Shared genetic architecture between DTI‐ALPS traits and neurodegenerative diseases Authors: Hong Zhou, Jialin Deng, Qian Wu, Yule Zeng, Tianyao Wang, Jiayi Wu, Xiaomei Duan, Beisha Tang, Xinhua Wei, Lu He, Feng Liu Published: 2026 ABSTRACT AbstractINTRODUCTIONDiffusion tensor image analysis along the perivascular space (DTI‐ALPS) index is associated with neurodegenerative diseases (NDDs), but its shared genetic basis with NDDs remains unclear.METHODSBy integrating genome‐wide association datasets for three DTI‐ALPS traits and seven NDDs, we quantified polygenic overlap using MiXeR, identified shared loci using conditional and conjunctional false discovery rate analyses, and performed gene mapping, enrichment, temporal expression, and transcriptome‐wide association analyses.RESULTSDTI‐ALPS traits showed widespread but heterogeneous polygenic overlap with NDDs. We identified 22 shared loci, including novel associations implicating GAK and SIAH3, with the strongest convergence at 17q21.31. Shared loci mapped to 183 protein‐coding genes enriched in the endolysosomal system and microtubule cytoskeleton. These genes showed similar temporal expression patterns, and 45 were associated with both DTI‐ALPS traits and NDDs.DISCUSSIONThese findings reveal a shared genetic architecture between DTI‐ALPS traits and NDDs, highlighting mechanisms that may contribute to their overlap.Keywords: cross‐trait genetic analysis, diffusion tensor image analysis along the perivascular space, glymphatic system, neurodegenerative diseases, polygenic architecture BACKGROUND Neurodegenerative diseases (NDDs), including Alzheimer's disease (AD), Parkinson's disease (PD), amyotrophic lateral sclerosis (ALS), multiple sclerosis (MS), frontotemporal dementia (FTD), Huntington's disease (HD), and Lewy body dementia (LBD), impose a substantial and growing global health burden. 1 , 2 Traditional pathogenic models have emphasized protein aggregation, mitochondrial dysfunction, impaired proteostasis, and neuroinflammation as key contributors to disease progression. 3 , 4 , 5 However, these mechanisms do not fully explain the convergence of pathological features observed across clinically distinct NDDs. 3 , 4 , 5 Increasing evidence suggests that alterations in brain fluid transport and metabolite clearance may represent an additional component of neurodegenerative vulnerability. 6 , 7 The glymphatic system, a brain‐wide perivascular pathway facilitating cerebrospinal fluid (CSF) and interstitial fluid exchange, has been proposed to contribute to the clearance of amyloid beta (Aβ), tau, α‐synuclein, and other metabolic byproducts. 8 , 9 Through these functions, glymphatic transport is thought to support brain homeostasis and waste removal. 10 , 11 Consequently, disruption of brain fluid transport and clearance processes has emerged as a potential mechanism contributing to protein accumulation, neuroinflammation, and progressive neuronal dysfunction. 12 , 13 Diffusion tensor image analysis along the perivascular space (DTI‐ALPS) has emerged as one of the most widely used neuroimaging approaches for investigating glymphatic‐related processes in vivo. 14 , 15 The ALPS index is derived from diffusion measurements in periventricular white matter regions where major fiber orientations are relatively orthogonal to perivascular pathways, thereby providing an imaging measure of diffusion characteristics along perivascular spaces. 14 , 15 Although originally proposed as a surrogate marker of glymphatic function, DTI‐ALPS more directly reflects perivascular diffusion properties and may capture a combination of biological processes related to brain fluid transport, white matter microstructure, vascular factors, and tissue integrity. 16 , 17 Nevertheless, a growing body of evidence has demonstrated associations between ALPS alterations and NDDs. 18 , 19 , 20 , 21 In AD, lower ALPS values have been associated with hippocampal atrophy, amyloid burden, and cognitive decline. 18 In PD, reduced ALPS values have been linked to dopaminergic dysfunction and motor impairment. 19 In ALS, decreased ALPS values are associated with clinical disability and sleep disturbances, 20 whereas in MS, ALPS abnormalities have been linked to demyelination and neurodegeneration. 21 Although these findings suggest that ALPS‐related imaging abnormalities may represent a common feature across multiple NDDs, 18 , 19 , 20 , 21 the biological mechanisms underlying these associations remain incompletely understood. Recent genome‐wide association studies (GWASs) have begun to characterize the genetic architecture of DTI‐ALPS traits, identifying loci and candidate genes involved in neurodevelopment and brain structure and further suggesting that ALPS‐associated genetic variation may be linked to neurodegeneration‐related biomarkers, such as CSF tau levels. 22 , 23 These observations raise the possibility that ALPS‐related imaging phenotypes and NDDs may share common genetic mechanisms. 22 , 23 However, existing studies have not systematically examined genetic sharing across multiple NDDs. Given the pronounced polygenicity and pleiotropy of both ALPS‐related traits and NDDs, 22 , 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 more powerful statistical frameworks are needed to comprehensively characterize their shared genetic architecture. The bivariate causal mixture model (MiXeR) quantifies genome‐wide polygenic overlap independent of overall genetic correlation, 31 , 32 whereas the conditional and conjunctional false discovery rate (cond/conjFDR) framework leverages cross‐trait enrichment to identify pleiotropic loci with substantially improved power. 