Multi-ancestry multi-trait analysis reveals shared genetics across major psychiatric disorders and Alzheimer’s disease Authors: Xin Zheng, Hongye Feng, Lijie Ren, Youcheng Zhang Published: 2026 ABSTRACT AbstractThe clinical overlap between major psychiatric disorders (MPDs) and Alzheimer’s disease (AD) implicates complex shared etiology. Previous studies demonstrated that both diseases are genetically complex and highly heritable, suggesting that more endeavors are necessary to be made from the very bottom to understand their genetic basis. With the advance of post-genomic analysis, multi-ancestry meta-analysis allows the generalizability of the genetic architecture across different populations to uncover ancestry-specific variants, while multi-trait analysis enables the discovery of the co-colocalized risk genomic regions across diseases. Therefore, in this study, we leveraged published GWAS summary statistics from European, East Asian, Hispanic and African American populations to report schizophrenia, major depressive disorders, and Alzheimer’s disease risk loci and further fine-mapping to credible sets with >95% PP inclusion of the causal variant. We distilled 2871 potential traits from publicly available and found 134 traits significantly genetically correlated with both MPDs and AD using batch LD score regression. We then prioritized the identified loci from multi-ancestry results for cross-trait colocalization analysis to assess shared genetic etiology and further nominated 2 colocalized loci across both conditions, including rs2532240 and rs6504163. In the end, we finalized our analysis by validation and functional inference of the underlying susceptibility genes as well as putative mechanisms using evidence from multiple resources, including FIVEx, Open Targets, and scQTLbase.Subject terms: Genomics, Psychiatric disorders, Clinical genetics Introduction Major psychiatric disorders (MPDs) and Alzheimer’s disease (AD) represent two critical domains of brain dysfunction that impose a substantial burden on individuals, families, and healthcare systems worldwide. Major psychiatric disorders, with schizophrenia (SCZ) and major depressive disorder (MDD) representing the two key conditions, often manifest with significant impairments in cognition, affect, and behaviour. Concurrently, Alzheimer’s disease, a progressive neurodegenerative disorder is marked by declining memory and impaired executive function [1–5]. The clinical overlap between major psychiatric disorders and AD is bilateral. Patients experiencing psychiatric conditions typically manifest in early adulthood or mid-adulthood with late-life neurodegenerative diseases with 4 times higher risk, whereas 65% of patients with Alzheimer’s disease noted a spectrum of neuropsychiatric symptoms during the course of the illness that can mirror or compound those seen in primary psychiatric illnesses [3, 6]. Previous studies have demonstrated AD as a most common type of dementia but also genetically complex, with heritability of up to 76%, while major psychiatric disorders such as schizophrenia were estimated in a Danish twin study with approximately 79% heritability [7, 8]. Other studies suggested genetic correlation with shared molecular mechanism across major psychiatric disorders and neurodegenerative diseases [3]. Thus, it is necessary to understand more from the very bottom their overlapping genetic basis. With recent advances in sequencing technology, genome-wide association studies (GWAS) that cover large-scale populations have been conducted on both major psychiatric disorders and Alzheimer’s disease, identifying novel genetic variants [9–11]. However, most studies predominantly focused on the European population or individuals from a single ancestry such as East Asian, Hispania, etc. This neglect of multi-ethnic effect not only limits the generalizability of the findings but also hinders our ability to uncover ancestry-specific genetic determinants that could contribute to health disparities. To address this limitation, the implementation of multi-ancestry meta-analyses such as GWAMA, MANTRA, METAL, PLINK and MR-MEGA, reveals the more precise disentanglement of the disease genomics [12–17]. Meanwhile, multi-trait analysis allows the identification of genomic regions (loci) that concordantlycontributes to different disorders, therefore providing chances to understand the overlapping