For many reasons, tests of the environment, and particularly of gene-by-environment (G×E) interactions, have been left out of genome-wide association (GWA) estimates of genetic main effects. A tutorial on the current study designs for mapping G×E interactions is provided by Duncan Thomas (Gene–environment-wide association studies: emerging approaches. Nature Reviews Genetics 11, 259–272 (2010))1. Some2 have argued that G×E interactions will contribute little to the missing heritability3 if G×E effects are small or if correlations between gene and environment (rGE) mask genotype-by-genotype (G×G) effects. However, G×E effects may be underestimated in behavioural genetic approaches when it is assumed that environments differ when they are in fact similar. The theoretical challenge is that, in most studies, the environment is self-selected, and so environment effects cannot be distinguished from unmeasured genetic effects. Furthermore, if the environment is poorly assessed, the G×E equation contains two components with highly divergent error variations, creating high risks for type 1 and type 2 errors4. Replication of G×E findings is therefore crucially dependent on accurate assessments of the environment5,6 and on the absence of rGE.

A promising avenue for circumventing the issues that are inherent in correlational G×E studies is the genetically informed experimental intervention7. In randomized control trials (RCTs), the environment is manipulated in standard ways, and the randomization breaks the potential rGE. Evidence in support of this approach comes from three pioneering G×E RCTs, all showing that intervention efficacy is genetically moderated by the 7-repeat allele of the dopamine receptor D4 (DRD4) gene, which contains 7 copies of a 48 bp tandem repeat. In one randomized experiment, toddlers who carried this allele showed a greater reduction in disruptive behaviour after parenting skill intervention than children who did not carry this allele7. In another experiment, preschoolers with the same genotype were more positively affected by being randomly assigned to exposure to computer games that targeted their emerging phoneme awareness skills than those children not carrying this allele were8. In a third trial, this one focused on African–American adolescents and their families, teenagers carrying the 7-repeat version of DRD4 were more positively affected than others by an intervention targeting substance use9.

These genetically moderated intervention effects are based on rather small samples (157–337 individuals) and need replication. However, they agree with meta-analytic evidence10 and provide experimental support for the potential importance of G×E11, suggesting that experimental methods are powerful strategies for examining candidate G×E effects. The observed effects are also consistent with the differential susceptibility claim that individuals differ in the extent to which they are affected — either positively or negatively — by environmental exposures12,13. One exciting implication of this perspective is that there should be genetically based heterogeneity in intervention efficacy and, as a corollary, that studies incorrectly estimate intervention efficacy systematically, overestimating it for some (less susceptible) individuals and underestimating it for other (more susceptible) individuals.

GWA studies (GWASs) have documented disappointingly small genetic associations with common human diseases, cognitive abilities and behavioural traits. For example, the first GWASs of reading ability14 and IQ15 explained less than 1% of the variance, whereas behavioural genetic (twin) studies have revealed much stronger genetic effects, explaining 50% or more of the variation between individuals2. We propose here that one way to bridge this gap — that is, to account for the missing heritability3 — is through G×E experiments.