Longitudinal studies to detect gene × environment interactions in common disease – Bang for...

7
Longitudinal studies to detect gene environment interactions in common disease – Bang for your buck? A commentary on Chaufan’s ‘‘How much can a large population study on genes, environments, their interactions and common diseases contribute to the health of the American people?’’ (65:8, 1730–1741(2007)) Stephen Robertson a, * , Richie Poulton b a Department of Paediatrics and Child Health, Dunedin School of Medicine, University of Otago, Dunedin 9054, New Zealand b Dunedin Multidisciplinary Health and Development Research Unit, Department of Preventive and Social Medicine, Dunedin School of Medicine, University of Otago, Dunedin, New Zealand article info Article history: Available online 26 May 2008 Keywords Gene–environment interactions Social inequalities in health Medicalization Type 2 diabetes Gene environment (G E) USA Over the last 30 years there have been intensive efforts to define the genetic causes of human disease. The last decade has seen increasing emphasis placed on identifying the genetic basis for susceptibility to the aetiologically complex conditions that are major causes of disease in developed nations today. The slow progress in this area is often attributed to methodological factors and/or miscon- ceptions surrounding the genetic architecture of disease susceptibility (Davey Smith et al., 2005). Even the latest, largest and most comprehensive study to date has still only identified genetic components of susceptibility to dis- eases like type 2 diabetes, bipolar disorder, hypertension and breast cancer that explain a small fraction of population attributable risk for these conditions (Wellcome Trust Case Control Consortium, 2007). Latterly the realiza- tion has dawned that perhaps gene–gene or gene–environ- ment interactions explain this gap. Enthusiasm has increased over the prospect of detecting such associations by the conduct of large prospective longitudinal studies of representative cross sections of communities in the USA and elsewhere (Collins, 2004; Davey Smith et al., 2005). Recently in Social Science & Medicine in her article ‘‘How much can a large population study on genes, environments, their interactions and common diseases contribute to the health of the American people?’’, Chaufan questioned the assumptions underlying such ventures by raising objec- tions surrounding the scientific basis, utility and ethics of conducting such studies (Chaufan, 2007). Her central claim is that the environmental factors underlying the most bur- densome diseases affecting developed nations today are * Corresponding author. Tel./fax: þ64 3 4797469. E-mail addresses: [email protected] (S. Robert- son), [email protected] (R. Poulton). Contents lists available at ScienceDirect Social Science & Medicine journal homepage: www.elsevier.com/locate/socscimed 0277-9536/$ – see front matter Ó 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.socscimed.2008.04.010 Social Science & Medicine 67 (2008) 666–672

Transcript of Longitudinal studies to detect gene × environment interactions in common disease – Bang for...

ilable at ScienceDirect

Social Science & Medicine 67 (2008) 666–672

Contents lists ava

Social Science & Medicine

journal homepage: www.elsevier .com/locate/socscimed

Longitudinal studies to detect gene� environment interactions incommon disease – Bang for your buck? A commentary on Chaufan’s‘‘How much can a large population study on genes, environments,their interactions and common diseases contribute to the healthof the American people?’’ (65:8, 1730–1741(2007))

Stephen Robertson a,*, Richie Poulton b

a Department of Paediatrics and Child Health, Dunedin School of Medicine, University of Otago, Dunedin 9054, New Zealandb Dunedin Multidisciplinary Health and Development Research Unit, Department of Preventive and Social Medicine, Dunedin School of Medicine,University of Otago, Dunedin, New Zealand

a r t i c l e i n f o

Article history:Available online 26 May 2008

KeywordsGene–environment interactionsSocial inequalities in healthMedicalizationType 2 diabetesGene� environment (G� E)USA

* Corresponding author. Tel./fax: þ64 3 4797469.E-mail addresses: [email protected]

son), [email protected] (R. Poulton

0277-9536/$ – see front matter � 2008 Elsevier Ltddoi:10.1016/j.socscimed.2008.04.010

Over the last 30 years there have been intensive effortsto define the genetic causes of human disease. The lastdecade has seen increasing emphasis placed on identifyingthe genetic basis for susceptibility to the aetiologicallycomplex conditions that are major causes of disease indeveloped nations today. The slow progress in this area isoften attributed to methodological factors and/or miscon-ceptions surrounding the genetic architecture of diseasesusceptibility (Davey Smith et al., 2005). Even the latest,largest and most comprehensive study to date has stillonly identified genetic components of susceptibility to dis-eases like type 2 diabetes, bipolar disorder, hypertensionand breast cancer that explain a small fraction of

tago.ac.nz (S. Robert-).

. All rights reserved.

population attributable risk for these conditions (WellcomeTrust Case Control Consortium, 2007). Latterly the realiza-tion has dawned that perhaps gene–gene or gene–environ-ment interactions explain this gap. Enthusiasm hasincreased over the prospect of detecting such associationsby the conduct of large prospective longitudinal studiesof representative cross sections of communities in theUSA and elsewhere (Collins, 2004; Davey Smith et al.,2005).

