Recent progress in the genetics of common obesity


Dr Ruth J. F. Loos, MRC Epidemiology Unit, Institute of Metabolic Science, Addenbrooke's Hospital – Box 285, Hills Road, Cambridge CB2 0QQ, UK.
Tel: + 44 0 12 2376 9139
Fax: + 44 0 12 2333 0316


The genetic contribution to interindividual variation in common obesity has been estimated at 40–70%. Yet, despite a relatively high heritability, the search for obesity susceptibility genes has been an arduous task. This paper reviews recent progress made in the obesity genetics field with an emphasis on established obesity susceptibility loci identified through candidate gene as well as genome-wide studies. For the last 15 years, candidate gene and genome-wide linkage studies have been the two main genetic epidemiological approaches to identify genetic loci for common traits, yet progress has been slow and success limited. Only recently have candidate gene studies started to succeed; by means of large-scale studies and meta-analyses at least five variants in four candidate genes have been found to be robustly associated with obesity-related traits. Genome-wide linkage studies, however, have so far not been able to pinpoint genetic loci for common obesity. The genome-wide association approach, which has become available in recent years, has dramatically changed the pace of gene discoveries for common disease, including obesity. Three waves of large-scale high-density genome-wide association studies have already discovered at least 15 previously unanticipated genetic loci incontrovertibly associated with body mass index and extreme obesity risk. Although the combined contribution of these loci to the variation in obesity risk at the population level is small and their predictive value is typically low, these recently discovered loci are set to improve fundamentally our insights into the pathophysiology of obesity.

The prevalence of obesity and overweight continues to rise worldwide, not only causing serious personal health problems but also imposing a substantial economic burden on societies [1]. By current estimates, nearly 70% of adults in the USA and >60% in the UK are overweight; half of these are obese [2, 3].

It is well established that rapid globalization of the westernized lifestyle is fuelling this emerging obesity epidemic. Yet, not everyone in the present-day obesogenic environment develops obesity, highlighting the multifactorial nature of the condition. Indeed, obesity arises through the joint actions of multiple genetic and environmental factors, i.e. the obesogenic environment increases the risk of obesity, in particular in those who are already genetically susceptible.

Family and twin studies have shown that genetic factors contribute 40–70% to the interindividual variation in common obesity [4]. Despite a relatively high heritability, the search for obesity susceptibility genes has not been trivial. For the last 15 years, candidate gene and genome-wide linkage studies have reported a multitude of genetic loci, but only a handful have been robustly confirmed in subsequent studies. In recent years, however, significant progress has been made through genome-wide association studies with the discovery of at least 15 genetic loci incontrovertibly associated with obesity.

This paper reviews recent advances made in the obesity genetics field with an emphasis on established obesity susceptibility loci identified through candidate gene as well as genome-wide studies. The significant progress made by genome-wide association studies warrants specific and more extensive attention. I will therefore discuss the chronological sequence of discoveries, their impact on public health and clinical practice, their potential to unravel the underlying pathophysiology, and the ways ahead to find more obesity susceptibility loci.

Gene identification before the genome-wide association era

Although recent success of genome-wide association studies has drawn a lot of attention, gene identification for the last 15 years has been based on two broad genetic epidemiological approaches, i.e. candidate gene and genome-wide linkage studies.

For over a decade, the scientific community interested in obesity genetics was privileged to have a Human Obesity Gene Map that catalogued all genetic variants and chromosomal loci ever associated with or linked to obesity-related traits [5]. Its first edition was published in 1996, followed by 10 yearly updates, the latest in 2006 covering the literature available as of the end of 2005. The map continues to be a useful resource, in print and web-based versions (, providing a comprehensive and easily accessible overview of the literature on candidate gene and genome-wide linkage studies published before 2006.

