Identification of pleiotropy at the gene level between psychiatric disorders and related traits

The study investigates whether psychiatric disorders and related endophenotypes share genetic architecture at the gene level, beyond known overlaps at the single-SNP level. Motivated by shared clinical features and prior reports of genetic correlations among psychiatric, cognitive, brain morphometric, and personality traits, the authors test the hypothesis that independent variants within the same gene can associate with different traits (gene-level pleiotropy) or with the same trait (allelic heterogeneity). Identifying such gene-level overlaps could refine biological understanding and inform functional follow-up and therapeutic targeting.

Prior work has established polygenic architectures and shared heritability among brain-related traits and disorders. The BrainStorm consortium reported SNP-level genetic correlations among ADHD, bipolar disorder, major depressive disorder, and schizophrenia, and correlations between psychiatric, cognitive, and personality traits. GWAS have identified variants for educational attainment, general cognitive function, and intelligence, with demonstrated overlaps with schizophrenia and bipolar disorder. Neuroimaging consortia (e.g., ENIGMA) have reported structural brain differences in schizophrenia and some polygenic overlap with brain volumes, though findings vary. Personality traits, including neuroticism, are risk factors or modifiers for psychiatric disorders, and genetic correlations have been reported between psychiatric diagnoses and personality or well-being. However, prior studies largely focused on SNP-level pleiotropy rather than gene-level analyses capturing independent signals within loci.

The authors compiled publicly available GWAS summary statistics for 27 phenotypes: seven psychiatric disorders (schizophrenia, autism spectrum disorder, major depressive disorder, anorexia nervosa, ADHD, bipolar disorder, anxiety), brain morphometric traits (e.g., subcortical volumes from ENIGMA), cognitive traits (educational attainment, general cognitive function), personality traits (subjective well-being, depressive symptoms, neuroticism, extraversion, openness, agreeableness, conscientiousness), children’s aggressive behaviour, and loneliness. All GWASs were of European ancestry. Unified quality control was applied to each GWAS summary set per cojo-GCTA recommendations: removal of poorly imputed variants (imputation score < 0.9), ambiguous SNPs, markers with minor allele frequency < 0.1, and indels; mapping to hg19 (liftover from hg18 when needed); analyses restricted to autosomes. Independent association signals per trait were identified using conditional and joint (cojo) stepwise regression (cojo-GCTA), using a merged Norwegian and German LD reference (~4678 individuals; ~7.1M genotyped/imputed markers). Cojo-GCTA threshold previously validated by the authors was used to define independent signals. Independent SNPs across all traits were annotated to RefSeq genes (2017 freeze) using LDsnpR, assigning SNPs within ±10 kb of gene boundaries. Overlapping or antisense genes sharing the same annotated markers were merged into gene_blocks, defining 1161 genes or gene-groups from 1410 annotated genes. Regions with >2 association signals after cojo-GCTA (p < 1×10⁻⁷) were further LD-pruned (PLINK, 10 Mb window, r² = 0.2) to derive 8772 independent markers from 177,344 SNPs mapped to 226 selected regions, setting an experiment-wide Bonferroni threshold of 0.05/8772 = 5.70×10⁻⁶. Three scenarios were defined: Scenario I (gene-level pleiotropy: multiple LD-independent variants within a gene/gene-group associated with at least two traits), Scenario II (allelic heterogeneity within one trait), Scenario III (SNP-level pleiotropy: same SNPs or LD-linked SNPs associated with multiple traits; LD defined as r² ≥ 0.2 in EUR 1000 Genomes). Gene-based analysis was also conducted: for all 1410 genes harboring at least one cojo-GCTA SNP, Brown’s method was used to compute gene scores per trait, incorporating all SNPs within gene ±10 kb and correcting for LD using the same reference. FDR correction within trait was applied; across-trait significance threshold set at 0.05/27 = 0.0018, and genes further filtered by minimal p ≤ 5.70×10⁻⁶. Haplotype analyses (Haploview, four-gamete rule, threshold 0.8, window 500 kb; haplotype frequency >10%) were performed for an illustrative locus (MAN2A1) using the Norwegian-German reference panel to explore whether independent signals reflected different haplotypes.