Recently in Social Science & Medicine in her article ‘‘Howmuch can a large population study on genes, environments,their interactions and common diseases contribute to thehealth of the American people?’’, Chaufan questioned theassumptions underlying such ventures by raising objec-tions surrounding the scientific basis, utility and ethics ofconducting such studies (Chaufan, 2007). Her central claimis that the environmental factors underlying the most bur-densome diseases affecting developed nations today are

S. Robertson, R. Poulton / Social Science & Medicine 67 (2008) 666–672 667

not only well defined but they also represent the major andmodifiable components of disease causation that primaryand secondary preventative measures can alreadyaddress. From this viewpoint she states that a longitudinalstudy such as that proposed by the U.S. Secretary of Healthand Human Services’ Advisory Committee on Genetics,Health, and Society (SACGHS, 2007) that aims to measuregenetic factors alongside potential environmental contrib-utors to common disease states, has ‘‘little to add to currentknowledge about how to prevent, treat, or decrease in-equalities in, common diseases’’ (Chaufan, 2007, p. 1730).Her call echoes that of others who believe that genomicshas little to offer for addressing disease burdens whichare primarily of environmental origin (Cooper & Psaty,2003). In this response to her thesis, we not only reiteratemuch of the previously developed justifications for longitu-dinal gene–environment (G � E) studies (Khoury, Davis,Gwinn, Lindegren, & Yoon, 2005) but re-examine theseideas in the light of recent empirical studies in the area.

Type 2 diabetes mellitus as an illustrative example

Part of the persuasiveness of Chaufan’s argument lies inher dependence on type 2 diabetes as her prime illustrativeexample. It is an unassailable truth that environmentalfactors account for 80–90% of the population attributablerisk for this condition (Cooper & Psaty, 2003). It may bethat in a profoundly diabetogenic environment such asexists in 21st century urban USA, the definition of gene–environment interactions add little per se to the manage-ment of an obese and inactive populace. If an environmentalcontributor is near ubiquitous and the genetic predisposi-tion common as well, interventions are most sensiblyweighted, as she asserts, towards environmental risk factormodification. Even here though there is room for furtherresearch since the aetiopathogenesis of type 2 diabetesmay not be as well understood as her critique implies.For instance, Chaufan argues that dietary interventions toprevent prenatal ‘‘programming’’ leading to a susceptibilityto develop type 2 diabetes (the fetal origins of adult onsetdisease hypothesis) is as evidence-based as dietary man-agement of the adult diabetic state. Many questions re-main in this area – at what time in gestation and whatnutritional/hormonal factors are involved in this program-ming? Is such susceptibility determined genetically? If so,is it the mother or the fetus that bears that predisposition?Some of the emerging answers are surprising and counter-intuitive, for example, indicating that periconceptualnutritional factors may play a role, as distinct from mid-gestational endocrine and metabolic influences (Gluckman& Hanson, 2004). Later on in gestation, evidence suggeststhat fetal and maternal genetic factors contribute to birthweight (Frayling & Hattersley, 2001; Hattersley et al.,1998). It would seem plausible that such factors could eas-ily operate at the very beginning of the human lifecoursebut their identity, and the environmental factors they syn-ergise with, are unknown (Bloomfield, Oliver, & Harding,2006). In short, it would come as little surprise that syner-gistic interactions between gene and environment (G� E)remain to be discovered in studies addressing the relativeimportance of genes and environment in the prenatal

antecedents to the diabetic state. Intelligent, targeted pub-lic health initiatives aimed at primary prevention will onlybe developed if a firm grasp of the pathophysiology of thiscondition, including its temporal evolution, is understood.The measurement of genetic variation has a large part toplay in this endeavour.

Discovering or addressing disease risk factors – whereshould public health priorities lie?

Also valid is the case Chaufan argues that inequalities inthe provision of health care and education are compound-ing the growing problem of type 2 diabetes in developed(and increasingly, less developed) nations today (Chaufan,2007). This important matter, which essentially addressesissues relating to resource allocation and distributive jus-tice, should not be conflated with arguments that addressthe purely scientific merit of conducting a longitudinalstudy into combined genetic and environmental risk fac-tors that predispose to common causes of ill-health. The ex-istence of unaddressed public health needs founded onexisting knowledge does not illegitimise ongoing researchwhich may continue to deliver such dividends in the future(Merikangas & Risch, 2003). The principal question posedby Chaufan, that must be answered, is do longitudinal stud-ies measuring gene–environment interactions present suchan opportunity?

