Novel Bipolar And Schizophrenia Risk Genes Identified Through Genic Associations In Transcriptome Imputation

Abstract

Background Genome-wide association studies (GWAS) have identified genomic regions harboring GWA-significant common variation. Undoubtedly many more remain to be identified. Large-scale transcriptomic datasets have pinpointed variants regulating gene expression (eQTLs) in specific tissues; moreover, these results can be used to impute tissue-specific gene expression levels (GREx) for any genetically-characterized sample. Building from these observations, recent research has pointed out that imputed gene expression in larger case-control samples can be tested to identify novel risk genes. Here, we use the largest existing transcriptomic database of brain tissue, along with data from ten GTEx brain regions, to impute GREx and test for association in GWAS datasets of schizophrenia (SCZ) and bipolar disorder (BIP) from the Psychiatric Genomics Consortium (PGC). Methods Following systematic comparison of prediction modelling techniques, models were created for 13,452 genes from 668 individuals with imputed genotype and RNA-seq data from the CommonMind Consortium(CMC). GREx was imputed in a novel cohort of ~400 individuals and had good prediction accuracy (~85% of genes have R2 0.01-0.5), in line with previous models and results. Our results were compared across ancestries and to individual- vs. summary-level data. Our analysis used CMC and 17 GTEx models to impute GREX in PGC SCZ (~34,000 cases/45,000 controls) and BIP (~20,000/31,000). Models corresponded to 10 brain regions, 6 heart tissues, and a whole blood sample. We tested imputed GREX for association with SCZ and BIP. Results We identified 302 genes with genome-wide significant (p Discussion We have used GREx imputation to harness large GWAS sample sizes and yield biologically relevant data about disease architecture, resulting in the first prediction models for the DLPFC. These predictors may identify novel genic associations, as shown for BIP and SCZ. We will expand on these analyses to probe the general relationship between GWAS-loci, differential gene expression, and eQTLs. We will further use these predictors to address questions about the role of gene expression in disease risk and heritability.