Chapter 10. the Interplay of Genes, Lifestyle, and Obesity

chapter 10. The interplay of genes, lifestyle, and obesity

Paul W. Franks CHAPTER 10 CHAPTER

This chapter reviews the evi- long term [1]. Success in pharmaco- marketed for treatment of diabetes: dence supporting a joint effect of therapeutics for weight loss has also (i) metformin, which reduces hepatic genes and lifestyle factors in obesi- been meagre, and in some instances gluconeogenesis (the production of ty, focusing mainly on evidence from disastrous. A handful of anti-obesity glucose in the liver); (ii) sodium-glu- epidemiological studies and clinical medications have been approved cose linked transporter 2 (SGLT2) trials research. by the European Medicines Agency inhibitors, such as empagliflozin, Obesity is the scourge of most (EMA) and the United States Food which reduce re-uptake of glucose contemporary societies; about 40% and Drug Administration (FDA). One in the kidneys and are diuretic; and of adults worldwide are overweight of the most successful of these is (iii) glucagon-like peptide-1 (GLP-1) and 13% are obese (http://www.who. the lipase inhibitor orlistat. Howev- agonists, such as exenatide, which int/mediacentre/factsheets/fs311/ er, because orlistat diminishes in- diminish appetite by delaying gastric en/). Much of the burden that obesity testinal fat absorption, a frequent emptying. However, because all of conveys arises from the life-threat- side-effect of the drug is fatty stool, these drugs can cause side-effects ening diseases it causes, although which many patients cannot toler- and they are not all reimbursable by there are also direct consequences, ate. Other weight-loss drugs, such health insurance providers for treat- because quality of life is often dimin- as rimonabant, are approved for use ment of obesity, they are rarely used ished in people with morbid obesity in the European Union but are not primarily for weight reduction. as a result of social stigma and other widely prescribed because of safety The third weapon in the anti-obe- societal challenges. concerns and limited effectiveness. sity arsenal is bariatric surgery. Un- Although intensive lifestyle mod- Many other weight-loss drugs have like drugs and lifestyle intervention, ification leads to short-term weight been brought to market in the past which perturb the disease process, loss in most people, weight regain few decades, only to be withdrawn surgery can permanently alter the typically begins within a year of in- because of serious side-effects, in- disease trajectory. Therefore, long- tensive intervention, and only a small cluding death [2]. There are other term weight loss through surgery minority of the target populations are drugs that do achieve safe weight is generally sustained to a much able to maintain reduced weight in the loss, primarily those approved and greater degree than weight loss from

Chapter 10. The interplay of genes, lifestyle, and obesity 79 lifestyle intervention or drug therapy. factors modulate a person’s response morphological features. Studies in However, like drug therapies, bariat- to weight-loss therapies might help adults have reported somewhat low- ric surgery is expensive – although to predict the response to different er heritability estimates for obesity. it is cost-effective for diabetes treat- types of intervention, by guiding One of the most eloquent adult ment compared with drug therapy therapeutic choices in ways that are twin studies assessed the heritability [3] – and is not risk-free; serious ad- more precise and effective than con- of body mass index (BMI) in several verse events [4] include about 4 in ventional approaches, thereby avoid- hundred male and female Swedish 1000 patients dying within 60 days of ing unnecessary side-effects and twins; about half of them had been surgery [5]. Thus, although surgery reducing costs. reared together, and the remainder is appropriate for a small minority of had been reared apart, having been morbidly obese patients, it is no pan- Why might gene–lifestyle adopted into different families soon acea for the obesity epidemic. interactions be relevant in after birth [8]. The study showed that Of the three core prevention and obesity? the concordance of BMI in mono- treatment options for obesity, behav- zygotic (identical) twins was about ioural interventions that favourably Germline DNA variants are especial- 70% regardless of whether the twins affect chronic energy balance are by ly appealing biomarkers for targeting had been reared apart or together, far the most compelling, not least be- obesity interventions, because they whereas the concordance in dizygot- cause diet and exercise are generally are randomly assigned at meiosis ic (fraternal) twins was substantially safe, are relatively inexpensive, and and are stable throughout a person’s less, suggesting a strong genetic convey numerous additional benefits life, rendering their associations with component to obesity. Importantly, to health and well-being that drugs phenotypes fairly robust to confound- however, as discussed later in this and surgery do not. However, the ing and reverse causation. DNA vari- chapter, the genetic aberrations that considerable therapeutic idiosyncra- ants are also the starting point of a cause obesity do so through a range sies of lifestyle therapy cause wide process called the central dogma of of diverse mechanisms, including variability in its effectiveness at a molecular biology [6], downstream of those that affect appetite, satiation, population level. Some of this varia- which a complex molecular cascade and energy expenditure. bility is due to the extent to which the ensues that translates the effects of However, knowing that obesity is participant adheres to the interven- extrinsic environmental exposures highly heritable does not necessarily tion, and some is due to differences (of which diet and exercise are major mean that it is the consequence of in the participant’s biology, which components in obesity) to the clinical gene–lifestyle interactions. To deter- modulates the effects of lifestyle in- phenotypes that characterize health mine this, one could test whether the terventions on rates of weight loss and disease. That molecular cas- obesogenic effects of lifestyle expo- and weight regain. cade is made up of gene transcripts sures (in epidemiological studies) or The common outcome variable in and proteins, as well as epigenomic response to weight-perturbing inter- obesity research and clinical practice features (in the form of methylation ventions (in clinical trials) are heri- is weight change, because it can be marks, open chromatin, histone table. In studies of twins in the USA assessed easily and inexpensively. modifications, etc.), small circulating exposed to long-term overfeeding Changes in the amount and deposi- molecules (metabolites), and an ar- [9] or exercise [10] interventions, the tion of adipose tissue and ectopic fat ray of peptide hormones and other concordance