33 , 34 Together, these approaches enable comprehensive assessment of both genome‐wide and locus‐specific genetic sharing between complex traits. In this study, we systematically investigated the shared genetic architecture between three DTI‐ALPS traits (mean, left, and right ALPS indices) and seven major NDDs using large‐scale GWAS summary statistics. We first quantified genome‐wide polygenic overlap using MiXeR and subsequently identified shared loci through cond/conjFDR analyses. To interpret the biological significance of these findings, we performed functional annotation, gene mapping, enrichment analyses, temporal expression profiling, and transcriptome‐wide association analyses. By integrating statistical genetics with neuroimaging‐derived ALPS phenotypes, this study aims to characterize the molecular pathways underlying the overlap between DTI‐ALPS traits and NDDs and to provide new insights into the biological processes linking perivascular diffusion‐related imaging signatures and neurodegenerative vulnerability. An overview of the study design is shown in Figure 1. Overview of study workflow and analytical framework. This study systematically investigated the shared genetic architecture between DTI‐ALPS traits derived from diffusion tensor imaging – mALPS, lALPS, and rALPS – and seven major neurodegenerative diseases (AD, PD, ALS, MS, FTD, HD, and LBD). The DTI‐ALPS index is calculated as the ratio of diffusivities perpendicular to fiber bundles and parallel to veins (Dxproj and Dxassoc) over those perpendicular to both fiber bundles and veins (Dyproj and Dzassoc). Using MiXeR, we first quantified genome‐wide polygenic overlap across trait pairs, followed by the identification of novel and shared loci via cond/conjFDR analyses. Candidate SNPs were functionally annotated and mapped to genes using positional, eQTL, and chromatin interaction approaches, and GO enrichment analyses were performed to characterize biological pathways. Finally, temporal gene‐expression trajectories and expression‐trait associations were analyzed to elucidate the molecular mechanisms linking ALPS traits to neurodegeneration. AD, Alzheimer's disease; ALPS, analysis along the perivascular space; ALS, amyotrophic lateral sclerosis; cond/conjFDR, conditional/conjunctional false discovery rate; eQTL, expression quantitative trait locus; FTD, frontotemporal dementia; GO, Gene Ontology; HD, Huntington's disease; LBD, Lewy body dementia; lALPS, left ALPS index; mALPS, mean ALPS index; MiXeR, causal mixture model; MS, multiple sclerosis; PCW, postconceptional week; PD, Parkinson's disease; PY, postnatal year; rALPS, right ALPS index; SNP, single nucleotide polymorphism. RESEARCH IN CONTEXT Systematic review: The authors searched PubMed for genome‐wide association studies of DTI‐ALPS traits and neurodegenerative diseases. Although prior studies identified several risk loci for DTI‐ALPS traits and neurodegenerative diseases, the shared genetic architecture between them remains unclear. Systematic review: The authors searched PubMed for genome‐wide association studies of DTI‐ALPS traits and neurodegenerative diseases. Although prior studies identified several risk loci for DTI‐ALPS traits and neurodegenerative diseases, the shared genetic architecture between them remains unclear. Interpretation: Our findings move beyond phenotypic and imaging associations by showing that DTI‐ALPS traits share genetic architecture with neurodegenerative susceptibility. The recurrent 17q21.31 signal highlights a major susceptibility hub, while enrichment of shared genes in the endolysosomal system and microtubule cytoskeleton points to potential biological mechanisms underlying this overlap. Similar developmental expression trajectories and transcriptome‐based associations further support the biological relevance of these loci. These findings suggest that ALPS‐related imaging signatures are genetically connected to neurodegenerative vulnerability rather than representing isolated imaging correlates. Interpretation: Our findings move beyond phenotypic and imaging associations by showing that DTI‐ALPS traits share genetic architecture with neurodegenerative susceptibility. The recurrent 17q21.31 signal highlights a major susceptibility hub, while enrichment of shared genes in the endolysosomal system and microtubule cytoskeleton points to potential biological mechanisms underlying this overlap. Similar developmental expression trajectories and transcriptome‐based associations further support the biological relevance of these loci. These findings suggest that ALPS‐related imaging signatures are genetically connected to neurodegenerative vulnerability rather than representing isolated imaging correlates. Future directions: These findings should be validated in larger, multi‐ancestry datasets and confirmed through mechanistic studies. Future directions: These findings should be validated in larger, multi‐ancestry datasets and confirmed through mechanistic studies. METHODS GWAS data Summary statistics for the three DTI‐ALPS traits, including mean ALPS (mALPS), left ALPS (lALPS), and right ALPS (rALPS), were obtained from a recent ALPS GWAS of 31,021 participants from the UK Biobank cohort. 22 Summary statistics for seven NDDs were retrieved from published GWASs, 24 , 25 , 26 , 27 , 28 , 29 , 30 including AD (40,918 cases, 358,140 controls), ALS (27,205 cases, 110,881 controls), FTD (3756 cases, 11,233 controls), MS (14,802 cases, 26,703 controls), PD (15,056 cases, 12,637 controls), HD (9064 cases), and LBD (2591 cases, 4027 controls). In addition, the HD summary statistics were derived from an age‐at‐onset modifier GWAS. 25 Because the ALPS GWAS was entirely derived from UK Biobank participants, individuals from the UK Biobank were removed from all NDD datasets to avoid sample overlap, thereby ensuring statistical independence across GWAS datasets and minimizing potential inflation of cross‐trait association signals. All datasets were restricted to individuals of European ancestry, and additional details of the GWAS datasets are provided in Table S1 and Supplementary Methods. Univariate and bivariate MiXeR analyses MiXeR was used to quantify the genome‐wide polygenic overlap between the three DTI‐ALPS traits (mALPS, lALPS, and rALPS) and seven NDDs. 31 , 32 First, we conducted univariate MiXeR analyses to estimate polygenicity, discoverability, and heritability for each trait. 