Candidate gene studies

Candidate gene studies are hypothesis-driven and rely on the current understanding of the biology and pathophysiology that underlies the susceptibility to obesity. Genes, for which there is evidence for a role in regulation of the energy balance in animal models or in extreme/monogenic forms of obesity, are tested for association with obesity-related traits at the population level. In the mid 1990s, when genotyping was still an expensive and tedious task, candidate gene studies typically examined only one or a few variants per gene, with a focus on nonsynonymous variants because of their potential functional implications. Over time, genotyping costs have come down substantially, and publicly available datasets, such as dbSNP and the International HapMap, have provided deeper insight into genetic variation in genes. This knowledge has led to more comprehensive studies that systematically examine the association of all common variation in a gene of interest by means of carefully selected tagSNPs and their haplotypes under the assumption that a causal variant would be in high linkage disequilibrium with one of the tagSNPs or at least captured by the haplotypes.

Since the first candidate gene studies for obesity-related traits, >15 years ago, the number of proposed obesity susceptibility genes has grown steadily. The latest update of the Human Obesity Gene Map reported 127 candidate genes for which at least one study reported a positive association with obesity-related traits [5]. This rapid expansion reflects, at least in part, the availability of cost-effective genotyping technologies to a wider scientific community.

However, for many of the proposed candidate genes replication in successive studies has been inconsistent, so that the overall conclusion on association remains ambiguous. The limited success of the candidate gene approach can mainly be ascribed to small sample sizes (n < 1000) that are insufficiently powered to identify modest effects that are expected for common obesity. However, recent years have seen a change as an increasing number of studies have tested for associations in larger populations (n > 5000) and more often the initiative has been taken to meta-analyse all available published (and unpublished) data. The results of these large-scale studies are summarized in Tables 1 and 2.

Table 1. 
Overview of large-scale (n > 5000) individual candidate gene studies for obesity-related traits (in alphabetical order)
GeneDesignVariantsIndividuals, nMain outcomeReference
  1. BMI, body mass index.

ADRB1 Arg389Gly7 677No association with obesity or BMIGjesing et al. (Diabet Med 2007) [91]
ADRB2 Arg16Gly7 808No association with obesityGjesing et al. (Diabetologia 2007) [92]
Gln27Glu No association with obesity   
ADRB3 Trp64Arg7 605No association with obesityGjesing et al. (Mol Genet Metab 2008) [93]
BDNF Val66Met10 109Val-allele carriers have a higher BMI (+0.76 kg m−2[95CI 0.036, 1.16]; P < 0.001) than Met-allele homozygotesShugart et al. (EJHG 2009) [20]
CNR1Systematic screen of gene26 SNPs5 750The rs806381 (P= 1.1 × 10−6) and rs2023239 (4.7 × 10−4) are associated with increased obesity risk and BMIBenzinou et al. (HMG 2008) [94]
DIO2 Thr92Ala7 342No association with obesityGrarup et al. (JCEM 2007) [95]
ENPP1 Lys121Gln5 863No association with BMI or waistGrarup et al. (Diabetologia 2006) [28]
ENPP1 Lys121Gln + 3 marker haplotype8 676No association with BMILyon et al. (Diabetes 2006) [25]
ENPP1 Lys121Gln + 3 marker haplotype8 089No association with obesityWeedon et al. (Diabetes 2006) [26]
ENPP1 Lys121Gln + 3 marker haplotype5 153No association with BMI, overweight, or obesity. Gln-allele nominally associated with morbid obesity (P= 0.02)Meyre et al. (Diabetologia 2007) [96]
ERRalpha Pro116Pro and ESRRA236 365No association with obesity or BMILarsen et al. (IJO 2006) [97]
FAAH Pro129Thr5 801No association with obesity or BMIJensen et al. (J Mol Medical 2007) [98]
IGF2Systematic screen of gene3 SNPs5 000No association with BMIHeude et al. (JCEM 2007) [99]
KLF7Systematic screen of gene4 SNPs14 818Minor allele of rs7568369 protected against obesity (OR = 0.90 [0.84, 0.96], P= 0.001), and is associated with lower BMI (P= 0.002) and waist circumference (P= 0.003)Zobel et al. (EJE 2009) [100]
LTA Thr60Asn5 630N-allele carriers have increased waist circumference (P= 0.009), but not BMIHamid et al. (Diabetologia 2005) [101]
MC4R V103I7 937Ile-allele carriers have a reduced risk of obesity (OR 0.69 [95% 0.50, 0.96], P= 0.03)Heid et al. (J Medical Genetics 2005) [9]
NOS1AP rs753849016 913No association with obesityAndreasen et al. (BMC Med Genetics 2008) [102]
NPY2RSystematic screen of gene3 SNPs5 971Minor-allele carriers of rs12649641 have increased risk of obesity (P= 0.02)Torekov et al. (Diabetologia 2006) [103]
NPY2RSystematic screen of gene5 SNPs>8 000No association with BMICampbell et al. (Diabetes 2007) [104]
PCSK1Systematic screen of geneN221D13 659Minor allele is associated with increased risk of obesity (OR 1.34 [95% 1.20, 1.49], P= 7.27 × 10−8)Benzinou et al. (Nature Genetics 2008) [16]
Q665E-S690T Minor allele is associated with increased risk of obesity (OR 1.22 [95% 1.15, 1.29], P= 2.31 × 10–12)   
PKLR rs302078116 801No association with obesityAndreasen et al. (BMC Med Genetics 2008) [102]
PPARA Leu162Val5 799No association with obesitySparso et al. (Mol Genet Metab 2007) [105]
PPARDSystematic screen of gene12 SNPs5 971No association with BMIGrarup et al. (Diabetologia 2007) [106]
PPARGC1BSystematic screen of geneAla203Pro7 790203Pro allele is associated wit a reduced obesity risk (P= 0.004)Andersen et al. (J Medical Genet 2005) [107]
PYY Arg72Thr6 022Thr-allele carriers have increased risk of overweight (P= 0.02)Torekov et al. (Diabetes 2005) [108]
Table 2. 
Overview of meta-analyses of candidate gene loci for obesity-related traits (in alphabetical order)
GeneVariantsStudies, nIndividuals, nOutcomeReference
  1. BMI, body mass index.

