A gene-level test for directional selection on gene expression

Human phenotypes and evolutionary fitness are influenced by both coding and non-coding genetic variants. However, most variants associated in genome-wide association studies and. scans for selection are non-coding. Interpretation of these variants' effects and understanding of the way in which they contribute to phenotypic variation and adaptation is therefore limited by our understanding of gene regulation and by the difficulty in confidently linking non-coding variants to genes. To overcome this, we developed a gene-by-gene test for population-specific selection based on combinations of regulatory variants. Specifically, we extended the Qx test for polygenic selection to gene expression models trained using joint-tissue imputation, a transcriptome-wide association method trained on paired genotype and RNA-seq data. We used effect sizes and variants from those models to calculate Qx statistics for 17,388 protein-coding genes based on allele frequencies in the twenty six 1000 Genomes populations. We identified 45 genes with significant evidence (FDR < 0.1) for selection, including FADS1, KHK, SULT1A2, ITGAM, and genes in the HLA region. We further confirm that significant selection signals do correspond to plausible population-level differences in predicted expression. Our gene-level Qx score is independent of other methods for detecting selection based on genomic variation, may therefore be useful when used in combination with more traditional selection tests to specifically identify selection on regulatory variation. However, we find that very few (0.2%) genes have strong evidence for directional, population-specific selection on the component of their expression that is predicted by cis-regulatory variants. While this is consistent with most cis-regulatory variation evolving under genetic drift or stabilizing selection, it is also possible that any effects are smaller than we can detect, or that population-specific selection is driven by tissue-specific or trans effects. Overall, our results demonstrate the utility of one approach to combining population-level information with functional data to understand the evolution of gene expression. ### Competing Interest Statement The authors have declared no competing interest.