31 Polygenicity was defined as the number of trait‐influencing variants required to explain 90% of single nucleotide polymorphism (SNP)‐based heritability. Discoverability represented the variance of additive genetic effects, with higher values indicating that a trait carries variants with larger effect sizes. Heritability, derived from both parameters, reflects the proportion of the phenotypic variance attributable to genetic factors. Second, based on the parameter estimates obtained from the univariate analyses, bivariate MiXeR analyses were performed to quantify the extent of polygenic overlap between DTI‐ALPS traits and NDDs. 32 In this framework, the additive genetic effects of two traits are modeled as a mixture of bivariate normal components representing variants that influence neither trait, one trait, or both traits. MiXeR estimates the numbers of trait‐specific and shared variants, as well as the proportion of shared variants with concordant effect directions. The degree of overlap was quantified using the Dice coefficient, which measures the proportion of shared variants relative to the total number of trait‐influencing variants across both traits. All linkage disequilibrium (LD) estimates were calculated using the 1000 Genomes Project Phase 3 European reference panel. 35 To minimize potential bias from a complex LD structure, SNPs within the major histocompatibility complex (MHC) region (chr6: 26 to 34 Mb) were excluded from all MiXeR analyses. Detailed statistical descriptions of MiXeR are provided in the Supplementary Methods. Cond/conjFDR analyses The condFDR framework, grounded in an empirical Bayesian approach, leverages association signals from a secondary trait to refine the statistical significance of SNPs in a primary trait, thereby increasing statistical power beyond conventional genome‐wide significance thresholds. 33 , 34 For each ALPS‐NDD pair, condFDR analyses were performed in both directions: (1) conditioning on each NDD to enhance the discovery of DTI‐ALPS‐associated variants and (2) conditioning on each DTI‐ALPS trait to improve the discovery of NDD‐associated variants. In each case, SNPs from the primary phenotype were re‐ranked according to their association strength with the secondary phenotype, and conditional quantile‐quantile (Q‐Q) plots were generated to evaluate cross‐trait enrichment. Increasing leftward deflection from the null line across secondary‐trait significance strata (e.g., p < 0.1, p < 0.01, and p < 0.001) was interpreted as evidence of stronger enrichment. Subsequently, conjFDR analysis was applied to identify shared genomic loci jointly associated with both traits. 33 , 34 The conjFDR value – defined as the maximum of the two corresponding condFDR values – provides a conservative estimate of the probability that a given SNP is associated with both traits simultaneously. Following previous studies, 36 , 37 the significance thresholds were set at 0.01 for condFDR and 0.05 for conjFDR. To mitigate potential confounding from complex LD patterns, SNPs within the MHC region (chr6: 26 to 34 Mb) and the 8p23.1 region (chr8: 7.2 to 12.5 Mb) were excluded from all cond/conjFDR analyses. 38 Additional information on the statistical framework of cond/conjFDR analysis is provided in the Supplementary Methods. Genomic locus definition Independent genomic loci for each trait pair identified by cond/conjFDR analyses were defined following the Functional Mapping and Annotation (FUMA) framework. 39 Specifically, SNPs were considered independent significant variants if they met the thresholds of condFDR < 0.01 and conjFDR < 0.05 in the respective analyses and were not in LD (r 2 < 0.6) with each other. 37 Lead SNPs were defined as a subset of independently significant variants that were mutually independent at r 2 < 0.1, while candidate SNPs referred to variants in LD (r 2 > 0.6) with at least one independent significant SNP. 40 All LD estimates were based on the 1000 Genomes Project Phase 3 European reference panel. 35 The boundary of each locus was delineated by including all corresponding candidate SNPs; within each cond/conjFDR analysis, loci located less than 250 kb apart were merged into a single locus, and the variant with the lowest cond/conjFDR value was retained as the final lead variant. 41 For loci identified through condFDR analyses, we applied two hierarchical merging strategies. First, for each NDD phenotype, loci obtained from its combinations with the three ALPS indices (e.g., AD‐mALPS, AD‐lALPS, and AD‐rALPS) were merged into a condFDR‐based NDD‐ALPS locus set (e.g., AD‐ALPS). Second, for each ALPS index, loci identified across the seven NDDs (e.g., mALPS‐AD, mALPS‐ALS, mALPS‐FTD, mALPS‐MS, mALPS‐PD, mALPS‐LBD, and mALPS‐HD) were merged into a condFDR‐based ALPS‐NDD locus set (e.g., mALPS‐NDD). Likewise, loci identified by conjFDR analyses across the three ALPS indices for a given NDD were integrated into a conjFDR‐based NDD‐ALPS locus set (e.g., AD‐ALPS). Novel loci were defined separately for each of the locus sets described above. Loci were considered novel if they met both of the following criteria: (1) they were located at least 500 kb from any locus reaching relaxed genome‐wide significance (p < 1 × 10−6) in the corresponding original GWAS dataset 22 , 24 , 25 , 26 , 27 , 28 , 29 , 30 (listed in Table S1) or in previously published studies 42 , 43 , 44 , 45 , 46 , 47 , 48 , 49 , 50 , 51 , 52 , 53 , 54 , 55 , 56 (listed in Table S2) and (2) none of the candidate SNPs within the locus had been annotated in the GWAS Catalog 57 as associated with the ALPS index or the corresponding NDD. For studies that did not provide explicit locus boundaries, genomic intervals were defined as ± 500 kb (1 Mb window) centered on the lead SNP. Functional annotation, gene mapping, and enrichment analysis Functional annotation was performed for candidate SNPs with conjFDR < 0.1 using the FUMA platform. 39 First, we assessed the deleterious potential of variants using the combined annotation‐dependent depletion (CADD) score, which integrates over 60 functional genomic features to predict the relative pathogenicity of single nucleotide variants; variants with CADD scores > 12.37 were classified as potentially deleterious. 58 Second, we evaluated the regulatory potential of SNPs using RegulomeDB, which assigns scores ranging from 1a to 7, with lower scores indicating stronger evidence for regulatory activity. 59 Finally, we characterized the chromatin state of each SNP using the 15‐state ChromHMM model, 60 which integrates multiple histone modification marks across 127 human reference epigenomes; SNPs annotated with open chromatin states (states 1 to 7) were considered more likely to reside in transcriptionally active regulatory regions such as promoters or enhancers. 