in adaptive response are probably more clinically impor- biochemical components. to the interventions was significant- tant phenotypes, but they are more Studies in twins provided some ly higher within twin pairs compared difficult to quantify. Beyond this, of the earliest compelling evidence with the concordance between unre- weight change should be more than that obesity is under a high degree lated participants for a range of body merely aesthetic; thus, the many of genetic control. A study of chil- composition measures, including clinical sequelae of weight change dren and adolescents showed that waist circumference, body fat per- should also be tracked. Neverthe- 82–90% of the phenotypic variance centage, and fat-cell diameter (for less, whether the outcome of lifestyle was explained by additive genetic exercise response). intervention trials is weight or a relat- factors [7]. The study used objective Collectively, there is compelling ed metabolic outcome, the response assessments of body composition evidence supporting the view that to the intervention is generally highly (dual-energy X-ray absorptiometry body corpulence in the free-liv- heterogeneous. [DEXA] and hydrostatic weighing), ing state and change in body cor- Therefore, quantifying and under- enabling the careful distillation of pulence with diet or exercise are standing the ways in which genetic body corpulence into its constituent governed to a considerable extent

80 by genetic factors. These are so- ments in expensive follow-up studies hensive analysis of FTO variation called quantitative genetics studies. (such as clinical trials) to test wheth- and explored interactions with ob- Unlike the molecular genetics stud- er a gene–lifestyle interaction has jectively assessed physical activity ies of the modern era, quantitative the potential for clinical translation. (via accelerometry). The trial tested genetics provides a broad-strokes The FTO example cited above for genotype–treatment interactions genome-wide overview of genet- was the first of several encouraging on changes in obesity-related traits ic influence on a phenotype but illustrations of gene–lifestyle inter- in the Diabetes Prevention Program offers no insights into the specific actions in obesity. The role of FTO (DPP), a randomized controlled tri- molecular aberrations (e.g. single variation in obesity was first deal (RCT) of intensive lifestyle mod- nucleotide polymorphisms [SNPs], scribed in three papers published in ification, metformin, and placebo insertions and deletions [indels], and close proximity in 2007. Two of the control interventions [16]. Although copy number variations [CNVs]) that studies made their discoveries using there was no evidence of an inter- cause obesity or modify the effects genome-wide association studies action between the rs9960939 FTO of exposures and interventions on (GWAS) [12, 13], whereas the third variant and lifestyle intervention on weight change. [14] serendipitously identified the weight change, there was nominal genetic association signal using a statistical evidence of an interaction Examples of gene–lifestyle set of 48 intergenic SNPs intended on change in subcutaneous adipose interactions for quality control, of which one was mass (assessed using computed strongly associated with morbid obe- tomography). The interaction effect CHAPTER 10 CHAPTER Population-level molecular genetics sity. Nevertheless, the three studies was consistent with the epidemiolog- studies of obesity, whether focused reached consistent conclusions and ical data reported in the Danish and on associations or interactions, were provided the first convincing evi- Amish studies. once hopelessly unreliable. The evidence of an association of common Many studies were published in dence reported in most such studies genetic variation and obesity. The the following year, each addressing before 2007 lacked any reasonable strongest signal for BMI emanated the FTO interaction hypothesis, but degree of replication. Sample sizes from the rs9960939 variant, which with mixed results. Given this state generally ranged from a few dozen per copy conveys an odds ratio of of equipoise, an analysis was un- to a few hundred participants, and 1.35 for obesity and amounted to a dertaken involving about 220 000 most of the studies that focused on difference in body weight for a per- adults and 20 000 children and ad- interactions lacked robust measures son 1.7 m tall of about 3 kg between olescents, to seek replication of the of lifestyle exposures. A recent the high-risk and low-risk homozy- original study’s findings. To do this, a systematic review [11] identified 212 gous genotype groups [12]. standardized analysis plan was exe- studies published between 1995 and Soon after the publication of these cuted in each of the 54 cohorts from mid-2012 that tested gene–lifestyle papers, studies began to emerge re- which the 240 000 participants em- interactions in obesity; the review porting evidence of gene–lifestyle in- anated. The meta-analysis of these found that only those studies that teractions at the FTO locus [12–17]. data yielded a statistically significant focused on gene–physical activity The first study to do so came from a interaction effect, one that was con- interactions at the FTO (rs9960939) Danish cohort study called Inter99 sistent in direction with the original locus and gene–diet interactions at [15]. The authors used a cross-sec- reports, although of a much smaller the PPARG (Pro12Ala) locus had tional subcohort of about 5500 Inter99 magnitude (about one sixth of the been independently replicated. As participants to show that the genetic magnitude of the original interaction is explained later in this chapter, effect of the rs9960939 FTO variant effect) [19]. replication studies of gene–lifestyle on BMI was about 2 kg/m2 in people interactions face many challenges reporting little or no physical activity Replication studies: relevance that extend beyond those faced by but was closer to 1 kg/m2 in those re- and challenges association studies. Therefore, the porting high levels of physical activity. absence of replication does not nec- Soon after this work was published, After the discovery of FTO in 2007, essarily mean that the initial finding a second observational study and a many subsequent GWAS analyses was false-positive. However, repli- clinical trial reported complementary were performed, each larger than cation studies are a sentinel feature results. The observational study was the last and each contributing to of science, and without replication of a population isolate of Amish in- the burgeoning array of obesity-as- results for interaction effects it would dividuals living in Pennsylvania [17]. sociated genetic variants [18]. With be difficult to justify major invest- The authors undertook a compre- the emergence of these data came

