March 6th, 2019
Gene-environment (G × E) interaction is important for many complex traits. For a case-control study of a disease trait, logistic regression is the standard approach to model disease as a function of a gene (G), environmental factor, G × E interaction, and adjustment covariates. We propose an alternative model with G as the outcome and show how it provides a unified framework for obtaining all of the common G × E tests. These include the 1-degree-of-freedom (df) test of G × E interaction, the 2-df joint test of G and G × E, the case-only and empirical-Bayes tests, and several two-step tests. In the context of this unified model, we propose a novel 3-df test and demonstrate that it provides robust power across a wide range of underlying G × E interaction models. We demonstrate the 3-df test in a genomewide scan of G × Sex interaction for childhood asthma using data from the Children’s Health Study. This scan identifies a strong G × Sex interaction at the phosphodiesterase (PDE) 4D locus, a known asthma-related locus, with a strong effect in males (odds ratio = 1.70 per allele, P = 3.8 × 10-8) and virtually no effect in females. We describe a software program, GxEScan, which can be used to fit standard and unified models for genomewide G × E studies.