mechanism. Therefore, in this study, we performed our multi-ancestry multi-trait (MAMT) analysis to further disentangle the shared genetics between major psychiatric disorders and Alzheimer’s disease. We first reported the large-scale multi-ancestry genome-wide meta-analysis results based on four different ancestries: European (EUR), East Asian (EA), Hispanic (HIS), and African American (AFM). This meta-analysis included a total of 169,417, 1,410,479, and 644,188 participants in schizophrenia, major depressive disorders, and Alzheimer’s disease, respectively. Using two meta-regression approaches, including MR-MEGA for heterogeneity and PLINK random effect model for homogeneity, we demonstrated an improved resolution of fine-mapping at SCZ, MDD, and AD loci contributing to the genetic ancestry effect. We then estimated the genetic correlation of MPDs and AD with 2871 potential traits from publicly available data using batch LD score regression [18, 19]. After that, we leveraged the identified loci from multi-ancestry analysis for colocalization with Hypothesis Prioritisation in multi-trait Colocalization (HyPrColoc) [20]. Our findings assessed shared genetic etiology across multiple related traits and elucidated several causal variants explaining potential overlapping mechanisms between MPDs and AD. In the end, we finalized our analysis by functional inference on the susceptibility genes associated with the identified loci using public evidence from multiple database sources including FIVEx, Open Targets, and scQTLbase [21].s Results Study design Multiple ancestry analysis reveals more precise disentanglement of the disease genomics. Results with information on the benefits of ethnic differences provide our additional consideration for a comprehensive view of the following analysis. Our multi-ancestry and multi-trait (MAMT) meta-analyses involved two broad classes of conditions and 3 minor classes of diseases, major psychiatric disorders (schizophrenia and major depressive disorders) and neurodegeneration disease (Alzheimer’s disease), with a total of 169,417 SCZ cases, 1,410,479 MDD and 644,188 AD participants (Fig. 1, Supplementary Table S1). All conditions employed GWAS data from 4 different ancestries, including European (EUR), Hispanic (HIS), East Asian (EA), and African American (AFM). For quality control, our analysis included variants that passed quality control with a minor allele frequency (MAF) > 0.01 to accurately quantify homogeneity and heterogeneity. To disentangle the heterogeneous and homogeneous effect, our multi-ancestry analysis implemented meta-regression of multi-ethnic genetic association (MR-MEGA) and the random effect model (PLINK meta-analysis). MR-MEGA incorporates genetic variation as covariates to better detect heterogeneous effects across different cohorts, whereas the random effect model demonstrated a greater ability to identify homogenous allelic effects. We considered significant variants for both analyses as passing the standard P-value threshold of 5 × 10−8, consistent with most GWAS meta-analyses as a general standard criterion [22]. Fine-mapping was then conducted to identify disease-associated loci. Later, we prioritized the identified loci to multi-trait colonization analysis to nominate shared loci across MPDs and AD. In the end, by leveraging multiple levels of evidence including eQTL, single cell eQTL and phenome-wide association study, we validated and performed functional inference of these colocalized loci to the putative mechanism across MPDs and AD. Study design. Outline of multi-ancestry multi-trait meta-analysis. Top Study subjects covered three conditions, including major psychiatric disorders (schizophrenia and major depressive disorders) and neurodegenerative disease (Alzheimer’s disease), and four ancestries (European, East Asian, African American, and Hispanic). Middle multi-ancestry meta-analysis was applied with two different approaches, including MR-MEGA for identifying heterogeneous effects and the PLINK random effect model for homogeneity. Fine-mapping was then performed to define the associated genomic risk loci. Bottom Multi-trait meta-analyses were conducted to compute the genetic correlation and detect the colocalization across traits. Multiple levels of genetic evidence were further leveraged for validation and functional inferences. Multi-ancestry meta-analysis identified heterogeneous and homogeneous