• Across 27 GWAS, conditional regression identified 2190 unique associated SNPs; 1415 were annotated to 1410 genes (merged into 1161 genes/gene-groups), while 775 SNPs were intergenic.
• 226 gene-groups displayed significant association to at least two traits; after applying the experiment-wide threshold (5.70×10⁻⁶), three scenarios emerged: • Scenario I (gene-level pleiotropy, LD-independent markers across traits): 25 genes. Examples include RERE, LOC102724552, GPM6A (schizophrenia, educational attainment); RBKS (schizophrenia, bipolar disorder); DLGAP2, ERICH1-AS1 (schizophrenia, ADHD); GLIS3 (schizophrenia, neuroticism); JADE2 (educational attainment, bipolar disorder); ASTN2 (autism spectrum disorder, bipolar disorder); FHIT (schizophrenia, major depressive disorder); TEAD1 (educational attainment, subjective well-being); TENM4, MACROD2 (general cognitive function, bipolar disorder); EXT1 (general cognitive function, autism); LRRC4C (general cognitive function, anorexia nervosa); CACNA1E (general cognitive function, neuroticism); CTNNA2, LRP1B, ATXN1, SNX29, CDH2 (educational attainment, general cognitive function); ZNF385B, MIPEPP3, PHACTR3 (general cognitive function, schizophrenia). RBFOX1 showed two independent signals with both schizophrenia and general cognitive function (four independent hits in one gene). • Scenario II (allelic heterogeneity within a trait): 9 genes, including BSN and CTNNA3 (educational attainment), RPS6KA2 (bipolar disorder), NKAIN2 and CDKAL1 (general cognitive function). For Alzheimer’s disease, allelic heterogeneity was observed at the NECTIN2–TOMM40–APOE–APOC1 locus. • Scenario III (SNP-level pleiotropy, LD-linked markers across traits): 47 loci (53 genes). For DCC and STK24, 5 and 4 SNPs were associated with 6 and 4 traits respectively. Comparison with prior marker-level overlap for schizophrenia and education replicated 8/10 scenario III markers.
• Mixed cases (both LD-dependent and -independent signals): multiple loci including CACNA1C (two independent schizophrenia signals; rs10744560 also associated with bipolar disorder).
• Gene-based (Brown score) analysis identified 379 genes significantly associated with at least two traits (study-wide), including 23 additional genes with gene-level pleiotropy not captured by SNP-based analysis: LRRN2, MAN2A1, EYS (schizophrenia, educational attainment); COL16A1, EFNA5, SHANK3 (educational attainment, general cognitive function); BRE-AS1, LINC01378 (schizophrenia, bipolar disorder); CLU, MIR6843 (schizophrenia, Alzheimer’s disease); SFXN5, SATB2, FXP1, CKB, TRMT61A, APOPT1 (schizophrenia, general cognitive function); TEX41 (ADHD, educational attainment); ZMIZ2, TMEM245, SLCO3A1 (educational attainment, general cognitive function); KCNC2 (educational attainment, neuroticism); CDH8 (ADHD, general cognitive function); TCF4 (schizophrenia, neuroticism). Additionally, 104 genes showed LD-linked marker pleiotropy and 48 showed mixed scenarios.
• Haplotype example (MAN2A1, chr5:109.03–109.22 Mb): rs4388249_T associated with schizophrenia (p=3.05×10⁻8) lies on a different common haplotype than rs1368357_T associated with lower educational attainment (p=3.37×10⁻9); the two SNPs are not in LD (r < 0.2), indicating distinct haplotypic associations within the same block.
• Overlap patterns were strongest among pairs such as schizophrenia/general cognitive function, schizophrenia/educational attainment, schizophrenia/bipolar disorder, and schizophrenia/neuroticism, possibly reflecting both true shared biology and greater statistical power in larger GWAS.

The findings demonstrate that pleiotropy among psychiatric and related traits exists not only at single-SNPs but also at the gene level, where distinct LD-independent variants within the same locus associate with different traits. This reveals biological relationships that SNP-level analyses can miss, refining candidate genes for functional studies and therapeutic development. Gene-level pleiotropy and allelic heterogeneity illuminate complex genomic architecture, including multiple functional variants, isoform-specific effects, or different haplotypes within a locus influencing separate phenotypes. The enrichment of large genes among pleiotropic loci may reflect both biological reality (more variants and isoforms increasing opportunities for pleiotropic effects) and statistical bias. Examples such as CLU (independent signals in schizophrenia and Alzheimer’s disease) underscore potential shared pathways (e.g., neuroinflammation, blood–brain barrier and amyloid pathways), highlighting the necessity to consider cross-trait effects when prioritizing drug targets to avoid unintended consequences. Haplotype analyses (e.g., MAN2A1) support that independent associations can map to distinct common haplotypes, suggesting follow-up via deep sequencing to detect potential synthetic associations or rare causal variants. While SNP-level pleiotropy is more prevalent overall, unique gene-level overlaps justify integrating both approaches for a comprehensive view of shared genetic architecture.

This study systematically identifies gene-level pleiotropy across psychiatric disorders and related cognitive, brain morphometric, and personality traits, alongside allelic heterogeneity within traits. By leveraging conditional regression and gene-based scoring, the authors highlight 48 genes with independent cross-trait signals and additional loci detectable only via gene-based analyses. These results prioritize genes for deep genotyping/sequencing and functional studies, and inform drug target selection by revealing cross-disorder implications (e.g., shared signals between schizophrenia and Alzheimer’s disease). Future work should integrate multiple genetic evidence layers (GWAS, rare variants, CNVs), perform fine-mapping and haplotype-aware analyses across larger, ancestrally diverse datasets, and develop robust regulatory-region annotations to refine causal mechanisms.

• Potential sample overlap across GWAS meta-analyses (especially among ENIGMA brain volumes, cognitive and personality datasets) could inflate apparent cross-trait pleiotropy via LD-linked signals despite efforts to exclude overlaps where possible.
• Annotations were limited to gene bodies ±10 kb; absence of comprehensive regulatory-region mapping may misattribute associations to nearby genes (e.g., GLIS3 versus regulatory effects on SLC1A1).
• More stringent multiple-testing corrections for gene-based analyses versus SNP-level analyses might undercount gene-level pleiotropy relative to SNP-level results.
• Heterogeneity in GWAS sample sizes and trait heritabilities affects power; larger studies (e.g., schizophrenia, educational attainment, general cognitive function) yielded more overlaps, while smaller brain-structure GWAS had limited power.
• LD reference comprised a merged Norwegian–German European sample, a proxy that may introduce noise if LD patterns differ from discovery cohorts, potentially affecting independent signal selection and haplotype inference.