61 , 62 After functional annotation, candidate SNPs with conjFDR < 0.1 were mapped to protein‐coding genes using three complementary strategies: positional mapping (assigning SNPs to genes within ± 10 kb), expression quantitative trait locus (eQTL) mapping (associating SNPs with expression of linked genes), and chromatin interaction mapping (connecting SNPs to regulatory targets through three‐dimensional chromatin contacts). Detailed information on the gene‐mapping procedure is provided in the Supplementary Methods. Gene Ontology (GO) enrichment analysis for shared genes mapped from candidate SNPs was conducted using g:Profiler. 63 The background gene set was defined using the default “only annotated genes” setting, corresponding to all human genes with at least one annotation in GO biological process (BP), molecular function (MF), or cellular component (CC). Multiple testing correction was performed using the Benjamini‐Hochberg FDR, and terms with q < 0.05 were considered statistically significant. Temporal gene expression trajectory analyses Developmental expression trajectories of shared genes were characterized using publicly available mRNA‐seq data from the PsychENCODE consortium. 64 This dataset includes 607 samples from 41 individuals ranging from eight postconceptional weeks (PCWs) to 40 postnatal years (PYs). Gene expression values, quantified as reads per kilobase of transcript per million mapped reads (RPKM), were divided into nine temporal windows covering developmental stages from early gestation to adulthood. Expression levels were log2‐transformed and standardized by subtracting the mean expression of each sample. 65 For each NDD‐ALPS pair, composite expression across developmental stages was calculated by averaging standardized expression values across all mapped genes. Locally weighted scatterplot smoothing was applied to visualize developmental expression trajectories across the lifespan for each disorder. Expression‐trait association analyses We used the MetaXcan approach 66 to assess whether genetically predicted gene expression levels are associated with ALPS indices and NDDs. First, S‑PrediXcan was applied to 13 brain tissues and whole blood using eQTL prediction models derived from Genotype‐Tissue Expression (GTEx) version 8. 67 The analyzed brain tissues included the amygdala, anterior cingulate cortex (BA24), caudate, cerebellar hemisphere, cerebellum, cortex, frontal cortex (BA9), hippocampus, hypothalamus, nucleus accumbens, putamen, cervical spinal cord (C1), and substantia nigra. Second, S‐MultiXcan was performed for multi‐tissue transcriptomic analyses by integrating tissue‐specific associations across all brain tissues and whole blood into a single multivariate model, thereby enhancing detection power compared with univariate analyses. Only genes with valid prediction models in at least one tissue were retained for downstream analyses. Expression‐trait associations were considered significant at a Benjamini‐Hochberg FDR‐corrected q < 0.05. Validation analyses To assess the robustness and reproducibility of the shared loci identified by the conjFDR analyses, we implemented two validation procedures. First, we applied the local analysis of [co]variant association (LAVA) 68 to test whether the NDD‐ALPS shared loci exhibited consistent local genetic correlations. In LAVA, local bivariate correlations were computed only for regions in which both traits demonstrated nominally significant local heritability in the univariate tests (p < 0.05). For these regions, local genetic correlations were estimated, and statistical significance was assessed using the Benjamini‐Hochberg FDR (q < 0.05). Second, to assess the robustness and reproducibility of the findings across alternative ALPS phenotypes, we used GWAS summary statistics for the anterior, middle, posterior, and total ALPS indices reported by Ran et al. 23 ConjFDR analyses were rerun for each NDD using these ALPS traits, applying the same analytical pipeline, significance thresholds, and locus‐definition criteria as in the primary analyses. Loci identified in the validation analyses were matched to those from the primary analyses based on physical overlap within 250 kb. RESULTS Quantification of genetic overlap between NDDs and ALPS indices Univariate MiXeR analyses 31 estimated polygenicity, discoverability, and heritability for the three ALPS indices (mALPS, lALPS, and rALPS) and seven NDDs (AD, PD, ALS, MS, FTD, HD, and LBD). As shown in Figure 2A, the three ALPS indices showed similar heritability (0.199 to 0.219), polygenicity (1929 to 2208 trait‐influencing variants), and discoverability (1.46 × 10−4 to 1.60 × 10−4), suggesting comparable genetic architectures across hemispheric measures. In contrast, NDDs showed more heterogeneous patterns, with estimated polygenicity ranging from two trait‐influencing variants in LBD to 595 in PD, discoverability from 1.75 × 10−4 to 5.36 × 10−2, and heritability from 0.044 to 0.331. Among them, PD exhibited the highest polygenicity, MS the greatest heritability, and LBD the strongest discoverability (Table S3). Quantification of shared polygenic architecture between the ALPS indices and NDDs. (A) Univariate MiXeR analyses estimating heritability, polygenicity, and discoverability for each trait. The y‐axis for discoverability is displayed on a ‐log10 scale. (B) Bivariate MiXeR results showing the estimated numbers (in thousands, with standard errors in parentheses) of trait‐specific and shared genetic variants between the three ALPS indices (mALPS, lALPS, rALPS) and five NDDs (AD, ALS, FTD, MS, PD). Circle colors distinguish traits, and relative circle size reflects the degree of polygenicity. (C) Dice coefficients and proportions of concordant shared variants for each trait pair, illustrating the extent and directionality of shared polygenic effects, respectively. AD, Alzheimer's disease; ALPS, analysis along the perivascular space; ALS, amyotrophic lateral sclerosis; FTD, frontotemporal dementia; HD, Huntington's disease; LBD, Lewy body dementia; lALPS, left ALPS index; mALPS, mean ALPS index; MiXeR, causal mixture model; MS, multiple sclerosis; NDD, neurodegenerative disease; PD, Parkinson's disease; rALPS, right ALPS index. Bivariate MiXeR analyses 32 further revealed varying degrees of polygenic overlap between ALPS indices and NDDs (Figure 2B and Table S4). Moderate overlap was observed for AD, ALS, FTD, MS, and PD, with 32 to 386 shared variants per trait pair, whereas HD and LBD showed minimal overlap, with no more than 12 shared variants identified for any trait pair. Among these NDDs, ALS showed the strongest overlap with mALPS, with ∼93% of ALS‐associated variants also influencing mALPS, whereas MS showed the weakest overlap with lALPS (∼16%). From the perspective of ALPS indices, mALPS shared the largest fraction of variants with ALS (∼18%), and lALPS the smallest with AD (∼1.5%) (Figure 2B and Table S4). Notably, Dice coefficients ranged from 2.8% (lALPS‐AD) to 30.4% (mALPS‐ALS), and the proportion of concordant effect directions varied widely (16.5% to 86.7%), indicating heterogeneous pleiotropic relationships between ALPS indices and NDDs (Figure 2C). Full MiXeR results are provided in Figures S1–S21 and Tables S3 and S4. Identification of shared loci between NDDs and ALPS indices Using cond/conjFDR analyses, 33 , 34 we leveraged cross‐trait enrichment and identified pleiotropic variants shared between the ALPS indices and NDDs. Conditioning on the ALPS traits yielded 778 NDD‐associated loci, which merged into 272 loci across disorders – 66 for AD, 24 for ALS, three for FTD, 137 for MS, 26 for PD, 10 for HD, and six for LBD – of which 45 were classified as novel (Figure 3A and Tables S5 and S6). Conversely, conditioning on the seven NDDs identified 349 DTI‐ALPS‐associated loci, merged into 89 loci (31 for mALPS, 21 for lALPS, and 37 for rALPS), including 27 novel loci (Figure 3A and Tables S7 and S8). Conditional Q‐Q plots demonstrated strong pleiotropic enrichment between the ALPS indices and five NDDs (AD, ALS, FTD, MS, and PD) (Figures S22–S36), whereas no significant enrichment was observed for HD or LBD (Figures S37–S42). ConjFDR analyses further detected 40 loci jointly associated with the ALPS indices and NDDs (Figure 3B and Table S9). After merging overlapping signals, 22 distinct shared loci remained (six for AD‐ALPS, six for ALS‐ALPS, two for FTD‐ALPS, six for MS‐ALPS, and two for PD‐ALPS), of which eight were novel for NDDs, 12 for ALPS traits, and five for both (Table S10). Notably, shared loci for AD, ALS, FTD, and PD converged within the 17q21.31 interval (chr17: 43.46 to 44.87 Mb), highlighting this region as a major susceptibility hub underlying the genetic overlap between ALPS traits and NDDs (Figure 3C). Shared genomic loci between ALPS indices and NDDs. (A) Chromosomal ideograms summarizing loci identified by condFDR and conjFDR analyses. CondFDR analyses detected 272 NDD‐associated loci (diamonds) and 89 ALPS‐associated loci (circles), whereas conjFDR analyses identified 22 shared loci (triangles). Different colors represent distinct NDD phenotypes. (B) Manhattan plots from conjFDR analyses showing genome‐wide shared loci between the three ALPS indices (mALPS, lALPS, rALPS) and five NDDs (AD, ALS, FTD, MS, PD). Panels from top to bottom correspond to mALPS, lALPS, and rALPS. Each point represents a SNP, with lead SNPs highlighted by larger black‐outlined dots. The y‐axis indicates −log10‐transformed conjFDR values, and the dashed line marks the significance threshold (conjFDR = 0.05). (C) LocusZoom plots depicting regional association signals at 17q21.31 (chr17:43.46‐44.87 Mb) from the conjFDR analyses of the AD‐mALPS, ALS‐mALPS, FTD‐mALPS, and PD‐mALPS pairs. SNPs are colored according to their linkage disequilibrium (r2) with the lead SNP, and the genomic positions of relevant genes (e.g., MAPT, LRRC37A, LRRC37A2, and PLEKHM1) are displayed below. AD, Alzheimer's disease; ALS, amyotrophic lateral sclerosis; ALPS, analysis along the perivascular space; condFDR, conditional false discovery rate; conjFDR, conjunctional false discovery rate; FTD, frontotemporal dementia; HD, Huntington's disease; LBD, Lewy body dementia; lALPS, left ALPS index; mALPS, mean ALPS index; MS, multiple sclerosis; NDD, neurodegenerative disease; PD, Parkinson's disease; rALPS, right ALPS index; SNP, single nucleotide polymorphism. Functional annotation, gene mapping, and enrichment analysis Across all shared loci, most candidate SNPs were located in non‐coding regulatory regions (50.87% intronic; 38.49% noncoding RNA introns) (Figure 4A and Table S11). A subset of variants showed evidence of functional relevance, including 3.12% with CADD > 12.37 (suggesting potential deleteriousness) and 16.03% with RegulomeDB < 3 (supporting regulatory activity). Nearly all candidate SNPs (93.85%) mapped to open chromatin states (ChromHMM states 1 to 7), consistent with active regulatory elements. SNP‐to‐gene mapping yielded a total of 596 protein‐coding genes across shared loci, corresponding to 183 unique genes, such as MAPT, LRRC37A, LRRC37A2, and PLEKHM1 (Table S12). Subsequently, GO enrichment analysis of the 183 unique genes identified 30 significant terms (Benjamini‐Hochberg FDR q < 0.05; Table S13). As shown in Figure 4B, these genes were predominantly enriched for cellular components related to the endolysosomal system and microtubule cytoskeleton, including lysosome, intracellular vesicle, and microtubule cytoskeleton. Functional annotation, enrichment, and expression analyses of shared loci between the ALPS indices and NDDs. (A) Functional categorization of candidate SNPs (conjFDR < 0.1) across all shared loci, including genomic location, RegulomeDB scores, and minimum chromatin states. The x‐axis shows the log10‐transformed SNP count in each category, and percentages indicate their relative proportions. Over half of the variants were intronic, with a substantial fraction located in introns of non‐coding RNAs. (B) Bubble chart showing significant GO cellular component terms (BH‐FDR q < 0.05) for genes mapped from candidate SNPs. The x‐axis indicates the gene count annotated to each term, and the y‐axis lists the enriched cellular components. Bubble color denotes enrichment significance as −log10‐transformed BH‐FDR q values. Enriched terms were mainly related to the endolysosomal system and microtubule cytoskeleton. (C) Developmental expression trajectories of shared genes across nine developmental stages from eight PCWs to 40 PYs. The y‐axis shows normalized log2‐transformed mean expression values. Shaded areas indicate the 95% confidence