ROS-induced lipid bilayer damage [8]Cellular swelling, intermediate levels of chromatin condensation, and MLKL-mediated pore formation [8]Cellular swelling, intermediate levels of chromatin condensation, and gasdermin D pore formation [8]Shrinkage of the cell, AIF-mediated mitochondrial dysfunction, condensed chromatin, extensive DNA fragmentation, and compromised membrane integrity [77]Therapeutic InhibitorsFerrostatin-1, Liproxstatin-1Necrostatin-1 (RIPK1 inhibitor)MCC950 (NLRP3 inhibitor), VX-765 (Caspase inhibitor)PARP inhibitors (Olaparib, Veliparib)Nomenclature Introduced2012 [7]2005 [7]2001 [7]2009 [7]", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Untitled", "text": "Comparison of mechanisms, biomarkers, and pharmacological agents\n\nPARP-1: Poly(ADP-ribose) polymerase 1, RSL3: RAS-selective lethal 3, TNF-\u03b1: Tumor necrosis factor alpha, TLRs: Toll-like receptors, LPS: Lipopolysaccharide, ROS: Reactive oxygen species, GPX4: Glutathione peroxidase 4, xCT/SLC7A11: Cystine/glutamate antiporter, ACSL4: Acyl-CoA synthetase long-chain family member 4, Nrf2: Nuclear factor erythroid 2-related factor 2, RIPK1: Receptor-interacting protein kinase 1, RIPK3: Receptor-interacting protein kinase 3, MLKL: Mixed lineage kinase domain-like pseudokinase, FADD: Fas-associated protein with death domain, NLRP3: NOD-like receptor family pyrin domain containing 3, GSDMD: Gasdermin D, AIF: Apoptosis-inducing factor, PARG: Poly(ADP-ribose) glycohydrolase, MDA: Malondialdehyde, p-MLKL: phosphorylated-Mixed lineage kinase domain-like pseudokinase, GSDMD-N: N-terminal fragment of Gasdermin D, IL-1\u03b2: Interleukin-1 beta, PAR: Poly(ADP-ribose), GSH: Glutathione, NADPH: Nicotinamide adenine dinucleotide phosphate, ATP: Adenosine triphosphate, NAD\u207a: Nicotinamide adenine dinucleotide, PARP: Poly(ADP-ribose) polymerase.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "The core mechanisms of necroptosis and its implications in neurodegeneration", "text": "Necroptosis is a caspase-independent, pro-inflammatory cell-death programme that becomes prominent when apoptotic signalling is blocked [44]. The loss of plasma membrane integrity and the induction of inflammatory responses link this process particularly to neurodegenerative diseases, infections, and inflammatory disorders. The RIPK1-RIPK3-MLKL axis constitutes the core signaling pathway of necroptosis [18]. Necroptosis is triggered by the activation of death receptors through the binding of death ligands such as TNF-\u03b1 to receptors like TNFR1. Following this binding, Complex I is formed, which includes proteins such as RIPK1, TRADD, TRAF2, and cIAP1/2 [45]. This complex initially transmits cell survival signals via NF-\u03baB [46]. Upon deubiquitination of RIP1 by CYLD, caspase-8, FADD, RIP1, and RIP3 assemble to form complex IIa. Within this complex, caspase-8 remains active and suppresses the necroptotic activity of RIP1 and RIP3, leading to apoptosis. However, when caspase-8 is inhibited or silenced, RIP1, RIP3, and MLKL associate to form the necrosome (complex IIb). In this context, the mutual direct or indirect phosphorylation between these components occurs [47]. Phosphorylated MLKL then oligomerises and inserts into the plasma membrane, disturbing ion homeostasis and causing swelling, membrane rupture and a burst of reactive oxygen species (ROS) [46] (Table 2).", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "The core mechanisms of necroptosis and its implications in neurodegeneration", "text": "Table 2Biomarkers of non-canonical cell death pathways in neurodegenerative diseasesDeath typeBiomarker typeBiomarkersRole / mechanismFerroptosisLipid peroxidation productsMDA, 4-HNEIndicators of oxidative stress and ferroptosisFerroptosisOther lipid markersF\u2082-isoprostanes, oxidized PE (PUFA-PE-OOH)Products of ACSL4/lipoxygenase activity; ferroptosis-specific lipid peroxidationFerroptosisNeuroimagingQSM, T2 MRIDetect iron deposits in brain regionsFerroptosisIron transport proteinsFerroportinRegulates iron export; dysfunction causes accumulationFerroptosisAntioxidant defensesGPX4Protects against lipid peroxidesFerroptosisTranscriptomic signaturesACSL4, SAT1, ALOX12, TrioSig, FerrDb/KEGG panelsReflect ferroptosis susceptibility and heterogeneityNecroptosisPhosphorylated proteinspRIPK1, pRIPK3, pMLKLMark sequential stages of death signalingNecroptosisNecrosome complexRIPK1-RIPK3-MLKL complexPlatform for necroptosis executionNecroptosis(non-canonical)Upstream mediatorsTRIF, ZBP1Bypass RIPK1, directly activate RIPK3-MLKL axisNecroptosis (non-canonical)Biomarker signaturepRIPK3, pMLKL, low/absent pRIPK1, TRIF/ZBP1Distinguishes non-canonical necroptosisNecroptosisDAMPsHMGB1Released during necroptosis, measurable in fluidsPyroptosis (non-canonical)Effector proteinsCleaved GSDMD (GSDMD-N)Pore formation and membrane rupturePyroptosis (non-canonical)Initiator caspasesCaspase-4/5 (human), Caspase-11 (mouse)Direct sensing of cytosolic danger signals; cleave GSDMDPyroptosisInflammatory cytokinesIL-1\u03b2, IL-18 (CSF)Released through GSDMD poresPyroptosis (alternative)Effector proteinsCleaved GSDME (GSDME-N)Caspase-3 cleavage switches apoptosis-pyroptosisPyroptosis (alternative)Initiator caspaseCaspase-8Cleaves GSDMD under inflammatory stressParthanatos (non-canonical)Nuclear eventsAIF-MIF translocationNuclear co-translocation mediates large-scale DNA fragmentationParthanatosPAR-protein complexesPAR-\u03b1-synucleinPAR interacts with \u03b1-syn aggregatesParthanatosRepair enzyme dysregulationARH3 deficiency, Ser-ADPr accumulationImpaired reversal of ADP-ribosylationParthanatosMetabolic markersSARM1 activation, NAD\u207a depletionAxonal degeneration through NAD\u207a collapseMDA: Malondialdehyde, 4-HNE: 4-Hydroxynonenal, PE: Phosphatidylethanolamine, PUFA-PE-OOH: Polyunsaturated fatty acid\u2013phosphatidylethanolamine hydroperoxide, ACSL4: Acyl-CoA synthetase long-chain family member 4, QSM: Quantitative susceptibility mapping, MRI: Magnetic resonance imaging, GPX4: Glutathione peroxidase 4, SAT1: Spermidine/spermine N1-acetyltransferase 1, ALOX12: Arachidonate 12-lipoxygenase, TrioSig: Ferroptosis-related gene signature (Trio signature), FerrDb: Ferroptosis Database, KEGG: Kyoto Encyclopedia of Genes and Genomes, pRIPK1: Phospho-receptor-interacting serine/threonine-protein kinase 1, pRIPK3: phosphorylated-Receptor-interacting protein kinase 3, pMLKL: phosphorylated-mixed lineage kinase domain-like pseudokinase, TRIF: TIR-domain-containing adapter-inducing interferon-\u03b2, ZBP1: Z-DNA-binding protein 1, HMGB1: High-mobility group box 1, DAMP: Damage-associated molecular pattern, GSDMD: Gasdermin D, GSDMD-N: N-terminal fragment of gasdermin D, IL-1\u03b2: Interleukin-1 beta, IL-18: Interleukin-18, CSF: Cerebrospinal fluid, GSDME: Gasdermin E, GSDME-N: N-terminal fragment of gasdermin E, AIF: Apoptosis-inducing factor, MIF: Macrophage migration inhibitory factor, PAR: Poly(ADP-ribose), ARH3: ADP-ribosylhydrolase 3, Ser-ADPr: Serine-ADP-ribosylation, SARM1: Sterile alpha and TIR motif-containing 1, NAD\u207a: Nicotinamide adenine dinucleotide", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "The core mechanisms of necroptosis and its implications in neurodegeneration", "text": "Biomarkers of non-canonical cell death pathways in neurodegenerative diseases\n\nMDA: Malondialdehyde, 4-HNE: 4-Hydroxynonenal, PE: Phosphatidylethanolamine, PUFA-PE-OOH: Polyunsaturated fatty acid\u2013phosphatidylethanolamine hydroperoxide, ACSL4: Acyl-CoA synthetase long-chain family member 4, QSM: Quantitative susceptibility mapping, MRI: Magnetic resonance imaging, GPX4: Glutathione peroxidase 4, SAT1: Spermidine/spermine N1-acetyltransferase 1, ALOX12: Arachidonate 12-lipoxygenase, TrioSig: Ferroptosis-related gene signature (Trio signature), FerrDb: Ferroptosis Database, KEGG: Kyoto Encyclopedia of Genes and Genomes, pRIPK1: Phospho-receptor-interacting serine/threonine-protein kinase 1, pRIPK3: phosphorylated-Receptor-interacting protein kinase 3, pMLKL: phosphorylated-mixed lineage kinase domain-like pseudokinase, TRIF: TIR-domain-containing adapter-inducing interferon-\u03b2, ZBP1: Z-DNA-binding protein 1, HMGB1: High-mobility group box 1, DAMP: Damage-associated molecular pattern, GSDMD: Gasdermin D, GSDMD-N: N-terminal fragment of gasdermin D, IL-1\u03b2: Interleukin-1 beta, IL-18: Interleukin-18, CSF: Cerebrospinal fluid, GSDME: Gasdermin E, GSDME-N: N-terminal fragment of gasdermin E, AIF: Apoptosis-inducing factor, MIF: Macrophage migration inhibitory factor, PAR: Poly(ADP-ribose), ARH3: ADP-ribosylhydrolase 3, Ser-ADPr: Serine-ADP-ribosylation, SARM1: Sterile alpha and TIR motif-containing 1, NAD\u207a: Nicotinamide adenine dinucleotide", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "The core mechanisms of necroptosis and its implications in neurodegeneration", "text": "Elevated membrane permeability and cytokine release place necroptosis at the intersection of neuroinflammation and neuronal loss [44]. Studies have guided treatments by reducing both neuronal loss and neuroinflammation.\n\nIn AD, elevated necroptotic signalling has been documented in patient brains and aligns with worsening cognition [48, 49]. Salvadores et al. emphasize in their study that amyloid- \u03b2 peptides can trigger necroptotic pathways through microglial activation, a process that further spreads neuroinflammation and accelerates neuronal deterioration [48]. Experimental blockade of the RIPK1-RIPK3-MLKL axis in AD models limits neuronal death and helps maintain cognitive performance underscoring the pathway\u2019s therapeutic appeal [50, 51]. Comparable observations have emerged in PD and vascular dementia: dopaminergic neurons and glia display necroptotic markers, and pathway inhibition affords measurable neuroprotection [52, 53]. The interplay between necroptosis and neuroinflammation is particularly critical; necroptosis in microglia promotes a chronic inflammatory phenotype [44, 54]. Necroptosis also intersects with oxidative and other cellular stress responses [55]. This suggests that multifaceted treatment strategies targeting necroptosis and associated inflammatory and oxidative cascades may be necessary for optimal benefit [56, 57].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "The core mechanisms of necroptosis and its implications in neurodegeneration", "text": "These insights have sparked interest in small-molecule inhibitors of RIPK1, RIPK3 and MLKL, several of which confer neuroprotection in pre-clinical models and are advancing toward clinical evaluation [58] (Table 3). Beyond therapy, necroptotic markers detected in biofluids or imaging may aid prognosis and treatment monitoring [55, 59]. A fuller understanding of how necroptosis interfaces with oxidative and inflammatory stress should therefore inform both biomarker development and multipronged therapeutic strategies for neurodegenerative disease. This profile predicts inter-pathway interactions detailed in Sect. 5, particularly where oxidative stress signaling converges.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "The core mechanisms of pyroptosis and its implications in neurodegeneration", "text": "Pyroptosis is an inflammatory form of programmed cell death. The term combines pyro (\u201cfire,\u201d reflecting its pro-inflammatory nature) and ptosis (\u201cfalling,\u201d as with other programmed deaths) [28]. It is typically initiated by innate immune stimuli that activate inflammasomes most prominently members of the NOD-like receptor (NLR) family and is associated with the rapid release of IL-1\u03b2 and IL-18. Morphologically, cells often swell, the plasma membrane ruptures, and intracellular contents spill into the extracellular space, which can amplify local inflammation [60, 61]. Two main routes have been described. In the canonical pathway, sensors such as NLRP3 assemble an inflammasome that activates caspase-1. Caspase-1 cleaves gasdermin D (GSDMD) to liberate its N-terminal fragment (GSDMD-N), which oligomerizes in the plasma membrane to form pores and thereby drives pyroptotic death [62]. In the non-canonical pathway, human caspase-4/5 (and murine caspase-11) are directly engaged by intracellular lipopolysaccharide; molecules such as oxPAPC may compete with LPS and blunt this activation. These caspases also cleave GSDMD to generate pore-forming GSDMD-N. While they do not directly mature IL-1\u03b2 or IL-18, the subsequent ionic fluxes (particularly K\u207a efflux) could permissively activate NLRP3 and lead to cytokine release [28].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "The core mechanisms of pyroptosis and its implications in neurodegeneration", "text": "Multiple factors appear to modulate pyroptosis. ROS and other stress signals can promote inflammasome priming or activation [63, 64], while NF-\u03baB signaling frequently serves as a transcriptional \u201cpriming\u201d step for pro-IL-1\u03b2 and inflammasome components. Pathogens, toxins, and cytokine milieus further tune the response, suggesting substantial context dependence across tissues and disease states [65\u201367].\n\nIn the nervous system, pyroptosis seems to play a dual role. As part of host defense it may help contain infection; however, dysregulated or chronic activation can contribute to sterile inflammation and tissue injury [67\u201369]. Emerging literature links inflammasome activity and pyroptotic signaling to neurodegenerative settings for example, amyloid-\u03b2 or tau in AD, dopaminergic vulnerability in PD, and related mechanisms in HD and multiple sclerosis [70\u201374]. In such contexts, sustained production of inflammatory mediators may create conditions that favor synaptic dysfunction, neuronal loss, and circuit impairment [75].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "The core mechanisms of pyroptosis and its implications in neurodegeneration", "text": "These insights have encouraged exploration of pharmacological modulators of inflammasome pyroptosis signaling. Inhibitors that target NLRP3, (e.g., MCC950, OLT1177/dapansutrile) have shown promise in experimental model systems and continue to be evaluated. Representative compounds and their reported effects are summarized in Table\u00a03. This profile also predicts intra-pathway interactions, specifically where inflammatory signaling converges, which are detailed in Sect.\u00a05.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "The core mechanisms of parthanatos and its implications in neurodegeneration", "text": "Parthanatos, derived from \u201cpoly(ADP-ribose) (PAR)\u201d and the Greek word \u201cThanatos\u201d (death), refers to a form of cell death caused by the accumulation of PAR polymers [76]. Parthanatos can be induced by various stressors, including ROS, DNA-damaging agents, nitric oxide (NO)-derived peroxynitrite, DNA alkylating compounds such as N-methyl-N\u2019-nitro-N-nitrosoguanidine (MNNG), ultraviolet radiation, and ionizing irradiation [20]. Calcium overload triggered by glutamate receptor activation, along with ROS produced under oxidative stress, leads to both single- and double-stranded DNA breaks. These lesions result in the excessive activation of Poly(ADP-ribose) polymerase 1 (PARP-1). Poly(ADP-ribose) glycohydrolase (PARG), the main enzyme responsible for degrading poly(ADP-ribose) (PAR) chains, cleaves and releases small PAR oligomers from PARP1 and associated proteins. These PAR fragments subsequently migrate to the cytoplasm, where they interact with mitochondria, promoting the release of a cleaved form of apoptosis-inducing factor (AIF) from mitochondrial membranes. AIF then translocates to the nucleus partly guided by its nuclear localization sequence where it facilitates large-scale DNA fragmentation by activating endonucleases. A downstream bioenergetic crisis often accompanies this pathway, with NAD\u207a depletion, ATP loss, and broader metabolic impairment frequently described as hallmarks [77]. Neurons may be particularly vulnerable given their high energetic demand and limited regenerative capacity, which could help explain why parthanatos-like signatures are discussed in neurodegenerative contexts [78]. In AD, for instance, amyloid-\u03b2 accumulation has been associated with oxidative stress, heightened PARP-1 activity, mitochondrial dysfunction, and indices of NAD\u207a/ATP depletion findings that together suggest a potential parthanatos component and motivate interest in targeting PARP-1 [77, 79].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "The core mechanisms of parthanatos and its implications in neurodegeneration", "text": "Experimental studies increasingly indicate that attenuating this pathway can be neuroprotective in vitro and in vivo models, although the magnitude and generalizability of benefit likely depend on context [80, 81]. Persistent PARP-1 activation under chronic oxidative stress and cumulative DNA damage has been linked to greater neuronal susceptibility, and dysregulated PARP-1 activity is sometimes correlated with disease severity, raising the possibility that it acts both as a marker and a mediator. These observations have encouraged exploration of PARP inhibitors in neurological settings [81\u201383]. Timing, dosing, and brain penetration appear critical variables, and several agents including olaparib and nicotinamide have been evaluated for effects consistent with dampening parthanatos-related injury (see Table 3 for examples in AD and PD models) [81, 84].\n\nLooking ahead, closing the translational gap will likely require finer mapping of cell-type specific vulnerabilities, clarification of context-dependent triggers, and a better understanding of how parthanatos intersects with other regulated death programs such as ferroptosis and necroptosis. Parallel advances in more selective, CNS-penetrant PARP inhibitors and in biomarker strategies that enable real-time pathway monitoring could improve therapeutic precision. If successful, parthanatos modulation may eventually align with multi-target strategies aimed at slowing progressive neuronal loss.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Disease focused summary of non-canonical cell death pathways in major neurodegenerative disorders", "text": "In AD, PD, ALS, and HD, non-canonical cell death converges on shared modules driven by iron-dependent lipid peroxidation in ferroptosis, RIPK1-RIPK3-MLKL signaling in necroptosis, NLRP3-GSDMD activation in pyroptosis, and PARP1-PAR-AIF following oxidative/DNA damage in parthanatos. Disease specific accents include A\u03b2-iron coupling in AD, \u03b1-synuclein lipid/iron interactions in PD, GPX4 dependency with RIPK1 regulated glial pathology in ALS, and striatal iron lipoxygenase programs in HD. These nodes frequently cross-regulate, so targeting one arm can shift others. Concise, disease-resolved summaries are provided in Sect.\u00a03.1\u20133.4, with pathway crosstalk detailed in Sect.\u00a05.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Disease focused summary of non-canonical cell death pathways in major neurodegenerative disorders", "text": "Ferroptosis is mechanistically driven by cortical/hippocampal iron dyshomeostasis, PUFA rich membrane vulnerability, and repression of the SLC7A11/GSH/GPX4 axis, which converge to heighten neuronal lipid-peroxidation stress and are further exacerbated by microglial neuron crosstalk. In AD models, iron modulating strategies, radical trapping antioxidants, and Nrf2/GPX4 supportive interventions blunt ferroptotic lipid ROS and mitigate synaptic injury [85]. Necroptosis proceeds via death-receptor/TLR signaling that activates RIPK1, recruits RIPK3, and phosphorylates MLKL to permeabilize membranes; elevations of phospho-RIPK1-RIPK3-MLKL in AD tissue and pathway blockade reducing neuronal/synaptic injury in models support causality [57]. Pyroptosis is a major driver of neuroinflammation in AD. The process is orchestrated by inflammasomes particularly NLRP1 and NLRP3 and is triggered by stressors such as amyloid-\u03b2 aggregates. A\u03b2 promotes NLRP3 assembly and caspase activation, leading to gasdermin D cleavage and formation of N-terminal pores in the plasma membrane. The ensuing ion flux and membrane rupture release IL-1\u03b2 and IL-18, amplifying local inflammation and contributing to neurodegeneration [73, 86]. The initiation of parthanatos occurs when neurons undergo oxidative stress, often driven by the accumulation of toxic metabolites such as amyloid-\u03b2 plaques a hallmark of AD [87]. These stressors cause DNA damage, leading to hyperactivation of PARP-1. Upon activation, PARP-1 synthesizes PAR polymers which serve as signals for DNA repair under physiological conditions but, when they accumulate excessively, act as triggers for programmed cell death [88, 89].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Disease focused summary of non-canonical cell death pathways in major neurodegenerative disorders", "text": "Ferroptosis in PD stems from substantia nigra iron overload and PUFA-rich membrane vulnerability together with collapse of anti-ferroptotic defenses System Xc\u2212-GSH-GPX4, FSP1/CoQ10 (and mitochondrial DHODH), and GCH1-BH4 and is exacerbated by DJ-1 dysfunction and reduced CoQ10/BH4. Iron accelerates \u03b1-synuclein aggregation and phosphorylation (via IRP/IRE control and kinase modulation), while \u03b1-synuclein reciprocally promotes Fe3+-Fe2+ reduction, lipid peroxidation, and mitochondrial injury through VDAC/mPTP interactions creating a ROS-driven feed-forward loop. Iron chelation, radical-trapping antioxidants, and Nrf2/GPX4-supportive strategies attenuate dopaminergic injury in models [90, 91]. Necroptosis proceeds via death-receptor/TLR signaling that activates RIPK1, recruits RIPK3, and phosphorylates MLKL to permeabilize membranes; elevated RIPK1-RIPK3-p-MLKL and pathway blockade reducing dopaminergic loss in experimental PD support causality [92]. Pyroptosis is fueled by \u03b1-synuclein aggregates that trigger NLRP3 assembly and caspase-1 activation in microglia, leading to GSDMD pore formation and IL-1\u03b2/IL-18 release; pharmacologic NLRP3 inhibition dampens inflammatory injury in PD models [93]. Parthanatos is initiated by oxidative/DNA-damage stress that hyperactivates PARP-1, driving PAR accumulation and AIF translocation; in PD models (MPP\u207a/MPTP; \u03b1-syn PFF), PARP-1 or PAAN/MIF-axis inhibition limits neurodegeneration, indicating a druggable PARP1, AIF death program [94\u201396].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Disease focused summary of non-canonical cell death pathways in major neurodegenerative disorders", "text": "Ferroptosis in HD, striatal vulnerability converges on iron dysregulation and PUFA rich membrane stress with impaired antioxidant defenses; imaging and cohort work indicate basal ganglia iron accumulation from premanifest stages, while mechanistic studies identify ALOX5 as a key driver of mHTT-linked, ACSL4 independent ferroptosis, nominating lipoxygenases and GPX4/SLC7A11 support as actionable nodes [97\u201399]. Necroptosis, HD-like toxin models (3-nitropropionic acid) show robust activation of the RIPK1-RIPK3-MLKL axis in striatum, and pathway interference lessens behavioral and histologic injury supporting a contributory role of necroptosis in HD relevant contexts [100]. Pyroptosis, striatal pyroptotic signaling (NLRP3-caspase-GSDMD) is reported in R6/2 mice, with additional work highlighting NLRP3-centered neuroinflammation as a disease mechanism and candidate target in HD [101]. Parthanatos biomarkers PARP1-PAR-AIF axis appears dysregulated rather than uniformly upregulated in HD. Recent clinical biomarker studies show reduced CSF (cerebrospinal fluid) PAR in mutation carriers, suggesting altered PARylation dynamics despite persistent DNA damage stress, and keeping PARP1-directed strategies under active debate. Mechanistic reviews still place PARP1 hyperactivation and AIF translocation at the core of parthanatos, underscoring disease and context dependence for HD [102, 103].