The role of interacting determinants in the localization of genes

We describe the potential gains in power for localizing disease genes that can be obtained by allowing for interactions with environmental agents or other genes. The focus is on linkage and association methods in nuclear families with dichotomous phenotypes. A logistic model incorporating various main effects and interactions is used for penetrance, but similar methods apply to censored age-at-onset or continuous phenotypes. We begin by discussing the influence of gene-environment interactions in segregation analysis, illustrated with analysis of smoking as a modifying factor for lung cancer. We then discuss a number of approaches to linkage analysis-model-free and model-based(including generalized estimating equations) incorporating interactions with environmental factors and other genes, either candidate genes or linked loci. We find that a test of heterogeneity in IBD sharing probabilities across strata defined by sharing of environmental factors can offer greater power for detecting linkage than the simple mean test, provided the interaction effect is sufficiently strong; we explore the conditions under which this gain in power occurs. Finally, we describe approaches for testing association and disequilibrium involving interactions, utilizing case-control, case-parent, and pedigree-based approaches. A technical problem that must be addressed in many analyses is the effect of missing data on environmental covariates; we use multiple imputation in an analysis of lung cancer segregation to illustrate an approach to this problem.