Generating A Database of Phenotypic Consequences of Gene Regulation Across 40 Human Tissues

Abstract GWAS and sequencing studies have yielded thousands of genetic variants robustly associated genetic complex traits. However, the underlying biology of those associations needs to be further elucidated. To address this problem we developed PrediXcan, a mechanistically driven test, which was motivated by the accumulating evidence that the regulation of gene expression levels as well as splicing events have an important role in the genetic control of complex phenotypes. By using genotype to predict expression (or other intermediate molecular traits) and correlating them with the trait of interest, we can assess the phenotypic consequences of genetic variation through different intermediate processes. By collapsing the associations into functionally relevant units such as genes, we reduce the multiple testing burden. The method also provides direction of the effects. To implement the tissue specific analyses, we have developed prediction models for gene expression in 40 human tissues using the GTEx Consortium and Depression Genes Network data. Despite these advantages, the genetic architecture of psychiatric phenotypes is largely polygenic with modest effect sizes making discoveries only possible for very large sample sizes. Therefore we have extended our method so that only summary statistics are needed to infer PrediXcan results. This allows us to leverage the large meta analysis efforts that have collected hundreds of thousands of samples. Using this new method called MetaXcan we have generated results for 117 phenotypes with publicly available GWAS meta-analysis results. We validate our approach by re-identifying many established genes but in many cases, we find evidence that genes in the vicinity of reported ones are more likely mediators of the phenotype. Furthermore, we make this results database publicly available (http://gene2pheno.org). The database should be a valuable resource for the community to explore the phenotypic consequences of gene regulation. Software and all prediction models necessary to reproduce them or apply to new datasets are made publicly available on https://github.com/hakyimlab/PrediXcan.