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Disease focused summary of non-canonical cell death pathways in major neurodegenerative disorders", "text": "Ferroptosis in ALS is supported by motor-neuron lipid-peroxidation vulnerability with GPX4 dependence, GPX4 loss or knockdown drives ALS like paralysis, while GPX4 overexpression or anti-ferroptotic interventions suppress lipid ROS and preserve motor neurons in models [104, 105]. Necroptosis features engagement of the RIPK1-RIPK3-MLKL axis and a RIPK1 regulated inflammatory glial state. Genetic or pharmacologic RIPK1 blockade mitigates neuroinflammation and slows disease in SOD1G93A mice, while human ALS tissues show elevated RIPK1 activity/signatures [106]. Pyroptosis related signals (inflammasome-NLRP3/GSDMD) have been detected in ALS spinal cord and glia; multiple studies suggest astrocytic involvement and microglial GSDMD activation, with NLRP3/GSDMD colocalization in ALS models and increased GSDMD in post-mortem tissue. However, GSDMD deletion in SOD1G93A mice yields only modest survival effects and minimal changes in gliosis or motor-neuron loss, indicating that pyroptosis contributes in a context-dependent manner and that upstream inflammasome nodes or alternative gasdermins may represent more relevant therapeutic targets [107\u2013109]. Reviews and experimental data in ALS highlight increased PARP1 activity/AIF signaling and DNA repair stress in motor neurons and glia, nominating the PARP1-PAR-AIF axis as a modifiable death program [89, 103, 110].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Alzheimer\u2019s disease", "text": "Ferroptosis is mechanistically driven by cortical/hippocampal iron dyshomeostasis, PUFA rich membrane vulnerability, and repression of the SLC7A11/GSH/GPX4 axis, which converge to heighten neuronal lipid-peroxidation stress and are further exacerbated by microglial neuron crosstalk. In AD models, iron modulating strategies, radical trapping antioxidants, and Nrf2/GPX4 supportive interventions blunt ferroptotic lipid ROS and mitigate synaptic injury [85]. Necroptosis proceeds via death-receptor/TLR signaling that activates RIPK1, recruits RIPK3, and phosphorylates MLKL to permeabilize membranes; elevations of phospho-RIPK1-RIPK3-MLKL in AD tissue and pathway blockade reducing neuronal/synaptic injury in models support causality [57]. Pyroptosis is a major driver of neuroinflammation in AD. The process is orchestrated by inflammasomes particularly NLRP1 and NLRP3 and is triggered by stressors such as amyloid-\u03b2 aggregates. A\u03b2 promotes NLRP3 assembly and caspase activation, leading to gasdermin D cleavage and formation of N-terminal pores in the plasma membrane. The ensuing ion flux and membrane rupture release IL-1\u03b2 and IL-18, amplifying local inflammation and contributing to neurodegeneration [73, 86]. The initiation of parthanatos occurs when neurons undergo oxidative stress, often driven by the accumulation of toxic metabolites such as amyloid-\u03b2 plaques a hallmark of AD [87]. These stressors cause DNA damage, leading to hyperactivation of PARP-1. Upon activation, PARP-1 synthesizes PAR polymers which serve as signals for DNA repair under physiological conditions but, when they accumulate excessively, act as triggers for programmed cell death [88, 89].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Parkinson\u2019s disease", "text": "Ferroptosis in PD stems from substantia nigra iron overload and PUFA-rich membrane vulnerability together with collapse of anti-ferroptotic defenses System Xc\u2212-GSH-GPX4, FSP1/CoQ10 (and mitochondrial DHODH), and GCH1-BH4 and is exacerbated by DJ-1 dysfunction and reduced CoQ10/BH4. Iron accelerates \u03b1-synuclein aggregation and phosphorylation (via IRP/IRE control and kinase modulation), while \u03b1-synuclein reciprocally promotes Fe3+-Fe2+ reduction, lipid peroxidation, and mitochondrial injury through VDAC/mPTP interactions creating a ROS-driven feed-forward loop. Iron chelation, radical-trapping antioxidants, and Nrf2/GPX4-supportive strategies attenuate dopaminergic injury in models [90, 91]. Necroptosis proceeds via death-receptor/TLR signaling that activates RIPK1, recruits RIPK3, and phosphorylates MLKL to permeabilize membranes; elevated RIPK1-RIPK3-p-MLKL and pathway blockade reducing dopaminergic loss in experimental PD support causality [92]. Pyroptosis is fueled by \u03b1-synuclein aggregates that trigger NLRP3 assembly and caspase-1 activation in microglia, leading to GSDMD pore formation and IL-1\u03b2/IL-18 release; pharmacologic NLRP3 inhibition dampens inflammatory injury in PD models [93]. Parthanatos is initiated by oxidative/DNA-damage stress that hyperactivates PARP-1, driving PAR accumulation and AIF translocation; in PD models (MPP\u207a/MPTP; \u03b1-syn PFF), PARP-1 or PAAN/MIF-axis inhibition limits neurodegeneration, indicating a druggable PARP1, AIF death program [94\u201396].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Huntington\u2019s disease", "text": "Ferroptosis in HD, striatal vulnerability converges on iron dysregulation and PUFA rich membrane stress with impaired antioxidant defenses; imaging and cohort work indicate basal ganglia iron accumulation from premanifest stages, while mechanistic studies identify ALOX5 as a key driver of mHTT-linked, ACSL4 independent ferroptosis, nominating lipoxygenases and GPX4/SLC7A11 support as actionable nodes [97\u201399]. Necroptosis, HD-like toxin models (3-nitropropionic acid) show robust activation of the RIPK1-RIPK3-MLKL axis in striatum, and pathway interference lessens behavioral and histologic injury supporting a contributory role of necroptosis in HD relevant contexts [100]. Pyroptosis, striatal pyroptotic signaling (NLRP3-caspase-GSDMD) is reported in R6/2 mice, with additional work highlighting NLRP3-centered neuroinflammation as a disease mechanism and candidate target in HD [101]. Parthanatos biomarkers PARP1-PAR-AIF axis appears dysregulated rather than uniformly upregulated in HD. Recent clinical biomarker studies show reduced CSF (cerebrospinal fluid) PAR in mutation carriers, suggesting altered PARylation dynamics despite persistent DNA damage stress, and keeping PARP1-directed strategies under active debate. Mechanistic reviews still place PARP1 hyperactivation and AIF translocation at the core of parthanatos, underscoring disease and context dependence for HD [102, 103].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Amyotrophic lateral sclerosis", "text": "Ferroptosis in ALS is supported by motor-neuron lipid-peroxidation vulnerability with GPX4 dependence, GPX4 loss or knockdown drives ALS like paralysis, while GPX4 overexpression or anti-ferroptotic interventions suppress lipid ROS and preserve motor neurons in models [104, 105]. Necroptosis features engagement of the RIPK1-RIPK3-MLKL axis and a RIPK1 regulated inflammatory glial state. Genetic or pharmacologic RIPK1 blockade mitigates neuroinflammation and slows disease in SOD1G93A mice, while human ALS tissues show elevated RIPK1 activity/signatures [106]. Pyroptosis related signals (inflammasome-NLRP3/GSDMD) have been detected in ALS spinal cord and glia; multiple studies suggest astrocytic involvement and microglial GSDMD activation, with NLRP3/GSDMD colocalization in ALS models and increased GSDMD in post-mortem tissue. However, GSDMD deletion in SOD1G93A mice yields only modest survival effects and minimal changes in gliosis or motor-neuron loss, indicating that pyroptosis contributes in a context-dependent manner and that upstream inflammasome nodes or alternative gasdermins may represent more relevant therapeutic targets [107\u2013109]. Reviews and experimental data in ALS highlight increased PARP1 activity/AIF signaling and DNA repair stress in motor neurons and glia, nominating the PARP1-PAR-AIF axis as a modifiable death program [89, 103, 110].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Comprehensive biomarkers of non-canonical cell death pathways in neurodegeneration", "text": "Pathway-specific biomarkers in biofluids or tissue can enable earlier diagnosis, sharper disease monitoring, and mechanism-guided therapy in neurodegeneration. Integrating molecular, metabolic, and transcriptomic readouts provides a multidimensional view of regulated cell death. Recent discoveries highlight distinct biomarker signatures for each pathway, offering tools for early intervention and personalized treatment. Table\u00a02 summarizes the major non-canonical cell death biomarkers in neurodegeneration and their mechanistic roles.\n\nA broad spectrum of ferroptosis-related biomarkers has been identified, reflecting different mechanistic layers of this iron-dependent form of regulated cell death. As summarized in Table\u00a01, these biomarkers span from classical lipid peroxidation products to molecular regulators.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Comprehensive biomarkers of non-canonical cell death pathways in neurodegeneration", "text": "Ferroptosis-related biomarkers provide valuable insight into the mechanisms of neurodegenerative diseases. Classic markers of lipid peroxidation, such as malondialdehyde (MDA) and 4-hydroxynonenal (4-HNE), are widely used indicators of oxidative stress and ferroptosis. Elevated MDA levels are linked to PD and AD, while 4-HNE accumulation is associated with synaptic and mitochondrial dysfunction, making both valuable markers for early diagnosis and disease monitoring [111\u2013113]. Other lipid markers, such as F\u2082-isoprostanes well-established indicators of in vivo lipid peroxidation and oxidized phosphatidylethanolamine (PE) species, particularly PUFA-containing PE-OOH, are closely associated with ferroptosis due to their mechanistic link to iron-dependent lipid peroxidation. These compounds, produced through ACSL4 and lipoxygenase activity, highlight specific lipid peroxidation pathways that drive cell death and could serve as precise markers in clinical trials [114]. Neuroimaging tools such as quantitative susceptibility mapping (QSM) and T2 MRI show increased iron deposits in regions like the hippocampus, cortex, and substantia nigra in AD, PD, and HD. These findings match pathological data and support the idea that excess iron accelerates ferroptosis and neuronal loss [115\u2013117]. From a molecular perspective, iron transporter proteins such as ferroportin are key factors regulating ferroptosis. Decreased ferroportin levels predict progression of AD, while ferroportin dysfunction leads to harmful iron accumulation in [118]. Antioxidant defenses also matter. GPX4 is a crucial enzyme that protects neurons from lipid peroxides. Loss of GPX4 causes rapid neurodegeneration [119]. Finally, transcriptomic studies have identified panels of ferroptosis-related genes (e.g., ACSL4, SAT1, ALOX12) that serve as diagnostic and prognostic markers in AD and ALS. These gene signatures reflect the molecular diversity of ferroptosis and may help in patient classification and therapy monitoring. Transcriptomic studies have identified ferroptosis-related gene signatures, including TrioSig and panels from FerrDb/KEGG, enriched in AD, while single-cell analyses show SAT1 and FTH1 upregulation in astrocytic ferroptosis [120, 121]. These gene signatures reflect the molecular diversity of ferroptosis and may help in patient classification and therapy monitoring.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Comprehensive biomarkers of non-canonical cell death pathways in neurodegeneration", "text": "In necroptosis, pRIPK1, pRIPK3, and pMLKL, in particular, mark sequential steps of death signaling, and their phosphorylated status has been confirmed in AD, multiple sclerosis, and ALS brain tissues [18]. The necrosome complex, found in granulovacuolar inclusions, further highlights necroptosis in AD and HD [122]. Non-canonical necroptosis proceeds through alternative pathways that bypass RIPK1 but center on the RIPK3-MLKL axis. In particular, TLR3/4-TRIF-mediated signaling and the activation of ZBP1 (Z-DNA binding protein 1) directly trigger RIPK3, thereby initiating necroptotic cell death. Therefore, TRIF and ZBP1 expression profiles are suggested as specific indicators of non-canonical necroptosis [123]. TRIF and ZBP1 expression profiles have been proposed as specific indicators of non-canonical necroptosis. In viral infection models, upregulation of ZBP1 has been confirmed, while in autoimmune and neuroinflammatory diseases, tissue-level activation of the TLR3/4-TRIF-RIPK3 axis has been validated [44]. pRIPK3 and pMLKL are important biomarkers, while RIPK1 phosphorylation remains absent or at low levels, providing a distinguishing clue from the canonical pathway. Therefore, the combination of \u2018pRIPK3 and pMLKL positivity, absence of pRIPK1, and upregulation of TRIF/ZBP1 constitutes a distinctive biomark [124]. HMGB1, a typical damage-associated molecular pattern (DAMP), provides a measurable fluid biomarker reflecting tissue damage and systemic inflammation [125].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Comprehensive biomarkers of non-canonical cell death pathways in neurodegeneration", "text": "Integrating these biomarker signatures into personalized medicine may enable the stratification of patients with neurodegenerative disorders in which non-canonical necroptosis predominates, thereby guiding prognosis and the development of targeted RIPK3/MLKL-based therapies.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Comprehensive biomarkers of non-canonical cell death pathways in neurodegeneration", "text": "Cleaved GSDMD (GSDMD-N) and activated caspase-4/5 in humans and caspase-11 in mice represent central indicators of neuronal pyroptosis triggered by cytosolic danger signals. While direct evidence from human AD or PD brain tissue is still emerging, foundational mechanistic work confirms that caspase-4/5/11 cleave GSDMD to GSDMD-N in non-canonical pyroptosis [126]. In AD, microglial caspase-4 is upregulated via DNA hypomethylation, linking this pathway to pathology [127, 128]. In PD, GSDMD contributes to dopaminergic neurodegeneration and glial activation [68]. Elevated cerebrospinal fluid levels of IL-1\u03b2 and IL-18 provide fluid-based evidence of this pathway [129]. In addition, caspase-3 mediated cleavage of GSDME (GSDME-N) links apoptosis-to-pyroptosis switching to synaptic dysfunction in AD [130], while caspase-8 mediated GSDMD cleavage has been demonstrated as an alternative pyroptotic route under inflammatory stress [131]. Together, these biomarkers reflect distinct but converging pathways of non-canonical pyroptosis in the brain. Importantly, their integration into personalized medicine frameworks may enable stratification of patients according to dominant pyroptotic mechanisms, facilitate early diagnosis through cerebrospinal fluid or imaging-based assays, and support tailored interventions aimed at selectively blocking caspase-4/5, GSDMD, or GSDME activation. Such biomarker-guided strategies hold potential to refine prognosis and optimize targeted therapies in AD, PD and related neurodegenerative diseases.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Comprehensive biomarkers of non-canonical cell death pathways in neurodegeneration", "text": "In neurodegenerative diseases, non-canonical branches of parthanatos define biomarker signatures with direct implications for personalized medicine. Nuclear AIF-MIF translocation, PAR-\u03b1-synuclein complexes, altered CSF PAR levels, ARH3 deficiency with serine-ADP-ribosylation, and SARM1-driven NAD\u207a depletion each represent distinct molecular readouts of parthanatos activity [76, 132\u2013134]. Importantly, these biomarkers may be instrumental in patient stratification. Individuals showing AIF-MIF positivity may benefit from MIF or PARP1 inhibitors, high PAR or PAR-\u03b1-synuclein signatures may indicate suitability for PARP-targeted therapies, and NAD\u207a/ARH3 imbalances may indicate SARM1 inhibition or NAD\u207a supplementation. Consequently, by linking biomarker profiles to treatment choices, it provides a framework for personalizing interventions, improving prognosis, and advancing precision medicine in diseases related to neurodegeneration.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Ferroptosis biomarkers", "text": "A broad spectrum of ferroptosis-related biomarkers has been identified, reflecting different mechanistic layers of this iron-dependent form of regulated cell death. As summarized in Table\u00a01, these biomarkers span from classical lipid peroxidation products to molecular regulators.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Ferroptosis biomarkers", "text": "Ferroptosis-related biomarkers provide valuable insight into the mechanisms of neurodegenerative diseases. Classic markers of lipid peroxidation, such as malondialdehyde (MDA) and 4-hydroxynonenal (4-HNE), are widely used indicators of oxidative stress and ferroptosis. Elevated MDA levels are linked to PD and AD, while 4-HNE accumulation is associated with synaptic and mitochondrial dysfunction, making both valuable markers for early diagnosis and disease monitoring [111\u2013113]. Other lipid markers, such as F\u2082-isoprostanes well-established indicators of in vivo lipid peroxidation and oxidized phosphatidylethanolamine (PE) species, particularly PUFA-containing PE-OOH, are closely associated with ferroptosis due to their mechanistic link to iron-dependent lipid peroxidation. These compounds, produced through ACSL4 and lipoxygenase activity, highlight specific lipid peroxidation pathways that drive cell death and could serve as precise markers in clinical trials [114]. Neuroimaging tools such as quantitative susceptibility mapping (QSM) and T2 MRI show increased iron deposits in regions like the hippocampus, cortex, and substantia nigra in AD, PD, and HD. These findings match pathological data and support the idea that excess iron accelerates ferroptosis and neuronal loss [115\u2013117]. From a molecular perspective, iron transporter proteins such as ferroportin are key factors regulating ferroptosis. Decreased ferroportin levels predict progression of AD, while ferroportin dysfunction leads to harmful iron accumulation in [118]. Antioxidant defenses also matter. GPX4 is a crucial enzyme that protects neurons from lipid peroxides. Loss of GPX4 causes rapid neurodegeneration [119]. Finally, transcriptomic studies have identified panels of ferroptosis-related genes (e.g., ACSL4, SAT1, ALOX12) that serve as diagnostic and prognostic markers in AD and ALS. These gene signatures reflect the molecular diversity of ferroptosis and may help in patient classification and therapy monitoring. Transcriptomic studies have identified ferroptosis-related gene signatures, including TrioSig and panels from FerrDb/KEGG, enriched in AD, while single-cell analyses show SAT1 and FTH1 upregulation in astrocytic ferroptosis [120, 121]. These gene signatures reflect the molecular diversity of ferroptosis and may help in patient classification and therapy monitoring.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Necroptosis biomarkers", "text": "In necroptosis, pRIPK1, pRIPK3, and pMLKL, in particular, mark sequential steps of death signaling, and their phosphorylated status has been confirmed in AD, multiple sclerosis, and ALS brain tissues [18]. The necrosome complex, found in granulovacuolar inclusions, further highlights necroptosis in AD and HD [122]. Non-canonical necroptosis proceeds through alternative pathways that bypass RIPK1 but center on the RIPK3-MLKL axis. In particular, TLR3/4-TRIF-mediated signaling and the activation of ZBP1 (Z-DNA binding protein 1) directly trigger RIPK3, thereby initiating necroptotic cell death. Therefore, TRIF and ZBP1 expression profiles are suggested as specific indicators of non-canonical necroptosis [123]. TRIF and ZBP1 expression profiles have been proposed as specific indicators of non-canonical necroptosis. In viral infection models, upregulation of ZBP1 has been confirmed, while in autoimmune and neuroinflammatory diseases, tissue-level activation of the TLR3/4-TRIF-RIPK3 axis has been validated [44]. pRIPK3 and pMLKL are important biomarkers, while RIPK1 phosphorylation remains absent or at low levels, providing a distinguishing clue from the canonical pathway. Therefore, the combination of \u2018pRIPK3 and pMLKL positivity, absence of pRIPK1, and upregulation of TRIF/ZBP1 constitutes a distinctive biomark [124]. HMGB1, a typical damage-associated molecular pattern (DAMP), provides a measurable fluid biomarker reflecting tissue damage and systemic inflammation [125].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Necroptosis biomarkers", "text": "Integrating these biomarker signatures into personalized medicine may enable the stratification of patients with neurodegenerative disorders in which non-canonical necroptosis predominates, thereby guiding prognosis and the development of targeted RIPK3/MLKL-based therapies.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Pyroptosis biomarkers", "text": "Cleaved GSDMD (GSDMD-N) and activated caspase-4/5 in humans and caspase-11 in mice represent central indicators of neuronal pyroptosis triggered by cytosolic danger signals. While direct evidence from human AD or PD brain tissue is still emerging, foundational mechanistic work confirms that caspase-4/5/11 cleave GSDMD to GSDMD-N in non-canonical pyroptosis [126]. In AD, microglial caspase-4 is upregulated via DNA hypomethylation, linking this pathway to pathology [127, 128]. In PD, GSDMD contributes to dopaminergic neurodegeneration and glial activation [68]. Elevated cerebrospinal fluid levels of IL-1\u03b2 and IL-18 provide fluid-based evidence of this pathway [129]. In addition, caspase-3 mediated cleavage of GSDME (GSDME-N) links apoptosis-to-pyroptosis switching to synaptic dysfunction in AD [130], while caspase-8 mediated GSDMD cleavage has been demonstrated as an alternative pyroptotic route under inflammatory stress [131]. Together, these biomarkers reflect distinct but converging pathways of non-canonical pyroptosis in the brain. Importantly, their integration into personalized medicine frameworks may enable stratification of patients according to dominant pyroptotic mechanisms, facilitate early diagnosis through cerebrospinal fluid or imaging-based assays, and support tailored interventions aimed at selectively blocking caspase-4/5, GSDMD, or GSDME activation. Such biomarker-guided strategies hold potential to refine prognosis and optimize targeted therapies in AD, PD and related neurodegenerative diseases.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Parthanatos biomarkers", "text": "In neurodegenerative diseases, non-canonical branches of parthanatos define biomarker signatures with direct implications for personalized medicine. Nuclear AIF-MIF translocation, PAR-\u03b1-synuclein complexes, altered CSF PAR levels, ARH3 deficiency with serine-ADP-ribosylation, and SARM1-driven NAD\u207a depletion each represent distinct molecular readouts of parthanatos activity [76, 132\u2013134]. Importantly, these biomarkers may be instrumental in patient stratification. Individuals showing AIF-MIF positivity may benefit from MIF or PARP1 inhibitors, high PAR or PAR-\u03b1-synuclein signatures may indicate suitability for PARP-targeted therapies, and NAD\u207a/ARH3 imbalances may indicate SARM1 inhibition or NAD\u207a supplementation. Consequently, by linking biomarker profiles to treatment choices, it provides a framework for personalizing interventions, improving prognosis, and advancing precision medicine in diseases related to neurodegeneration.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Untitled", "text": "The non-canonical forms of programmed cell death ferroptosis, necroptosis, pyroptosis, and parthanatos exhibit distinct molecular mechanisms; however, an increasing number of studies suggest that these pathways may dynamically interact with one another. This crosstalk plays a significant role in the pathophysiology of various diseases, particularly neurodegenerative disorders [135, 136].