Chapter 10. The interplay of genes, lifestyle, and obesity 81 studies modelling the combined ef- this hypothesis in 111 000 adults, the founding exists. Although BMI is fects of these loci (as genetic risk authors were able to reproduce the probably the most common estimate scores) and their interactions with original finding, albeit with an inter- of adiposity in research studies and lifestyle. Of the many that have now action effect of substantially smaller clinical practice, it is a proxy for the been published, three epidemiologi- magnitude [11]. underlying degree of adiposity. In cal studies stand out. A third major study, performed general, people with higher BMI The first study examined the in three epidemiological cohorts in scores are also fatter, but this is by interaction of 12 obesity loci and the USA, focused on the interaction no means always true. Consider, for physical activity in 20 000 adults in of a genetic risk score comprising example, muscular athletes such the United Kingdom [19a]. The study 32 obesity-associated loci and con- as major league basketball players, was a textbook analysis of gene–life- sumption of sugar-sweetened bever- many of whom have BMI scores that style interaction effects and yielded a ages [20]. These analyses showed classify them as “overweight” [22]. highly statistically significant interac- that the genetic predisposition to In population-based cohorts, one tion effect, which showed that physi- obesity tended to be stronger in peo- should expect there to be a subpop- cal activity appeared to diminish the ple who consumed higher volumes ulation of people who are heavy and effect of the genetic loci on BMI. of sugar-sweetened beverages. In a lean, in part because they are more These exciting results were pub- field that is plagued by a dearth of physically active. One would expect lished in one of the leading general replication data, this study stands few physically inactive people to be medical journals, and the study was out as one of very few to report novel heavy and lean, and even fewer phys- clearly well conducted, but without findings on gene–lifestyle interac- ically active people to be fat. Thus, if replication data the possibility that tions alongside robust replication one were to model the association of these findings might be popula- data from independent cohorts. obesogenic gene variants with BMI tion-specific or false-positive could Although replication is important, in the subpopulation of physically in- not be ruled out. Therefore, a study it provides no assurance of cause active people, one would anticipate attempted to replicate these find- and effect in observational studies. a strong relationship, but if one were ings in a combined sample of about There are many alternative explana- to model the same association in the 40 000 Swedish adults, but initially tions for why two variables might be physically active subpopulation, one failed. For reasons outlined in de- associated with another that do not should expect this relationship to be tail by Ahmad et al. [11], a series of include causality, because the fac- weaker, because BMI is a weaker factors were identified that inhibited tors that might confound these rela- proxy for total adiposity in physically replication of gene–lifestyle interac- tionships in one cohort could easily active people compared with physi- tion effects (listed in the “Key fac- do so in others. Gene–lifestyle in- cally inactive people. Thus, because tors” box at the end of this chapter). teraction studies are more prone to the interaction tests outlined in the It was subsequently determined that confounding and bias than studies studies discussed above compare these factors are features that are that test the marginal associations the magnitude of the association be- likely to affect other replication stud- of lifestyle or genetic exposures in tween genotypes and BMI by strata ies of gene–lifestyle interactions, in- disease. This is because interaction of physical activity, statistically sig- cluding the large study of interaction studies are prone to all of the major nificant interaction tests could be between FTO variation and phys- sources of bias and confounding driven entirely by confounding. Thus, ical activity discussed above [19]. that plague conventional associa- when outcomes are assessed using Hence, although replication is the tion studies, as well as types of con- imperfect proxies and the validity of bulwark against false discovery, it is founding and bias that are specific to that proxy varies across the distri- important to ensure that replication interaction effects. For example, the bution of the lifestyle exposure, this studies that fail to support the initial way in which data are distributed can type of confounding, which is specif- discoveries do so for the right rea- undermine the credibility of statisti- ic to interaction analyses, should be sons. In the replication study [11], it cal interactions [21]. carefully considered. was shown that when inhibiting fac- In the examples discussed above tors are considered, the sample size of gene–physical activity interac- Clinical trials required to achieve sufficient power tions in obesity (assessed using to test the hypothesis amounts to a BMI, which is calculated as the Epidemiology is a powerful tool for cohort collection of more than about weight in kilograms divided by the generating hypotheses about gene– 100 000 adults, about 5 times as square of the height in metres), a lifestyle interactions, but it is prone large as the original study. By testing further potential source of con- to bias, confounding, and reverse

82 causality. RCTs of lifestyle inter- prove dramatically in some partici- assumption that is supported by her- ventions are more tightly controlled pants (super-responders), whereas itability analyses undertaken in the and monitored than epidemiologi- in other participants these same HERITAGE Family Study and else- cal studies; they are prospective in phenotypes do not improve (non-re- where, showing that the variability in design (most published epidemio- sponders), or even worsen. trait response is larger between fam- logical studies of gene–lifestyle in- It is often the case that research- ilies than within families [25]. Studies teractions have been performed in ers interpret “phenotypic response” of gene–treatment interactions per- cross-sectional data sets), thereby data from the HERITAGE Family formed in RCTs have the potential to permitting the assessment of tempo- Study and elsewhere as compelling identify specific genetic variants that ral relationships, and are less prone evidence of biologically (genetically) underlie treatment response. Two of to confounding, because treatment encoded exercise-response poten- the largest and most comprehensive (lifestyle vs control) is randomly as- tial [23]. However, genetic predispo- RCTs of lifestyle interventions were signed and hence should not be cor- sition is only one of many plausible performed in the USA. The first, the related with other factors that under- explanations for these results. For DPP, was