loci across populations In schizophrenia, 304 loci and 219 loci were identified using MR-MEGA and the random effect model with P < 5 × 10−8, respectively (Fig. 2, Supplementary Table S2 and S3). Among these, 54 and 78 loci were novel compared to previous findings (Nature, 2022 [10]). Of the loci detected by MR-MEGA, 46 demonstrated significant ancestral heterogeneity (P-value_ancestry_het < 0.05 and P-value_residual_het > 0.05), with allelic effects predominantly positive in the African cohort (e.g., RP11-557C9.1, PDE4D, ZAN, MARK2, CACNA2D2, and STAG1). Conversely, 149 loci exclusively identified by the random effect model exhibited homogeneous effects across the four ancestries (Supplementary Table S2,a and S3,a). Focusing on the novel discoveries, 22 of the 54 MR-MEGA loci were heterogeneous across ancestries. This included strongly significant signals (P-value_ancestry_het < 0.01) at WBSCR17, SCRIB, DCC, MAN2A1, LINC00320, and CNTN4, supporting structural population differences over confounding factors. Specifically, CNTN4 effects were driven by the European and Hispanic cohorts, whereas the remaining loci predominantly showed positive effects in the African population. The Manhattan plot for the MR-MEGA results for major psychiatric disorders (schizophrenia and major depressive disorder) and Alzheimer’s disease. The x-axis showed chromosome and base pair positions of the SNPs tested in the meta-analyses. Variants above the horizontal red line indicate that the significant SNPs reached a P-value < 5 × 10−8, while the whole plot is truncated at |−log10(Pval)| < 50. Color represented different diseases as in legend where MPDs were placed on the upper part of the plot ( − log10(Pval) > 0) while AD was on the bottom part ( − log10(Pval) < 0). Lead SNPs were annotated with the nearest gene and labeled with corresponding colors. Specifically, for MPDs, only the distinct lead SNPs (compared with the two large multi-ancestry studies Trubetskoy, V. et al., 2022 for SCZ and Adams, MJ. et al., 2025 for MDD) were highlighted. For major depressive disorder, 18 total risk loci were identified using MR-MEGA and the random effect model (Supplementary Table S2,b and S3,b). Compared to recent findings (Cell, 2025 [23]), 16 loci from MR-MEGA and 1 locus from the random effect model were novel. Among the overall findings, 17 loci displayed ancestral heterogeneity and 1 locus showed consistent, homogeneous effects across ancestries. Within the novel discoveries, 12 loci were heterogeneous, such as PGBD1, LRFN5, MAD1L1, NEGR1, ASTN2, ADARB1, ERBB4, CAMTA1, etc.. Notably, CAMTA1 (1:7316114; rs7549870) was detected by both methods but exhibited a distinct positive effect specific to the Hispanic cohort. In Alzheimer’s disease, 47 loci and 51 loci were identified using MR-MEGA and the random effect model, respectively, based on summary statistics from Lake, J. 2023 [24]. Among the loci detected by MR-MEGA, 29 were method-specific. In terms of cross-ancestry effects, 8 loci (TREML2, RP3-399L15.3, JAZF1, FERMT2, SLC24A4, RP11-30K9.6, WNT3, and CASS4) demonstrated significant ancestral heterogeneity, where JAZF1 locus presented the strongest heterogeneity signal (P-value_ancestry_het = 0.001015; P-value_residual_het = 0.926054). The remaining loci exhibited consistent, homogeneous effects across groups (Supplementary Table S2,c and S3,c). Fine-mapping generated 95% and 99% credible sets with restricted number of causal variants We performed fine-mapping on the loci identified by MR-MEGA and the random effect model to calculate posterior probabilities (PPs) and prioritize putative causal variants (see Method). In schizophrenia, we resolved 46 loci to highly tractable 95% credible sets with less than 5 likely causal variants such as KANSL1, NEBL, GMEB1, EPN2, SMPD3, LRRC4B, GRM3, etc. (Supplementary Table S2 and S3). 