Commentary: Mendelian randomization and gene-environment interaction

Genetic association studies for non-familial diseases typically focus on a particular candidate gene and one or more environmental exposures (gene–environment interaction). The long term rationale of such studies appears to be that an accumulation of knowledge regarding susceptibility genes will allow us to identify high risk population subgroups. This in turn may subsequently allow the development of intervention strategies aimed at such high risk groups including modification of lifestyle habits and increased surveillance for those at most risk. However, the feasibility of this strategy is uncertain and indeed it may ultimately prove to be unfeasible, especially if there is a strong stochastic component inherent in the development of individual cases of a particular disease. An alternative rationale for conducting genetic susceptibility studies is to use genetics to test specific hypotheses regarding the role of non-genetic exposures. The concept of ‘Mendelian randomization’ has recently been discussed in some detail in the International Journal of Epidemiology,1 although the idea has been around for some time.2 As illustrated in the accompanying article, Katan pointed out over 15 years ago that the observed association between low cholesterol levels and increased cancer rates may have arisen through reverse causation.3 Specifically, pre-symptomatic cancers may have resulted in a reduction in cholesterol levels. The solution that Katan proposed for testing the causality of this association was to compare different polymorphisms of the apolipoprotein E (Apo E) gene between cases and controls. Apo E is partly responsible for clearance of cholesterol from plasma and the efficiency of this clearance varies according to different Apo E genotypes.4 Consequently, individuals with a genotype including the E2 allele have, on average, lower levels of cholesterol than individuals with genotypes comprising of other alleles. If low cholesterol levels are associated with cancer then we would expect cancer cases to be more likely to possess the E2 allele than non-cancer controls. Even if the observed association was due to confounding from other sources (e.g. cigarette smoking or as components of a low fat diet), the proposed Mendelian randomization analysis would be able to rule out this possibility.