Not all common diseases are equally well understood

It may be that merit lies with spending public money toaddress the well-recognized environmental determinantsof type 2 diabetes instead of on the conduct of hi-tech lon-gitudinal epidemiological studies, but it would be wrong togeneralize this to many other common and disablingdisease states. The pathophysiology of some common dis-orders, amongst them breast cancer, prostate cancer, de-pression and inflammatory bowel disease is much morepoorly understood than type 2 diabetes mellitus. Well-founded, evidence-based interventions to modify theenvironment of individuals at risk for all these conditionsin order to significantly reduce the incidence of these dis-eases are lacking. Chaufan’s statement that ‘‘.we alreadyknow which environments are optimal to prevent most ofthese common conditions.’’ (Chaufan, 2007, p. 1733) canbe directly challenged. What environments are best toavoid prostate or breast cancer? Inflammatory bowel dis-ease? Bipolar disorder? Many large epidemiological studieshave been performed in an endeavour to find risk factorsbut have met with only modest success in explaining thedisease susceptibility attributable to environmental factors.There is certainly no evidence that Chaufan’s panacea –‘‘reasonable access to healthy lifestyles, proper housing,living wages, safe neighborhoods, well-funded schools, eq-uitable medical care’’ (Chaufan, 2007, p. 1739) will impactsignificantly on the incidence of any of the diseases listedabove. This is precisely the scenario where incorporatinggenetics into the mix of measured co-variables could im-pact very favourably on our ability to effectively interveneto reduce the incidence and morbidity of these conditions.

S. Robertson, R. Poulton / Social Science & Medicine 67 (2008) 666–672668

Genetics as a lens – finding new environmentaldeterminants of disease

Missing in Chaufan’s critique is an acknowledgementthat such longitudinal studies can contribute to the identi-fication of new environmental determinants of ill-healthand in addition can offer an enhanced understanding ofhow (and in whom), modifiable environmental factorscan cause disease. A deeper understanding of both will of-fer the chance to deliver enhanced and improved primarypreventative strategies. Unlike Chaufan, we suggest thatall the answers to public health crises like, but not limitedto, the current type 2 diabetes epidemic, are not currentlyknown.

In the past, among the variables examined and con-trolled for in epidemiological studies seeking risk factorsfor the development of these conditions, are many thatare unmodifiable – sex, stature, skin pigmentation, birthorder, ethnicity all being examples. Their unmodifiablecharacter has not prevented their inclusion as risk factorsand co-variables for the simple reason that controlling forthem can bring other risk factors to attention during anal-ysis. Evidence for a main effect (the association of a variableor risk factor to the disease state at a statistically significantlevel over the entire study population) is frequently absentinitially; the variable being sought only comes to attentionduring stratification analysis as being significant for a frac-tion of the study sample. Why should stratifying for sex, forexample, be a legitimate and powerful tool (Rutter, Caspi, &Moffitt, 2003), but accounting for the effect of measuredgenotype be considered a folly that is sure to fail? Simplyput, the failure to launch appropriately powered epidemio-logical studies in this, the post-genome era, is setting one-self up to fail to identify modifiable risk factors that couldsignificantly reduce morbidity and mortality for some ofthe more common diseases afflicting people today. Geneticfactors should be treated no differently to sex, ethnicity,birth order and the like. Commonly arising genetic variantsin the population can define significant sub-populationsthat an environmental risk factor can operate on to predis-pose to disease with that same risk factor not conferringrisk to those of a different genotype. The epidemiologicalapproach used to find these genetic factors has beendubbed Mendelian randomization (Davey Smith & Ebra-him, 2003, 2005; Davey Smith et al., 2005; Khoury et al.,2005) and it arguably presents one of the most powerfulapplications applied to population-based studies that in-corporate the measurement of genotype in their design.Essentially the approach involves the measurement ofdisease incidence over a range of environmental exposurefor a subgroup with a given genotype, or conversely mea-surement of the degree of susceptibility (or resilience)conferred by genetic variants in a subgroup exposed toa uniform environmental factor. The detection of sucheffects is the demonstration of a G� E interaction. By its na-ture it possesses a powerful ability to internally control forconfounding effects although caveats must still be recog-nized and accounted for (Rutter, 2007).

There is ample precedent for the success of this generalapproach. One well-known example from the behaviouralliterature illustrates the value of combining information