performed in about 3000 lie an association between exposure example, exercise intervention stud- prediabetic overweight adults [26], and outcomes. However, because ies have shown that when people are whereas the second, the Action for it is usually not possible to blind a encouraged to undertake structured Health in Diabetes (Look AHEAD) participant to treatment allocation in exercise, non-exercise activity ther- trial, focused on about 5000 people a lifestyle trial (i.e. trials focused on mogenesis (the component of total with clinically manifest type 2 diabe- CHAPTER 10 CHAPTER changing diet and exercise behav- physical activity that is not struc- tes [27]. The clinical interventions iours), and because there is no pla- tured) decreases on average [24], a in both trials focused on inducing cebo that can be given for exercise concept sometimes termed behav- weight loss of about 7% of body and most dietary factors, lifestyle ioural compensation. Importantly, weight through structured and per- trials are less robust to confounding because the time spent undergoing sonalized diet and exercise regimes, than, say, placebo-controlled drug the lifestyle intervention in a trial as well as comparison arms that pro- trials. (often about 150 minutes per week) vided standard of care; the DPP also It is important to keep this in is a very small proportion (< 2%) of included two drug arms (metformin mind, because in lifestyle interven- the overall waking hours, a partic- and troglitazone). Extensive genet- tion studies behavioural compen- ipant’s behaviour during the hours ic analyses, including those relating sation is known to occur, and this when they are not participating in to the FTO locus (discussed above might lead participants assigned to the intervention sessions will affect for the DPP) have been performed treatment or control interventions the extent to which their health phe- in both trials, with several analyses to modify behaviours outside the notypes change during the trial, irre- focusing on weight change. hours of the intervention and thereby spective of the intervention’s inten- The first analysis of this nature affect the trial’s outcomes. There- sity or how faithfully the participant in the DPP focused on the PPARG fore, although much is made of the has adhered to it. Moreover, varia- Pro12Ala locus [28]. These analy- variability in treatment response in bility in measurement precision and ses in the DPP tested a hypothesis lifestyle intervention trials, perhaps accuracy (error), which are inherent set forth by earlier epidemiological most notably in the Health, Risk features of all clinical trials and ob- studies relating to the interaction Factors, Exercise Training, and Ge- servational studies, causes a phe- of dietary fats. PPARG is a nuclear netics (HERITAGE) Family Study nomenon called regression dilution, receptor that regulates many genes [23], it is reasonable to assume that which contributes to the apparent and pathways involved in energy some of the variability in response is variability in phenotypic response to metabolism, adipogenesis, and oth- due to behavioural compensation. interventions. Thus, measurement er metabolic processes. Long-chain The HERITAGE Family Study was error is usually most abundant at unsaturated fatty acids bind with high an intervention-only exercise train- the extremes of a trait’s distribution, affinity toPPARG , as do thiazolidine- ing (aerobic and resistance training) and this should be considered when diones, a class of drugs used to im- study administered over 20 weeks in using data from lifestyle intervention prove peripheral insulin sensitivity. about 1000 participants. The results trials to understand human biology. Therefore, the authors tested wheth- from the study seem to suggest that Nevertheless, some of the inter- er the Pro12Ala variant modified the there are “responders” and “non-re- individual variability in response to weight-loss effects of (i) lifestyle sponders” to exercise interventions, lifestyle interventions is likely to be intervention per se, (ii) dietary fat- causing clinical phenotypes to im- under biological/genetic control, an ty acid consumption, and (iii) the

Chapter 10. The interplay of genes, lifestyle, and obesity 83 thiazolidinedione drug troglitazone. findings is a large N( = 67 000), in- across multiple human tissues [38], No statistical interaction was ob- dependent analysis of gene–diet in- which indicated an overrepresenta- served with lifestyle, but with both teractions in BMI in a cross-sectional tion of several of these loci across dietary fats and troglitazone, the hy- cohort collection [35]. This analysis neural pathways involved in satia- pothesized interaction effects were of 32 of the 92 loci studied in the tion and appetite. Those analyses observed. DPP and Look AHEAD trials found were extended in the most recent Many subsequent studies were that the strongest evidence of inter- GIANT publication on BMI-associat- conducted in the DPP; some in- action between a gene variant and ed variants to include about 60 further volved detailed explorations of can- diet was at the MTIF3 locus. Obvious variants [18]. Using the DEPICT soft- didate genes, such as MC4R [29], differences in study designs and out- ware [39], the authors provided fur- ADIPOQ [30], TCF7L2 [31], and comes make determining the com- ther evidence of enrichment across PPARGC1A [32], and others fo- parability of the interaction effects central nervous system pathways cused on polygenic risk scores [33]. across these studies challenging. (i.e. synaptic function, long-term The most recent of these studies [34] MTIF3 is involved in forming the potentiation, and neurotransmitter assessed the effects of 92 variants initiation complex of the mitochon- signalling) but also found that some that were recently reported for their drial 55S ribosome [36, 37], which in of the newly discovered loci mapped associations with BMI by the Genet- turn synthesizes 13 of the inner mi- to pathways implicated in movement ic Investigation of Anthropometric tochondrial membrane proteins. The behaviour (physical activity and co- Traits (GIANT) consortium [18]. Joint regulation of MTIF3 plays a key role ordination) in mouse models. These analyses were conducted in the DPP in mitochondrial energy metabolism intriguing functional implications add and Look AHEAD trials to determine and reactive oxygen species produc- further support to the potential role of whether these variants, singly or in tion as part of the electron transport GWAS-derived loci in gene–lifestyle combination, modified the effects chain [37]. As was reported previ- interactions. Although most common of lifestyle interventions focused on ously [34], although rs1885988 is an variants have roughly comparable weight loss or prevention of weight intronic variant, its close proximity effect sizes, FTO stands out given regain. Overall, little evidence was (411 bp) to a triallelic missense SNP that the association and