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Untitled", "text": "Ferroptosis and necroptosis interactions seem to coalesce at ROS and iron management. Ferroptosis induces iron-dependent lipid peroxidation and membrane damage, whereas necroptosis proceeds via RIPK1-RIPK3-MLKL signaling. Iron overload can increase ROS and sensitize cells to both processes, and lipid peroxidation products formed during ferroptosis may also prime cells for activation of the RIPK1-RIPK3-MLKL axis. Necroptotic damage may also release labile iron, which could feedback to promote ferroptotic sensitization [137, 138]. Pharmacological data (e.g., necrostatin-1 cross-effects) further illustrate this axis, although off-target actions complicate interpretation Notably, neuronal death after hemorrhagic stroke engages both ferroptosis and necroptosis, with inhibition of either pathway conferring more than 80% neuroprotection [139]. During early reperfusion after ischemic stroke, treatment with the ferroptosis inhibitor liproxstatin-1 suppressed necroptotic signaling, while the necroptosis inhibitor necrostatin-1 reduced ferroptosis-related protein alterations. Similarly, the iron chelator deferoxamine mesylate inhibited both processes. Together, these findings underscore a therapeutically exploitable interplay that may be addressed through combined strategies, such as pairing lipid peroxidation inhibitors (ferrostatin-1 or iron chelators) with RIPK1/RIPK3 blockade. Still, pharmacological specificity is essential, as RIPK3 expression varies across cell types and may influence necroptotic responses [137].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Untitled", "text": "Links to pyroptosis have likewise been proposed. Inflammation and oxidative stress can accompany ferroptosis, and lipid peroxidation together with glutathione depletion may favor inflammasome activation and GSDMD-mediated pyroptosis [140]. In traumatic brain injury, PAR1 signaling emerges as a shared upstream node linking ferroptotic lipid-peroxidation/iron-redox stress with NLRP3 inflammasome mediated pyroptosis. PAR1 antagonism (SCH79797) concurrently reduces ferroptosis (via PPAR-\u03b3/Nrf2-HO-1, preserving GPX4/SOD) and suppresses NLRP3-ASC-caspase-1 and downstream cytokines, highlighting PAR1 as a tractable lever to co-modulate both pathways [141].\n\nFerroptosis-related oxidative stress can also produce DNA damage, which may hyperactivate PARP1 and engage parthanatos [142, 143]. This DNA-damage link highlights how ferroptotic processes may spill over into additional death programs under conditions of severe redox imbalance.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Untitled", "text": "Additional connections are noted between necroptosis and pyroptosis: activated MLKL disrupts the plasma membrane and promotes K\u207a efflux, a well-recognized permissive signal for NLRP3 inflammasome activation and subsequent pyroptotic responses [144]. This MLKL-K\u207a-NLRP3 relay has been demonstrated across multiple systems and refined in recent structural/functional syntheses [145, 146]. In the pyroptosis arm, GSDMD pores drive rapid ionic dysregulation most notably K\u207a efflux, which further licenses NLRP3 activation; downstream, the membrane-rupture protein NINJ1 oligomerizes to execute terminal plasma-membrane rupture across lytic deaths (pyroptosis, necroptosis), thereby amplifying DAMP release and inflammation [147, 148]. Consistently, in an OGD cell model, the necroptosis inhibitor ZJU-37 (RIP1/RIP3 dual inhibitor) suppressed NLRP3 inflammasome activity [149].\n\nFinally, ROS generated during necroptosis or pyroptosis can exacerbate DNA damage and PARP1 activation, linking these processes to parthanatos-like injury via NAD\u207a/ATP depletion and AIF translocation [150, 151]. Supporting this overlap, inflammation-associated ROS in pyroptosis may also engage PARP1-dependent cell death [143, 151]. In a HD mouse model, PARP-1 inhibition was shown to reduce caspase-1 dependent pyroptosis, lowering NLRP3 expression and influencing neuronal morphology [152].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Untitled", "text": "Taken together, these multi-directional interactions highlight therapeutically targetable nodes across death pathways. While each pathway offers potential intervention points, overlapping mechanisms also pose challenges, including pharmacological specificity and context-dependent signaling. Overall, regulated death programs do not function in isolation but rather form a dynamic network that shapes disease progression (Fig.\u00a03). Understanding this crosstalk may open new therapeutic avenues, particularly in conditions where multiple cell death mechanisms are simultaneously activated.\n\nFig. 3Interaction network among non-canonical cell death pathways\n\nInteraction network among non-canonical cell death pathways\n\nThis schematic diagram illustrates the molecular interactions of ferroptosis, necroptosis, pyroptosis, and parthanatos, along with their connections to shared pathological processes. Solid lines represent direct molecular interactions and biochemical relationships. Dashed lines indicate indirect associations or the influence of one pathway on another. Created in https://BioRender.com/1bv771o.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Emerging therapeutic strategies and pharmacological agents for non-canonical cell death mechanisms", "text": "Pharmacological suppression of ferroptosis has emerged as a potential strategy for neuroprotection, supported by preclinical studies that identify small-molecule inhibitors targeting key nodes in ferroptotic signaling [153, 154]. In neuronal cultures, such agents appear to curb lipid peroxidation and help preserve mitochondrial function, in part by limiting free-radical propagation and interrupting the self-amplifying loop of lipid damage; together, these effects may reduce ferroptosis-linked cell death [155\u2013157]. These properties suggest possible therapeutic value in complex brain disorders where oxidative stress and inflammation frequently coexist.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Emerging therapeutic strategies and pharmacological agents for non-canonical cell death mechanisms", "text": "The complex biochemistry underlying ferroptosis provides multiple potential targets for drug discovery, including enzymes and ion channels involved in iron transport and lipid oxidation. Research indicates that iron chelators and lipoxygenase inhibitors can significantly mitigate ferroptotic cascades by reducing free iron availability and limiting enzymatic lipid peroxidation [158]. Agents that support GPX4 activity or otherwise bolster antioxidant defenses have also been explored [158]. Because mitochondrial dysfunction features prominently in ferroptosis and often tracks with neurodegenerative pathology, it represents an attractive therapeutic focus [159]. Across models, mitochondrial injury coincides with elevated ROS, further fueling lipid peroxidation and death [160, 161]. Inhibitors that blunt ferroptosis have been reported to improve mitochondrial respiration and lower ROS generation, with accompanying signals of better energy homeostasis [162]. and associations with enhanced cell survival and reduced degenerative markers [163]. Thus, targeting mitochondrial impairment via ferroptosis-directed agents may help sustain neuronal viability.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Emerging therapeutic strategies and pharmacological agents for non-canonical cell death mechanisms", "text": "Preclinical work also notes that ferroptosis antagonists can lower lipid-peroxidation readouts in neural tissues [153], while small molecules that restore glutathione reinforce endogenous brakes on ferroptotic progression [164]. These synergistic effects suggest that a combined approach of ferroptosis inhibitors and redox modulators may provide robust neuroprotection.\n\nPharmacological approaches aimed at necroptosis have shown encouraging signals in preclinical neurodegeneration models. Small molecules such as necrostatin-1 have been widely studied for interference with RIPK1 signaling, thereby limiting downstream RIPK3 and MLKL activation [11, 114]. Agents such as dabrafenib and ponatinib have been reported to reduce tau pathology and amyloid burden while preserving synaptic function in AD models, alongside attenuation of neuroinflammation and maintenance of neuronal integrity [51, 59]. Even so, concerns regarding target specificity and off-target liabilities persist, underscoring the need for further optimization of dosing and selectivity [50, 165].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Emerging therapeutic strategies and pharmacological agents for non-canonical cell death mechanisms", "text": "Beyond direct cytoprotection, necroptosis inhibition may also temper inflammation, as membrane rupture can release damage-associated molecular patterns (DAMPs) that amplify neuroinflammatory cascades [50, 59, 166]. Consistent with this, RIPK1 inhibitors have been shown to reduce neuronal loss together with pro-inflammatory cytokine production in experimental systems, supporting combined strategies that address both cell-death signaling and inflammation [167, 168].\n\nA substantial body of evidence indicates that modulation of pyroptotic pathways holds significant potential for mitigating neurodegenerative pathologies. Pharmacological inhibition of caspase activation, particularly targeting caspase-4/5/11, has been demonstrated to attenuate pyroptosis-associated neurotoxicity and ameliorate neuropathological features in various preclinical models [169, 170]. A number of more recent initiatives included both natural products and synthetic products, including alkaloids from plant producers that favor antioxidative/anti-inflammatory pathways that may still serve to indirectly inhibit pyroptotic signaling but also products that target inflammasome-related pathways and upstream players [170, 171].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Emerging therapeutic strategies and pharmacological agents for non-canonical cell death mechanisms", "text": "The link between pyroptosis and neuroinflammation appears especially tight in glial circuits. Microglia and astrocytes are essential for CNS homeostasis, yet their dysregulated activation during neurodegeneration can sustain inflammatory cascades and worsen neuronal stress. Accordingly, strategies that jointly modulate glial activation states and inhibit pyroptotic signaling could offer dual benefits for neuroprotection [172]. In practical terms, careful tuning of microglial phenotypes together with targeted suppression of pyroptosis may help preserve neuronal integrity in disorders such as AD and PD, though the optimal combinations remain to be defined [73, 173].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Emerging therapeutic strategies and pharmacological agents for non-canonical cell death mechanisms", "text": "Given PARP1\u2019s central role in parthanatos, direct PARP1 inhibition has attracted interest as a complementary approach. Because parthanatos has been implicated in neuronal loss across several neurodegenerative contexts, targeting PARP1 could limit downstream injury; its involvement in NAD\u207a/ATP depletion, DNA-damage responses, and AIF translocation further supports its relevance [80, 174]. At the same time, PARP1 is integral to DNA repair and cellular homeostasis, raising concerns about adverse effects with prolonged inhibition [174]. This caveat underscores the need for selective, context-aware inhibitors that balance neuroprotection with preservation of essential cellular functions, an aim highlighted in recent work [80, 174]. Representative therapeutic strategies directed at ferroptosis, necroptosis, pyroptosis, and parthanatos are summarized in Table 3, which collates findings from diverse preclinical neurodegeneration models.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Emerging therapeutic strategies and pharmacological agents for non-canonical cell death mechanisms", "text": "Rigorous pharmacodynamic evaluation of therapeutic agents is central to clinical translation, because it verifies on-target pathway engagement, links exposure to effect, and informs dose selection and patient stratification. Integrated in silico modeling and high-throughput screening prioritize chemotypes by predicted binding and pharmacokinetics; the same pharmacodynamic benchmarks in cell systems then verify on-pathway activity and help set exposure targets ferroptosis: lower C11-BODIPY/LiperFluo oxidation and 4-HNE/MDA adducts with support of GPX4/system Xc\u2212 and preserved mitochondrial respiration/ATP [156, 157, 175]; necroptosis: decreases in p-RIPK1/p-MLKL with DAMP/cytokine panels [58]; pyroptosis: attenuation of GSDMD-N and IL-1\u03b2/IL-18 [28]; parthanatos: reduced PAR accumulation, less AIF nuclear translocation, and partial recovery of NAD\u207a/ATP [80, 174]. Clinically, relevance may be greatest in phenotypes with combined oxidative-inflammatory burden and multimodal cell-death engagement, where these panels can document on-target engagement and inform early-phase dosing. BBB penetration remains a practical bottleneck; approaches under evaluation include CNS-activated prodrugs, efflux-bypassing chemotypes, and nano/EV-based carriers to raise brain exposure while limiting off-target effects [176, 177]. Key constraints to acknowledge include, for ferroptosis, variable inducer specificity and adaptive resistance (e.g., SLC7A11/GPX4/ACSL4 dependence, transsulfuration/MT1G) alongside off-target antioxidant/chelator liabilities [178]; for necroptosis, species-dependent differences across the RIPK1-RIPK3-MLKL axis and the practical challenge of directly drugging the pseudokinase MLKL (with risks of off-target or even paradoxical effects), together with the pathway\u2019s context-dependent immunomodulatory roles [179, 180]; for pyroptosis, selectivity host-defense trade-offs with NLRP3 blockade (dosing must avoid broad IL-1\u03b2/IL-18 suppression), translational gaps due to human caspase-4/5 vs. murine caspase-11 differences, and infection surveillance concerns when inflammasome/caspase signaling is broadly inhibited and for NLRP3 lingering safety concerns around covalent inhibitors (irreversibility/off-target reactivity) [181\u2013183]; and for parthanatos, chronic PARP1 blockade can disrupt DNA repair (e.g., PARP-trapping), cause off-target effects on other PARPs, and be constrained by CNS exposure favoring selective, CNS-penetrant agents and careful dosing [22, 79]. Given the likelihood of long-term treatment in neurodegeneration, pharmacodynamic confirmation should be paired with longitudinal safety/tolerability monitoring and, where feasible, EV-based biomarkers, to guide progression to later-phase trials [184, 185].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Emerging therapeutic strategies and pharmacological agents for non-canonical cell death mechanisms", "text": "Table 3Summary of therapeutic compounds in preclinical models of neurodegenerative diseasesCell death mechanismsRelated diseaseStudy modelCompound, pharmacological agent, drugTherapeutic targetReferencesFerroptosisADIn vitroEdaravoneTLR4, lipid peroxidation[186]FerroptosisADIn vitroSenegeninGPX4, ACSL4[187]FerroptosisADIn vivoGinkgolide BNrf2/GPX4[188]FerroptosisADIn vivoTetrahydroxy stilbene glycosideNrf2/GPX4[189]FerroptosisADIn vivoDeferoxamineGSH, MDA, ROS[82]FerroptosisADIn vivo\u03b3-glutamylcysteineGSH, GPX4[190]FerroptosisADIn vivo\u03b1-Lipoic acidGPx4/iron[191]FerroptosisADIn vivoGW7647GPX4, iron[192]FerroptosisADIn vivo1,6-O, O-diacetylbritannilactoneGSH, MDA[193]FerroptosisADIn vitro, in vivoInsamgobonhwan (contents: Liriope platyphylla, Asparagus cochinchinensis, Rehmania radix preparata, Ginseng radix)GPX4/HO-1/COX-2[194]FerroptosisADIn vitro, in vivoForsythoside ANrf2/GPX4[195]FerroptosisADIn vitro, in vivoSalidrosideNrf2/HO1/GPX4/ ACSL4, iron[196, 197]FerroptosisADIn vitro, in vivoForsythoside ANrf2, GPX4, iron[195]FerroptosisADIn vitro, in vivoEriodictyolGPX4, Nrf2, HO-1[198]FerroptosisPDIn vitroIdebenoneGPX4, Lipid peroxidation[199]FerroptosisPDIn vitro\u03b1-Lipoic acidPI3K/Akt/Nrf2, GPX4, SLC7A11[112]FerroptosisPDIn vitroDeferoksaminGPX4, FTH1, DMT1, TfR1[200]FerroptosisPDIn vitroPaeoniflorinAkt/Nrf2/Gpx4[201]FerroptosisPDIn vitroBafilomycin A1FTH1, GPX4[202]FerroptosisHDIn vitroFerrostatin-1/ DeferoksaminNOX2, iron-dependent lipid peroxidation[203]FerroptosisALSIn vitro, in vivoRTA-408Nrf2/Gpx4, SLC7A11[204]NecroptosisADIn vitroNecrostatin-1RIP1[205]NecroptosisADIn vivoNecrosulfonamideMLKL[206]NecroptosisADIn vivoPazopanibRIPK1, RIPK3, MLKL[207]NecroptosisADIn vivoDabrafenib - PonatinibpRIPK1, pRIPK3, pMLKL[51]NecroptosisPDIn vivoNecrosulfonamideMLKL[208]NecroptosisPDIn vitro, in vivoNecrostatin-1RIP1, RIP3, MLKL[209, 210]PyroptosisADIn vitro, in vivoMCC950NLRP3[211, 212]PyroptosisPDIn vitro, in vivoMCC950NLRP3[213\u2013217]PyroptosisPDIn vitro, in vivoOLT1177NLRP3[218]ParthanatosADIn vivoOlaparib and MC2050PARP-1[219]ParthanatosHDIn vivoOlaparibPARP-1[220]", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Emerging therapeutic strategies and pharmacological agents for non-canonical cell death mechanisms", "text": "Summary of therapeutic compounds in preclinical models of neurodegenerative diseases\n\nAD: Alzheimer\u2019s disease, PD: Parkinson\u2019s disease, ALS: amyotrophic lateral sclerosis, HD: Huntington\u2019s disease, TLR4: Toll-like receptor 4, GPX4: Glutathione peroxidase 4, ACSL4: Acyl-CoA synthetase long-chain family member 4, Nrf2: Nuclear factor erythroid 2\u2013related factor 2, HO-1: Heme oxygenase 1, COX-2: Cyclooxygenase-2, GSH: Glutathione, MDA: Malondialdehyde, ROS: Reactive oxygen species, SLC7A11: Solute carrier family 7 member 11 (xCT), FTH1: Ferritin heavy chain 1, DMT1: Divalent metal transporter 1, TfR1: Transferrin receptor 1, AKT: Protein kinase B (PKB), NOX2: NADPH oxidase 2, RTA-408: Synthetic triterpenoid, Nrf2 activator, RIP1 (RIPK1): Receptor-interacting protein kinase 1, RIP3 (RIPK3): Receptor-interacting protein kinase 3, MLKL: Mixed lineage kinase domain-like pseudokinase, pRIPK1 / pRIPK3 / pMLKL: Phosphorylated forms of RIPK1, RIPK3, MLKL, NLRP3: NOD-like receptor family pyrin domain containing 3.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Integrative framework and future perspectives", "text": "Building on the pathway-level pharmacodynamic anchors and the crosstalk nodes outlined above, this framework links shared drivers (oxidative stress, lipid peroxidation, iron imbalance, mitochondrial dysfunction, ion flux, DNA damage) to cell-type programs and clinical phenotypes We propose a multi-scale framework that integrates ferroptosis, necroptosis, pyroptosis, and parthanatos into a single, action-oriented network; this approach is consistent with the inter-pathway interactions identified in recent reviews [6, 221, 222]. Oxidative stress/lipid peroxidation, iron imbalance, mitochondrial dysfunction, ion flux disorders and DNA damage/PARP1 hyperactivation stand out as common drivers of this network [178, 223]. At the level of execution modules, ferroptosis is characterized by the GPX4/SLC7A11/ACSL4 axes, necroptosis by the RIPK1-RIPK3-MLKL axes, pyroptosis by the NLRP3-caspase-4/5/11-GSDMD axes, and parthanatos by the PARP1-AIF axes [20]. A critical example of crosstalk is that MLKL-dependent K\u207a efflux facilitates the pyroptotic response by lowering the NLRP3 threshold [146, 224]. Furthermore, lipid peroxidation/mitochondrial ROS can potentiate necroptosis, highlighting the decisive role of lipid oxidation in necroptosis regulation [30]. In parallel, excessive PARP1 activation deepens mitochondrial failure and creates conditions that may increase ferroptotic sensitivity [225, 226]. Neurons are particularly prone to ferroptosis due to their iron-rich lipidome and energy fragility [227]. In microglia, the NLRP3 inflammasome distinctly shapes pyroptotic threshold and neuroinflammation [172]. Astrocytes contribute to the maintenance of GPX4 activity by regulating glutathione/selenium metabolism, thereby gatering ferroptosis susceptibility [228, 229]. Triggers such as amyloid or \u03b1-synuclein accumulation, excitotoxicity, and environmental toxins engage this network through different entry points, leading to convergent outcomes such as synaptic loss, axonal degeneration, and maladaptive neuroinflammation [77, 230\u2013233].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Integrative framework and future perspectives", "text": "For translational roadmap efforts, we recommend a composite biomarker approach that maps a panel of pathway-proximal reads to EV-based signatures [234]. Recent studies that activation of non-canonical regulated cell death programs generates pathway-proximal markers that are actively packaged into extracellular vesicles (EVs) via endo-lysosomal remodeling [234\u2013236]. These EV-encapsulated markers can cross the blood-brain barrier and provide an accessible window into neurodegenerative processes measurable in plasma or cerebrospinal fluid. Proof-of-concept studies have documented p-MLKL-positive EVs released during necroptosis, GSDMD-N bearing EVs arising from pyroptotic cells, and 4-HNE\u2013modified proteins within circulating EVs after oxidative injury [57, 237\u2013239]. Complementing these mechanistic insights, a recent meta-analysis suggests that points to the potential diagnostic utility of EV-based assays [240]. Incorporating pathway-proximal cell-death biomarkers into these panels could enable clinicians to (i) stratify patients by the dominant death pathway involved, (ii) match them with pathway-targeted interventions, and (iii) monitor on-target drug engagement in near real time capabilities often cited as prerequisites for precision medicine. Accordingly, recent reviews tend to frame EV-derived biomarkers as a promising bridge between molecular diagnosis and patient-tailored therapy in neurodegeneration.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Integrative framework and future perspectives", "text": "On the therapeutic front, several complementary or sequential approaches could be explored: (i) mitigating ferroptosis with radical-trapping antioxidants such as ferrostatin-1 or liproxstatin-1, together with selenium-GPX4 support and programmes that up-regulate SLC7A11; (ii) tempering necroptotic escalation via RIPK1 inhibitors; (iii) easing pyroptosis through NLRP3 modulators (e.g., MCC950); and (iv) restoring metabolic resilience with context-specific PARP inhibitors (e.g., Deferiprone) [172, 209, 241\u2013244]. The therapeutic compounds summarised in Table 3 each mapped to disease-specific biomarkers provide further support for this integrated intervention framework. Moving forward, it would be helpful to: (i) gather clinical evidence to support composite biomarker panels, (ii) refine the dosing and sequencing principles for combination therapies, and (iii) promote consistent reporting of study outcomes. By harmonising molecular read-outs with cell-type context and patient presentation, this integrative framework may gently guide the progression from early exploratory studies to thoughtfully crafted clinical trial designs [221, 240].", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Conclusion", "text": "Non-canonical cell death pathways such as ferroptosis, necroptosis, pyroptosis, and parthanatos are increasingly recognized as important elements in the complex landscape of neurodegenerative diseases. Current evidence suggests that these processes may act both as contributors to neuronal injury and as potential therapeutic entry points. While encouraging observations have been made in preclinical models, the extent to which this transpires in clinical benefit is yet to be determinedIn this regard, the incorporation of cell death signatures into biomarker-guided methods, could be useful for patient stratification into more appropriately defined therapy decisions. Multi-target strategies employing simultaneous means of addressing oxidative stress, inflammation and metabolic dysregulation, may provide a wider benefit but must be further evaluated for safety and long-term efficacy. Improvements in drug delivery systems that can cross the blood-brain barrier and developments in omics-based translational models may increase avenues for success in clinical practice. In addition, knowing the connections between regulated forms of cell death may promote combination therapies among these pathways, but this also complicates clinical strategies.", "title": "" }, { "pmid": 41699331, "pmcid": "PMC12909434", "section": "Conclusion", "text": "Taken together, while these emerging insights are promising, further efforts are needed to validate mechanistic findings in human studies and to develop safe, precise, and durable interventions. A systems level perspective that integrates molecular mechanisms with clinical context may eventually contribute to the development of disease-modifying strategies for neurodegenerative disorders.