3 loci (SYNGAP1, LINC00502, and ZNF823) were identified in both methods, whereas 4 including PRKD1, FURIN, PCGF3 and ATP2A2 were specifically reported in MR-MEGA model. We also identified a set of single-variant 99% credible sets at 8 loci, including RP11-557C9.1, PDE4D, SIKE1, LPP, AC092846.1, SYNGAP1, RP11-399D6.2, and ZNF823 (Supplementary Tables S3,a). Notably, RP11-557C9.1, PDE4D, SIKE1, and LPP also exhibited significant ancestral heterogeneity (P-value_ancestry_het <0.05; P-value_residual_het > 0.05), highlighting them as robust ancestry-specific causal variants. In major depressive disorder, we reported only 2 loci (CAMTA1 and RP11-390P2.4) to variants under both 95% and 99% PP thresholds combining results from MR-MEGA and PLINK. Within these, RP11-390P2.4 showed a strong signal driven by a single putative causal variant, while CAMTA1 demonstrated significant ancestral heterogeneity (P-value_ancestry_het = 3.492195 × 10−7, P-value_residual_het = 0.383512). For Alzheimer’s disease, we applied our fine-mapping pipeline to summary statistics from Lake J, 2023 [24]. To maintain analytical consistency across our three trait cohorts and preserve loci for downstream cross-trait analyses, we applied a unified genome-wide significance threshold (P-value <5 × 10−8). This yielded single-variant 95% credible sets at 10 loci via MR-MEGA (9 exceeding 99% PP) and 11 loci via the random effect model (all exceeding 99% PP). CTSB showed a single putative causal variant across both methods. 7 credible sets (BIN1, INPP5D, RHOH, APH1B, FAM157C, GRN, and TOMM40) were specific to MR-MEGA, with INPP5D exhibiting nominally significant ancestral effects (P-value_ancestry_het = 0.079017; P-value_residual_het = 0.115467). Multi-trait analysis discovered shared traits and loci across MPDs and AD To investigate common genetically correlated traits between major psychiatric disorders (MPDs) and Alzheimer’s disease (AD), we performed batch linkage disequilibrium score regression (batchLDSC) on across 2,871 publicly available phenotypes with MPDs and AD respectively. In the European cohort, following multiple-testing correction, we identified 72 traits significantly genetically correlated with both SCZ and AD, while 121 common traits correlated with MDD and AD (Fig. 3a, b, Supplementary Table S5). Notably, majority of common traits (n = 60) with SCZ and AD demonstrated consistent directionality, whereas 114 traits with MDD and AD exhibited opposite correlation (Supplementary Table S5,d-e). Notably, the shared genetic architecture demonstrated highly consistent directionality across both disease comparisons. For SCZ and AD, lipoprotein cholesterol exhibited a consistent positive genetic correlation, indicating that risk alleles for these disorders concurrently act to increase cholesterol levels. Psychiatric and cognitive phenotypes, including memory loss or impairment, nervous feelings, and worrier/anxious feelings, similarly showed consistent positive genetic correlations across both SCZ and AD. Furthermore, we observed consistent negative genetic correlations for weight, obesity but also hypertension, suggesting the susceptibility may be genetically linked to reduced metabolic traits on both disorders, despite the increased clinical risk. In the MDD and AD comparison, only 7 traits displayed consistent genetic correlations. Psychiatric traits and memory impairment were reported positively correlated, whereas blood urea nitrogen showed negative directionality with both conditions. In contrast, genetic liability to MDD is strongly positively correlated, but negative in AD with mid-life systemic morbidities such as cardiovascular disease. Conversely, susceptibility to AD is positively correlated with extended longevity phenotypes that are distinctly negatively correlated with MDD. Overall, these findings suggest that the genetic basis of MDD s is closely tied to systemic somatic disease and metabolic stress. Meanwhile, genetic risk for AD is associated with mid-life physiological resilience, which may facilitate survival to the advanced ages where neurodegeneration primarily occurs. Cross-trait analysis between major psychiatric disorders and Alzheimer’s disease. a Pairwise genetic correlations and direction between cohorts (SCZ, AD) and 2871 traits from publicly available data were estimated using batch LDSC. b Pairwise genetic correlations and direction between cohorts (MDD, AD) and traits. c, d Colocalizations across schizophrenia and Alzheimer’s disease by locus comparison illustrated the shared genomic loci, where the large subplot on the left provided combined view of colocalized candidate SNPs of the -log10(P) value in SCZ (y-axis) and AD (x-axis), while two regional plots on the right showed the location and -log10(P) of the SNPs in SCZ (up) and AD (bottom) respectively. To further determine shared genomic regions across MPDs and AD, we conducted multi-trait colocalization using ancestry-matched summary statistics and ethnicity-specific