Disorder-specific and shared genetic architecture underlying schizophrenia and bipolar disorder.

Schizophrenia (SCZ) and bipolar disorder (BIP) share substantial genetic liability, yet exhibit heterogeneity in cognition, illness course and cardiometabolic health. Conventional disorder-level genome-wide association studies (GWASs) aggregate genetic liability obscuring this heterogeneity. We applied zero-phase component analysis whitening and common factor GWAS to SCZ and BIP summary statistics, deriving a shared component (PSYshared) and two mutually orthogonal, disorder-predominant components (SCZ~BIP and BIP~SCZ). We then examined whether the three profiles differed in locus architecture, pathway enrichment, developmental expression and intermediate-phenotype associations. The decomposition identified 261 loci, of which 201 (77%) were component-specific and 16 loci were not reported in any previous larger SCZ and BIP GWASs or multivariate psychosis GWASs. PSYshared was enriched for synaptic architecture, nuclear organisation and protein trafficking, with developmental signals spanning early- and late-mid-prenatal stages, late infancy and young adulthood. SCZ~BIP implicated cortical neuronal differentiation, DNA/RNA regulation and protein turnover, with developmental associations in late infancy and young adulthood. BIP~SCZ showed synaptic release and transport signals, alongside exploratory associations with gene/protein regulation and mitochondrial translation (nominal P < 0.001), with significant developmental enrichment in late infancy. Correlations of SCZ and BIP with intermediate traits were asymmetric, whereas SCZ~BIP and BIP~SCZ diverged in opposite directions across cognitive performance, educational attainment, metabolic syndrome, and systemic inflammatory markers. In contrast, both orthogonal components correlated positively with risk taking, with a larger estimate for BIP~SCZ; PSYshared showed the largest estimate among the derived components. Steiger tests showed that the components differed significantly across all six phenotypes. Component contrasts were 1.63-2.07-fold greater than corresponding SCZ-BIP contrasts with the increase reaching significance for cognitive performance and metabolic syndrome. These findings support a multidimensional representation of SCZ-BIP liability and provide a framework for testing whether relative component burdens account for variation in cognition, cardiometabolic and inflammatory burden, illness course, and treatment response beyond diagnosis.