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Introduction", "text": "Neurodegenerative disorders, including Alzheimer\u2019s disease (AD), Parkinson\u2019s disease (PD), and Huntington\u2019s disease (HD), are characterized by progressive neuronal dysfunction and cell death, leading to irreversible cognitive and motor decline (1\u20133). These conditions not only severely impair quality of life but also place a growing burden on global healthcare systems (4). Despite substantial advances in understanding their pathophysiology, effective disease-modifying therapies remain limited, highlighting the urgent need for novel therapeutic strategies.\n\nEmerging evidence underscores the central role of nutrient metabolism in maintaining neuronal integrity. Neurons rely heavily on glucose and lipid metabolism to meet high energy demands, while amino acid metabolism supports neurotransmitter synthesis and synaptic plasticity (5\u20137). In neurodegenerative diseases, these metabolic processes are frequently disrupted. For instance, impaired glucose utilization and insulin resistance are hallmarks of AD, contributing to energy deficits and oxidative stress (8\u201311). Similarly, mitochondrial dysfunction and lipid peroxidation are key pathogenic features in PD and HD (12\u201317).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Introduction", "text": "In this context, metabolic reprogramming, the adaptive rewiring of cellular metabolic pathways in response to stress or disease, has gained recognition as a critical driver of neurodegeneration (18, 19). Metabolic reprogramming refers to the cellular adaptation of metabolic pathways in response to specific pathological conditions to satisfy energy requirements or to cope with environmental stressors (20, 21). These alterations not only influence the energy status of neurons but are also intricately associated with critical pathological mechanisms such as neuroinflammation and oxidative stress.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Introduction", "text": "Nutritional intervention is emerging as a promising therapeutic strategy, garnering increasing attention. The Mediterranean diet, abundant in antioxidants, omega-3 polyunsaturated fatty acids, and fiber, has demonstrated efficacy in enhancing cognitive function and alleviating symptoms in individuals with neurodegenerative diseases (22). Nevertheless, the clinical application of nutritional interventions faces numerous challenges, including the complexity of nutrient action mechanisms, patient adherence, and the long-term efficacy and safety of such interventions (23). To address these challenges, future research must delve deeper into the mechanisms underlying nutritional interventions and develop more precise assessment tools and personalized treatment approaches. The integrated application of metabolomics, genomics, and epigenetics will provide a theoretical foundation for tailored nutritional interventions. Furthermore, the incorporation of probiotics and epigenetic strategies, along with combined nutrient and pharmacological treatments, has opened new avenues for the management of neurodegenerative diseases (24).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Introduction", "text": "This review systematically searched the PubMed, Web of Science, and CNKI databases. The search keywords included \u201cneurodegenerative disorders,\u201d \u201cmetabolic reprogramming,\u201d \u201cnutritional therapy,\u201d \u201cMediterranean diet,\u201d \u201cmitochondrial dysfunction,\u201d \u201cAlzheimer\u2019s disease,\u201d \u201cParkinson\u2019s disease,\u201d and \u201cHuntington\u2019s disease,\u201d among others. The objective is to investigate the role of nutrition and cellular metabolic pathways in the pathogenesis of neurodegenerative conditions, elucidate their molecular mechanisms, and propose novel therapeutic interventions. By synthesizing findings across multiple scientific disciplines, the goal is to enable targeted treatment strategies for neurodegenerative diseases and to develop effective therapeutic modalities. Achieving this necessitates not only advances in basic research but also multidisciplinary collaboration among clinical medicine, nutritional science, molecular biology, and related fields to foster ongoing progress in disease management.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrition and neuronal metabolism", "text": "The pivotal role of nutrients in neuronal metabolism is undeniable; they not only supply energy but also contribute to the maintenance and repair of neuronal functions (25, 26). Figure 1 summarizes nutrient effects on neuronal metabolism. Through this pathway, Neurons oxidize to 5.8\u202fkg glucose per day (27). Through glycolytic pathways, glucose undergoes catabolism to form pyruvate, which then translocates into mitochondria to participate in the tricarboxylic acid cycle, culminating in ATP synthesis through oxidative phosphorylation; this metabolic process is essential for supplying the energy required for neuronal function (28). However, in the context of neurodegenerative diseases, disruptions in glucose metabolic pathways can hinder energy supply to neurons, resulting in dysfunction and cellular damage (29).\n\nThe influence of nutrients on neuronal metabolism. Glucose, fat, and amino acids in the extracellular environment enter neurons through the cell membrane. Glucose is converted into pyruvic acid through glycolysis. After entering the mitochondria, pyruvate participates in the tricarboxylic acid cycle, generating NADH and FADH2. These substances transfer electrons to oxygen through the electron transport chain, eventually producing ATP, which provides the main energy for the cell and generates key metabolites such as acetyl coenzyme A and citric acid. Neurons transmit signals through axons and release neurotransmitters powered by ATP at synapses.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrition and neuronal metabolism", "text": "Lipids, particularly omega-3 polyunsaturated fatty acids like docosahexaenoic acid (DHA), are vital for maintaining the fluidity of neuronal membranes and enhancing the efficacy of nerve signal transmission (30). A deficiency in DHA has been linked to increased neuroinflammation and the progression of neurodegenerative conditions, while DHA supplementation has been shown to bolster the stability of neuronal membranes and promote neuroprotection (31\u201333). Furthermore, disturbances in lipid metabolism, such as impaired fatty acid oxidation, may exacerbate neurodegenerative disease progression; thus, regulating lipid metabolism could potentially alleviate neuronal damage (34).\n\nAmino acid metabolism is also integral to neuronal function. Amino acids serve not only as the fundamental components of proteins but also as precursors for neurotransmitter synthesis (35, 36). For instance, tryptophan and tyrosine are converted into neurotransmitters like serotonin and dopamine in the brain, which are crucial for mood regulation and motor control (37\u201339). In neurodegenerative diseases, imbalances in amino acid metabolism can result in abnormal neurotransmitter levels, adversely affecting inter-neuronal signaling.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrition and neuronal metabolism", "text": "Within the nutrient metabolism pathways, glycolysis, the tricarboxylic acid cycle, and oxidative phosphorylation are essential processes for neuronal energy production (40, 41). Glycolysis occurs in the cytoplasm, converting glucose into pyruvate and yielding a modest amount of energy rapidly (42). The tricarboxylic acid cycle, located in the mitochondria, further metabolizes pyruvate, and integrates with oxidative phosphorylation to produce substantial ATP quantities (43). Oxidative phosphorylation, the final stage of neuronal energy metabolism, transpires in the inner mitochondrial membrane, generating ATP through electron transport chains and proton gradients (44). Disruptions in these metabolic pathways, particularly in neurodegenerative diseases, can lead to insufficient energy supply, neuronal dysfunction, and accelerated neurodegeneration.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrition and neuronal metabolism", "text": "Nutrients not only furnish energy for neurons but also play a role in preserving neuronal structure and facilitating signal transmission (45). For example, vitamins C and E function as antioxidants, shielding neurons from oxidative damage by neutralizing free radicals (46, 47). The B vitamin complex, especially folate, vitamin B6, and vitamin B12, is crucial for neurotransmitter synthesis and the formation of the myelin sheath, thereby supporting neuronal function (48, 49). Consequently, adequate nutrient intake and metabolic equilibrium are vital for sustaining neuronal health. We present a synthesis of the impact of nutrients on neuronal metabolism in Figure 1.\n\nThe interplay between nutrient metabolism and neuronal function represents a multifaceted biological mechanism wherein nutrients serve not only as energy sources but also as direct modulators of nerve signaling and synaptic plasticity (50). The equilibrium of metabolic pathways for nutrients significantly influences the energy status of neurons, thereby impacting their normal functionality and the stability of neural networks (51).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrition and neuronal metabolism", "text": "Nerve signaling relies on the release and reception of neurotransmitters, a process that is energetically demanding (52). Nutrients, particularly amino acids, are critical precursors in the synthesis of neurotransmitters (53). For example, the amino acids tryptophan and tyrosine are essential precursors in the biosynthesis of serotonin and dopamine, respectively; these neurotransmitters play a crucial role in modulating affective states, cognitive processes, and motor control (54\u201356). Disruptions in amino acid metabolism, whether through deficiency or excess, can lead to imbalances in neurotransmitter levels, impairing the normal transmission of nerve signals and potentially contributing to the pathogenesis of neurodegenerative disorders (57, 58).\n\nThe B vitamin complex, encompassing folic acid, vitamin B6, and vitamin B12, is vital for neurotransmitter synthesis and the formation of the myelin sheath (59, 60). Deficiencies in these vitamins can result in neuronal signaling disruptions, hindering inter-neuronal communication, and prolonged deficiencies may exacerbate neurodegenerative processes.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrition and neuronal metabolism", "text": "Synaptic plasticity, defined as the modulation of synaptic strength between neurons, is essential for learning and memory. Nutrient metabolism plays a crucial role in sustaining synaptic plasticity (61). For example, omega-3 polyunsaturated fatty acids, particularly DHA, directly influence the fluidity of neuronal membranes, thereby affecting the localization of synaptic proteins and the fusion of synaptic vesicles, which in turn impacts synaptic plasticity (62, 63). A deficiency in DHA has been associated with synaptic dysfunction, while DHA supplementation has demonstrated potential in enhancing learning and memory, offering avenues for early intervention and prevention of neurodegenerative diseases (64).\n\nNeurons exhibit high energy demands, with nutrient metabolism serving as the primary energy source (65). Glucose and ketone bodies are the principal energy substrates for the brain, generating ATP through glycolysis, the tricarboxylic acid cycle, and oxidative phosphorylation (66). In neurodegenerative conditions, such as AD, there is a marked decline in cerebral glucose metabolism, leading to insufficient energy supply, neuronal dysfunction, and heightened cellular damage (67). Nutritional strategies, such as the Mediterranean diet, rich in antioxidants and omega-3 fatty acids, have the potential to enhance neuronal metabolism and augment energy supply, thereby decelerating disease progression (68).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrition and neuronal metabolism", "text": "The role of mitochondrial quality control (MQC) in maintaining neuronal health is increasingly recognized as a critical factor in the prevention and management of neurodegenerative diseases. Recent research has highlighted the intricate mechanisms by which MQC is regulated and its implications for neuronal function. Notably, the modulation of mitochondrial motility and the expression of specific proteins such as Syntaphilin and Mitofusin 2 have been shown to play pivotal roles in coordinating MQC dynamics, thereby promoting neuronal health (69). These proteins act as mitochondrial motility modulators, facilitating the appropriate quality control pathways in response to mitochondrial damage, which is crucial for neuronal survival and function.\n\nIn addition to the structural and functional maintenance of mitochondria, the role of exercise in enhancing MQC has been explored, particularly in the context of cerebral ischemia. Exercise training has been demonstrated to upregulate the expression of sirtuin-3 (SIRT3), a mitochondrial protein that is essential for maintaining mitochondrial integrity and function. This upregulation is associated with improved neural function, reduced neuronal apoptosis, and re-established MQC following ischemic events (70). The study underscores the potential of exercise as a non-pharmacological intervention to optimize mitochondrial function and suggests that SIRT3 could be targeted for therapeutic strategies aimed at mitigating brain ischemia.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrition and neuronal metabolism", "text": "Abnormal nutrient metabolism, particularly in metabolic disorders like diabetes and obesity, significantly influences neuronal function and accelerates neurodegenerative processes (71\u201373). These metabolic conditions not only disrupt the normal metabolism of glucose and lipids but also initiate a cascade of events that result in neuronal damage and dysfunction. The interplay between metabolic stress and neuronal injury primarily involves oxidative stress, inflammatory responses, and mitochondrial dysfunction, all of which collectively exacerbate the degeneration of neurons.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrition and neuronal metabolism", "text": "There exists a robust correlation between diabetes, especially type 2 diabetes, and neurodegenerative diseases such as AD (74). In individuals with diabetes, the hyperglycemic state induces irregular energy metabolism in neurons, while the accumulation of advanced glycation end products (AGEs) intensifies oxidative stress and inflammatory responses, thereby impairing neuronal structure and function (75). Prolonged hyperglycemia can also lead to vascular complications, diminishing blood flow to neurons and further exacerbating neuronal injury (76). In diabetic patients, insulin resistance and \u03b2-cell dysfunction result in irregularities in insulin signaling, adversely affecting glucose utilization and energy production in neurons (77). Insulin plays a crucial role in regulating neuronal function, influencing neuronal growth, differentiation, and synaptic plasticity. Insulin resistance not only diminishes glucose uptake and utilization by neurons but also inhibits critical insulin-mediated signaling pathways such as PI3K/Akt and mTOR, which are essential for neuronal survival and functionality (77).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrition and neuronal metabolism", "text": "Obesity, particularly abdominal obesity, is linked to chronic low-grade inflammation and a state of metabolic stress that can potentially harm neurons (78). In obese individuals, various inflammatory mediators produced by adipose tissue, such as TNF-\u03b1 and IL-6, in-filtrate the brain via the bloodstream, triggering neuroinflammatory responses that damage neurons and glial cells, thereby disrupting normal neuronal function. Furthermore, obesity is associated with lipid metabolism abnormalities, including impaired fatty acid oxidation and lipid accumulation, which exacerbate oxidative stress in neurons and promote neurodegenerative processes (79).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrition and neuronal metabolism", "text": "Mitochondrial dysfunction represents a critical mechanism underlying neuronal injury in the context of metabolic stress. Mitochondria serve as the hub of cellular energy metabolism, and their impairment leads to reduced efficiency in oxidative phosphorylation and excessive generation of reactive oxygen species (ROS) (80, 81). The elevation of ROS not only inflicts damage on mitochondrial DNA but also disrupts proteins and lipids, instigating oxidative stress that compromise\u2019s neuronal structure and function. Additionally, mitochondrial damage activates the neuronal apoptotic pathway, further hastening neurodegeneration (82). Metabolic disorders also incite neuroinflammation, with the activation of microglia and astrocytes releasing a plethora of inflammatory mediators, such as IL-1\u03b2, TNF-\u03b1, and nitric oxide (NO), which not only exacerbate oxidative stress but also directly harm neurons and trigger neuronal death (83). Neuroinflammation further disrupts normal neuronal signaling, impairing synaptic function, and aggravating the progression of neurodegenerative diseases.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Effects of nutrients on neuronal metabolism", "text": "The pivotal role of nutrients in neuronal metabolism is undeniable; they not only supply energy but also contribute to the maintenance and repair of neuronal functions (25, 26). Figure 1 summarizes nutrient effects on neuronal metabolism. Through this pathway, Neurons oxidize to 5.8\u202fkg glucose per day (27). Through glycolytic pathways, glucose undergoes catabolism to form pyruvate, which then translocates into mitochondria to participate in the tricarboxylic acid cycle, culminating in ATP synthesis through oxidative phosphorylation; this metabolic process is essential for supplying the energy required for neuronal function (28). However, in the context of neurodegenerative diseases, disruptions in glucose metabolic pathways can hinder energy supply to neurons, resulting in dysfunction and cellular damage (29).\n\nThe influence of nutrients on neuronal metabolism. Glucose, fat, and amino acids in the extracellular environment enter neurons through the cell membrane. Glucose is converted into pyruvic acid through glycolysis. After entering the mitochondria, pyruvate participates in the tricarboxylic acid cycle, generating NADH and FADH2. These substances transfer electrons to oxygen through the electron transport chain, eventually producing ATP, which provides the main energy for the cell and generates key metabolites such as acetyl coenzyme A and citric acid. Neurons transmit signals through axons and release neurotransmitters powered by ATP at synapses.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Effects of nutrients on neuronal metabolism", "text": "Lipids, particularly omega-3 polyunsaturated fatty acids like docosahexaenoic acid (DHA), are vital for maintaining the fluidity of neuronal membranes and enhancing the efficacy of nerve signal transmission (30). A deficiency in DHA has been linked to increased neuroinflammation and the progression of neurodegenerative conditions, while DHA supplementation has been shown to bolster the stability of neuronal membranes and promote neuroprotection (31\u201333). Furthermore, disturbances in lipid metabolism, such as impaired fatty acid oxidation, may exacerbate neurodegenerative disease progression; thus, regulating lipid metabolism could potentially alleviate neuronal damage (34).\n\nAmino acid metabolism is also integral to neuronal function. Amino acids serve not only as the fundamental components of proteins but also as precursors for neurotransmitter synthesis (35, 36). For instance, tryptophan and tyrosine are converted into neurotransmitters like serotonin and dopamine in the brain, which are crucial for mood regulation and motor control (37\u201339). In neurodegenerative diseases, imbalances in amino acid metabolism can result in abnormal neurotransmitter levels, adversely affecting inter-neuronal signaling.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Effects of nutrients on neuronal metabolism", "text": "Within the nutrient metabolism pathways, glycolysis, the tricarboxylic acid cycle, and oxidative phosphorylation are essential processes for neuronal energy production (40, 41). Glycolysis occurs in the cytoplasm, converting glucose into pyruvate and yielding a modest amount of energy rapidly (42). The tricarboxylic acid cycle, located in the mitochondria, further metabolizes pyruvate, and integrates with oxidative phosphorylation to produce substantial ATP quantities (43). Oxidative phosphorylation, the final stage of neuronal energy metabolism, transpires in the inner mitochondrial membrane, generating ATP through electron transport chains and proton gradients (44). Disruptions in these metabolic pathways, particularly in neurodegenerative diseases, can lead to insufficient energy supply, neuronal dysfunction, and accelerated neurodegeneration.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Effects of nutrients on neuronal metabolism", "text": "Nutrients not only furnish energy for neurons but also play a role in preserving neuronal structure and facilitating signal transmission (45). For example, vitamins C and E function as antioxidants, shielding neurons from oxidative damage by neutralizing free radicals (46, 47). The B vitamin complex, especially folate, vitamin B6, and vitamin B12, is crucial for neurotransmitter synthesis and the formation of the myelin sheath, thereby supporting neuronal function (48, 49). Consequently, adequate nutrient intake and metabolic equilibrium are vital for sustaining neuronal health. We present a synthesis of the impact of nutrients on neuronal metabolism in Figure 1.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrient metabolism and neuronal function", "text": "The interplay between nutrient metabolism and neuronal function represents a multifaceted biological mechanism wherein nutrients serve not only as energy sources but also as direct modulators of nerve signaling and synaptic plasticity (50). The equilibrium of metabolic pathways for nutrients significantly influences the energy status of neurons, thereby impacting their normal functionality and the stability of neural networks (51).