Polygenic Liability for Major Psychiatric Disorders and Clinical Features of Schizophrenia

Polygenic scores (PGS) for schizophrenia (SCZ), major depressive disorder (MDD), and bipolar disorder (BD) are used to explain heterogeneity in schizophrenia, but European-derived scores applied in ancestrally mixed samples risk confounding by population structure. In the CATIE trial, we tested 3 a priori directional hypotheses under one rule: covariate-adjusted model, one-tailed p, within-family false-discovery-rate correction. The primary sample was 407 participants of European ancestry, with principal components recomputed within it; scores were computed with PRS-CS. Higher BD-PGS predicted less improvement in PANSS Total (β = 1.69 per SD, one-tailed p = 0.036, q = 0.036) and earlier treatment discontinuation (HR = 1.16, 95% CI 1.004-1.340, q = 0.036); joint models with all three scores did not account for either effect. MDD-PGS and SCZ-PGS were unassociated with baseline symptoms. The four motor/behavioral item-level associations with SCZ-PGS detected in the ancestrally mixed sample (ρ = 0.11-0.14) did not survive European-only analysis, were not reproduced by removing 72 participants at random (permutation p ≤ 0.002), and reflect a score that differs by 1.5 SD between ancestry groups: population stratification. Effects are small and not clinically actionable; PGS-symptom associations from mixed samples should be treated as provisional until replicated within one ancestry.