LD reference panels (See Method). We prioritized and aggregated all the loci identified from all our previous analysis on the three different traits as our genomic region of interest. We nominated 2 colocalized loci exhibiting strong evidence of shared causal variants (PP > 0.75), including KANSL1 (17:44265839; rs2532240) and CTD − 2501B8.5 (17:61545779; rs6504163) (Fig. 3c, d, Supplementary Table S6). It is demonstrated that the KANSL1 involves the autophagic process damaged mitochondria and accumulation of reactive oxygen species, thereby resulting in defective memory processes and cardiac functions [25]. Meanwhile for CTD-2501B8.5, we highlighted its connection across MPDs and AD related to metabolic syndromes traits identified in previous LDSC analysis, including high blood pressure and high cholesterol and hypertension. Further elucidation of shared colocalized loci showed the putative mechanism between MPDs and AD To determine the functional implications of the nominated colocalized loci, we integrated variant-to-gene (V2G) mapping and expression quantitative trait locus (eQTL) data from FIVEx, the Open Targets Genetics Platform, scQTLbase, and phenome-wide association studies (PheWAS) (Supplementary Table S7 and S8). For the KANSL1 locus (17:44265839; rs2532240), V2G summaries identified 13 significantly associated genes. Among these, ARL17A, ARL17B, STH, CR936218.1, LRRC37A, KANSL1, and KANSL1-AS1 exhibited significant brain-specific eQTLs (Fig. 4a-c, Supplementary Table S7 and S8). Tissue-level eQTLs localized KANSL1 expression to the substantia nigra, while single-cell mapping specifically linked it to oligodendrocytes (Pval = 1.99350 × 10−6), given that oligodendrocytes represent the predominant cell population in this brain region. KANSL1 encodes a component of the nonspecific lethal (NSL) complex, which facilitates the acetylation of histone H4 on lysine 16 (H4K16ac) to promote transcriptional activation. Its deficiency triggers neuronal dysfunction involving memory loss by oxidative stress-induced autophagy [26]. In humans, KANSL1 is integral to histone acetyltransferase (HAT) activity and the formation of WDR5-containing histone-modifying complexes, while in mice models, its dysregulation yields abnormal axonal transport and hyperactive neurological phenotypes. Dysregulation of the epigenetic landscape of normal aging has been observed in AD, and prior studies demonstrate that targeting WDR5 and histone K4 trimethylation recover prefrontal cortex synaptic function and memory-related behaviors in AD mouse models [27]. Elucidation of colocalized loci of major psychiatric disorders and AD with multiple levels of evidence from FIVEx, Open Target Genetics and scQTLbase. a, b left Forest-like plots revealed the associated traits and corresponding beta of the colocalized SNPs (KANSL1) under the category of nervous system disease and phenotype provided in Open Target Genetics Platform. right Heatmap summarized the eQTL evidence and aggregated the score of associated genes from Open Target. c, d The associated traits, eQTL mapping, and functionally implicated genes of colocalized SNPs (CTD-2501B8.5). The second colocalized locus, CTD − 2501B8.5 (17:61545779; rs6504163), maps upstream of the ACE gene. Open Targets analysis supported robust brain-specific eQTL associations for rs6504163 with ACE, KCNH6, MAP3K3, DCAF7, and TACO1 across structures including the cerebellum, caudate, and anterior cingulate cortex (Fig. 4d-f, Supplementary Table S7 and S8). Cell-type-specific eQTL further highlighted significant expression in CD8+ naive T cells. PheWAS analysis confirmed that this variant is broadly associated with nervous system diseases, including various MPDs and AD. More importantly, it is reported significant associations with phenotypes such as high blood pressure, high cholesterol, and responses to renin-angiotensin system agents [28, 29]. These locus-specific functional annotations support our previous findings, a pleiotropic link between MPDs and AD, involving shared cardiovascular and metabolic pathways in the same genomic region. This shared genetic liability aligns with established clinical comorbidities, reinforcing the genomic basis for the high prevalence of hypertension in schizophrenia (~39%) and AD-inclusive dementia (~16%) [30, 31]. Discussion In this study, we performed a large-scale multi-ancestry multi-trait GWAS disentangle the