about known environment risk exposures with informa-tion about measured genes. Findings from the DunedinMultidisciplinary Health and Development Study revealedhow a functional polymorphism in the promoter region ofthe serotonin transporter gene (5-HTTLPR) interacted withlife stress to predict depression (Caspi et al., 2003). Specif-ically, the 5-HTTLPR short allele conferred risk in the pres-ence of life stress whereas carriers of the long alleleappeared resilient to adverse life events. At the time thisstudy was conducted there was not a great deal of interestin the links between 5-HTTLPR and depression, mainly be-cause the results from simple genotype–phenotype associ-ation studies had been equivocal at best. Consistent withthis, 5-HTTLPR on its own was not associated with depres-sion (i.e. there was no main effect for 5-HTTLPR); ratherthe improved prediction of depression, beyond knowingabout stress exposure alone, only came via the combinationof the 5-HTTLPR polymorphism interacting with the envi-ronmental risk factor. Thus, the involvement of 5-HTTLPRin depression was confirmed by taking an established envi-ronmental risk factor of modest effect size and stratifyingby genotype. The ubiquity of stressful life events coupledwith the significant burden of disease association with de-pression makes improving treatment for this disordera public health priority. This G� E has helped spur new re-search aimed at understanding the mechanism(s) associ-ated with both risk and protective alleles. Ultimately thismay lead to the development of new and hopefully moreeffective treatments for this common disorder.

The Dunedin Study researchers have used the same basicapproach to explain how response to environmental patho-gens is moderated by genotype for two other importantpublic health concerns. In the first study, adult violenceand antisocial behaviour was predicted by an interaction be-tween childhood maltreatment and a functional polymor-phism in the promoter region of the gene encoding theenzyme monoamine oxidase A (MAOA) (Caspi et al., 2002).This was the first report to demonstrate the operation ofG� E, for a behavioural disorder using a measured gene. Dis-parate findings were brought together to formulate a plausi-ble G� E hypothesis in which the MAOA gene appearedrelated to both the outcome of interest (antisocial behav-iour) and to the exposure (maltreatment). One contributionof this study was to demonstrate that G� E do indeed oper-ate in psychiatry, much as they do in other branches of med-icine. In so doing it challenged the conventional wisdom ofthe day which was that G� E interactions were probablyrare and of little importance in the behavioural sciences.

The second study showed how the development of psy-chosis following exposure to cannabis during adolescencewas moderated by a functional polymorphism in the cate-chol-O-methyltransferase (COMT) gene (Caspi et al.,2005). Interestingly, this latter association was age-dependent – cannabis use in young adulthood did notelevate risk for developing psychosis, even in the presenceof genetic vulnerability in the form of the COMT valineallele. Determining the timing and magnitude of the riskposed by cannabis for psychosis informs a sometimesheated debate about the harms associated with cannabisuse. A sound evidence-base is a necessary pre-requisite topolicy that balances citizens’ rights and autonomy while

S. Robertson, R. Poulton / Social Science & Medicine 67 (2008) 666–672 669

protecting their health. Because cannabis sits at the nexusof health and legal policy it presents special challenges tolawmakers and public health officials. Thus, knowledgeabout the COMT gene can help us to understand the basicpathophysiology of schizophrenia, thereby opening upnew treatment possibilities, as well as informing nationaldrug and health policy and legislative reform (Fergusson,Poulton, Smith, & Boden, 2006).

When genes masquerade as environmentalrisk factors

The depression finding (Caspi et al., 2003) also providesa good example of how to check for spurious G� E due togene–environment correlation (hereafter rGE). rGE de-scribes how individual genetic differences can effectively‘drive’ differential environmental risk exposure. In otherwords, exposure to environmental risk is not a randomphenomenon; rather it stems in part from differences in ge-netic endowment (Plomin, DeFries, & Loehlin, 1977). rGEscome in three main forms: passive rGE refers to environ-mental influences linked to genetic effects that are externalto the person. For example, parents create the early child-rearing environment, as well as providing genetic materialto their offspring. In contrast, active rGE, which can take theform of selecting specific environments or ‘niche picking’,and evocative rGEs arise largely from factors within the per-son (Rutter, Moffitt, & Caspi, 2006). Examples of active rGEare seen in preference for sporting versus artistic activitiesand environments. Different responses from the social en-vironment elicited by shy versus gregarious people are anexample of evocative interaction. The key observation forthe current discussion is that environmentally mediatedrisk cannot always be assumed, despite appearances.From a public health perspective, the independence of en-vironment from gene may not seem especially important –whatever the genesis of the environmental risk, the effectsare presumably the same. However, this overlooks the pos-sibility that unmeasured genes ‘driving’ the environment,via intermediate behaviours, may be suitable targets for in-tervention. New, albeit more distal targets are of value forno less reason than the effectiveness of current interven-tions for many chronic diseases is moderate at best.

