NATURE REVIEWS | GENETICS VOLUME 6 | APRIL 2005 | 287REVIEWSSince the days of Archibold Garrod1, it has beenincreasingly accepted that the aetiology of most com-mon diseases involves not only discrete genetic and envi-ronmental causes, but also interactions between thetwo. Garrod suggested that “the influences of diet anddiseases” might “mask” some of the “inborn errors ofmetabolism” that he proposed, and that “idiosyncracies asregards drugs” were presumably due to inherited differ-ences — thereby presaging the field ofPHARMACOGENETICSby more than half a century.In the context of medical genetics and EPIDEMIOLOGY,the study of gene–environment interactions is useful forseveral reasons (BOX 1).Ifwe estimate only the separatecontributions of genes and environment to a disease,and ignore their interactions, we will incorrectly esti-mate the proportion of the disease (the ‘populationattributable risk’) that is explained by genes, the envi-ronment, and their joint effect. Restricting analysis ofenvironmental factors in epidemiological studies toindividuals who are genetically susceptible to the expo-sure should increase the magnitude of relative risks,increasing our confidence that the observed associationsare not due to chance. The identification of susceptibilityand/or resistance alleles in CANDIDATE-GENE STUDIES providesdirect evidence that these genes and their biologicalpathways are relevant to specific diseases in humans.Understanding these pathways might help to determinewhich compounds in a complex mixture cause disease.Ultimately, understanding gene–environment interac-tions might allow us to give individualized preventiveadvice before disease diagnosis, in addition to offer-ing personalized treatment after a disease, or diseasesusceptibility, has been diagnosed.Some gene–environment interactions can be identi-fied without any molecular analysis; one example is themuch stronger effect of sunlight exposure on skin can-cer risk in fair-skinned humans than in individuals withdarker skin2.Others can be observed as a reproducibleeffect of an environmental exposure on a susceptibleindividual; for example, the flushing response seen afteralcohol ingestion in individuals with low-activity poly-morphisms in the aldehyde dehydrogenase gene3.However, our rapidly expanding ability, particularlyafter the completion of the Human Genome Project,to define genetic differences at the DNA-sequence levelis opening up a vast new terrain in the search forgene–environment interactions.Although the phrase ‘gene–environment interaction’is frequently used to imply a specific relationshipbetween genes and the environment, the many existingdisease models differ with respect to the statistical asso-ciation between genes and the environment.At least inpart because of the many potential models of interac-tion, a gene–environment interaction will only beGENE–ENVIRONMENTINTERACTIONS IN HUMAN DISEASESDavid J. HunterAbstract | Studies of gene–environment interactions aim to describe how genetic andenvironmental factors jointly influence the risk of developing a human disease. Gene–environmentinteractions can be described by using several models, which take into account the various waysin which genetic effects can be modified by environmental exposures, the number of levels ofthese exposures and the model on which the genetic effects are based. Choice of study design,sample size and genotyping technology influence the analysis and interpretation of observedgene–environment interactions. Current systems for reporting epidemiological studies make itdifficult to assess whether the observed interactions are reproducible, so suggestions are madefor improvements in this area.PHARMACOGENETICSThe study of drug responses thatare related to inherited geneticdifferences.EPIDEMIOLOGYA discipline that seeks to explainthe extent to which factors thatpeople are exposed to(environmental or genetic)influence their risk of disease, bymeans of population-basedinvestigations.Program in Molecular andGenetic Epidemiology,Department ofEpidemiology, HarvardSchool of Public Health,677 Huntington Avenue,Boston, Massachusetts02115, USA, and theChanning Laboratory,Brigham and Women’sHospital, Boston,Massachusetts 02115, USA.e-mail: [email protected]:10.1038/nrg1578© 2005 Nature Publishing GroupCANDIDATE-GENE STUDIESStudies of specific genes inwhich variation might influencethe risk of a specific disease,usually because the gene is partof a biological pathway that isplausibly related to the disease.BIOMARKERA molecular marker of abiological function or externalexposure.ASSOCIATION STUDYAn approach to gene mappingthat looks for associationsbetween a particular phenotypeand allelic variation in apopulation.PRIOR PROBABILITYAn attempt to distinguishbetween more likely and lesslikely interactions on the basis ofknowledge of biologicalmechanisms, before aninteraction is observed.288 | APRIL 2005 | VOLUME 6 www.nature.com/reviews/geneticsREVIEWSthe two can be examined, is becoming more common,and greater use of population-based designs in geneticepidemiology has been advocated4.As studies ofgenetic susceptibility and environmental exposureshave been largely pursued by different groups of inves-tigators, multidisciplinary collaboration is necessary togenerate the best studies in the field.However, even with well-designed studies, there aremany ways of declaring ‘success’ in the search for inter-actions. This is largely because of variability in thequalitative and statistical models of interaction, andthe difficulty of assessing biological plausibility (either a priori — that is, when trying to prioritize PRIOR PROBA-BILITIES in these analyses — or once an interaction hasbeen observed). These factors are described below.Qualitative models. In the simplest case of dichotomousgenotype (such as carriers versus non-carriers of a genevariant) and dichotomous exposure (for example,exposed versus non-exposed), the four possible combi-nations of genotype and exposure can be displayed in a 2 × 4 table5, and the relative risks can be shown in agraph such as that shown in FIG. 1A.However, even in thissimplest case there are several models for describinginteractions between genetic susceptibility and environ-mental exposures in different diseases6,7(BOX 2).Thepossibilities are more numerous if there are many cate-gories of environmental exposure (for example, three ormore categories of