shared genetic architecture between MPDs and AD. By employing MR-MEGA to model ancestral heterogeneity and a random-effect model via PLINK to capture homogeneous effects, our study identified 201 SCZ-associated risk loci, isolating 46 loci driven by strong heterogeneous effects across populations. The integration of diverse ancestries significantly expanded the discovery of disease-associated loci and enabled robust cross-ancestry causal architecture comparisons. However, cross-ancestry analyses remain inherently constrained by variable sample sizes, structural differences in linkage disequilibrium (LD), and disparities in statistical power. While our dual-modeling approach mitigates stratification bias, certain variants may remain intrinsically specific to one population, potentially escaping detection in smaller, non-European cohorts. Subsequent fine-mapping refined these signals, allowing us to nominate credible sets for functional prioritization. We successfully identified loci with less than 5 causal variants under 95% credible set at robust loci for SCZ (e.g., KANSL, PDE4D, SIKE1, LPP), MDD (e.g. RP11-390P2.4), and AD (e.g. BIN1, GRN, CTSB). We also prioritized genomic regions resolved to causal variants such as KANSL1. Including these narrow credible sets accounts for local allelic heterogeneity and expands the potential to uncover complex and novel mechanisms. This study aimed to characterize the pleiotropy between psychiatric and neurodegenerative domains. At the phenotype-level, LDSC analysis revealed the majority of their shared traits (60 of 72) demonstrated consistent directionality, including positive correlations for lipoprotein cholesterol and negative correlations for obesity and weight. In contrast, MDD and AD predominantly demonstrated antagonistic pleiotropy. Of their 121 shared traits, 114 exhibited opposite genetic correlations, wherein genetic liability to MDD positively correlates with metabolic relevant traits, while AD susceptibility correlates negatively with mid-life physiological resilience. Although we nominated and validated several candidates across SCZ and AD, we did not find the shared loci across MDD and AD overall. Further inclusion of more MDD cohorts could be one option to improve the analysis. Bridging these phenotype -level to the variant-level findings, multi-trait colocalization identified specific genomic regions driving this overlap. We nominated two loci (KANSL1 and CTD − 2501B8.5) shared between SCZ and AD (PP > 0.75). Although MDD and AD share broad systemic correlations, we did not detect significant locus-level colocalization between them. This absence reflects the extreme polygenicity and diffuse effect sizes inherent to MDD, suggesting that the clinical and antagonistic genetic overlap between MDD and AD might converge at a broader polygenic network level, rather than through shared single-nucleotide variations. Integrating multi-omics functional evidence including V2G mapping and single-cell eQTLs from FIVEx, Open Targets, and scQTLbase, we studied the mechanisms underlying the colocalized SCZ-AD loci. KANSL1 exhibits significant brain-specific expression, particularly in substantia nigra oligodendrocytes. Its dysregulation impairs epigenetic landscapes, triggers oxidative stress-induced autophagy, and disrupts memory that closely tied to AD progression. CTD − 2501B8.5 was revealed its strong association with cardiovascular and metabolic phenotypes, consistent with our LDSC findings. Despite these findings across MPDs and AD, several limitations should still be kept in mind. First, this study did not explore the casual relationship, thereby further functional study on these variants is needed. Second, the psychiatric symptoms comorbid in AD patients might not be biological equivalent to the psychiatric diseases. Future genomic study restricting to only patients with comorbidity would be necessary. In conclusion, we demonstrated the shared genetics across MPDs and AD, with a combination of transferability analyses covering various levels of evidence genomics, eQTL, single cell eQTL, etc. We identified colocalized loci and revealed the putative overlapping mechanisms across conditions. Overall, our study presented an extensive cross-trait meta-analyses taking multiple ancestries into consideration, disentangling conditions with complex etiology while ensuring the discovery of shared genomic regions in