Genes can improve phenotyping of complex disorders

Among multifactorial psychiatric disorders AttentionDeficit Hyperactivity Disorder (ADHD) stands out due toits marked heterogeneity in clinical presentation. Childrenwith a diagnosis of ADHD vary in a number of importantways including: intellectual function; presence/absence ofcomorbid conduct disorder; differential response to stimu-lant drug regimens; natural history (course) of symptoms;and in long-term prognosis. This begs the question: do thedifferent features (and combinations thereof) have differ-ent etiologies? And if so, is the current ADHD diagnosismixing apples with oranges, and conflating several distinctdisorders that would benefit from quite different treatmentapproaches? This question is particularly relevant in disci-plines like psychiatry for which diagnosis, in the absence ofphysical tests, relies upon symptom syndromes. Recent

research has shown that polymorphisms in dopaminegenes (DRD4 and DAT1) are associated with differences inintellectual functioning among children diagnosed withADHD, after controlling for the severity of symptoms. Thesesame genes also predicted who was at risk for the worstoutcomes in adulthood (Mill et al., 2006). A more recentstudy in three different samples provided evidence for mo-lecular-genetic subtyping of antisocial behaviour withinADHD (Caspi et al., 2008), reinforcing how genes can actas effect modifiers (Fanous & Kendler, 2005). Together,these studies demonstrate how genetic information canbe used to resolve clinical heterogeneity and refine diagno-sis (Krishnan, 2005).

Proof in principle of how knowledge about genetic var-iability can inform psychiatric nosology, and might evenhave clinical testing utility, as well as elucidating pathogen-esis, emphasizes the value of genetic information. There areparallels in respiratory medicine with respect to the asthmaphenotype, another clinical syndrome marked by consider-able heterogeneity (Wenzel, 2006). Genes can also help im-prove phenotypic definition in other branches of medicine.For example, the cancer field is beginning to witness theemergence of a new (yet to be fully realised) disease taxon-omy based on molecular profiles, with greater precisionand subtyping of diseases that foreshadows more effectivepersonalized treatment (Potter, 2005).

Reaction norms

Chaufan argues at a theoretical level that reaction normswill ultimately thwart the ability to draw associations be-tween certain susceptibility alleles and disease traits, sim-ply because accounting for all possible environments andtheir interrelationships with genotype is methodologicallyimpossible. To the contrary, the study of reaction normssits comfortably with the detection and measurement ofG� E interactions (Mackay & Anholt, 2007). The argumentsthat counter her assertion include firstly, that no attemptwill be made to account for all environmental variablesand secondly, to point out that clear gene–environment in-teractions for significant disease states have been detectedin populations with diverse life courses. The aim is not to becomprehensive; the skill will lie in measuring the relevantenvironmental exposures.

The objection that these research endeavours may beexpensive pre-judges that their conclusions will have littlepublic health utility – this pessimism has little evidence tojustify it. Who is to say that the failure to find a significantmain effect between, for example, oral contraceptive useand breast cancer will not be re-defined by stratificationof the population by genotype and reanalysis – a findingthat would be altogether unsurprising to the swelling ranksof pharmacogeneticists? Counterintuitive findings of a sim-ilar nature have already been found in the field of breastcancer. However, the difference in breast cancer risk con-ferred by parity in carriers of highly penetrant susceptibil-ity alleles of the genes, BRCA1 and BRAC2 (an elevated riskfor BRCA1 carriers and a reduced risk for carriers of BRCA2mutations) (Narod, 2006; Narod et al., 1995) would obvi-ously not have been detected without stratification forthese genotypes. The observation that the penetrance of

S. Robertson, R. Poulton / Social Science & Medicine 67 (2008) 666–672670

BRCA1 mutation carriers seems to have altered over studiedbirth cohorts (Narod et al., 1995) strongly implies to somethe presence of undefined environmental variables ina G� E interaction at this locus, a parallel to an effectwell demonstrated in animal models (Cook et al., 2005).Could not other effects mediated by allelic variation atother less penetrant loci mediate similar effects? Whatseems unarguable is the value of such discoveries. Fewwould contend that clinical practice would not be changedby such findings.

Moving beyond self-apparent environmentalcontributors to disease

So far many G� E studies have examined the relation-ship between genetic variation with rather self-apparentenvironmental measures which present little ambiguityor difficulty in their measurement. Such measures are thosethat have frequently been recorded in study cohorts in thepast simply because they represent known risk factors orrepresent clear-cut biologically adverse influences. Theprime example is tobacco exposure. A gathering body of ev-idence is accruing to show synergistic interactions withsmoke exposure and alleles in various genes responsiblefor metabolism of xenobiotics, DNA mismatch repair andinflammatory response. The phenotypic endpoints rangefrom craniofacial clefting (Shi et al., 2007) to cancer (Clavel,2007) and asthma (Ober & Thompson, 2005). The salientobservation to make here is not so much that this repre-sents yet more evidence to back the well-acknowledgedevil that tobacco represents, but more what other harder-to-measure, more subtle and unconsidered environmentalinfluences are still waiting to be evaluated by the samemethodology to be revealed as similarly pervasive determi-nants of ill-health (or resilience) (Rothman et al., 2001).Chaufan’s conviction that all factors in the G� E equationwill be self-obvious and familiar may in time prove to beill-founded.

