MODELING COMMON AND DISTINCT GENETIC LIABILITY TO SUICIDALITY PHENOTYPES AND PSYCHIATRIC DISORDERS USING GENOMIC STRUCTURAL EQUATION MODELING AND MULTIVARIATE GWAS

Background Suicidality phenotypes, specifically suicidal ideation (SI), non-fatal suicide attempt (SA) and suicide death (SD), are substantially heritable, with twin and family studies estimating heritability at 30-55%. Recently GWAS have identified both shared and phenotype-specific loci across suicidality phenotypes, which also exhibit strong but incomplete genetic correlations, underscoring the presence of both common and distinct genetic contributions. Here, we present results from the Psychiatric Genomics Consortium Suicide Working Group’s (PGC SUI) first multivariate GWAS of a common factor underlying suicidality, alongside gwas-by-subtraction analyses to identify phenotype-specific genetic effects independent of other suicidality phenotypes and psychiatric disorders. Methods Data comprise summary statistics from PGC SUI’s most recent European-ancestry GWAS of SI (N cases=176,147, N controls=1,010,300), SA (N cases=53,919, N controls=1,063,988), and SD (N cases=7,584, N controls=652,070). Genomic structural equation modeling was used to model a common latent factor capturing shared genetic liability across SI, SA, and SD. SNP effects were estimated at the level of this latent suicidality factor to conduct a common factor GWAS and identify shared genetic risk loci as well as QSNPs (i.e., SNPs with phenotype-specific effects). Local genetic correlations were estimated using SUPERGNOVA. GWAS-by-subtraction was additionally applied to parse out genetic effects specific to each suicidality phenotype, after conditioning on (1) the other suicidality phenotypes and (2) 14 psychiatric disorders. Results The suicidality common factor GWAS identified 37 genome-wide significant loci, including seven not identified in the univariate GWAS or previous suicidality GWAS. Eight significant QSNPs, mapping to two genomic risk loci in the CFH and ARMS2 genes, indicated unique associations with SD; these findings were replicated in GWAS-by-subtraction analyses of SD after removing shared genetic liability with SI and SA. Local genetic correlation analyses revealed 10, 3, and 1 loci with significant positive genetic overlap between SI-SA, SI-SD, and SA-SD, respectively. Conditional models demonstrated significant residual variance remaining in all suicidality phenotypes after accounting for psychiatric disorders, as well as heterogeneous contributions of specific psychiatric disorders across suicidality phenotypes. GWAS-by-subtraction are underway to calculate SNP effects on suicidality phenotypes independent of 14 psychiatric disorders. Discussion Our findings indicate that suicidality reflects both a shared genetic liability and phenotype-specific effects. Multivariate GWAS increased locus discovery relative to univariate approaches, while QSNPs and GWAS-by-subtraction highlight heterogeneity, particularly for SD. Local genetic correlations suggest that overlap across phenotypes may cluster in specific regions. Importantly, residual genetic variance after conditioning on psychiatric disorders supports partial genetic specificity in the heritability of suicidality, and that it is not merely a result of the inheritance of psychiatric disorders. Together, these results refine our understanding of the genetic architecture of the suicidality spectrum, supporting its conceptualization as both a shared liability and a set of related but distinct phenotypes that extend beyond its psychiatric comorbidities.