Functional and molecular characterization of suicidality factors using phenotypic and genome-wide data.

Genome-wide association studies (GWAS) of suicidal thoughts and behaviors support the existence of genetic contributions. Continuous measures of psychiatric disorder symptom severity can sometimes model polygenic risk better than binarized definitions. We compared two severity measures of suicidal thoughts and behaviors at the molecular and functional levels using genome-wide data. We used summary association data from GWAS of four traits analyzed in 122,935 individuals of European ancestry: thought life was not worth living (TLNWL), thoughts of self-harm, actual self-harm, and attempted suicide. A new trait for suicidal thoughts and behaviors was constructed first, phenotypically, by aggregating the previous four traits (termed "suicidality") and second, genetically, by using genomic structural equation modeling (gSEM; termed S-factor). Suicidality and S-factor were compared using SNP-heritability (h 2 ) estimates, genetic correlation (r g ), partitioned h 2 , effect size distribution, transcriptomic correlations (ρ GE ) in the brain, and cross-population polygenic scoring (PGS). The S-factor had good model fit (χ 2  = 0.21, AIC = 16.21, CFI = 1.00, SRMR = 0.024). Suicidality (h 2  = 7.6%) had higher h 2 than the S-factor (h 2  = 2.54, P diff  = 4.78 × 10 -13 ). Although the S-factor had a larger number of non-null susceptibility loci (π c  = 0.010), these loci had small effect sizes compared to those influencing suicidality (π c  = 0.005, P diff  = 0.045). The h 2 of both traits was enriched for conserved biological pathways. The r g and ρ GE support highly overlapping genetic and transcriptomic features between suicidality and the S-factor. PGS using European-ancestry SNP effect sizes strongly associated with TLNWL in Admixed Americans: Nagelkerke's R 2  = 8.56%, P = 0.009 (PGS suicidality ) and Nagelkerke's R 2  = 7.48%, P = 0.045 (PGS S-factor ). An aggregate suicidality phenotype was statistically more heritable than the S-factor across all analyses and may be more informative for future genetic study designs interested in common genetic factors among different suicide related phenotypes.