Defining pathogenesis – the first step for evidence-based public health medicine

For some conditions genetic epidemiology has signifi-cantly contributed to the definition of the pathophysiologyof diseases and there are many indicators that measure-ment of gene–environment interactions will guide preven-tion and treatment. Inflammatory bowel disease (IBD),a group of pathologically heterogeneous conditions withsignificant genetic underpinnings, have for a long timebeen considered to be disorders of immune dysfunction.However, treatment approaches have been generic andnon-specific, largely owing to a lack of understanding aboutthe nature of the immune defects. The recent identificationof several genetic variants underlying susceptibility tothese conditions as belonging to genes encoding compo-nents of the innate immunity pathway has significantly im-proved understanding of the pathophysiology of thecondition. These findings have refined the nature of thedysregulated relationship between the gut immune systemand intestinal microflora as central to the pathogenesis ofIBD. Further understanding will require measurement of

the environment – for example, the impact of compositionof the gut microflora and host nutritional factors, includingdiet – and the way that they impact on the activity of thedisease (Ferguson, Shelling, Browning, Huebner, & Peter-mann, 2007). It would be nonsensical to study these factorsindependent of the various predisposing genotypes thathave shown main effects in genome screens to date.

Defining and refining the nature of environmentalrisk factors

A further dividend of incorporating the measurementof genotype in association studies is that they canfurther resolve the nature of the environmental suscepti-bility factor by implicating a pathophysiological mecha-nism. This confirmation of biological plausibility bringsconfidence to public health initiatives that address riskfactors that have modest main effects in conferring sus-ceptibility to common disabling conditions. Hunter(2005) reviewed the example of the association of redmeat with colorectal cancer and the subsequent observa-tion that variation within the NAT2 gene implicates het-erocyclic polyamines as the responsible carcinogenicmoiety. Such reductionistic insights have the potentialto inform studies aimed at detecting other potentialsources of environmental carcinogens – a formidablechallenge if methodological approaches were restrictedto conventional case–control study designs owing tothe complexity of environmental exposures under study(Rothman et al., 2001).

It must be emphasized that the detection of a G� E as-sociation does not necessarily imply that the appropriateintervention will require the genotyping of the populationin which that intervention will be applied. For example,the consumption of red meat is shown to exhibit a moder-ate main effect on the risk of developing colorectal cancer,and subsequent work has shown that the most stronglypredisposed fraction of the population is defined by thepossession of ‘‘fast metaboliser alleles’’ in the gene NAT2.In an instance like this, the discovery of the G� E interac-tion has added biological plausibility to the associationand defined the mechanism underlying the initiallydetected, relatively weak, main effect (Khoury et al.,2005). Under such circumstances it would be reasonableto promote an intervention to the population as a whole(such as moderation of red meat intake) as a reasonablepreventative strategy if no significant harm accrued to the70% of the population in whom the strategy would beless effective. The alternative (not conducting G� E studiesto identify such strategies) is to fail to identify (or validate)a modifiable factor that significantly affects mortality andmorbidity. The inclusion of genetics in the epidemiologicalstudies has enhanced the science underpinning the publichealth intervention.

In this respect, Chaufan concedes that even the patho-physiology of type 2 diabetes mellitus is likely to be hetero-geneous and that G� E studies may contribute to definingthese differences. Implicit in this admission is that someenvironmental variables may be more important thanothers depending on genotype. In a health system that isincreasingly demanding cost efficiency, the utility of

S. Robertson, R. Poulton / Social Science & Medicine 67 (2008) 666–672 671

targeted management (whether it be pharmaceuticals, life-style interventions, nutritional management) can onlymean a more efficient use of the health dollar and sparingindividuals needless or less fruitful interventions. To positthat all interventions can be homogeneously applied,with equal weight, to a population with a disorder that iscausally heterogeneous, implies some virtue in therapeuticimprecision.

Whither genotyping?

Two statements must be made to counter the miscon-ception that measuring genotypes in such studies impliesthat those same genetic determinants themselves mustbe modified if they are found to constitute an importantcomponent of the population attributable risk forcommon diseases. Firstly, it is unlikely that public healthinitiatives leveraged from G� E studies will require geno-typing large swathes of the population. Evaluating familyhistory to identify who may be at risk of an adverse G� Eoutcome is increasingly being viewed as the cost-effectiveand feasible tool with which to apply these new findings(Khoury et al., 2005; Scheuner, Wang, Raffel, Larabell, &Rotter, 1997). Secondly, the demonstration of a gene–environment interaction does not mean that genotypeneeds to be ‘‘modified’’, or even measured – it is clearthat it is the environment that needs to be addressed.Although gene therapy may represent a promising ap-proach for rare and very disabling genetic diseases, fewwould argue that genetic modification on a large scalerepresents a plausible approach to control of commondisease states in the 21st century.

The real danger in being nihilist about the utility ofG� E studies is that modifiable environmental factors willbe dismissed as irrelevant simply because the relevantG� E interaction was not looked for and detected by an ap-propriately designed study like that proposed by theSACGHS.