effect-mod- found of interactions between life- with a DNase peak indicates that the ifying roles (of lifestyle exposures) style and these genetic variants, rs1885988 variant is a marker for a in obesity are now well defined. suggesting that GWAS-derived ge- chromatin site involved in transcrip- However, despite huge efforts, the netic loci for obesity have no clinical- tion factor binding regulation. mechanisms through which FTO ly meaningful impact on response to acts remain unclear. What is clear is lifestyle interventions. However, one Functional implications that these mechanisms are complex, variant (at MTIF3) yielded a statis- involving long-range interactions with tically significant interaction effect Notwithstanding the limitations of other loci (e.g. IRX3 [40]), and may on weight loss that was consistent focusing on GWAS-derived (margin- be triggered by epigenomic factors in direction and magnitude in the al-effect) loci for interaction analy- (e.g. TRIM28 [41]). DPP and Look AHEAD trials. The ses, the approach has the advantage interaction manifested through a that huge efforts have been invested Conclusions slightly elevated risk of weight gain in determining the functional basis of in carriers of the G allele (the allele the genes to which the index maps. There is an abundance of pub- associated with higher BMI in the Importantly, locus mapping is still a lished evidence, predominantly from GIANT consortium meta-analysis fairly imprecise affair, and in many in- cross-sectional epidemiological stud- [18]) who were assigned to the con- stances the region to which a variant ies, that supports the notion that life- trol intervention, which contrasted with the strongest association sig- style and genetic factors interact to with the genetic effect in those as- nal in a GWAS maps spans several cause obesity. However, few studies signed to the lifestyle interventions genes. Thus, leaping from an associ- have been adequately replicated, (where the G allele was associated ation signal to functional prognosis is and functional validation and specif- with greater weight loss). The very fraught with caveats. ically designed intervention studies similar results in the two trials rep- Nevertheless, in silico function- are rarely undertaken; both of these resent some of the most robust evi- al annotation performed by the GI- are necessary to determine whether dence of a gene–lifestyle interaction ANT consortium mapped BMI-as- observations of gene–lifestyle inter- in weight change published to date. sociated loci to putative functional action in obesity are causal and of Adding further credence to these variants and transcription profiles clinical relevance.

84 Key points • The patterns and distributions of obesity within and between ethnically diverse populations living in similar and contrasting environments suggest that some ethnic groups are more susceptible to obesity than others. Generally, when exposed to environments typical of industrialized countries, aboriginal peoples appear to be highly susceptible, whereas populations of European ancestry appear to be far less prone to obesity. • More than 150 common loci have been robustly associated with measures of body composition. • Evidence from several behavioural intervention studies suggests that response to caloric manipulation brought about by fasting, overfeeding, or exercise is heritable. • There is now convincing epidemiological evidence of interactions between common variants at FTO and lifestyle on obesity. Almost all of these data are from cross-sectional studies, and temporal relationships are not clear. There are large studies supporting gene–lifestyle interactions at several other common loci, but the burden of evidence is far less for these loci than for FTO. • The evidence from clinical trials supporting gene–lifestyle interactions at FTO or other loci is relatively weak compared with the epidemiological evidence. • The magnitude of the interaction effects reported for FTO (or other common variants) is insufficient to warrant the use of those data for clinical translation. CHAPTER 10 CHAPTER Key factors The following key factors affect the detection and replication of gene–lifestyle interaction effects. • Exposure variance. When all else holds equal, statistical power is usually inversely related to the variance (usually expressed as standard deviation) of the exposure variable. • Outcome variance. When all else holds equal, statistical power is usually positively related to the variance (usually expressed as standard deviation) of the outcome variable. • Categorization of variables. For exposures (or outcomes) that are normally distributed and bear linear relationships with outcomes (or exposures), data stratification tends to reduce power [42]. Moreover, a variable that is stratified at the median point of its distribution will tend to yield higher statistical power than one that is stratified at other points in its distribution. • Measurement error. Error in the assessment of exposure or outcome variables has a profound impact on statistical power, such that sample size requirements to detect interactions may differ by several orders of magnitude, depending on the quality of exposure and outcome measures [43]. • Differential confounding. Interaction effects detected in observational studies are prone to confounding. However, confounding variables often differ between populations. Thus, if an interaction effect that is detected in one population is driven by confounding, and the confounding variables are absent in a replication cohort, then the replication cohort is likely to fail to reproduce the results of the initial study. However, successful replication does not necessarily exclude the possibility that interaction effects are confounded, because confounding factors may be simultaneously present in the discovery and replication cohorts. • Publication bias. Publishing negative findings, whether from studies of interaction or not, is generally more challenging than publishing results that appear statistically significant. Thus, the absence of negative- outcome replication studies in the literature may not mean that replication studies have not been performed. • Winner’s curse. The interaction effects featured in high-impact journals are often among the most striking. However, striking effects are sometimes overestimates of the true latent effect; thus, the results of subsequent studies are likely to be weaker, which in turn limits the statistical power of those later studies. This concept is often referred to as the “winner’s curse”. • Population-specific effects. Although the logical conclusion when an adequately powered replication study fails is that the original discovery may be false-positive, one cannot exclude the possibility that the original finding was true-positive and population-specific. Further studies that explore three-way interactions (gene × lifestyle × population-specific parameters) would be needed to model these effects.