\n\nNerve signaling relies on the release and reception of neurotransmitters, a process that is energetically demanding (52). Nutrients, particularly amino acids, are critical precursors in the synthesis of neurotransmitters (53). For example, the amino acids tryptophan and tyrosine are essential precursors in the biosynthesis of serotonin and dopamine, respectively; these neurotransmitters play a crucial role in modulating affective states, cognitive processes, and motor control (54\u201356). Disruptions in amino acid metabolism, whether through deficiency or excess, can lead to imbalances in neurotransmitter levels, impairing the normal transmission of nerve signals and potentially contributing to the pathogenesis of neurodegenerative disorders (57, 58).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrient metabolism and neuronal function", "text": "The B vitamin complex, encompassing folic acid, vitamin B6, and vitamin B12, is vital for neurotransmitter synthesis and the formation of the myelin sheath (59, 60). Deficiencies in these vitamins can result in neuronal signaling disruptions, hindering inter-neuronal communication, and prolonged deficiencies may exacerbate neurodegenerative processes.\n\nSynaptic plasticity, defined as the modulation of synaptic strength between neurons, is essential for learning and memory. Nutrient metabolism plays a crucial role in sustaining synaptic plasticity (61). For example, omega-3 polyunsaturated fatty acids, particularly DHA, directly influence the fluidity of neuronal membranes, thereby affecting the localization of synaptic proteins and the fusion of synaptic vesicles, which in turn impacts synaptic plasticity (62, 63). A deficiency in DHA has been associated with synaptic dysfunction, while DHA supplementation has demonstrated potential in enhancing learning and memory, offering avenues for early intervention and prevention of neurodegenerative diseases (64).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrient metabolism and neuronal function", "text": "Neurons exhibit high energy demands, with nutrient metabolism serving as the primary energy source (65). Glucose and ketone bodies are the principal energy substrates for the brain, generating ATP through glycolysis, the tricarboxylic acid cycle, and oxidative phosphorylation (66). In neurodegenerative conditions, such as AD, there is a marked decline in cerebral glucose metabolism, leading to insufficient energy supply, neuronal dysfunction, and heightened cellular damage (67). Nutritional strategies, such as the Mediterranean diet, rich in antioxidants and omega-3 fatty acids, have the potential to enhance neuronal metabolism and augment energy supply, thereby decelerating disease progression (68).\n\nThe role of mitochondrial quality control (MQC) in maintaining neuronal health is increasingly recognized as a critical factor in the prevention and management of neurodegenerative diseases. Recent research has highlighted the intricate mechanisms by which MQC is regulated and its implications for neuronal function. Notably, the modulation of mitochondrial motility and the expression of specific proteins such as Syntaphilin and Mitofusin 2 have been shown to play pivotal roles in coordinating MQC dynamics, thereby promoting neuronal health (69). These proteins act as mitochondrial motility modulators, facilitating the appropriate quality control pathways in response to mitochondrial damage, which is crucial for neuronal survival and function.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrient metabolism and neuronal function", "text": "In addition to the structural and functional maintenance of mitochondria, the role of exercise in enhancing MQC has been explored, particularly in the context of cerebral ischemia. Exercise training has been demonstrated to upregulate the expression of sirtuin-3 (SIRT3), a mitochondrial protein that is essential for maintaining mitochondrial integrity and function. This upregulation is associated with improved neural function, reduced neuronal apoptosis, and re-established MQC following ischemic events (70). The study underscores the potential of exercise as a non-pharmacological intervention to optimize mitochondrial function and suggests that SIRT3 could be targeted for therapeutic strategies aimed at mitigating brain ischemia.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Abnormal nutrient metabolism and neuronal injury", "text": "Abnormal nutrient metabolism, particularly in metabolic disorders like diabetes and obesity, significantly influences neuronal function and accelerates neurodegenerative processes (71\u201373). These metabolic conditions not only disrupt the normal metabolism of glucose and lipids but also initiate a cascade of events that result in neuronal damage and dysfunction. The interplay between metabolic stress and neuronal injury primarily involves oxidative stress, inflammatory responses, and mitochondrial dysfunction, all of which collectively exacerbate the degeneration of neurons.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Abnormal nutrient metabolism and neuronal injury", "text": "There exists a robust correlation between diabetes, especially type 2 diabetes, and neurodegenerative diseases such as AD (74). In individuals with diabetes, the hyperglycemic state induces irregular energy metabolism in neurons, while the accumulation of advanced glycation end products (AGEs) intensifies oxidative stress and inflammatory responses, thereby impairing neuronal structure and function (75). Prolonged hyperglycemia can also lead to vascular complications, diminishing blood flow to neurons and further exacerbating neuronal injury (76). In diabetic patients, insulin resistance and \u03b2-cell dysfunction result in irregularities in insulin signaling, adversely affecting glucose utilization and energy production in neurons (77). Insulin plays a crucial role in regulating neuronal function, influencing neuronal growth, differentiation, and synaptic plasticity. Insulin resistance not only diminishes glucose uptake and utilization by neurons but also inhibits critical insulin-mediated signaling pathways such as PI3K/Akt and mTOR, which are essential for neuronal survival and functionality (77).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Abnormal nutrient metabolism and neuronal injury", "text": "Obesity, particularly abdominal obesity, is linked to chronic low-grade inflammation and a state of metabolic stress that can potentially harm neurons (78). In obese individuals, various inflammatory mediators produced by adipose tissue, such as TNF-\u03b1 and IL-6, in-filtrate the brain via the bloodstream, triggering neuroinflammatory responses that damage neurons and glial cells, thereby disrupting normal neuronal function. Furthermore, obesity is associated with lipid metabolism abnormalities, including impaired fatty acid oxidation and lipid accumulation, which exacerbate oxidative stress in neurons and promote neurodegenerative processes (79).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Abnormal nutrient metabolism and neuronal injury", "text": "Mitochondrial dysfunction represents a critical mechanism underlying neuronal injury in the context of metabolic stress. Mitochondria serve as the hub of cellular energy metabolism, and their impairment leads to reduced efficiency in oxidative phosphorylation and excessive generation of reactive oxygen species (ROS) (80, 81). The elevation of ROS not only inflicts damage on mitochondrial DNA but also disrupts proteins and lipids, instigating oxidative stress that compromise\u2019s neuronal structure and function. Additionally, mitochondrial damage activates the neuronal apoptotic pathway, further hastening neurodegeneration (82). Metabolic disorders also incite neuroinflammation, with the activation of microglia and astrocytes releasing a plethora of inflammatory mediators, such as IL-1\u03b2, TNF-\u03b1, and nitric oxide (NO), which not only exacerbate oxidative stress but also directly harm neurons and trigger neuronal death (83). Neuroinflammation further disrupts normal neuronal signaling, impairing synaptic function, and aggravating the progression of neurodegenerative diseases.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrition and neurodegenerative diseases", "text": "AD and metabolic dysregulation have emerged as a prominent focus in neurodegenerative research. Empirical evidence indicates that insulin resistance and mitochondrial impairment are critical pathogenic factors contributing to cognitive deterioration in AD. Insulin resistance not only disrupts cerebral metabolic homeostasis but also impairs synaptic plasticity, thereby facilitating neurodegeneration (84). Furthermore, mitochondrial dysfunction results in excessive reactive oxygen species (ROS) generation, intensifying neuronal injury (85).\n\nInsulin resistance significantly mediates the link between AD and type 2 diabetes mellitus (T2DM). Epidemiological data reveal that over 80% of AD patients concurrently exhibit T2DM or dysglycemia, implying overlapping pathogenic pathways [2]. Insulin resistance exacerbates neuroinflammation, promotes \u03b2-amyloid aggregation, and induces abnormal tau phosphorylation, thereby accelerating AD progression (86, 87).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrition and neurodegenerative diseases", "text": "Additionally, the relationship between mitochondrial dysfunction, metabolic syndrome (MetS), and neurodegenerative disorders has garnered considerable scientific attention. Research suggests that MetS and diabetes share common mechanistic pathways in the pathogenesis of AD and related neurodegenerative conditions, with mitochondrial dynamics serving as a pivotal factor (85). Mitochondrial DNA mutations, environmental toxin exposure, and excessive caloric intake can precipitate mitochondrial impairment, subsequently leading to insulin resistance and elevated ROS production (85).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrition and neurodegenerative diseases", "text": "The Mediterranean diet, characterized by an abundance of vegetables, fruits, whole grains, nuts, olive oil, and fish, has been associated with a decreased risk of developing AD (88). This dietary pattern is rich in antioxidants, including vitamins C and E, as well as omega-3 polyunsaturated fatty acids, which possess anti-inflammatory and antioxidant properties that may mitigate neuronal damage linked to oxidative stress and neuroinflammation in AD (89). Additionally, the Mediterranean diet is high in fiber, which supports the health of the gut microbiome; the interplay between the gut microbiome and brain health is increasingly recognized and may influence AD progression by modulating inflammatory and metabolic pathways (89). Omega-3 polyunsaturated fatty acids, particularly DHA, hold significant promise in the prevention and management of AD (90). DHA is a crucial component of neuronal membranes, and its deficiency is associated with heightened neuroinflammation and neurodegenerative processes (91). Supplementation with DHA has been shown to enhance the stability of neuronal membranes, promote neuroprotection, and potentially improve cognitive function (92).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrition and neurodegenerative diseases", "text": "Parkinson\u2019s disease, as a complex neurodegenerative disorder, involves a pathogenic mechanism characterized by multifactorial processes, including oxidative stress, mitochondrial dysfunction, and lipid peroxidation. Recent research has increasingly elucidated the link between nutritional factors and neurodegeneration, particularly highlighting the impact of impaired oxidative phosphorylation and lipid peroxidation in PD pathogenesis.\n\nPrimarily, metabolic dysregulation in PD is closely associated with disturbances in glucose and lipid homeostasis. Evidence indicates that mitochondrial impairment may be connected to disruptions in lipid and glucose metabolism, as well as insulin resistance, collectively contributing to abnormal \u03b1-synuclein aggregation and degeneration of substantia nigra dopaminergic neurons (93). Furthermore, oxidative stress is a pivotal factor in neurodegeneration, with mitochondrial damage serving as a primary contributor to increased reactive oxygen species production and disease progression (94). Consequently, targeting these metabolic pathways presents promising therapeutic avenues for PD intervention.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrition and neurodegenerative diseases", "text": "Secondly, dysregulation of lipid metabolism significantly influences PD pathophysiology. Lipid peroxidation is a prevalent pathological hallmark in neurodegenerative conditions. Strategies aimed at inhibiting neuronal lipid oxidation may decelerate disease progression and mitigate symptom severity (95). Dietary antioxidants, such as polyphenols and vitamin-based antioxidants, may alleviate PD pathology by restoring redox homeostasis at the cellular level (96). Therefore, nutritional modulation and antioxidant supplementation could serve as effective strategies to delay PD progression.\n\nFinally, nutritional interventions hold potential therapeutic value in PD management. Emerging studies demonstrate that nanocarrier systems, such as organic\u2013inorganic hybrid mesoporous silica nanoparticles functionalized with bioactive compounds like curcumin and lactoferrin, can enhance neuroprotection and improve therapeutic efficacy through targeted delivery (97). Additionally, lipid-based nutrients play crucial roles in neuronal survival and neuroprotection in PD and AD, with prophylactic lipid supplementation potentially slowing disease advancement (98).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrition and neurodegenerative diseases", "text": "HD is a hereditary neurodegenerative condition resulting from mutations in the gene responsible for encoding the Huntington protein, predominantly impacting the striatum region of the brain (99). This disorder leads to impairments in motor coordination, cognitive deterioration, and emotional disturbances (99). A significant factor in the pathogenesis of HD is the disruption of nutrient metabolism, particularly the abnormalities in energy metabolism, which not only compromise neuronal function but also correlate with the rate and severity of disease progression (100). Nutritional interventions, especially those aimed at modulating energy metabolism and oxidative stress, exhibit promising therapeutic potential for HD (101).\n\nIn HD patients, abnormal energy metabolism is characterized by diminished glucose metabolic efficiency, mitochondrial dysfunction, and impaired fatty acid oxidation (101, 102). These metabolic anomalies result in inadequate energy supply to neurons, further aggravating cellular damage. Mitochondria, which serve as the hub of energy metabolism, experience severe impairment in HD, evidenced by reduced oxidative phosphorylation efficiency and heightened reactive oxygen species (ROS) production, which damages mitochondrial DNA, instigates oxidative stress, and compromises neuronal structure and function (103, 104). Additionally, weight loss is frequently observed in HD patients, potentially linked to energy metabolism disorders and nutrient malabsorption (105).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrition and neurodegenerative diseases", "text": "Amyotrophic Lateral Sclerosis (ALS) is characterized by the degeneration of motor neurons, leading to muscle weakness and atrophy (106). In individuals with ALS, there is a notable disruption in energy metabolism, evidenced by heightened energy expenditure coupled with diminished energy intake, which may contribute to muscle wasting and the elevated demands of neuronal function (107). Disorders in nutrient metabolism, particularly those affecting protein and amino acid pathways, play a crucial role in the progression of ALS (108). Implementing a high-protein diet and administering specific amino acid supplements, such as leucine, have demonstrated enhancements in muscle functionality and overall quality of life for ALS patients. Microbial therapy targeting the gut-brain axis (GBA) has also become an emerging personalized treatment approach (109). Furthermore, the incorporation of antioxidants and omega-3 fatty acids has been shown to confer neuroprotective benefits by mitigating oxidative stress and inflammation in this population (109).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutrition and neurodegenerative diseases", "text": "Frontotemporal Dementia (FTD) encompasses a spectrum of disorders that primarily impact the frontal and temporal lobes of the brain, significantly influencing behavior, language, and cognitive abilities (110). In FTD patients, metabolic irregularities are evident through altered glucose metabolism, particularly within the frontotemporal regions (111). This metabolic dysfunction is linked to inadequate energy supply to neurons, thereby impairing neuronal activity and synaptic plasticity.\n\nFurthermore, conditions such as Depression and Multiple Sclerosis (MS) are prevalent (112, 113). We have compiled a summary of nutrient metabolism irregularities and nutritional intervention strategies across various neurodegenerative disorders, as presented in Table 1.\n\nNutrient metabolism abnormalities and nutritional intervention strategies in different neurodegenerative diseases.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "AD", "text": "AD and metabolic dysregulation have emerged as a prominent focus in neurodegenerative research. Empirical evidence indicates that insulin resistance and mitochondrial impairment are critical pathogenic factors contributing to cognitive deterioration in AD. Insulin resistance not only disrupts cerebral metabolic homeostasis but also impairs synaptic plasticity, thereby facilitating neurodegeneration (84). Furthermore, mitochondrial dysfunction results in excessive reactive oxygen species (ROS) generation, intensifying neuronal injury (85).\n\nInsulin resistance significantly mediates the link between AD and type 2 diabetes mellitus (T2DM). Epidemiological data reveal that over 80% of AD patients concurrently exhibit T2DM or dysglycemia, implying overlapping pathogenic pathways [2]. Insulin resistance exacerbates neuroinflammation, promotes \u03b2-amyloid aggregation, and induces abnormal tau phosphorylation, thereby accelerating AD progression (86, 87).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "AD", "text": "Additionally, the relationship between mitochondrial dysfunction, metabolic syndrome (MetS), and neurodegenerative disorders has garnered considerable scientific attention. Research suggests that MetS and diabetes share common mechanistic pathways in the pathogenesis of AD and related neurodegenerative conditions, with mitochondrial dynamics serving as a pivotal factor (85). Mitochondrial DNA mutations, environmental toxin exposure, and excessive caloric intake can precipitate mitochondrial impairment, subsequently leading to insulin resistance and elevated ROS production (85).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "AD", "text": "The Mediterranean diet, characterized by an abundance of vegetables, fruits, whole grains, nuts, olive oil, and fish, has been associated with a decreased risk of developing AD (88). This dietary pattern is rich in antioxidants, including vitamins C and E, as well as omega-3 polyunsaturated fatty acids, which possess anti-inflammatory and antioxidant properties that may mitigate neuronal damage linked to oxidative stress and neuroinflammation in AD (89). Additionally, the Mediterranean diet is high in fiber, which supports the health of the gut microbiome; the interplay between the gut microbiome and brain health is increasingly recognized and may influence AD progression by modulating inflammatory and metabolic pathways (89). Omega-3 polyunsaturated fatty acids, particularly DHA, hold significant promise in the prevention and management of AD (90). DHA is a crucial component of neuronal membranes, and its deficiency is associated with heightened neuroinflammation and neurodegenerative processes (91). Supplementation with DHA has been shown to enhance the stability of neuronal membranes, promote neuroprotection, and potentially improve cognitive function (92).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "PD", "text": "Parkinson\u2019s disease, as a complex neurodegenerative disorder, involves a pathogenic mechanism characterized by multifactorial processes, including oxidative stress, mitochondrial dysfunction, and lipid peroxidation. Recent research has increasingly elucidated the link between nutritional factors and neurodegeneration, particularly highlighting the impact of impaired oxidative phosphorylation and lipid peroxidation in PD pathogenesis.\n\nPrimarily, metabolic dysregulation in PD is closely associated with disturbances in glucose and lipid homeostasis. Evidence indicates that mitochondrial impairment may be connected to disruptions in lipid and glucose metabolism, as well as insulin resistance, collectively contributing to abnormal \u03b1-synuclein aggregation and degeneration of substantia nigra dopaminergic neurons (93). Furthermore, oxidative stress is a pivotal factor in neurodegeneration, with mitochondrial damage serving as a primary contributor to increased reactive oxygen species production and disease progression (94). Consequently, targeting these metabolic pathways presents promising therapeutic avenues for PD intervention.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "PD", "text": "Secondly, dysregulation of lipid metabolism significantly influences PD pathophysiology. Lipid peroxidation is a prevalent pathological hallmark in neurodegenerative conditions. Strategies aimed at inhibiting neuronal lipid oxidation may decelerate disease progression and mitigate symptom severity (95). Dietary antioxidants, such as polyphenols and vitamin-based antioxidants, may alleviate PD pathology by restoring redox homeostasis at the cellular level (96). Therefore, nutritional modulation and antioxidant supplementation could serve as effective strategies to delay PD progression.\n\nFinally, nutritional interventions hold potential therapeutic value in PD management. Emerging studies demonstrate that nanocarrier systems, such as organic\u2013inorganic hybrid mesoporous silica nanoparticles functionalized with bioactive compounds like curcumin and lactoferrin, can enhance neuroprotection and improve therapeutic efficacy through targeted delivery (97). Additionally, lipid-based nutrients play crucial roles in neuronal survival and neuroprotection in PD and AD, with prophylactic lipid supplementation potentially slowing disease advancement (98).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "HD", "text": "HD is a hereditary neurodegenerative condition resulting from mutations in the gene responsible for encoding the Huntington protein, predominantly impacting the striatum region of the brain (99). This disorder leads to impairments in motor coordination, cognitive deterioration, and emotional disturbances (99). A significant factor in the pathogenesis of HD is the disruption of nutrient metabolism, particularly the abnormalities in energy metabolism, which not only compromise neuronal function but also correlate with the rate and severity of disease progression (100). Nutritional interventions, especially those aimed at modulating energy metabolism and oxidative stress, exhibit promising therapeutic potential for HD (101).\n\nIn HD patients, abnormal energy metabolism is characterized by diminished glucose metabolic efficiency, mitochondrial dysfunction, and impaired fatty acid oxidation (101, 102). These metabolic anomalies result in inadequate energy supply to neurons, further aggravating cellular damage. Mitochondria, which serve as the hub of energy metabolism, experience severe impairment in HD, evidenced by reduced oxidative phosphorylation efficiency and heightened reactive oxygen species (ROS) production, which damages mitochondrial DNA, instigates oxidative stress, and compromises neuronal structure and function (103, 104). Additionally, weight loss is frequently observed in HD patients, potentially linked to energy metabolism disorders and nutrient malabsorption (105).