Pathway-Based Polygenic Risk Scores for Schizophrenia and Associations With Reported Psychotic-like Experiences and Neuroimaging Phenotypes in the UK Biobank

BACKGROUND: Schizophrenia is a heritable psychiatric disorder with a polygenic architecture. Genome-wide association studies have reported that an increasing number of risk-associated variants and polygenic risk scores (PRSs) explain 17% of the variance in the disorder. Substantial heterogeneity exists in the effect of these variants, and aggregating them based on biologically relevant functions may provide mechanistic insight into the disorder. METHODS: Using the largest schizophrenia genome-wide association study conducted to date, we associated PRSs based on 5 gene sets previously found to contribute to schizophrenia pathophysiology—postsynaptic density of excitatory synapses, postsynaptic membrane, dendritic spine, axon, and histone H3-K4 methylation—along with respective whole-genome PRSs, with neuroimaging (n > 29,000) and reported psychotic-like experiences (n > 119,000) variables in healthy UK Biobank subjects. RESULTS: Several variables were significantly associated with the axon gene-set (psychotic-like communications, parahippocampal gyrus volume, fractional anisotropy thalamic radiations, and fractional anisotropy posterior thalamic radiations (β range −0.016 to 0.0916, false discovery rate–corrected p [p(FDR)] ≤ .05), postsynaptic density gene-set (psychotic-like experiences distress, global surface area, and cingulate lobe surface area [β range −0.014 to 0.0588, p(FDR) ≤ .05]), and histone gene set (entorhinal surface area: β = −0.016, p(FDR) = .035). From these, whole-genome PRSs were significantly associated with psychotic-like communications (β = 0.2218, p(FDR) = 1.34 × 10(−7)), distress (β = 0.1943, p(FDR) = 7.28 × 10(−16)), and fractional anisotropy thalamic radiations (β = −0.0143, p(FDR) = .036). Permutation analysis revealed that these associations were not due to chance. CONCLUSIONS: Our results indicate that genetic variation in 3 gene sets relevant to schizophrenia may confer risk for the disorder through effects on previously implicated neuroimaging variables. Because associations were stronger overall for whole-genome PRSs, findings here highlight that selection of biologically relevant variants is not yet sufficient to address the heterogeneity of the disorder.