Transcriptome Wide Association Study of Individually Imputed Genetically Regulated Gene Expression in the Million Veteran Program

ABSTRACT

Background: Most neuropsychiatric disorders are moderately heritable but characterized by many genetic risk variants with weak effects. As such, it is difficult to point to direct causes or elucidate mechanisms of action. Despite the ease in gathering genetic data from humans, genetic data does not easily explain mechanistic effects. Gene expression on the other hand, which can more easily explain mechanistic effects, is harder to gather, especially in brain regions that are critical to the understanding of neuropsychiatric disease. To address this, we developed methods to impute genetically regulated gene expression (GReX) from genotypes and imputed GReX in over 440,000 European individuals in the Million Veteran Program (MVP) for a wide variety of tissues and cell types. Method(s) We use EpiXcan (based on PrediXcan) to develop machine learning models from training genotype, expression, and epigenetic data. We use custom scripts to impute individual GReX and perform a variety of downstream association analyses, including GReX Phenome Wide Association Studies (PheWAS) and Transcriptome Wide Association Studies (TWAS). Result(s) Results show an overlap in Schizophrenia genes identified by individual level TWAS and those identified by summary level TWAS informed by GWAS. TWASs for neuropsychiatric phenotypes identify genes established in the literature, but also novel targets. Inverse-variance meta-analyzed single gene imputation efforts across ancestries confirm clinical results obtained from COVID-19 positive individuals in both IL10RB and IFNAR2. GReX PheWAS for these particular genes using a novel negative binomial distribution for phecodes confirm COVID-19 related phenotypes. Finally, we describe various enriched pathways found in a COVID-19 TWAS, including immunological pathways. Discussion(s) GReX presents a unique solution to integrate effects across the genome and increase sample size in gene expression analyses. We are pursuing the creation of additional EpiXcan models, improved statistical methods for downstream association analyses, and replication efforts across biobanks. We plan to perform these analyses in all ancestries, available EpiXcan and PrediXcan models, and phenotypes. Disclosure Nothing to disclose. Copyright © 2022