ADRB2Arg16Gly114 328No association with obesityJalba et al. (Obesity 2008) [109]
ADRB2Glu27Gln2310 404Glu27-allele carriers have an increased risk of obesity (OR 1.566; [95% CI 1.07, 2.29]; P= 0.02) in Asians, Pacific Islanders, and American Indians only. No association in WhitesJalba et al. (Obesity 2008) [109]
ADRB3Arg64Trp367 399Arg-carriers tend to have an increased BMI (+0.19 kg m−2[95% CI –0.026, 0.41], P= 0.07)Allison et al. (IJO 1998) [110]
ADRB3Arg64Trp489 236Arg-carriers have an increased BMI (+0.30 kg m−2[95% CI 0.13, 0.47])Fujisawa et al. (Clin Endocrinol & Metab 1998) [111]
ADRB3Arg64Trp356 582Arg-carriers have an increased BMI (+0.26 kg m−2[95% CI 0.18, 0.42]; P= 0.001) in JapaneseKurokawa et al. (Obes Res 2001) [112]
ADRB3Arg64Trp9744 833Arg-carriers have an increased BMI (+0.24 kg m−2[95% CI 0.12, 0.37]; P= 0.0002) in East Asians only. No association in WhitesKurokawa et al. (IJO 2008) [24]
GRLAsn363Ser124 792No association with BMI or obesityMarti et al. (BMC Med Genetics 2006) [113]
IL6-174G→C2526 944No association with BMIQi et al. (JCEM 2007) [114]
LEPRK109R83 012No association with BMIHeo et al. (IJO 2002) [115]
LEPRQ223R93 263No association with BMIHeo et al. (IJO 2002) [115]
LEPRK656N93 263No association with BMIHeo et al. (IJO 2002) [115]
MC4RVal103Ile127 713Ile-allele carriers have a reduced risk of obesity (OR 0.69; [95% CI 0.50, 0.96]; P= 0.03)Geller et al. (AJHG 2004) [10]
MC4RVal103Ile2529 563Ile-allele carriers have a reduced risk of obesity (OR 0.82; [95% CI 0.70, 0.96]; P= 0.015)Young et al. (IJO 2007) [11]
MC4RVal103Ile2939 879Ile-allele carriers have a reduced risk of obesity (OR 0.80; [95% CI 0.70, 0.92]; P= 0.002)Stutzmann et al. (HMG 2007) [12]
MC4RIle251Leu1211 435Leu-allele carriers have a reduced risk of obesity (OR 0.52; [95% CI 0.38, 0.71]; P= 3.58 × 10−5)Stutzmann et al. (HMG 2007) [12]
PPARgPro12Ala5029 424Ala-allele carriers have an increased BMI in Whites (P= 0.015)Tonjes et al. (Diabetes Care 2006) [116]

Robust associations have been observed for nonsynonymous variants in the melanocortin 4 receptor (MC4R), prohormone convertase 1/3 (PCSK1), brain-derived neurotrophic factor (BDNF), and β-adrenergic receptor 3 (ADRB3) genes.