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Others", "text": "Amyotrophic Lateral Sclerosis (ALS) is characterized by the degeneration of motor neurons, leading to muscle weakness and atrophy (106). In individuals with ALS, there is a notable disruption in energy metabolism, evidenced by heightened energy expenditure coupled with diminished energy intake, which may contribute to muscle wasting and the elevated demands of neuronal function (107). Disorders in nutrient metabolism, particularly those affecting protein and amino acid pathways, play a crucial role in the progression of ALS (108). Implementing a high-protein diet and administering specific amino acid supplements, such as leucine, have demonstrated enhancements in muscle functionality and overall quality of life for ALS patients. Microbial therapy targeting the gut-brain axis (GBA) has also become an emerging personalized treatment approach (109). Furthermore, the incorporation of antioxidants and omega-3 fatty acids has been shown to confer neuroprotective benefits by mitigating oxidative stress and inflammation in this population (109).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Others", "text": "Frontotemporal Dementia (FTD) encompasses a spectrum of disorders that primarily impact the frontal and temporal lobes of the brain, significantly influencing behavior, language, and cognitive abilities (110). In FTD patients, metabolic irregularities are evident through altered glucose metabolism, particularly within the frontotemporal regions (111). This metabolic dysfunction is linked to inadequate energy supply to neurons, thereby impairing neuronal activity and synaptic plasticity.\n\nFurthermore, conditions such as Depression and Multiple Sclerosis (MS) are prevalent (112, 113). We have compiled a summary of nutrient metabolism irregularities and nutritional intervention strategies across various neurodegenerative disorders, as presented in Table 1.\n\nNutrient metabolism abnormalities and nutritional intervention strategies in different neurodegenerative diseases.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Cell metabolic reprogramming and neurodegenerative diseases", "text": "Metabolic reprogramming refers to the stimulus-induced rewiring of intracellular metabolic networks. Unlike tumor cells\u2014where reprogramming supports biomass accumulation and uncontrolled proliferation\u2014neurons prioritize survival and bioenergetic stability in response to chronic stress. In neurons, this process manifests as three core events: (i) a glycolytic shift toward aerobic glycolysis, (ii) impaired mitochondrial biogenesis and oxidative phosphorylation, and (iii) loss of redox homeostasis with sustained reactive oxygen species (ROS) generation. Collectively, these changes constitute an adaptive yet maladaptive response to persistent energy stress\u2014initially protective, but ultimately driving oxidative damage and apoptotic signaling (18, 114\u2013117).\n\nHowever, neuronal metabolic reprogramming exhibits distinct characteristics (9). In contrast to tumor and immune cells, neurons are highly specialized and possess limited proliferative capacity, resulting in significant differences in the objectives and mechanisms underlying their metabolic reprogramming. In neurodegenerative diseases, this reprogramming is primarily characterized by aberrant energy metabolism, mitochondrial dysfunction, and alterations in nutrient utilization and demand, all of which directly impact neuronal viability and functionality (118).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Cell metabolic reprogramming and neurodegenerative diseases", "text": "In AD, a marked reduction in cerebral glucose metabolism serves as a critical indicator of metabolic reprogramming. This phenomenon not only signifies a diminished energy supply to the brain but is also intricately linked to neuronal injury and apoptosis (119). The decline in oxidative phosphorylation efficiency, attributed to mitochondrial dysfunction, is a pivotal contributor to the aberrant energy metabolism observed in neurons (120). Furthermore, insulin resistance plays a crucial role in AD, impairing glucose uptake and utilization by neurons, thereby exacerbating metabolic disturbances (121).\n\nIn PD, mitochondrial dysfunction and oxidative stress are central to the metabolic reprogramming process, resulting in the accumulation of reactive oxygen species (ROS) that further compromise neuronal integrity (122). In HD, abnormalities in energy metabolism are evident through diminished glucose metabolism efficiency, mitochondrial dysfunction, and impaired fatty acid oxidation, culminating in inadequate energy supply to neurons and exacerbating cellular damage.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Cell metabolic reprogramming and neurodegenerative diseases", "text": "The distinctiveness of neuronal metabolic reprogramming is also underscored by its intricate relationship with the pathological mechanisms of neurodegenerative diseases. For instance, in AD, the accumulation of \u03b2-amyloid plaques and neurofibrillary tangles not only disrupts neuronal structure and function but also interferes with normal metabolic pathways, instigating metabolic reprogramming (123). Similarly, in PD, the aggregation of \u03b1-synuclein adversely affects mitochondrial function, triggering metabolic reprogramming that intensifies oxidative stress and neuronal damage (124).\n\nInvestigating the mechanisms of neuronal metabolic reprogramming is crucial for comprehending the fundamental nature of neurodegenerative diseases and formulating therapeutic interventions. The mechanisms involved encompass, but are not limited to, mitochondrial biogenesis and dysfunction, altered oxidative phosphorylation efficiency, modulation of glycolysis and tricarboxylic acid cycles, irregular fatty acid oxidation, and variations in nutrient utilization and demand by neurons. A thorough examination of these mechanisms will elucidate the specific contributions of metabolic reprogramming to neurodegenerative diseases, offering novel insights for diagnosis, prevention, and treatment strategies. We encapsulate the function of metabolic reprogramming in neurodegenerative disorders in Figure 2.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Cell metabolic reprogramming and neurodegenerative diseases", "text": "The mechanism of metabolic reprogramming in neurodegenerative diseases. Glucose enters neurons through the GLUT4 transporter, is converted into glucose-6-phosphate and pyruvic acid through glycolysis, provides initial energy, and can be converted into glycogen for storage. High levels of mitochondrial stress led to the production of reactive oxygen species (ROS), activate the STING pathway, trigger an inflammatory response, and cause the death and abnormal movement of dopamine-secreting neurons, especially evident in Parkin-deficient mice. Microglia are inactive at rest, but inflammatory stimulation or brain injury can activate them, promoting the expression of inflammatory genes through the c/EBP\u03b2 pathway and leading to the formation of neurofibrillary tangles.\n\nThe potential of metabolic reprogramming as a regulatory and therapeutic target offers extensive research opportunities within the domain of neurodegenerative diseases (125). Both nutritional and pharmacological interventions can significantly influence neuronal survival and functionality by modulating critical metabolic pathways, as well as regulating neuroinflammation and oxidative stress, thereby potentially decelerating or reversing neurodegenerative processes (18, 126).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Cell metabolic reprogramming and neurodegenerative diseases", "text": "Nutritional interventions, particularly those involving diets abundant in antioxidants, omega-3 polyunsaturated fatty acids, and specific amino acids, exert a profound regulatory influence on metabolic reprogramming (127, 128). The Mediterranean diet, characterized by its rich variety of vegetables, fruits, whole grains, nuts, olive oil, and fish, has demonstrated efficacy in enhancing energy metabolism, mitigating oxidative stress and inflammation, and providing neuroprotective benefits (129). Antioxidants found in the Mediterranean diet, such as vitamin C, vitamin E, and selenium, play a crucial role in neutralizing free radicals, alleviating oxidative stress, and safeguarding neuronal integrity (130). Omega-3 polyunsaturated fatty acids, particularly DHA and EPA, not only diminish the synthesis of inflammatory mediators but also enhance mitochondrial functionality, improve neuronal membrane fluidity, facilitate signaling processes, and hold potential therapeutic value for neurodegenerative conditions (131). Tailored nutritional interventions, customized according to individual metabolic profiles and disease conditions, can more effectively modulate neuronal metabolism, decrease oxidative stress and inflammation, and foster neuroprotection by optimizing dietary components, such as increasing the intake of antioxidants and fatty acids. For instance, in patients with neurodegenerative diseases characterized by mitochondrial dysfunction, the supplementation of antioxidants and omega-3 fatty acids may yield superior neuroprotective outcomes (132).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Cell metabolic reprogramming and neurodegenerative diseases", "text": "Pharmacological interventions, particularly those aimed at specific metabolic pathways, also exhibit promising regulatory effects on metabolic reprogramming. For example, insulin sensitizers like thiazolidinediones confer neuroprotective benefits in neurodegenerative diseases by enhancing insulin sensitivity and glucose metabolism (133). These agents facilitate glucose uptake and utilization, thereby improving the energy supply to neurons and potentially slowing disease progression. Antioxidants and anti-inflammatory medications, such as N-Acetylcysteine amide and non-steroidal anti-inflammatory drugs, provide protective effects against neurodegenerative diseases by diminishing oxidative stress and curtailing inflammation (134, 135). These pharmacological agents can lower the production of reactive oxygen species (ROS), inhibit the release of inflammatory mediators, and mitigate neuronal damage.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Cell metabolic reprogramming and neurodegenerative diseases", "text": "The pivotal function of metabolic reprogramming in neurodegenerative pathologies underscores its significance as a therapeutic target. Notably, modulation of the AMPK/PGC-1\u03b1 signaling axis has been extensively investigated. Evidence indicates that Dihuang Yinzi enhances mitochondrial biogenesis via this pathway, thereby ameliorating cognitive deficits and mitochondrial ultrastructural damage in transgenic Alzheimer\u2019s disease models (136). Furthermore, Sestrin2 overexpression mitigates osteoarthritic pain and suppresses neuroinflammatory responses through activation of the AMPK/PGC-1\u03b1 pathway (137). These findings elucidate the therapeutic potential of targeting the AMPK/PGC-1\u03b1 signaling cascade in neurodegenerative and neuroinflammatory disorders.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Concept and mechanism of metabolic reprogramming", "text": "Metabolic reprogramming refers to the stimulus-induced rewiring of intracellular metabolic networks. Unlike tumor cells\u2014where reprogramming supports biomass accumulation and uncontrolled proliferation\u2014neurons prioritize survival and bioenergetic stability in response to chronic stress. In neurons, this process manifests as three core events: (i) a glycolytic shift toward aerobic glycolysis, (ii) impaired mitochondrial biogenesis and oxidative phosphorylation, and (iii) loss of redox homeostasis with sustained reactive oxygen species (ROS) generation. Collectively, these changes constitute an adaptive yet maladaptive response to persistent energy stress\u2014initially protective, but ultimately driving oxidative damage and apoptotic signaling (18, 114\u2013117).\n\nHowever, neuronal metabolic reprogramming exhibits distinct characteristics (9). In contrast to tumor and immune cells, neurons are highly specialized and possess limited proliferative capacity, resulting in significant differences in the objectives and mechanisms underlying their metabolic reprogramming. In neurodegenerative diseases, this reprogramming is primarily characterized by aberrant energy metabolism, mitochondrial dysfunction, and alterations in nutrient utilization and demand, all of which directly impact neuronal viability and functionality (118).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Concept and mechanism of metabolic reprogramming", "text": "In AD, a marked reduction in cerebral glucose metabolism serves as a critical indicator of metabolic reprogramming. This phenomenon not only signifies a diminished energy supply to the brain but is also intricately linked to neuronal injury and apoptosis (119). The decline in oxidative phosphorylation efficiency, attributed to mitochondrial dysfunction, is a pivotal contributor to the aberrant energy metabolism observed in neurons (120). Furthermore, insulin resistance plays a crucial role in AD, impairing glucose uptake and utilization by neurons, thereby exacerbating metabolic disturbances (121).\n\nIn PD, mitochondrial dysfunction and oxidative stress are central to the metabolic reprogramming process, resulting in the accumulation of reactive oxygen species (ROS) that further compromise neuronal integrity (122). In HD, abnormalities in energy metabolism are evident through diminished glucose metabolism efficiency, mitochondrial dysfunction, and impaired fatty acid oxidation, culminating in inadequate energy supply to neurons and exacerbating cellular damage.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Concept and mechanism of metabolic reprogramming", "text": "The distinctiveness of neuronal metabolic reprogramming is also underscored by its intricate relationship with the pathological mechanisms of neurodegenerative diseases. For instance, in AD, the accumulation of \u03b2-amyloid plaques and neurofibrillary tangles not only disrupts neuronal structure and function but also interferes with normal metabolic pathways, instigating metabolic reprogramming (123). Similarly, in PD, the aggregation of \u03b1-synuclein adversely affects mitochondrial function, triggering metabolic reprogramming that intensifies oxidative stress and neuronal damage (124).\n\nInvestigating the mechanisms of neuronal metabolic reprogramming is crucial for comprehending the fundamental nature of neurodegenerative diseases and formulating therapeutic interventions. The mechanisms involved encompass, but are not limited to, mitochondrial biogenesis and dysfunction, altered oxidative phosphorylation efficiency, modulation of glycolysis and tricarboxylic acid cycles, irregular fatty acid oxidation, and variations in nutrient utilization and demand by neurons. A thorough examination of these mechanisms will elucidate the specific contributions of metabolic reprogramming to neurodegenerative diseases, offering novel insights for diagnosis, prevention, and treatment strategies. We encapsulate the function of metabolic reprogramming in neurodegenerative disorders in Figure 2.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Concept and mechanism of metabolic reprogramming", "text": "The mechanism of metabolic reprogramming in neurodegenerative diseases. Glucose enters neurons through the GLUT4 transporter, is converted into glucose-6-phosphate and pyruvic acid through glycolysis, provides initial energy, and can be converted into glycogen for storage. High levels of mitochondrial stress led to the production of reactive oxygen species (ROS), activate the STING pathway, trigger an inflammatory response, and cause the death and abnormal movement of dopamine-secreting neurons, especially evident in Parkin-deficient mice. Microglia are inactive at rest, but inflammatory stimulation or brain injury can activate them, promoting the expression of inflammatory genes through the c/EBP\u03b2 pathway and leading to the formation of neurofibrillary tangles.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Regulation of metabolic reprogramming", "text": "The potential of metabolic reprogramming as a regulatory and therapeutic target offers extensive research opportunities within the domain of neurodegenerative diseases (125). Both nutritional and pharmacological interventions can significantly influence neuronal survival and functionality by modulating critical metabolic pathways, as well as regulating neuroinflammation and oxidative stress, thereby potentially decelerating or reversing neurodegenerative processes (18, 126).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Regulation of metabolic reprogramming", "text": "Nutritional interventions, particularly those involving diets abundant in antioxidants, omega-3 polyunsaturated fatty acids, and specific amino acids, exert a profound regulatory influence on metabolic reprogramming (127, 128). The Mediterranean diet, characterized by its rich variety of vegetables, fruits, whole grains, nuts, olive oil, and fish, has demonstrated efficacy in enhancing energy metabolism, mitigating oxidative stress and inflammation, and providing neuroprotective benefits (129). Antioxidants found in the Mediterranean diet, such as vitamin C, vitamin E, and selenium, play a crucial role in neutralizing free radicals, alleviating oxidative stress, and safeguarding neuronal integrity (130). Omega-3 polyunsaturated fatty acids, particularly DHA and EPA, not only diminish the synthesis of inflammatory mediators but also enhance mitochondrial functionality, improve neuronal membrane fluidity, facilitate signaling processes, and hold potential therapeutic value for neurodegenerative conditions (131). Tailored nutritional interventions, customized according to individual metabolic profiles and disease conditions, can more effectively modulate neuronal metabolism, decrease oxidative stress and inflammation, and foster neuroprotection by optimizing dietary components, such as increasing the intake of antioxidants and fatty acids. For instance, in patients with neurodegenerative diseases characterized by mitochondrial dysfunction, the supplementation of antioxidants and omega-3 fatty acids may yield superior neuroprotective outcomes (132).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Regulation of metabolic reprogramming", "text": "Pharmacological interventions, particularly those aimed at specific metabolic pathways, also exhibit promising regulatory effects on metabolic reprogramming. For example, insulin sensitizers like thiazolidinediones confer neuroprotective benefits in neurodegenerative diseases by enhancing insulin sensitivity and glucose metabolism (133). These agents facilitate glucose uptake and utilization, thereby improving the energy supply to neurons and potentially slowing disease progression. Antioxidants and anti-inflammatory medications, such as N-Acetylcysteine amide and non-steroidal anti-inflammatory drugs, provide protective effects against neurodegenerative diseases by diminishing oxidative stress and curtailing inflammation (134, 135). These pharmacological agents can lower the production of reactive oxygen species (ROS), inhibit the release of inflammatory mediators, and mitigate neuronal damage.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Regulation of metabolic reprogramming", "text": "The pivotal function of metabolic reprogramming in neurodegenerative pathologies underscores its significance as a therapeutic target. Notably, modulation of the AMPK/PGC-1\u03b1 signaling axis has been extensively investigated. Evidence indicates that Dihuang Yinzi enhances mitochondrial biogenesis via this pathway, thereby ameliorating cognitive deficits and mitochondrial ultrastructural damage in transgenic Alzheimer\u2019s disease models (136). Furthermore, Sestrin2 overexpression mitigates osteoarthritic pain and suppresses neuroinflammatory responses through activation of the AMPK/PGC-1\u03b1 pathway (137). These findings elucidate the therapeutic potential of targeting the AMPK/PGC-1\u03b1 signaling cascade in neurodegenerative and neuroinflammatory disorders.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutritional interventions and therapeutic strategies", "text": "Currently, there is an increasing focus on the role of nutritional interventions in the management of neurodegenerative diseases, highlighting their significant potential (138). Nutrients, particularly antioxidants, omega-3 polyunsaturated fatty acids, vitamins, minerals, and specific amino acids, exhibit neuroprotective properties by enhancing energy metabolism and mitigating oxidative stress and inflammation. The Mediterranean diet, abundant in these advantageous components, has demonstrated improvements in cognitive function and a reduction in disease symptoms among individuals with neurodegenerative conditions (139). Tailored nutritional intervention strategies, which adjust dietary plans to align with the unique metabolic requirements of patients, further amplify the efficacy of these interventions. Nonetheless, nutritional interventions encounter several challenges and limitations in the context of neurodegenerative disease treatment. Therefore, we summarized the advantages and challenges of nutritional therapy in the treatment of neurodegenerative diseases in Table 2, and we present customized nutritional intervention protocols depicted in Figure 3.\n\nAdvantages and challenges of nutritional intervention in the treatment of neurodegenerative diseases.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutritional interventions and therapeutic strategies", "text": "In the context of neurodegenerative diseases, the utilization of probiotics seeks to restore balance to the gut microbiome, thereby influencing central nervous system health via the microbial-gut-brain axis (140). Probiotics, including Lactobacillus and Bifidobacterium, exhibit neuroprotective properties by inhibiting pathogenic growth, enhancing intestinal barrier integrity, modulating immune responses, and alleviating oxidative stress (141). Furthermore, they play a crucial role in the synthesis and metabolism of neurotransmitters like serotonin and dopamine, which are vital for mood regulation, memory, and motor function (142). Postbiotic metabolites, derived from dead probiotics, continue to offer health benefits. In contrast to live probiotics, postbiotics demonstrate enhanced stability and safety, particularly following antibiotic therapy or in individuals with compromised immune systems. These metabolites, such as short-chain fatty acids (SCFAs), positively influence gut and brain health (143).