Testing for gene-environment (GxE) interaction using p-value aggregation identifies many GxE loci

Abstract

Genome-wide gene-environment (GxE) interaction studies have seen limited success in detecting reliable GxE signals. A standard genome-wide GxE scan assumes a single genetic mode of inheritance, such as an additive model. It can lead to reduced statistical power when the true genetic model is non-additive, such as a recessive model. We propose a robust GxE testing approach that uses Cauchy p-value aggregation. It combines the p-values from GxE tests based on the additive, dominant, and recessive genetic models. Using extensive simulation studies, we demonstrate that the p-value combination strategy offers a robust and powerful approach to identifying GxE interactions regardless of the underlying genetic model. The method is substantially more powerful than the additive model when the true genetic model is recessive. It is also more powerful than the general two-degree-of-freedom genotypic test for GxE interaction. We apply our approach to analyze GxE interactions in the UK Biobank data across several combinations of phenotypes and environmental factors. For glycated hemoglobin (HbA1c) level, treating cumulative smoking exposure as the lifestyle factor, our approach identified 82 independent GxE loci while controlling FDR at 5%. The GxE test based on the additive genetic model detected 24 loci. For type 2 diabetes with sleep duration as a lifestyle factor, the proposed approach detected 563 independent GxE loci at 5% FDR, substantially exceeding the number of discoveries by the other approaches.