In summary, efforts to find genetic factors conferringsusceptibility to disease to date have mostly looked formain effects (Wellcome Trust Case Control Consortium,2007). Emerging genetic epidemiology suggests thatgene–environment interactions can identify new risk fac-tors; some of which may be modifiable. The implicationsfor preventative public health strategies are obvious. Thebest method to discover them is via longitudinal studiesthat include the measurement of relevant genes in the pop-ulation under study. Historically epidemiology, clinical tri-als and studies on the efficacy of pharmacological andnon-pharmacological therapies have treated study popula-tions as genetically homogeneous. This age began to closewith the sequencing of the human genome in 2001, andthe increasing affordability of wide-scale measurement ofgenetic variation will see its end.

We do not accept the contention that incorporating themeasurement of genotype into longitudinal epidemiologi-cal studies is wasteful or unlikely to yield benefit. Currentevidence from similar but smaller studies to that proposedby the SACGHS indicates substantial benefit can accruefrom such an approach. Genotypes can be the lenses

through which new environmental risk factors can befound, priorities on interventions refined and substantialhealth benefit accrued to all strata of society.

References

Bloomfield, F. H., Oliver, M. H., & Harding, J. E. (2006). The late effects offetal growth patterns. Archives of Diseases in Childhood Fetal NeonatalEdition, 91(4), F299–F304.

Caspi, A., Langley, K., Milne, B., Moffitt, T. E., O’Donovan, M., Owen, M.,et al. (2008). A replicated molecular genetic basis for subtypingantisocial behavior in children with attention-deficit/hyperactivitydisorder. Archives of General Psychiatry, 65(2), 203–210.

Caspi, A., McClay, J., Moffitt, T. E., Mill, J. S., Martin, J., Craig, I., et al. (2002).Role of genotype in the cycle of violence in maltreated children. Sci-ence, 297, 851–854.

Caspi, A., Moffitt, T. E., Cannon, M., McClay, J., Murray, R. M.,Harrington, H. L., et al. (2005). Moderation of the effect ofadolescent-onset cannabis use on adult psychosis by a functionalpolymorphism in the COMT gene: longitudinal evidence of a gene� environment interaction. Biological Psychiatry, 57, 1117–1127.

Caspi, A., Sugden, K., Moffitt, T. E., Taylor, A., Craig, I., Harrington, H. L., etal. (2003). Influence of life stress on depression: moderation by a poly-morphism in the 5-HTT gene. Science, 301, 386–389.

Chaufan, C. (2007). How much can a large populationstudy ongenes, environ-ments, their interactions and common diseases contribute to the health ofthe American people? Social Science & Medicine, 65(8), 1730–1741.

Clavel, J. (2007). Progress in the epidemiological understanding of gene–environment interactions in major diseases: cancer. Comptes RendusBiologies, 330(4), 306–317.

Collins, F. S. (2004). The case for a US prospective cohort study of genesand environment. Nature, 429(6990), 475–477.

Cook, J. D., Davis, B. J., Cai, S. L., Barrett, J. C., Conti, C. J., & Walker, C. L. (2005).Interaction between genetic susceptibility and early-life environmen-tal exposure determines tumor-suppressor-gene penetrance. Proceed-ings of the National Academy of Sciences U S A, 102(24), 8644–8649.

Cooper, R. S., & Psaty, B. M. (2003). Genomics and medicine: distraction,incremental progress, or the dawn of a new age? Annals of InternalMedicine, 138(7), 576–580.

Davey Smith, G., & Ebrahim, S. (2003). ‘Mendelian randomization’: can ge-netic epidemiology contribute to understanding environmental deter-minants of disease? International Journal of Epidemiology, 32(1), 1–22.

Davey Smith, G., & Ebrahim, S. (2005). What can Mendelian randomisa-tion tell us about modifiable behavioural and environmental expo-sures? British Medical Journal, 330(7499), 1076–1079.

Davey Smith, G., Ebrahim, S., Lewis, S., Hansell, A. L., Palmer, L. J., &Burton, P. R. (2005). Genetic epidemiology and public health: hope,hype, and future prospects. Lancet, 366(9495), 1484–1498.

Fanous, A. H., & Kendler, K. S. (2005). Genetic heterogeneity, modifiergenes, and quantitative phenotypes in psychiatric illness: searchingfor a framework. Molecular Psychiatry, 10, 6–13.

Ferguson, L. R., Shelling, A. N., Browning, B. L., Huebner, C., & Petermann, I.(2007). Genes, diet and inflammatory bowel disease. MutationResearch, 622(1–2), 70–83.

Fergusson, D. M., Poulton, R., Smith, P. F., & Boden, J. M. (2006). Cannabisand psychosis: a summary and synthesis of the evidence. British Med-ical Journal, 332, 172–175.