intervals, and the vertical dotted line marks birth. (D) Heatmap showing significant expression‐trait associations (BH‐FDR q < 0.05) for the 45 genes whose genetically predicted expression is associated with both ALPS indices and NDDs and that overlap with shared genes. The magnitudes of the −log10‐transformed BH‐FDR q values are coded using a red‐to‐blue color scale. AD, Alzheimer's disease; ALS, amyotrophic lateral sclerosis; ALPS, analysis along the perivascular space; BH‐FDR, Benjamini‐Hochberg false discovery rate; conjFDR, conjunctional false discovery rate; FTD, frontotemporal dementia; GO, Gene Ontology; HD, Huntington's disease; LBD, Lewy body dementia; lALPS, left ALPS index; mALPS, mean ALPS index; MS, multiple sclerosis; ncRNA, noncoding RNA; NDD, neurodegenerative disease; PD, Parkinson's disease; PCW, postconceptional week; PY, postnatal year; rALPS, right ALPS index; SNP, single nucleotide polymorphism. Temporal expression trajectories of shared genes Using mRNA‐seq data from PsychENCODE, 64 temporal expression trajectory analysis indicated that shared genes showed broadly similar developmental expression profiles across disorders, with differences primarily observed in the magnitude of expression rather than the overall developmental pattern (Figure 4C). Across all NDD‐ALPS pairs, expression rose gradually from early fetal stages (8 to 9 PCWs), reached its highest levels between mid‐gestation and late gestation (16 to 22 PCWs to ∼35 PCWs), and then stabilized from childhood through adolescence and adulthood (2.8 to 40 PYs). Among the disorders, genes shared with MS displayed the highest expression throughout development, whereas those shared with PD showed the lowest. Genes shared with AD exhibited a slightly later prenatal peak near ∼35 PCWs, followed by a continuous postnatal decrease. In contrast, genes associated with ALS, FTD, and PD followed a similar pattern, with a mid‐gestational peak, a reduction after birth, and relatively stable expression from early childhood onward (0.5 to 2.6 PYs). Expression‐trait associations Using the MetaXcan framework, 66 we identified 1799 genes whose genetically predicted expression levels were significantly associated with the ALPS indices and 5697 genes associated with NDDs (Benjamini‐Hochberg FDR q < 0.05; Table S14). The intersection of these two sets yielded 752 genes jointly associated with the ALPS indices and NDDs (Table S15). Among them, 213 genes were associated with two or more disorders, and 45 also overlapped with pleiotropic genes mapped from shared loci (Figure 4D and Figure S43). Notably, several 17q21.31 genes – including MAPT, LRRC37A, LRRC37A2, PLEKHM1, CRHR1, and ARL17B – showed expression‐trait associations with the ALPS indices and at least five NDDs and were independently supported by SNP‐to‐gene mapping, providing convergent genetic and transcriptomic evidence. Validation of shared loci To evaluate the robustness of the shared loci identified by the conjFDR analyses, we first assessed whether these loci exhibited consistent local genetic correlations using LAVA. 68 Among the 40 loci jointly associated with the ALPS indices and the five NDDs, 17 demonstrated nominally significant local heritability in both traits (p < 0.05; Table S16), and 15 retained significant bivariate correlations after Benjamini‐Hochberg FDR correction (q < 0.05). To further assess the robustness and reproducibility of the findings across alternative ALPS phenotypes, we re‐ran conjFDR analyses using the anterior, middle, posterior, and total ALPS indices from Ran et al. 23 A total of 27 distinct shared loci were identified across the NDD‐ALPS combinations, of which nine overlapped with the shared loci identified in the primary analyses within 250 kb (Table S17). Notably, four replicated loci fell within the chr17: 43.46 to 44.87 Mb region corresponding to 17q21.31 – a well‐established susceptibility region for multiple NDDs – providing partial support for the robustness of this signal across different ALPS phenotypic definitions. DISCUSSION In this study, we systematically characterized the shared genetic architecture between DTI‐ALPS traits and multiple NDDs. By integrating GWAS datasets for three ALPS imaging indices with seven major NDDs, we identified widespread yet heterogeneous polygenic overlap, indicating that ALPS‐related imaging phenotypes share common genetic influences with diverse neurodegenerative disorders. Through cond/conjFDR analyses, we identified 22 pleiotropic loci, including several previously unreported associations. Notably, multiple shared signals converged at the 17q21.31 region, a well‐established susceptibility locus in neurodegeneration. Functional annotation and gene expression analyses further demonstrated enrichment of shared genes in the endolysosomal system and microtubule cytoskeleton. Collectively, these findings provide evidence for a shared genetic architecture between DTI‐ALPS traits and NDDs and highlight biological pathways that may contribute to their overlap. Based on the univariate MiXeR estimates, discoverability and polygenicity were negatively correlated across all analyzed traits, suggesting that traits influenced by fewer variants may be driven by variants with larger genetic effects (Figure S44). The three ALPS traits showed consistency in heritability, polygenicity, and discoverability, indicating a similar underlying genetic architecture across them. Compared with NDDs, the ALPS traits exhibited greater polygenicity, generally higher heritability except for MS, and lower discoverability, suggesting a complex genetic architecture that may require larger cohorts to fully capture genetic signals of ALPS traits. Bivariate MiXeR analyses further revealed moderate polygenic sharing between the ALPS traits and five NDDs, with the highest Dice coefficient observed in the mALPS‐ALS pair (30.4%). Notably, the proportion of concordant effect directions varied across trait pairs, ranging from 16.5% to 86.7%. This pattern suggests that the genetic overlap between DTI‐ALPS traits and NDDs may involve both concordant and discordant biological pathways, reflecting complex pleiotropic relationships. CondFDR analyses identified several novel loci associated with ALPS indices and NDDs. When conditioning on NDD phenotypes, 27 novel loci were identified in association with ALPS indices. Conversely, conditioning on the ALPS indices yielded 45 novel loci associated with NDDs. ConjFDR analyses further identified 22 distinct loci jointly associated with the ALPS indices and the five NDDs. Among them, the locus at 4p16.3 (lead SNP: rs3755963) in the PD‐ALPS pair was newly associated with ALPS traits and mapped to GAK, which encodes cyclin G‐associated kinase, a serine/threonine kinase involved in regulating α‐synuclein expression and toxicity, processes implicated in PD pathogenesis. 