Chapter 10. The interplay of genes, lifestyle, and obesity 85 References

1. Delahanty LM, Pan Q, Jablonski KA, Aroda 11. Ahmad S, Varga TV, Franks PW (2013). 19. Kilpeläinen TO, Qi L, Brage S, Sharp VR, Watson KE, Bray GA, et al.; Diabetes Pre- Gene × environment interactions in obesity: SJ, Sonestedt E, Demerath E, et al. (2011). vention Program Research Group (2014). Ef- the state of the evidence. Hum Hered. 75(2– Physical activity attenuates the influence fects of weight loss, weight cycling, and weight 4):106–15. http://dx.doi.org/10.1159/000351070 of FTO variants on obesity risk: a me- loss maintenance on diabetes incidence and PMID:24081226 ta-analysis of 218,166 adults and 19,268 change in cardiometabolic traits in the Dia- children. PLoS Med. 8(11):e1001116. http:// betes Prevention Program. Diabetes Care. 12. Frayling TM, Timpson NJ, Weedon MN, dx.doi.org/10.1371/journal.pmed.1001116 37(10):2738–45. http://dx.doi.org/10.2337/ Zeggini E, Freathy RM, Lindgren CM, et al. PMID:22069379 dc14-0018 PMID:25024396 (2007). A common variant in the FTO gene is associated with body mass index and predis- 19a. Li S, Zhao JH, Luan J, Ekelund U, Luben 2. Colman E (2012). Food and Drug Administra- poses to childhood and adult obesity. Science. RN, Khaw KT, et al. (2010). Physical activity tion’s Obesity Drug Guidance Document: a short 316(5826):889–94. http://dx.doi.org/10.1126/ attenuates the genetic predisposition to history. Circulation.125(17):2156–64. http://dx. science.1141634 PMID:17434869 obesity in 20,000 men and women from EPIC- doi.org/10.1161/CIRCULATIONAHA.111.028381 Norfolk prospective population study. PLoS PMID:22547756 13. Scuteri A, Sanna S, Chen WM, Uda M, Med. 7(8):e1000332. http://dx.doi.org/10.1371/ Albai G, Strait J, et al. (2007). Genome-wide journal.pmed.1000332 PMID:20824172 3. Colquitt JL, Pickett K, Loveman E, Frampton association scan shows genetic variants in GK (2014). Surgery for weight loss in adults. the FTO gene are associated with obesity-re- 20. Qi Q, Chu AY, Kang JH, Jensen MK, Cochrane Database Syst Rev. 8:CD003641. lated traits. PLoS Genet. 3(7):e115. http:// Curhan GC, Pasquale LR, et al. (2012). Sug- PMID:25105982 dx.doi.org/10.1371/journal.pgen.0030115 ar-sweetened beverages and genetic risk PMID:17658951 of obesity. N Engl J Med. 367(15):1387–96. 4. Keating C, Neovius M, Sjöholm K, Pelto- http://dx.doi.org/10.1056/NEJMoa1203039 nen M, Narbro K, Eriksson JK, et al. (2015). 14. Dina C, Meyre D, Gallina S, Durand E, PMID:22998338 Health-care costs over 15 years after bariatric Körner A, Jacobson P, et al. (2007). Vari- surgery for patients with different baseline glu- ation in FTO contributes to childhood obe- 21. Hunter DJ (2005). Gene-environment in- cose status: results from the Swedish Obese sity and severe adult obesity. Nat Genet. teractions in human diseases. Nat Rev Genet. Subjects study. Lancet Diabetes Endocrinol. 39(6):724–6. http://dx.doi.org/10.1038/ng2048 6(4):287–98. http://dx.doi.org/10.1038/nrg1578 3(11):855–65. http://dx.doi.org/10.1016/ PMID:17496892 PMID:15803198 S2213-8587(15)00290-9 PMID:26386667 15. Andreasen CH, Stender-Petersen KL, 22. McArdle WD, Katch FI, Katch VL (2014). 5. Morino M, Toppino M, Forestieri P, An- Mogensen MS, Torekov SS, Wegner L, An- Exercise physiology: nutrition, energy, and grisani L, Allaix ME, Scopinaro N (2007). dersen G, et al. (2008). Low physical activity human performance. 8th ed. Baltimore (MD), Mortality after bariatric surgery: analysis of accentuates the effect of the FTO rs9939609 USA: Wolters Kluwer Health, Lippincott Wil- 13,871 morbidly obese patients from a na- polymorphism on body fat accumulation. Dia- liams & Wilkins. tional registry. Ann Surg. 246(6):1002–7, betes. 57(1):95–101. http://dx.doi.org/10.2337/ discussion 1007–9. http://dx.doi.org/10.1097/ db07-0910 PMID:17942823 23. Roth SM (2008). Perspective on the SLA.0b013e31815c404e PMID:18043102 future use of genomics in exercise prescrip- 16. Franks PW, Jablonski KA, Delahanty tion. J Appl Physiol (1985). 104(4):1243–5. 