the multi-trait scenario. Methods Study populations and GWAS studies In our multi-ancestry multi-trait (MAMT) genome-wide association meta-analysis, we included GWAS summary statistics from multiple cohorts covering four different ancestry groups: European (EUR), East Asian (EA), Hispanic (HIS), and African American (AFM). The meta-analysis included a total of 169,417, 1,410,479, and 644,188 participants in schizophrenia, major depressive disorders, and Alzheimer’s disease cohorts under case and control settings retrieved from public sources (Supplementary Table S1). These public data have been previously pre-processed, including aligning to GRCh37, removing multi-allelic, filtering rare variants (minor allele frequency MAF < 0.01), and excluding the overlapping participants across datasets upon their release to ensure the bias not being introduced into the meta-analyses. Meta-analysis and defining associated loci Two approaches were applied for meta-analyses, including Meta-Regression of Multi-Ethnic Genetic Association (MR-MEGA v0.2) for heterogeneity analysis and the random effect model from PLINK 1.9 for homogeneity discovery. We used k = 3 as the optimal number of PCs to control overfitting. The random effect analysis in PLINK 1.9 was used to perform random effect meta-analysis to detect homogenous allelic effects. For quality control, all datasets were harmonized, standardized, and columns were re-formatted manually. Significance threshold was set to 5 × 10−8 for all multi-ancestry meta-analysis to keep consistency across different ancestral cohorts. Functional Mapping and Annotation (FUMA GWAS) v1.6.0 was used to find maximal LD blocks around loci that reached P < 5 × 10−8 in the specified multi-ancestry meta-analysis [32]. LD blocks of independent significant SNPs (R2 > 0.6, 1000 G Phase 3 ALL population) with less than 250 kb interval distance were integrated as single genomic regions (loci). Lead SNPs in the loci are determined by further clumping the independent significant variants within the genomic locus (r2 ≥ 0.1). We applied positional mapping, brain specific eQTL mapping and chromatin interaction mapping to define the final genomic risk loci. Fine-mapping Fine-mapping was performed to identify the credible sets from the identified loci, using approximate Bayes’ factors (BF) in favour of association from the meta-regression model implemented. In MR-MEGA, BF was computed during the multi-ancestry meta-analysis, Posterior probabilities (PP) were calculated using single-SNP Bayes factors, and credible sets were generated for each locus (with genomic intervals defined as described above) until the cumulative PP exceeded the specific threshold. In PLINK, BF was computed using the “Sum of Single Effects” (SuSiE) model following the formula and subsequently calculated PP as in MR-MEGA [33, 34]. In practice, we computed the LD matrix from the reference panel (EUR and EA) and used the loci as the input to fit the SuSiE regression. We repeated the other procedure with the same criterion and threshold as the comparison. Genetic correlation analysis across traits using batch LDSC Genetic correlation between our cohorts (SCZ, MDD, and AD) and potential traits (in total, 2871 traits from publicly available data) was estimated by Cross-trait linkage disequilibrium score regression (LDSC) using software package LDSC (v1.0.0) with default settings [18, 19]. We utilized the CTG-VL platform and performed the batch LDSC using from publicly available trait-level data [35, 36]. We submitted the data after harmonizing the column information of each SNP with munge_sumstats package to rs_id, effect allele, non-effect allele, sample size, P-value and beta [37]. Then, we performed stepwise LDSC on every pair of input summary statistics. The results of genetic correlation were extracted and passed through the FDR correction with the threshold P adjusted < 0.05. We then examined the overlapped significant traits across major psychiatric disorders and Alzheimer’s disease, and in the end, visualized with ggplot in R43. Shared genetics identification via multi-trait colocalization analysis Multi-trait colocalization was performed to assess shared genetic aetiology across multiple related traits (e.g. molecular traits, metabolic pathways and complex diseases) by identifying causal pathways, prioritize causal variants and evaluate pleiotropy. Bayesian-based colocalization