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutritional interventions and therapeutic strategies", "text": "The strategy of combining nutrients and pharmaceuticals aims to synergistically amplify therapeutic outcomes, mitigate drug side effects, and enhance patient quality of life. We present a synthesis of nutritional interventions alongside treatment strategies for neurodegenerative diseases in Figure 4. Additionally, we compiled a summary of the emerging nutritional intervention strategies along with their mechanisms of action in Table 3. Advancements in targeted delivery systems for neurodegenerative disorders have garnered considerable research focus, particularly in the utilization of nanotechnology for precision therapeutics. The development of diverse nanocarriers\u2014including polymeric nanoparticles, liposomes, inorganic nanomaterials, and biomimetic vectors\u2014has enabled brain-specific drug delivery and localized release via stimulus-responsive nanocarrier platforms (144). Furthermore, mitochondrial-targeted nanotherapeutics demonstrate substantial promise in neurodegeneration management; employing carriers such as liposomes, DQAsomes, and polymeric nanoparticles enhances mitochondrial drug delivery, thereby increasing bioavailability and therapeutic efficacy (145). Recent progress has also been achieved in overcoming the blood\u2013brain barrier (BBB). Techniques such as ligand-functionalized nanoparticles, bispecific antibody shuttles, focused ultrasound-mediated BBB modulation, intranasal administration, exosomes, and mRNA lipid nanoparticles facilitate effective central nervous system drug delivery (146). The naso-brain delivery pathway, an innovative approach, enables direct transport of therapeutics to the brain via olfactory or trigeminal nerve pathways, bypassing the BBB, which enhances pharmacological efficacy and minimizes systemic adverse effects. This method supports the integration of nanotheranostic systems for more precise diagnosis and personalized treatment of neurodegenerative diseases (147).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutritional interventions and therapeutic strategies", "text": "Therapeutic strategies for neurodegenerative diseases combined with nutritional intervention.\n\nEmerging nutrition intervention strategies and their mechanisms of action.\n\nTo validate the efficacy of nutritional interventions for neurodegenerative diseases, it is imperative that clinical trials are meticulously structured to guarantee the precision and dependability of the findings. A well-defined research objective is essential, specifying the intended outcomes of the nutritional intervention, such as enhancing cognitive function, alleviating disease symptoms, or slowing disease progression. Subsequently, an appropriate control group must be established to evaluate the true impact of the intervention. This control group may consist of patients receiving standard treatment or those undergoing a simulated nutritional intervention (e.g., a placebo) to eliminate the effects of extraneous variables.\n\nFurthermore, the design of clinical trials must consider the intervention\u2019s duration, dosage, and the specific nutrients involved. The timing of the intervention should be in-formed by prior research or theoretical frameworks to identify the optimal moment for nutritional intervention. Dosage selection should rely on safety and efficacy data from earlier studies to ensure both safety and effectiveness. Additionally, the choice of nutrients should be guided by their mechanisms of action in neurodegenerative conditions and findings from previous investigations.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutritional interventions and therapeutic strategies", "text": "Recent evidence has further substantiated the potential of dietary interventions in modulating neurodegenerative pathways. For instance, Mafe and B\u00fcsselberg demonstrated that adherence to a Mediterranean diet (MedDiet) significantly altered gut microbiota composition and associated metabolite signatures in patients with mild cognitive impairment (MCI), suggesting a plausible mechanistic link between diet-induced microbial changes and attenuated Alzheimer\u2019s disease (AD) progression (148). Their findings align with our proposed framework that MedDiet-mediated metabolic reprogramming may exert neuroprotective effects via gut\u2013brain axis modulation.\n\nSimilarly, Orywal et al. provided comprehensive preclinical and clinical data supporting the role of B-group vitamins, particularly B6, B12, and folate, in mitigating oxidative stress and homocysteine-mediated neurotoxicity across AD and Parkinson\u2019s disease (PD) cohorts (149). These results reinforce our advocacy for personalized nutritional strategies that integrate micronutrient status assessment, especially in early-stage neurodegeneration.\n\nCollectively, these studies underscore the necessity of designing multi-arm, diet-centric clinical trials that not only assess cognitive endpoints but also incorporate multi-omics readouts (microbiome, metabolome, epigenome) to capture the full therapeutic trajectory of nutritional interventions.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Nutritional interventions and therapeutic strategies", "text": "To enhance the scientific rigor and applicability of the study, a multicenter, randomized, double-blind, placebo-controlled design is recommended. A multicenter approach can augment sample size and the generalizability of results, while randomization minimizes selection bias. The double-blind methodology helps to mitigate subjective influences from both investigators and participants, and placebo controls are essential for assessing the genuine effects of the intervention. Moreover, incorporating biomarkers as objective measures of intervention outcomes, such as neuronal metabolites and oxidative stress markers, can provide more robust evidence to support the findings. To ensure adequate statistical power, future trials should pre-calculate sample size on the basis of clinically relevant effect sizes of cognitive or imaging endpoints, typically requiring 50\u2013100 participants per arm for phase IIa studies. Core biomarker panels ought to include plasma A\u03b242/40 ratio, p-tau-181, neurofilament light chain, and short-chain fatty acids quantified by targeted metabolomics, which have shown responsiveness to Mediterranean diet or omega-3 interventions and correlate with cognitive change.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Emerging interventions integrate dietary modification, probiotics, and pharmacologic agents", "text": "Currently, there is an increasing focus on the role of nutritional interventions in the management of neurodegenerative diseases, highlighting their significant potential (138). Nutrients, particularly antioxidants, omega-3 polyunsaturated fatty acids, vitamins, minerals, and specific amino acids, exhibit neuroprotective properties by enhancing energy metabolism and mitigating oxidative stress and inflammation. The Mediterranean diet, abundant in these advantageous components, has demonstrated improvements in cognitive function and a reduction in disease symptoms among individuals with neurodegenerative conditions (139). Tailored nutritional intervention strategies, which adjust dietary plans to align with the unique metabolic requirements of patients, further amplify the efficacy of these interventions. Nonetheless, nutritional interventions encounter several challenges and limitations in the context of neurodegenerative disease treatment. Therefore, we summarized the advantages and challenges of nutritional therapy in the treatment of neurodegenerative diseases in Table 2, and we present customized nutritional intervention protocols depicted in Figure 3.\n\nAdvantages and challenges of nutritional intervention in the treatment of neurodegenerative diseases.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Emerging interventions integrate dietary modification, probiotics, and pharmacologic agents", "text": "In the context of neurodegenerative diseases, the utilization of probiotics seeks to restore balance to the gut microbiome, thereby influencing central nervous system health via the microbial-gut-brain axis (140). Probiotics, including Lactobacillus and Bifidobacterium, exhibit neuroprotective properties by inhibiting pathogenic growth, enhancing intestinal barrier integrity, modulating immune responses, and alleviating oxidative stress (141). Furthermore, they play a crucial role in the synthesis and metabolism of neurotransmitters like serotonin and dopamine, which are vital for mood regulation, memory, and motor function (142). Postbiotic metabolites, derived from dead probiotics, continue to offer health benefits. In contrast to live probiotics, postbiotics demonstrate enhanced stability and safety, particularly following antibiotic therapy or in individuals with compromised immune systems. These metabolites, such as short-chain fatty acids (SCFAs), positively influence gut and brain health (143).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Emerging interventions integrate dietary modification, probiotics, and pharmacologic agents", "text": "The strategy of combining nutrients and pharmaceuticals aims to synergistically amplify therapeutic outcomes, mitigate drug side effects, and enhance patient quality of life. We present a synthesis of nutritional interventions alongside treatment strategies for neurodegenerative diseases in Figure 4. Additionally, we compiled a summary of the emerging nutritional intervention strategies along with their mechanisms of action in Table 3. Advancements in targeted delivery systems for neurodegenerative disorders have garnered considerable research focus, particularly in the utilization of nanotechnology for precision therapeutics. The development of diverse nanocarriers\u2014including polymeric nanoparticles, liposomes, inorganic nanomaterials, and biomimetic vectors\u2014has enabled brain-specific drug delivery and localized release via stimulus-responsive nanocarrier platforms (144). Furthermore, mitochondrial-targeted nanotherapeutics demonstrate substantial promise in neurodegeneration management; employing carriers such as liposomes, DQAsomes, and polymeric nanoparticles enhances mitochondrial drug delivery, thereby increasing bioavailability and therapeutic efficacy (145). Recent progress has also been achieved in overcoming the blood\u2013brain barrier (BBB). Techniques such as ligand-functionalized nanoparticles, bispecific antibody shuttles, focused ultrasound-mediated BBB modulation, intranasal administration, exosomes, and mRNA lipid nanoparticles facilitate effective central nervous system drug delivery (146). The naso-brain delivery pathway, an innovative approach, enables direct transport of therapeutics to the brain via olfactory or trigeminal nerve pathways, bypassing the BBB, which enhances pharmacological efficacy and minimizes systemic adverse effects. This method supports the integration of nanotheranostic systems for more precise diagnosis and personalized treatment of neurodegenerative diseases (147).", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Emerging interventions integrate dietary modification, probiotics, and pharmacologic agents", "text": "Therapeutic strategies for neurodegenerative diseases combined with nutritional intervention.\n\nEmerging nutrition intervention strategies and their mechanisms of action.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Clinical trials and future directions", "text": "To validate the efficacy of nutritional interventions for neurodegenerative diseases, it is imperative that clinical trials are meticulously structured to guarantee the precision and dependability of the findings. A well-defined research objective is essential, specifying the intended outcomes of the nutritional intervention, such as enhancing cognitive function, alleviating disease symptoms, or slowing disease progression. Subsequently, an appropriate control group must be established to evaluate the true impact of the intervention. This control group may consist of patients receiving standard treatment or those undergoing a simulated nutritional intervention (e.g., a placebo) to eliminate the effects of extraneous variables.\n\nFurthermore, the design of clinical trials must consider the intervention\u2019s duration, dosage, and the specific nutrients involved. The timing of the intervention should be in-formed by prior research or theoretical frameworks to identify the optimal moment for nutritional intervention. Dosage selection should rely on safety and efficacy data from earlier studies to ensure both safety and effectiveness. Additionally, the choice of nutrients should be guided by their mechanisms of action in neurodegenerative conditions and findings from previous investigations.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Clinical trials and future directions", "text": "Recent evidence has further substantiated the potential of dietary interventions in modulating neurodegenerative pathways. For instance, Mafe and B\u00fcsselberg demonstrated that adherence to a Mediterranean diet (MedDiet) significantly altered gut microbiota composition and associated metabolite signatures in patients with mild cognitive impairment (MCI), suggesting a plausible mechanistic link between diet-induced microbial changes and attenuated Alzheimer\u2019s disease (AD) progression (148). Their findings align with our proposed framework that MedDiet-mediated metabolic reprogramming may exert neuroprotective effects via gut\u2013brain axis modulation.\n\nSimilarly, Orywal et al. provided comprehensive preclinical and clinical data supporting the role of B-group vitamins, particularly B6, B12, and folate, in mitigating oxidative stress and homocysteine-mediated neurotoxicity across AD and Parkinson\u2019s disease (PD) cohorts (149). These results reinforce our advocacy for personalized nutritional strategies that integrate micronutrient status assessment, especially in early-stage neurodegeneration.\n\nCollectively, these studies underscore the necessity of designing multi-arm, diet-centric clinical trials that not only assess cognitive endpoints but also incorporate multi-omics readouts (microbiome, metabolome, epigenome) to capture the full therapeutic trajectory of nutritional interventions.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Clinical trials and future directions", "text": "To enhance the scientific rigor and applicability of the study, a multicenter, randomized, double-blind, placebo-controlled design is recommended. A multicenter approach can augment sample size and the generalizability of results, while randomization minimizes selection bias. The double-blind methodology helps to mitigate subjective influences from both investigators and participants, and placebo controls are essential for assessing the genuine effects of the intervention. Moreover, incorporating biomarkers as objective measures of intervention outcomes, such as neuronal metabolites and oxidative stress markers, can provide more robust evidence to support the findings. To ensure adequate statistical power, future trials should pre-calculate sample size on the basis of clinically relevant effect sizes of cognitive or imaging endpoints, typically requiring 50\u2013100 participants per arm for phase IIa studies. Core biomarker panels ought to include plasma A\u03b242/40 ratio, p-tau-181, neurofilament light chain, and short-chain fatty acids quantified by targeted metabolomics, which have shown responsiveness to Mediterranean diet or omega-3 interventions and correlate with cognitive change.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Discussion", "text": "This review systematically elucidates the critical function of nutritional modulation and cellular metabolic reprogramming in neurodegenerative pathologies: dysregulation of glucose, lipid, and amino acid metabolism collectively precipitates energy deficits, oxidative stress, and neuroinflammatory responses, establishing a \u201cmetabolism-inflammation\u201d axis. Interventions such as the Mediterranean diet, omega-3 polyunsaturated fatty acids, and B vitamin supplementation can markedly enhance cognitive performance and attenuate peripheral inflammatory biomarkers by restoring mitochondrial bioenergetics, bolstering antioxidant defenses, and modulating the gut-brain axis, thereby forming a closed-loop mechanistic framework for the \u201cnutrition-metabolism-neuroprotection\u201d triad.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Discussion", "text": "Current evidence is limited by four primary deficiencies: \u2460 substantial heterogeneity across studies due to variable dosing, intervention durations, and baseline nutritional statuses, complicating meta-analyses; \u2461 small sample sizes, often fewer than 50 participants per group, lacking statistical power calculations and increasing the risk of Type II errors; \u2462 reliance on cognitive assessment scales such as MMSE and MoCA without concurrent biomarker evaluation (e.g., A\u03b242/40 ratios, phosphorylated tau-181, neurofilament light chain, short-chain fatty acids); \u2463 paucity of long-term safety and adherence data, with research predominantly focused on Alzheimer\u2019s disease, leaving mechanisms in other neurodegenerative disorders insufficiently characterized.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Discussion", "text": "Future research should prioritize multicenter, randomized, double-blind, placebo-controlled trials with pre-specified sample size calculations based on clinically meaningful endpoints (e.g., CDR-SB\u202f\u2265\u202f0.5 or hippocampal atrophy \u22653% annually), ensuring at least 50\u2013100 participants per arm in phase IIA studies. Furthermore, developing a standardized core biomarker panel comprising plasma A\u03b242/40 ratio, phosphorylated tau-181, neurofilament light chain (NfL), and targeted short-chain fatty acid metabolomics will function as dual endpoints for assessing therapeutic response and mechanistic validation. Integrating multi-omics data and artificial intelligence predictive modeling will facilitate personalized nutritional strategies, with extended follow-up periods of \u226524\u202fmonths to assess disease-modifying potential. Ultimately, the goal is to translate the concept of \u201cnutrition-driven metabolic reprogramming\u201d from observational phenomena into a clinically applicable intervention paradigm for neurodegenerative disease management.", "title": "" }, { "pmid": 41693955, "pmcid": "PMC12893954", "section": "Conclusion", "text": "Neurodegenerative disorders are increasingly understood as complex, multifactorial pathologies driven by metabolic dysregulation, oxidative stress, and neuroinflammatory processes. This review emphasizes the critical influence of nutritional status and cellular metabolic reprogramming on the etiology and progression of diseases such as AD, PD, and HD. Growing evidence indicates the therapeutic potential of dietary interventions\u2014particularly those enriched with antioxidants, omega-3 polyunsaturated fatty acids, and B vitamins\u2014in modulating neurobiological metabolic pathways, improving mitochondrial bioenergetics, and reducing neuroinflammation. Despite encouraging preclinical and clinical results, the clinical integration of nutritional therapeutic strategies faces challenges due to variability in study methodologies, lack of standardized intervention protocols, and inadequate biomarker incorporation. Future investigations should focus on large-scale, multicenter randomized controlled trials utilizing multi-omics technologies and personalized nutrition approaches. Tailoring dietary regimens to individual metabolic phenotypes holds promise for developing safe, effective, and sustainable adjunctive treatments in neurodegenerative disease management. Ultimately, elucidating the connections between nutrition and brain energy metabolism is essential for advancing mechanistic understanding and for establishing precision nutrition as a cornerstone of neurodegeneration therapeutics.", "title": "" }, { "pmid": "40522761", "pmcid": null, "section": "Abstract", "text": "With the gradual advancement of research methods and technologies, various biological processes have been identified as playing roles in the pathogenesis of neurodegenerative diseases. However, current descriptions of these biological processes do not fully explain the onset, progression, and development of these conditions. Therefore, exploration of the pathogenesis of neurodegenerative diseases remains a valuable area of research. This review summarizes the potential common pathogeneses of Alz", "title": "Potential common pathogenesis of several neurodegenerative diseases." }, { "pmid": "40650880", "pmcid": null, "section": "Abstract", "text": "Anti-IgLON5 disease is an autoimmune encephalitis that presents with a heterogenous clinical phenotype, including sleep disorders, movement abnormalities and bulbar involvement. It is characterized by autoantibodies against IgLON5, 85% association with HLA-DQB1*05:\u223c and a brainstem-dominant tauopathy. Cellular and murine models report pathogenic effects of the autoantibodies, and neurodegenerative factors suggest progressive atrophy as a common sequela. However, evidence from in vivo patient dat", "title": "Brain atrophy patterns in anti-IgLON5 disease." }, { "pmid": "41813136", "pmcid": null, "section": "Abstract", "text": "Amyotrophic lateral sclerosis (ALS) and Huntington disease (HD) are lethal neurodegenerative diseases affecting motor function. Though their etiology and pathology are distinct, recent evidence suggests commonalities between TAR DNA-binding protein (TDP-43), which is associated with 97% of ALS cases, and huntingtin (HTT), the causative protein of HD. ALS is a heterogeneous, lethal neurodegenerative disease characterized by the progressive loss of upper and lower motor neurons, as well as brainst", "title": "ALS and Huntington Disease: Unraveling the Connections between TDP-43 and Huntingtin." }, { "pmid": "40576748", "pmcid": null, "section": "Abstract", "text": "Neurodegenerative disorders, including Huntington's disease, Amyotrophic lateral sclerosis, Alzheimer's disease, and Parkinson's disease, create more challenges as the population gets older and there are no curative therapies available. Recent advances in gut microbiome research have spotlighted postbiotics and psychobiotics as innovative therapeutic strategies targeting the gut-brain axis to alleviate neurodegenerative symptoms and slow disease progression. Postbiotics, which are metabolites an", "title": "Innovative Interventions: Postbiotics and Psychobiotics in Neurodegenerative Disease Treatment." }, { "pmid": "41295902", "pmcid": null, "section": "Abstract", "text": "In this edition of the Huntington's Disease Clinical Trials Update, we expand on the launch of the phase II/III clinical trial of SKY-0515 from Skyhawk Therapeutics and the phase I/II clinical trial of SPK-10001 from Spark Therapeutics. We also report positive topline data from uniQure's phase I/II clinical trial of AMT-130 after 36 months of follow-up. Further updates include recent developments in Roche's tominersen programme within GENERATION HD2, progress with votoplam (PTC518) in PIVOT-HD b", "title": "Huntington's disease clinical trials update: October 2025." }, { "pmid": "41841355", "pmcid": null, "section": "Abstract", "text": "Huntington's disease (HD) is a progressive neurodegenerative disorder caused by a CAG-repeat expansion in the Huntington gene (HTT). Herein, we describe the discovery of a series of HTT pre-mRNA-splicing modulators that promote the inclusion of a cryptic stop codon that in turn lowers levels of mutant Huntington protein (mHTT). Optimization of the starting thienopyridine amide core resulted in the discovery of the potent, CNS-penetrant, selective, and orally bioavailable HTT-splicing modulator B", "title": "Discovery and Optimization of Thienopyrazine RNA-Splicing Modulators for the Treatment of Huntington's Disease." }, { "pmid": "41318997", "pmcid": null, "section": "Abstract", "text": "Although an emerging body of evidence suggests that abnormal forms of tau are present and contribute to Huntington's disease (HD)-a genetic neurodegenerative disorder primarily characterized by the aggregation of the mutant huntingtin (mHtt) protein affecting cognitive, motor, and psychiatric function-it is not clear to what extent this is relevant to the disease phenotype, and hence future treatments. We therefore generated a novel murine model by crossing heterozygous zQ175 knockin HD mice wit", "title": "The impact of tau