69 , 70 Notably, eight of the 22 shared loci represent novel loci for NDDs. For example, the locus at chr13: 46.37 to 46.39 Mb (lead SNP: rs9526109) from the AD‐ALPS pair is located in the SIAH3 gene. Although SIAH3 has not been directly linked to AD genetic risk, its variants have been associated with accelerated ventricular enlargement, an AD‐related imaging phenotype, as well as cognitive decline and reduced cerebral metabolism. 71 The locus at chr6: 157.27 to 157.38 Mb (lead SNP: rs6904955) from the ALS‐ALPS pair maps to ARID1B, for which peripheral blood methylation has been positively correlated with disease duration in ALS, supporting ARID1B as a candidate gene underlying the shared genetic signal. 72 In addition, a substantial proportion of candidate SNPs within shared loci were located in non‐coding RNA intronic regions, and nearly all were annotated as open chromatin states. This distribution suggests that some shared genetic signals may involve regulatory effects in non‐coding regions, potentially influencing nearby genes or non‐coding transcripts. 73 , 74 Further functional studies are needed to clarify the precise mechanisms. Moreover, several shared loci clustered within 17q21.31, a well‐established susceptibility locus for NDDs. Among genes in this region, MAPT is a well‐known gene involved in tau biology and microtubule regulation. 75 , 76 LRRC37A and LRRC37A2 have been implicated in PD pathology, with high expression in substantia nigra astrocytes and potential roles in astroglial inflammatory responses and α‐synuclein‐related processes. 77 These findings suggest that 17q21.31 may represent a major convergence region linking DTI‐ALPS traits and NDDs, potentially through pathways related to microtubule regulation and astrocyte‐associated processes. Using three complementary gene‐mapping strategies, we identified 183 unique protein‐coding genes, which were predominantly enriched in the endolysosomal system and the microtubule cytoskeleton. These findings are consistent with evidence that endolysosomal dysfunction and microtubule abnormalities are common features across multiple NDDs, potentially contributing to impaired protein clearance, disrupted axonal transport, and neuronal vulnerability. 78 , 79 These enrichment findings suggest that genetic regulation of endolysosomal function and microtubule organization may contribute to the biological overlap between DTI‐ALPS traits and NDDs. In addition, the temporal expression trajectory analysis provided developmental‐stage context for these shared genes rather than direct mechanistic evidence. Given the relatively limited range of normalized expression values and the broadly similar trajectories commonly observed among brain‐expressed genes, this result should be interpreted cautiously and is discussed in detail in the Supplementary Discussion. Using MetaXcan, we identified 752 genes with expression‐trait associations for both the ALPS indices and NDDs, including 45 genes that also had locus‐to‐gene support. Notably, several of these 45 genes were located within the 17q21.31 region, including MAPT, LRRC37A, and LRRC37A2, which were discussed earlier, 75 , 76 , 77 as well as CRHR1 and ARL17B. CRHR1, a key gene involved in the hypothalamic‐pituitary‐adrenal axis, has also been implicated in Aβ accumulation and tau hyperphosphorylation. 80 Brain ARL17B expression and methylation were associated with risk SNPs for progressive supranuclear palsy. 81 Together with the locus‐mapping results, these expression‐trait associations highlight a set of genes that may contribute to the shared genetic architecture between DTI‐ALPS traits and NDDs. An important consideration when interpreting the present findings is the biological significance of DTI‐ALPS. Since its introduction, DTI‐ALPS has become one of the most widely used imaging approaches in studies of the glymphatic system because it provides a non‐invasive measure of diffusion properties along perivascular pathways and has demonstrated associations with a broad range of neurodegenerative disorders. 14 , 15 , 18 , 19 , 20 , 21 However, the extent to which ALPS specifically reflects glymphatic function remains an area of active debate. 16 , 17 , 82 Accumulating evidence suggests that ALPS is influenced by multiple biological and methodological factors beyond putative glymphatic activity, including white matter microstructure, age‐related tissue changes, cerebral small vessel disease burden, and differences in image acquisition and processing. 16 , 17 , 83 Consequently, alterations in ALPS cannot be interpreted as a direct or specific measure of glymphatic dysfunction. Rather, ALPS likely represents a composite imaging phenotype reflecting perivascular diffusion characteristics together with other structural and physiological properties of the surrounding brain tissue. This distinction is particularly important for interpreting the results presented here. The shared genetic architecture identified in this study should not be viewed as evidence that NDD susceptibility and glymphatic dysfunction necessarily share the same genetic basis. Instead, our findings indicate that DTI‐ALPS traits, as currently quantified in large‐scale population imaging studies, share genetic influences with multiple neurodegenerative disorders. The biological pathways highlighted by our analyses – including endolysosomal function, microtubule organization, and genes within the 17q21.31 region – may therefore contribute to imaging signatures captured by ALPS as well as to neurodegenerative vulnerability. Despite the ongoing debate regarding its biological interpretation, 16 , 17 , 82 DTI‐ALPS remains the most extensively studied and largest‐scale imaging phenotype currently available for investigating brain fluid transport‐related processes at the population level. 