Mental disorders are among the top causes of death and disability in the United States and are shaped by genetic influences. The development of polygenetic risk scores (PRSs) has advanced our understanding of both the etiology and prediction of mental illness. However, PRS have largely been generated from case- control studies of diagnoses. Consequently, the predictive utility of PRS has been affected by diagnostic unreliability, within-disorder heterogeneity, and definitional overlap among diagnoses. These problems limit predictive power and specificity of genetic findings. These problems may be addressed using empirical, dimensional phenotypes that are reliable, homogeneous, and distinct. This R21 proposes to test this hypothesis by deriving PRS for both diagnoses and empirical phenotypes developed by the Hierarchical Taxonomy of Psychopathology consortium, a group of over 100 scientists. Genome wide association studies of empirical and diagnostic phenotypes will be performed in two discovery cohorts (UK Biobank, N=500,000, and the Philadelphia Neurodevelopmental Cohort, N=10,000). The resulting PRS will be evaluated in three replication cohorts (Minnesota Twin Family Study, N=8,900, Tracking Adolescents? Individual Lives Study, N=2,200, and the Adolescent Brain Cognitive Development study, N=10,600). We will test whether HiTOP PRS outperform diagnostic PRS in power and precision, as indexed by (1) greater heritability and lower pleiotropy and (2) greater predictive power for target phenotype and specificity from non-target phenotypes. Success of this project will result in more powerful and precise targets for future genetic research. Empirical phenotypes can be scaled to large genetic data collection efforts, improving the efficiency of psychiatric genetic research, accelerating the rate of gene discovery, and strengthening genetic prediction.
Mental illnesses are highly heritable, but current genetic association research is based on categorical diagnoses, and are subsequently limited by diagnostic heterogeneity and comorbidity. We propose to use empirically-derived phenotypes for genetic association, improving the power and precision of gene-based prediction.