deletion on Huntington's disease: An in vivo perspective." }, { "pmid": "41846052", "pmcid": null, "section": "Abstract", "text": "Huntington's disease (HD) is driven by expanded CAG repeats in the huntingtin gene, resulting in mutant huntingtin (mHtt) aggregation, mitochondrial dysfunction, neuronal loss, and neuroinflammation. Although stem cell-based therapies provide potent regenerative, anti-inflammatory, and neuroprotective effects, their clinical translation remains constrained by tumorigenicity and poor engraftment efficiency. Therefore, we investigated whether exosomes derived from human induced pluripotent stem ce", "title": "Human iPSC-derived exosomes for amelioration of Huntington's disease through mitochondrial, synaptic, and anti-apoptotic mechanisms." }, { "pmid": "41238102", "pmcid": null, "section": "Abstract", "text": "The human brain, despite accounting for only 2\u00a0% of total body weight, exhibits an exceptionally high lipid content (approximately 20\u00a0% of its mass), highlighting the critical role of lipid metabolism in maintaining neural homeostasis and function. Neurodegenerative diseases-including Alzheimer's disease (AD), Parkinson's disease (PD), stroke, Huntington's disease (HD), and amyotrophic lateral sclerosis (ALS)-are characterized by progressive neuronal dysfunction and myelin degeneration. These co", "title": "Targeting lipid metabolism in neurodegenerative diseases: From experimental to clinical." }, { "pmid": "41827217", "pmcid": null, "section": "Abstract", "text": "Recent years have seen rapid progress in biological treatments for genetic diseases, as well as conditions like type 1 diabetes that lack an obvious genetic component. The authors sought to explain why this progress has emerged at this particular moment. The best way to illustrate this is by showcasing a wide range of therapies targeting diverse diseases. This progress has been driven by technological advances in genetically modified CAR-T and CAR-NK cells (e.g., using CRISPR or transgenes), whi", "title": "The New Wave of Gene and Cell Therapies Across Diseases." }, { "pmid": "41783572", "pmcid": null, "section": "Abstract", "text": "Neurological disorders such as Alzheimer's disease (AD) and Parkinson's disease (PD) pose a serious global public health threat, with complex etiologies involving genetic, environmental, and metabolic factors. Current data indicate that the prevalence of these disorders is rapidly increasing with the aging population, resulting in a growing economic and healthcare burden worldwide. In recent years, the imbalance of copper homeostasis has been increasingly implicated in the pathogenesis of neurol", "title": "Copper Homeostasis and Cuproptosis in Neurological Disorders." }, { "pmid": "40421472", "pmcid": null, "section": "Abstract", "text": "While neuropsychiatric symptoms are common in Huntington's Disease (HD), there is a dearth of evidence about the effectiveness of psychotropic medication for treating behavioral and cognitive symptoms. This article systematically reviews and aggregates the evidence of the effects of antidepressants on individuals with early HD. A systematic review and random-effects meta-analysis of RCTs comparing antidepressants to placebo in individuals with HD was performed, with a focus on outcomes of execut", "title": "Use of antidepressants in early Huntington's Disease: a systematic review and meta-analysis." }, { "pmid": "41690900", "pmcid": null, "section": "Abstract", "text": "Huntington's disease (HD) is usually described as rare, tragic, and intractable. Yet, HD offers a strategically unique entry point for neuroscience. With its genetic clarity, relatively predictable course, and organized global community, HD provides the clearest path to advancing brain repair.", "title": "Huntington's disease is the best investment in neuroscience today." }, { "pmid": "41663779", "pmcid": null, "section": "Abstract", "text": "Neurodegenerative diseases like Alzheimer's, Parkinson's, Huntington's, and amyotrophic lateral sclerosis are growing more common worldwide, yet treatment is still poor. Conventional therapies can have unforeseen side effects, produce poor medication reactions, and take longer to work. This persistent treatment gap highlights the need for novel approaches to these disorders' complex distinctions. Pharmacogenomics, which examines how genetic differences affect drug response, is a promising new su", "title": "Tailoring treatments: pharmacogenomics in the management of neurodegenerative diseases." }, { "pmid": "41376158", "pmcid": null, "section": "Abstract", "text": "Adeno-associated viral vectors have become a leading choice for gene therapy in the central nervous system due to their safety profile, efficient neuronal transduction, and capacity for sustained transgene expression. We previously reported that AAV2-derived capsids developed using the BRAVE (Barcoded Rational AAV Vector Evolution) approach have enhanced retrograde transport properties in the rodent brain, compared with parental AAV2. Retrograde transport enables broader coverage of connected br", "title": "Administration of a barcoded AAV capsid library to the putamen of non-human primates identifies variants with efficient retrograde transport." }, { "pmid": "41770933", "pmcid": null, "section": "Abstract", "text": "Huntington's disease (HD) is a fatal neurodegenerative disorder caused by inheriting an expanded CAG repeat tract in the huntingtin gene (", "title": "Huntington's disease LIG1 modifier variant increases ligase fidelity and suppresses somatic CAG repeat expansion." }, { "pmid": "41622913", "pmcid": null, "section": "Abstract", "text": "The mutation that causes Huntington's disease is a CAG repeat expansion in exon 1 of the huntingtin gene (HTT) that leads to an abnormally long polyglutamine tract in the huntingtin protein (HTT). Mutant CAG repeats are unstable and increase in size in specific neurons and brain regions with age, a phenomenon that constitutes the first step in the pathogenesis of the disease. In the presence of an expanded CAG repeat, cryptic polyA sites in intron 1 of the HTT pre-mRNA can become activated leadi", "title": "The HTT1a protein initiates HTT aggregation in a knock-in mouse model of Huntington's disease." }, { "pmid": "41755664", "pmcid": null, "section": "Abstract", "text": "Huntington's disease (HD) is a neurodegenerative disorder triggered by an unstable expansion of CAG repeats in the coding sequence of the ", "title": "Contribution of neuroepigenetics to HD - developmental and aging-related signatures." }, { "pmid": "41841481", "pmcid": null, "section": "Abstract", "text": "Since 1993, when a polymorphic CAG trinucleotide repeat was established as the cause of Huntington's disease (HD), fragment analysis has been the first-tier test in genetic diagnosis of HD. However, in recent years it has become increasingly clear that, in addition to the repeat length itself, sequence variations within the Sequence interruptions or non-canonical repeat motifs have historically been difficult to identify, as PCR cannot distinguish an interrupted repeat tract when assessed by ca", "title": "Challenges facing genetic diagnostics of Huntington's disease: an update." }, { "pmid": "41849583", "pmcid": null, "section": "Abstract", "text": "Lowering ", "title": "Lowering the " }, { "pmid": "41849580", "pmcid": null, "section": "Abstract", "text": "Huntington's disease (HD) arises from the toxic gain of function caused by a CAG expansion in the coding region of the ", "title": "Selective targeting of mutant " }, { "pmid": "41738896", "pmcid": null, "section": "Abstract", "text": "Neurological diseases such as Alzheimer's disease, Parkinson's disease, anxiety, depression, and Huntington's disease pose significant challenges due to their complex pathophysiology and limited treatment options. Traditional mammalian models, although widely used, often face limitations in cost, ethical concerns, and scalability. Zebrafish have become an excellent animal model for neurological research in recent years because of their remarkable genetic similarity to humans, well-preserved neur", "title": "Zebrafish as a Versatile Screening Model for Neurological Diseases: Insights Into Biology, Drug Delivery and Therapeutic Discovery." }, { "pmid": "41879495", "pmcid": null, "section": "Abstract", "text": "Neurodegenerative disorders such as Alzheimer's disease, Parkinson's disease, amyotrophic lateral sclerosis, and Huntington's disease involve progressive neuronal dysfunction driven by mechanisms including protein aggregation, oxidative stress, mitochondrial impairment, and neuroinflammation. Current treatments are largely symptomatic, with limited disease-modifying options. Advances in medicinal chemistry have led to the development of small-molecule derivatives targeting specific pathological ", "title": "Emerging Roles of Small-molecule Derivatives in Modulating Neurodegenerative Pathways: From Molecular Targets to Therapeutic Applications." }, { "pmid": "41639620", "pmcid": null, "section": "Abstract", "text": "Computational approaches involving complex data structures (e.g. machine learning, knowledge graphs) have been more prominent in biological studies for the last two decades. Due to increasingly larger amounts of data collected with modern omics techniques, there is a need for methods that can process such data quickly and thoroughly. In addition, those techniques can be applied to extrapolate results from a limited number of observations. Rare disease research benefits particularly from those ne", "title": "Design and evaluation of semantically-valid negative samples integration techniques for scalable semi-automated drug repurposing prediction pipelines in rare disease research." }, { "pmid": "41492395", "pmcid": null, "section": "Abstract", "text": "Cognitive decline is a core feature of Huntington's disease (HD), often preceding motor symptoms and progressing with disease severity. While several neuropsychological tests track cognitive changes, few studies have examined the biological correlates of brief screening tools adapted for HD. This study investigates the neuroanatomical and fluid biomarker correlates of performance on the Parkinson's Disease-Cognitive Rating Scale (PD-CRS), aiming to validate it as a clinically and biologically gr", "title": "Neural and biomarker correlates of the Parkinson's Disease-Cognitive Rating Scale in Huntington's disease." }, { "pmid": "41699789", "pmcid": null, "section": "Abstract", "text": "Huntington's disease (HD) is a progressive neurodegenerative disorder characterized by motor dysfunction, cognitive decline, and psychiatric disturbances. It is caused by the abnormal expansion of cytosine-adenine-guanine triplet repeats in the HTT gene, leading to the production of mutant huntingtin (mHTT) protein. This protein contributes to neuronal dysfunction and cell death, particularly in the striatum and cortex. The review examines recent research focused on identifying and assessing pot", "title": "Huntington's Disease: An Overview of Biomarkers in Prognostic, Diagnostic, and Therapeutic Management." }, { "pmid": "41527739", "pmcid": null, "section": "Abstract", "text": "The dysregulation of phosphorylation networks plays a critical role in the pathogenesis of a wide spectrum of brain disorders, including Alzheimer's disease, Parkinson's disease, Huntington's disease, amyotrophic lateral sclerosis, traumatic brain injury, ischemic stroke, drug abuse, major depressive disorder, and schizophrenia. Notably, phosphorylated isoforms of microtubule-associated protein tau and neurofilament heavy polypeptide are already utilized in clinical diagnostics, highlighting the", "title": "Insights into phosphoproteomic studies and prospects of phosphoproteins as biomarkers for brain disorders." }, { "pmid": "41828602", "pmcid": null, "section": "Abstract", "text": "Huntington's disease (HD) is characterized by progressive striatal atrophy and complex proteomic changes in the central nervous system. Using the ultrasensitive Next-Gen Ultra-Sensitive Immunoassay (NULISA) proteomic platform, we analyzed cerebrospinal fluid (CSF) from 88 persons with HD to dissect the biological correlates of gray matter loss. Our findings reveal a distinct \"Two-Track\" model of pathology. The first track, marked by the axonal damage protein neurofilament light chain (NEFL), sho", "title": "A Two-Track Model of Huntington's Disease Pathology: Striatal Atrophy Mediates Maladaptive Immune Dysregulation." }, { "pmid": "41762390", "pmcid": null, "section": "Abstract", "text": "The Unified Multiple System Atrophy Rating Scale (UMSARS) is widely used as an outcome measure in MSA trials, but it has limitations\u00a0for clinical trial use. To address these, we developed the Multiple System Atrophy Combined Outcome Assessment (MuSyCA), a comprehensive multimodal tool for disease-modifying MSA\u00a0trials. The purpose of this manuscript is to describe the development and validation plan for MuSyCA, with emphasis on its structure, intended use, and assessment of reliability, validity,", "title": "Multiple System Atrophy Combined Outcome Assessment (MuSyCA): process, format, and validation plan." }, { "pmid": "41406849", "pmcid": null, "section": "Abstract", "text": "Molecular diagnostics have transformed clinical microbiology by enabling rapid, accurate, and highly sensitive detection of infectious agents, significantly improving patient outcomes and public health response. Over the past decade, advances in polymerase chain reaction (PCR), next-generation sequencing (NGS), digital PCR, isothermal amplification, and CRISPR-based assays have enhanced pathogen identification, antimicrobial resistance profiling, and outbreak investigation. These technologies ha", "title": "Molecular diagnostics in clinical microbiology: Advances, applications, and future directions." }, { "pmid": "41674784", "pmcid": null, "section": "Abstract", "text": "CRISPR-Cas genome-editing technologies have emerged as powerful tools for precise DNA and RNA modulation, offering promising therapeutic strategies for neurodegenerative disorders such as Alzheimer's disease (AD), Parkinson's disease (PD), Huntington's disease (HD), and amyotrophic lateral sclerosis (ALS). This review critically evaluates current CRISPR/Cas applications in neurodegeneration, with emphasis on mechanistic insights, therapeutic outcomes, and translational feasibility. Preclinical a", "title": "CRISPR-Cas technologies in neurodegenerative disorders: mechanistic insights, therapeutic potential, and translational challenges." }, { "pmid": "41840824", "pmcid": null, "section": "Abstract", "text": "Incurable hereditary diseases such as Duchenne muscular dystrophy (DMD), Huntington's disease (HD), and myotonic dystrophy type 1 (DM1) fall into the nucleotide repeat expansion disorder (NRED) category. The discovery of CRISPR-Cas genome editing has paved the way toward hopeful strategies for accurate DNA-level repair. This systematic review presents preclinical data on the efficacy, molecular effects, and limitations of CRISPR-based treatments for NREDs. As per Preferred Reporting Items for Sy", "title": "CRISPR Gene Editing for Nucleotide Repeat Expansion Disorders: A Systematic Review of Preclinical and Clinical Evidence." }, { "pmid": "41784089", "pmcid": null, "section": "Abstract", "text": "BackgroundReliable detection of huntingtin (HTT) is essential for understanding Huntington's disease (HD) biology and for evaluating therapeutic strategies. However, high-quality monoclonal antibodies (mAbs) against the HTT C-terminal domain remain limited.ObjectiveWe sought to generate and validate novel monoclonal antibodies targeting the HTT C-terminal HEAT-containing domain to better detect HTT independently of potential effects of polyglutamine length that can impact some N-terminally targe", "title": "Novel monoclonal antibodies against the C-terminal HEAT domain of huntingtin." }, { "pmid": "41708002", "pmcid": null, "section": "Abstract", "text": "Protein-encoding nucleotide repeat expansion diseases, including polyglutamine (polyQ) and polyglycine (polyG) diseases, are characterized by the accumulation of aggregation-prone proteins. In the polyQ diseases, including Huntington's disease and several spinocerebellar ataxias, substantial prior evidence supports a pathogenic role for mutant polyQ-expanded protein misfolding and aggregation, with molecular chaperones showing promise in suppressing disease phenotypes in cellular and animal mode", "title": "The Hsp40 cochaperone DNAJC7 regulates polyglutamine aggregation and exhibits context-dependent effects on polyglycine aggregation." }, { "pmid": "41849610", "pmcid": null, "section": "Abstract", "text": "Huntington's disease (HD) is a monogenic autosomal dominant neurodegenerative disorder caused by a CAG repeat expansion in exon 1 of the ", "title": "Self-inactivating AAV-CRISPR at different ages enables sustained amelioration of Huntington's disease deficits in BAC226Q mice." }, { "pmid": "41618467", "pmcid": null, "section": "Abstract", "text": "Neurodegenerative disorders involve the gradual breakdown of neurons, leading to problems with thinking, movement, and mental health. More evidence is emerging about the important roles of non-coding RNAs (ncRNAs). CircRNAs are a type of ncRNA formed through back-splicing and are widely present in the mammalian brain. They play crucial roles in brain development and mainly regulate gene expression and post-transcriptional processes by acting as molecular sponges for miRNAs and RBPs. eRNAs are an", "title": "Emerging roles of circular RNAs and enhancer RNAs: new insights into the development and management of neurodegenerative disorders." }, { "pmid": "41833626", "pmcid": null, "section": "Abstract", "text": "Neurodegenerative diseases (NDs), including Alzheimer's, Parkinson's, Huntington's, amyotrophic lateral sclerosis, and multiple sclerosis, share a common pathogenic signature: disrupted proteostasis driven by impaired autophagy and maladaptive exosome dynamics. Under normal conditions, autophagy maintains neuronal homeostasis by clearing misfolded proteins and damaged organelles, while exosomes mediate neuroglial communication. When autophagic flux is impaired or lysosomal function is compromise", "title": "Autophagy-exosome crosstalk in neurodegeneration: Mechanisms and therapeutic opportunities." }, { "pmid": "41407165", "pmcid": null, "section": "Abstract", "text": "Nutraceuticals, bioactive compounds derived from food sources, are emerging as promising agents for neuroprotection, particularly through their modulation of gut health. Unlike conventional single-molecule therapeutics that often target isolated pathways, nutraceuticals offer a multi-targeted approach by influencing the gut-brain axis, a bidirectional communication network linking the gut microbiota and the central nervous system. Key nutraceuticals such as probiotics, prebiotics, polyphenols, o", "title": "Nutraceutical interventions for neuroprotection: a comprehensive review." }, { "pmid": "41353868", "pmcid": null, "section": "Abstract", "text": "Neuroinflammation is a hallmark of many neurodegenerative diseases, including Alzheimer's, Parkinson's and Huntington's disease, multiple sclerosis, and infantile neuronal ceroid lipofuscinosis. Dynamic protein S-acylation, a reversible lipid post-translational modification, is an important regulator in these processes. S-acylation is catalysed by the zDHHC palmitoyl acyltransferases, and removal of the acyl groups is mediated by acyl-protein thioesterases. S-acylation controls the localisation,", "title": "S-acylation and neuroinflammation: the therapeutic potential of zDHHC and deacylase modulation." }, { "pmid": "41689141", "pmcid": null, "section": "Abstract", "text": "Mitochondrial DNA (mtDNA) plays a pivotal role in the regulation of neuroinflammation, acting as a potent trigger of innate immune responses when released into the cytoplasm or extracellular space. mtDNA is structurally similar to bacterial DNA, containing unmethylated CpG motifs that are readily recognized by immune sensors. Under conditions of cellular stress, injury, or mitochondrial dysfunction, mtDNA can escape into the cytoplasm, where it activates the cGAS (cyclic GMP-AMP synthase)-STING ", "title": "Mitochondrial DNA: a molecular switch driving sterile neuroinflammation." }, { "pmid": "41699331", "pmcid": null, "section": "Abstract", "text": "Neurodegenerative diseases, specifically Alzheimer's disease (AD), Parkinson's disease (PD), Huntington's disease (HD), and Amyotrophic Lateral Sclerosis (ALS) are defined by progressively increased neuronal loss that lacks curative therapies. Increasing evidence supports that non-canonical regulated cell death pathways including ferroptosis, necroptosis, pyroptosis, and parthanatos, are implicated in pathological mechanisms of neuroinflammation, and oxidative stress, and mitochondrial dysfuncti", "title": "Non-canonical cell death in neurodegeneration: emerging mechanisms and therapeutic Frontiers." }, { "pmid": "41672134", "pmcid": null, "section": "Abstract", "text": "In this review we explore the potential of P2X7 receptor blockers to elicit neuroprotection. This conjecture is based on a reasonably well-established role of this receptor in activating glial cells to maintain a chronic low-level neuroinflammatory state in the brain of patients suffering some neurodegenerative diseases (NDDs). In this context we briefly discuss evidence supporting the role of P2X7 receptors (P2X7) in the pathogenesis of Alzheimer's disease, Parkinson's disease, amyotrophic late", "title": "P2X7 receptors as targets for neuroprotection." }, { "pmid": "41693955", "pmcid": null, "section": "Abstract", "text": "Neurodegenerative diseases, characterized by their insidious onset and progressive neuronal degeneration, present significant challenges in the fields of neuroscience and medicine. We elucidate the critical role of nutrition and cellular metabolism in the pathogenesis and progression of these disorders, with a particular focus on Alzheimer's disease (AD), Parkinson's disease (PD), and Huntington's disease (HD). We demonstrate that fundamental nutrients such as glucose, lipids, and amino acids ar", "title": "Dietary and metabolic reprogramming alleviates neurodegeneration: a review of mechanisms and clinical implications." }, { "pmid": "41636944", "pmcid": null, "section": "Abstract", "text": "Neurodegenerative diseases impose a substantial and growing global burden, affecting millions worldwide and leading to high medical, social, and economic costs. These are characterized by progressive neuronal dysfunction and loss, leading to cognitive, motor, and behavioral impairments. Neurodegenerative diseases such as Alzheimer's disease, Parkinson's disease, and Huntington's disease are driven by intertwined mechanisms of oxidative stress, neuroinflammation, protein aggregation, and neuronal", "title": "Phloretin as a Multitarget Neuroprotective Agent: Mechanistic Insights into the Modulation of Oxidative Stress, Inflammation, and Apoptosis." }, { "pmid": "41787540", "pmcid": null, "section": "Abstract", "text": "Randomized, double-blind, placebo-controlled trials are the gold standard for evaluating treatment effects, as they minimize biases. However, they raise ethical concerns regarding participant exposure to potentially ineffective interventions. For rare neurodegenerative diseases, these trials are further challenged by the limited number of patients. Our goal is to leverage recent advancements in statistical methodologies to enhance the statistical power of these trials. We aim to test these metho", "title": "How to improve statistical power in a trial with SCA2 patients using natural history data." } ], "built": "2026-03-29T11:11:22.057974", "stats": { "total_papers": 46, "full_text_papers": 20, "abstract_only_papers": 26, "total_chunks": 1571, "total_characters": 1899603 } }