Frayling, T. M., & Hattersley, A. T. (2001). The role of genetic susceptibilityin the association of low birth weight with type 2 diabetes. BritishMedical Bulletin, 60, 89–101.

Gluckman, P. D., & Hanson, M. A. (2004). Living with the past: evolution,development, and patterns of disease. Science, 305(5691), 1733–1736.

Hattersley, A. T., Beards, F., Ballantyne, E., Appleton, M., Harvey, R., &Ellard, S. (1998). Mutations in the glucokinase gene of the fetus resultin reduced birth weight. Nature Genetics, 19(3), 268–270.

Hunter, D. J. (2005). Gene–environment interactions in human diseases.Nature Reviews Genetics, 6(4), 287–298.

Khoury, M. J., Davis, R., Gwinn, M., Lindegren, M. L., & Yoon, P. (2005). Dowe need genomic research for the prevention of common diseaseswith environmental causes? American Journal of Epidemiology,161(9), 799–805.

Krishnan, K. R. (2005). Psychiatric disease in the genomic era: rationalapproach. Molecular Psychiatry, 10, 978–984.

Mackay, T. F., & Anholt, R. R. (2007). Ain’t misbehavin’? Genotype-environment interactions and the genetics of behavior. Trends inGenetics, 23(7), 311–314.

S. Robertson, R. Poulton / Social Science & Medicine 67 (2008) 666–672672

Merikangas, K. R., & Risch, N. (2003). Genomic priorities and publichealth. Science, 302(5645), 599–601.

Mill, J., Caspi, A., Williams, B. S., Craig, I., Taylor, A., Polo-Tomas, M., et al.(2006). Prediction of heterogeneity in intelligence and adult progno-sis by genetic polymorphisms in the dopamine system among chil-dren with attention-deficit/hyperactivity disorder: evidence from 2birth cohorts. Archives of General Psychiatry, 63(4), 462–469.

Narod, S. A. (2006). Modifiers of risk of hereditary breast cancer. Onco-gene, 25(43), 5832–5836.

Narod, S. A., Goldgar, D., Cannon-Albright, L., Weber, B., Moslehi, R.,Ives, E., et al. (1995). Risk modifiers in carriers of BRCA1 mutations.International Journal of Cancer, 64(6), 394–398.

Ober, C., & Thompson, E. E. (2005). Rethinking genetic models of asthma:the role of environmental modifiers. Current Opinion in Immunology,17(6), 670–678.

Plomin, R., DeFries, J. C., & Loehlin, J. C. (1977). Genotype–environmentinteraction and correlation in the analysis of human behaviour.Psychological Bulletin, 84, 309–322.

Potter, J. D. (2005). Epidemiology informing clinical practice: from bills ofmortality to population laboratories. Nature Clinical Practice Oncology,2, 625–633.

Rothman, N., Wacholder, S., Caporaso, N. E., Garcia-Closas, M.,Buetow, K., & Fraumeni, J. F., Jr. (2001). The use of commongenetic polymorphisms to enhance the epidemiologic study ofenvironmental carcinogens. Biochimica et Biophysica Acta, 1471(2),C1–10.

Rutter, M. (2007). Gene–environment interdependence. DevelopmentalScience, 10(1), 12–18.

Rutter, M., Caspi, A., & Moffitt, T. E. (2003). Using sex differences in psy-chopathology to study causal mechanisms: unifying issues and re-search strategies. Journal of Child Psychology and Psychiatry, 44(8),1092–1115.

Rutter, M., Moffitt, T. E., & Caspi, A. (2006). Gene–environment interplayand psychopathology: multiple varieties but real effects. Journal ofChild Psychology and Psychiatry, 47(3–4), 226–261.

Scheuner, M. T., Wang, S. J., Raffel, L. J., Larabell, S. K., & Rotter, J. I. (1997).Family history: a comprehensive genetic risk assessment method forthe chronic conditions of adulthood. American Journal of MedicalGenetics, 71(3), 315–324.

SACGHS (Secretary’s Advisory Committee on Genetics, Health and Soci-ety). (March 2007). Policy issues associated with undertaking a newlarge U.S. population cohort study of genes, environment, and diseases.U.S. Department of Health and Human Services.

Shi, M., Christensen, K., Weinberg, C. R., Romitti, P., Bathum, L., Lozada, A.,et al. (2007). Orofacial cleft risk is increased with maternal smokingand specific detoxification-gene variants. American Journal of HumanGenetics, 80(1), 76–90.

Wellcome Trust Case Control Consortium. (2007). Genome-wide associa-tion study of 14,000 cases of seven common diseases and 3,000shared controls. Nature, 447(7145), 661–678.

Wenzel, S. E. (2006). Asthma: defining of the persistent adult phenotypes.Lancet, 368(9537), 804–813.