method Hypothesis Prioritisation in multi-trait Colocalization (HyPrColoc) was applied to discover the shared loci, which were then ranked by the posterior probability (PP) and the ranking of the number of variants under 95% credible set. [20, 38]. The analysis was conducted on different ancestries including European and East Asian separately with ethnicity-specific LD reference panels. Supplementary information Supplementary Figures Supplementary Tables Acknowledgements This work was supported by the Shenzhen Science and Technology program (Grant No. RCBS20231211090801006; JCYJ20240813140400001; JCYJ20250604180726034) funded to Xin Zheng; Noncommunicable Chronic Diseases-National Science and Technology Major Project (Grant No. 2023ZD0504800, 2023ZD0504802), Shenzhen Clinical Research Center for Neurological Diseases (Grant No. LCYSSQ20220823091204009) and Natural Science Foundation of Guangdong Province (2023A1515010169) funded to Lijie Ren. YZ and XZ contributed to conceptualization of the study. YZ, HF and QH contributed to analysis of data. YZ and LR contributed to supervision of analyses. XZ contributed to the initial draft of the manuscript. All authors contributed to critical revisions of the manuscript and approved of the final draft for submission. Data availability All GWAS summary statistics are retrieved from publicly available sources (Supplementary Table S1). For Alzheimer’s disease, the MR-MEGA and PLINK meta-analysis were downloaded via the Broad’s Neurodegenerative Disease Knowledge Portal (https://ndkp.hugeamp.org/dinspector.html?dataset=Lake2023_AD_Mixed). The multi-ancestry meta analysis for comparison to get the distinct variants were retrieved from https://www.nature.com/articles/s41586-022-04434-5#Sec2 for schizophrenia, and https://www.cell.com/cell/fulltext/S0092-8674(24)01415-6#app-1 for major depressive disorder. The datasets for multi-trait genetic correlation analysis were all publicly available and internally stored in the CTG virtual lab platform (https://vl.genoma.io/). Code availability Software and packages used in this study include MR-MEGA (https://genomics.ut.ee/en/tools), PLINK v1.9 (https://www.cog-genomics.org/plink/), LD score regression software LDSC (https://github.com/bulik/ldsc), hyprcoloc (https://jrs95.github.io/hyprcoloc/), FUMA GWAS (https://fuma.ctglab.nl/, CTG-VL (https://vl.genoma.io/), Open Target Platform (https://platform.opentargets.org/), GWASATLAS (https://atlas.ctglab.nl/), FIVEx (https://fivex.sph.umich.edu/), scQTLbase (http://bioinfo.szbl.ac.cn/scQTLbase/Home/), R (https://cran.r-project.org/), and Python (https://www.python.org/). Competing interests The authors declare no known conflict of interest that could have appeared to influence the work reported in this paper. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Lijie Ren, Email: renlijie72@126.com. Youcheng Zhang, Email: yc.zhang3@siat.ac.cn. Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41398-026-04213-0. References Associated Data Supplementary Materials Supplementary Figures Supplementary Tables Data Availability Statement All GWAS summary statistics are retrieved from publicly available sources (Supplementary Table S1). For Alzheimer’s disease, the MR-MEGA and PLINK meta-analysis were downloaded via the Broad’s Neurodegenerative Disease Knowledge Portal (https://ndkp.hugeamp.org/dinspector.html?dataset=Lake2023_AD_Mixed). The multi-ancestry meta analysis for comparison to get the distinct variants were retrieved from https://www.nature.com/articles/s41586-022-04434-5#Sec2 for schizophrenia, and https://www.cell.com/cell/fulltext/S0092-8674(24)01415-6#app-1 for major depressive disorder. The datasets for multi-trait genetic correlation analysis were all publicly available and internally stored in the CTG virtual lab platform (https://vl.genoma.io/). Software and packages used in this study include MR-MEGA (https://genomics.ut.ee/en/tools), PLINK v1.9 (https://www.cog-genomics.org/plink/), LD score regression software LDSC (https://github.com/bulik/ldsc), hyprcoloc (https://jrs95.github.io/hyprcoloc/), FUMA GWAS (https://fuma.ctglab.nl/, CTG-VL (https://vl.genoma.io/), Open Target Platform (https://platform.opentargets.org/), GWASATLAS (https://atlas.ctglab.nl/), FIVEx (https://fivex.sph.umich.edu/), scQTLbase (http://bioinfo.szbl.ac.cn/scQTLbase/Home/), R (https://cran.r-project.org/), and Python (https://www.python.org/).