6. Crick F (1970). Central dogma of molecu- LM, McAteer JB, Kahn SE, Knowler WC, et PMID:18048588 lar biology. Nature. 227(5258):561–3. http:// al.; Diabetes Prevention Program Research dx.doi.org/10.1038/227561a0 PMID:4913914 Group (2008). Assessing gene-treatment 24. Goran MI, Poehlman ET (1992). Endur- interactions at the FTO and INSIG2 loci on ance training does not enhance total energy 7. Franks PW, Ravussin E, Hanson RL, Harp- obesity-related traits in the Diabetes Preven- expenditure in healthy elderly persons. Am J er IT, Allison DB, Knowler WC, et al. (2005). tion Program. Diabetologia. 51(12):2214–23. Physiol. 263(5 Pt 1):E950–7. PMID:1443128 Habitual physical activity in children: the role http://dx.doi.org/10.1007/s00125-008-1158-x 25. Bouchard C, Rankinen T (2001). Indi- of genes and the environment. Am J Clin Nutr. PMID:18839134 82(4):901–8. PMID:16210723 vidual differences in response to regular 17. Rampersaud E, Mitchell BD, Pollin TI, Fu physical activity. Med Sci Sports Exerc. 33(6 8. Stunkard AJ, Harris JR, Pedersen NL, M, Shen H, O’Connell JR, et al. (2008). Phys- Suppl):S446–51, discussion S452–3. http:// McClearn GE (1990). The body-mass in- ical activity and the association of common dx.doi.org/10.1097/00005768-200106001-00013 dex of twins who have been reared apart. FTO gene variants with body mass index and PMID:11427769 N Engl J Med. 322(21):1483–7. http://dx. obesity. Arch Intern Med. 168(16):1791–7. 26. Knowler WC, Barrett-Connor E, Fowler doi.org/10.1056/NEJM199005243222102 http://dx.doi.org/10.1001/archinte.168.16.1791 PMID:2336075 SE, Hamman RF, Lachin JM, Walker EA, et PMID:18779467 al.; Diabetes Prevention Program Research 9. Bouchard C, Tremblay A, Després JP, 18. Locke AE, Kahali B, Berndt SI, Justice Group (2002). Reduction in the incidence of Nadeau A, Lupien PJ, Thériault G, et al. AE, Pers TH, Day FR, et al.; LifeLines Cohort type 2 diabetes with lifestyle intervention or (1990). The response to long-term over- Study; ADIPOGen Consortium; AGEN-BMI metformin. N Engl J Med. 346(6):393–403. feeding in identical twins. N Engl J Med. Working Group; CARDIOGRAMplusC4D http://dx.doi.org/10.1056/NEJMoa012512 322(21):1477–82. http://dx.doi.org/10.1056/ Consortium; CKDGen Consortium; GLGC; PMID:11832527 NEJM199005243222101 PMID:2336074 ICBP; MAGIC Investigators; MuTHER Con- 27. Wing RR, Bolin P, Brancati FL, Bray GA, 10. Bouchard C, Tremblay A, Després JP, Thé- sortium; MIGen Consortium; PAGE Consor- Clark JM, Coday M, et al.; Look AHEAD Re- riault G, Nadeau A, Lupien PJ, et al. (1994). The tium; ReproGen Consortium; GENIE Con- search Group (2013). Cardiovascular effects response to exercise with constant energy in- sortium; International Endogene Consortium of intensive lifestyle intervention in type 2 take in identical twins. Obes Res. 2(5):400–10. (2015). Genetic studies of body mass index diabetes. N Engl J Med. 369(2):145–54. http://dx.doi.org/10.1002/j.1550-8528.1994. yield new insights for obesity biology. Nature. http://dx.doi.org/10.1056/NEJMoa1212914 tb00087.x PMID:16358397 518(7538):197–206. http://dx.doi.org/10.1038/ PMID:23796131 nature14177 PMID:25673413

86 28. Franks PW, Jablonski KA, Delahanty L, 33. Delahanty LM, Pan Q, Jablonski KA, Wat- 39. Pers TH, Karjalainen JM, Chan Y, Westra HJ, Hanson RL, Kahn SE, Altshuler D, et al.; Di- son KE, McCaffery JM, Shuldiner A, et al.; Di- Wood AR, Yang J, et al.; Genetic Investigation abetes Prevention Program Research Group abetes Prevention Program Research Group of ANthropometric Traits (GIANT) Consortium (2007). The Pro12Ala variant at the peroxi- (2012). Genetic predictors of weight loss and (2015). Biological interpretation of genome-wide some proliferator-activated receptor gamma weight regain after intensive lifestyle modifi- association studies using predicted gene gene and change in obesity-related traits in cation, metformin treatment, or standard care functions. Nat Commun. 