14 , 15 , 22 , 23 In the absence of comparably powered genetic datasets for more direct measures of brain fluid dynamics, DTI‐ALPS provides a valuable opportunity to explore the genetic basis of imaging phenotypes linked to perivascular diffusion and their relationship to NDD susceptibility. Several limitations should be acknowledged. First, all analyses were conducted in individuals of European ancestry, which may limit the generalizability of the findings and warrants caution when extrapolating the results to other populations. Second, although GWASs provide substantial power for detecting common variants, they are less sensitive to rare variants, which may also contribute to NDD susceptibility. 84 The reliance on GWAS summary statistics further restricts our ability to evaluate the contribution of rare variation to the shared genetic architecture identified in this study. Third, although DTI‐ALPS traits represent the largest currently available imaging phenotypes related to perivascular diffusion, their biological interpretation remains incompletely understood. As discussed earlier, ALPS may reflect multiple biological processes beyond putative glymphatic transport. Consequently, the shared genetic architecture identified here should be interpreted in the context of ALPS‐related imaging phenotypes. Fourth, as discussed in the Supplementary Discussion, the primary and validation analyses used different region‐of‐interest definitions and processing pipelines and exhibited only moderate genetic correlations (Table S18). This suggests that these ALPS phenotypes may capture partially distinct regional or methodological aspects of perivascular diffusion, and consistency across datasets should therefore be interpreted cautiously. Finally, the findings for HD and LBD should be interpreted cautiously, with further details provided in the Supplementary Discussion. Taken together, our findings reveal a shared genetic architecture between DTI‐ALPS traits and NDDs. The shared loci and pathway signatures identified here point to convergent molecular themes involving endolysosomal regulation and microtubule organization, which may contribute to the overlap between ALPS‐related imaging phenotypes and neurodegenerative vulnerability. By leveraging complementary analytical frameworks across diverse traits, this study provides a scalable approach for dissecting genetic links between neuroimaging‐derived phenotypes and disease susceptibility. These results support the value of DTI‐ALPS as a population‐scale imaging phenotype for investigating biological processes related to perivascular diffusion, while underscoring the need for future studies integrating complementary measures of brain fluid transport to clarify the mechanisms underlying its association with neurodegeneration. AUTHOR CONTRIBUTIONS Conceptualization: Hong Zhou, Qian Wu, Xinhua Wei, Feng Liu. Methodology: Jialin Deng, Qian Wu. Formal analysis: Jialin Deng, Qian Wu, Yule Zeng, Xiaomei Duan. Data curation: Jialin Deng, Qian Wu, Tianyao Wang, Jiayi Wu. Writing – original draft: Jialin Deng, Qian Wu, Feng Liu. Writing – review and editing: Jialin Deng, Qian Wu, Beisha Tang, Xinhua Wei, Feng Liu. Funding acquisition: Hong Zhou. Supervision: Hong Zhou, Xinhua Wei, Lu He, Feng Liu. All authors critically reviewed and approved the final manuscript. CONFLICT OF INTEREST STATEMENT The authors declare no competing interests. Author disclosures are available in the Supporting Information. CONSENT STATEMENT All GWAS summary statistics used in this study were obtained from previously published studies. Informed consent was obtained from all participants in the original studies, and the corresponding ethical approvals are detailed in those publications. Supporting information Supporting Information Supporting Information Supporting Information ACKNOWLEDGMENTS The authors thank the participants and investigators of the UK Biobank and the original GWAS consortia for making these data publicly available. This study was supported by the National Science and Technology Major Project (Grant/Award Numbers: 2024ZD0528203), the Clinical Research 4310 Program of the First Affiliated Hospital of the University of South China (Grant/Award Numbers: 20224310NHYCG03), and Innovation Platform and Talent Program (Grant/Award Numbers: 2023TP1047). Contributor Information Xinhua Wei, Email: eyxinhuawei@163.com. Lu He, Email: luhe1989@126.com. Feng Liu, Email: fengliu@tmu.edu.cn. DATA AVAILABILITY STATEMENT All datasets used in this study are publicly accessible. Specifically, the GWAS summary data for the ALPS indices based on UK Biobank participants are available at https://figshare.com/articles/dataset/GWAS_summary_statistics_for_ALPS‐index/25515331, and GWAS summary data for NDDs are provided in Table S1. The processed mRNA‐seq data were obtained from http://development.psychencode.org/. The analytical code employed in this study is available through publicly accessible repositories: MiXeR (https://github.com/precimed/mixer), cond/conjFDR (https://github.com/precimed/pleiofdr), FUMA (https://fuma.ctglab.nl/), g:Profiler (https://biit.cs.ut.ee/gprofiler/gost), MetaXcan (https://github.com/hakyimlab/MetaXcan), LAVA (https://github.com/josefin‐werme/LAVA), and LDSC (https://github.com/bulik/ldsc). REFERENCES Associated Data Supplementary Materials Supporting Information Supporting Information Supporting Information Data Availability Statement All datasets used in this study are publicly accessible. Specifically, the GWAS summary data for the ALPS indices based on UK Biobank participants are available at https://figshare.com/articles/dataset/GWAS_summary_statistics_for_ALPS‐index/25515331, and GWAS summary data for NDDs are provided in Table S1. The processed mRNA‐seq data were obtained from http://development.psychencode.org/. The analytical code employed in this study is available through publicly accessible repositories: MiXeR (https://github.com/precimed/mixer), cond/conjFDR (https://github.com/precimed/pleiofdr), FUMA (https://fuma.ctglab.nl/), g:Profiler (https://biit.cs.ut.ee/gprofiler/gost), MetaXcan (https://github.com/hakyimlab/MetaXcan), LAVA (https://github.com/josefin‐werme/LAVA), and LDSC (https://github.com/bulik/ldsc).