6:5890. http://dx.doi. the Diabetes Prevention Program. Diabetolo- in the Diabetes Prevention Program. Diabetes org/10.1038/ncomms6890 PMID:25597830 gia. 50(12):2451–60. http://dx.doi.org/10.1007/ Care. 35(2):363–6. http://dx.doi.org/10.2337/ s00125-007-0826-6 PMID:17898990 dc11-1328 PMID:22179955 40. Smemo S, Tena JJ, Kim KH, Gamazon ER, Sakabe NJ, Gómez-Marín C, et al. (2014). Obe- 29. Pan Q, Delahanty LM, Jablonski KA, 34. Papandonatos GD, Pan Q, Pajewski NM, sity-associated variants within FTO form long- Knowler WC, Kahn SE, Florez JC, et al.; Di- Delahanty LM, Peter I, Erar B, et al.; GIANT range functional connections with IRX3. Nature. abetes Prevention Program Research Group Consortium; Diabetes Prevention Program and 507(7492):371–5. http://dx.doi.org/10.1038/ (2013). Variation at the melanocortin 4 re- the Look AHEAD Research Groups (2015). nature13138 PMID:24646999 ceptor gene and response to weight-loss in- Genetic predisposition to weight loss & re- terventions in the Diabetes Prevention Pro- gain with lifestyle intervention: analyses from 41. Dalgaard K, Landgraf K, Heyne S, Lempradl gram. Obesity (Silver Spring). 21(9):E520–6. the Diabetes Prevention Program & the Look A, Longinotto J, Gossens K, et al. (2016). Trim28 PMID:23512951 AHEAD randomized controlled trials. Diabe- haploinsufficiency triggers bi-stable epigenet- tes. 64(12):4312–21. http://dx.doi.org/10.2337/ ic obesity. Cell. 164(3):353–64. http://dx.doi. 30. Mather KJ, Christophi CA, Jablonski KA, db15-0441 PMID:26253612 org/10.1016/j.cell.2015.12.025 PMID:26824653 Knowler WC, Goldberg RB, Kahn SE, et al.; Di- abetes Prevention Program Research Group 35. Nettleton JA, Follis JL, Ngwa JS, Smith 42. Ragland DR (1992). Dichotomizing continu- (2012). Common variants in genes encod- CE, Ahmad S, Tanaka T, et al. (2015). Gene × ous outcome variables: dependence of the mag- ing adiponectin (ADIPOQ) and its receptors dietary pattern interactions in obesity: analysis nitude of association and statistical power on the (ADIPOR1/2), adiponectin concentrations, and of up to 68 317 adults of European ancestry. cutpoint. Epidemiology. 3(5):434–40. http:// diabetes incidence in the Diabetes Prevention Hum Mol Genet. 24(16):4728–38. http://dx. dx.doi.org/10.1097/00001648-199209000-00009 Program. Diabet Med. 29(12):1579–88. http:// doi.org/10.1093/hmg/ddv186 PMID:25994509 PMID:1391136 dx.doi.org/10.1111/j.1464-5491.2012.03662.x 36. Christian BE, Spremulli LL (2009). Evidence 43. Wong MY, Day NE, Luan JA, Chan KP, PMID:22443353 Wareham NJ (2003). The detection of gene- 10 CHAPTER for an active role of IF3mt in the initiation of trans- 31. McCaffery JM, Jablonski KA, Franks PW, lation in mammalian mitochondria. Biochemis- environment interaction for continuous traits: Dagogo-Jack S, Wing RR, Knowler WC, et al.; try. 48(15):3269–78. http://dx.doi.org/10.1021/ should we deal with measurement error Diabetes Prevention Program Research Group bi8023493 PMID:19239245 by bigger studies or better measurement? (2011). TCF7L2 polymorphism, weight loss and Int J Epidemiol. 32(1):51–7. http://dx.doi. proinsulin:insulin ratio in the Diabetes Preven- 37. Behrouz B, Vilariño-Güell C, Heckman org/10.1093/ije/dyg002 PMID:12690008 tion Program. PLoS One. 6(7):e21518. http:// MG, Soto-Ortolaza AI, Aasly JO, Sando S, et dx.doi.org/10.1371/journal.pone.0021518 al. (2010). Mitochondrial translation initiation PMID:21814547 factor 3 polymorphism and Parkinson’s disease. Neurosci Lett. 486(3):228–30. http:// 32. Franks PW, Christophi CA, Jablonski KA, dx.doi.org/10.1016/j.neulet.2010.09.059 Billings LK, Delahanty LM, Horton ES, et al.; PMID:20887776 Diabetes Prevention Program Research Group (2014). Common variation at PPARGC1A/B 38. Speliotes EK, Willer CJ, Berndt SI, Mon- and change in body composition and meta- da KL, Thorleifsson G, Jackson AU, et al.; bolic traits following preventive interventions: MAGIC; Procardis Consortium (2010). As- the Diabetes Prevention Program. Diabetolo- sociation analyses of 249,796 individuals re- gia. 57(3):485–90. http://dx.doi.org/10.1007/ veal 18 new loci associated with body mass s00125-013-3133-4 PMID:24317794 index. Nat Genet. 42(11):937–48. http://dx.doi. org/10.1038/ng.686 PMID:20935630

Chapter 10. The interplay of genes, lifestyle, and obesity 87