The MC4R gene has a strong biological candidacy; MC4R is widely expressed in the central nervous system and plays a key role in the regulation of food intake and energy homeostasis [6]. Rare functional mutations in MC4R are the commonest monogenic cause of severe early-onset obesity [7]. However, until recently its role as a common obesity susceptibility gene has been unconvincing. The two most common MC4R variants, V103I and I251L, each result in a nonsynonymous change with potential functional implications [8]. Numerous studies have examined these MC4R variants, but none has found significant association with obesity-related traits, apart from one sizeable population-based study that observed a significant protective effect of the 103I-allele (frequency: 2–3% of the population) on obesity risk [9] (Table 1). Subsequently, three large-scale meta-analyses confirmed that 103I-allele carriers have a 20% lower risk of obesity than V103V homozygotes [10–12] (Table 2). In addition, a meta-analysis of data on the I251L MC4R variant provided strong evidence for a protective effect with a nearly 50% reduced risk of obesity for carriers of the 251L-allele (frequency: 1–2%) [12] (Table 2).

The PCSK1 gene is another strong candidate, as it encodes an enzyme that converts pro-hormones into hormones involved in energy metabolism regulation. Individuals with rare mutations in PCSK1 are born with a PC1/3 deficiency resulting in a syndrome characterized by extreme childhood obesity [13–15]. In a comprehensive large-scale study, the role of common variants in the PCSK1 gene was studied in relation to the risk of obesity [16]. After sequencing coding regions in a small sample of obese individuals, nine variants that captured the common genetic variation in PCSK1 were genotyped in 13 659 individuals of European ancestry. Two nonsynonymous variants, N221D and the Q665E-S690T pair, were consistently associated with obesity in adults and children (Table 1). Each additional minor allele (frequency: 4–7%) of the N221D variant increased the risk of obesity 1.34-fold, while each additional minor allele (frequency: 25–30%) of the Q665E-S690T pair increased the risk 1.22-fold. Functional characterization of these variants suggested a modest deleterious effect of the N221D variant, but no significant functional role of the Q665E-S690T amino acid substitutions [16].

While BDNF has been mainly studied for its presumed role in the regulation of development, stress response, survival, and mood disorders, rodent studies have found BDNF to be implicated in eating behaviour, body weight regulation and hyperactivity [17, 18]. Similar to MC4R and PCSK1, a rare mutation in BDNF probably causes severe obesity and hyperphagia [19]. A recent large-scale study, including 10 109 women, found that minor-allele homozygotes (Met66Met; frequency: 4.5%) of the Val66Met variant have a significantly lower body mass index (BMI) (−0.76 kg m−2) than Val66-allele carriers [20] (Table 1).

The Arg64Trp ADRB3 variant is one of the first genetic variants for which association with obesity was reported [21–23]. ADRB3 is an obvious candidate gene given its involvement in the regulation of lipolysis and thermogenesis. Following the first reports in 1995, >100 studies have been published on the association between the Arg64Trp variant and obesity-related traits, but results have been inconsistent. However, a recent meta-analysis that combined data of 44 833 individuals found significant association between the Arg64Trp variant and BMI in East Asians, with Arg64-allele carriers having a 0.31 kg m−2 higher BMI compared with the Arg64Arg homozygotes [24] (Table 2). No associations were observed in Whites.

Large-scale studies are also powered to prove that an association is truly negative, such as, for example, for the Lys121Gln ENPP1 variant. Four studies, each with >5000 participants and a combined sample size of 27 781 individuals, found no association between the Lys121Gln variant and obesity-related traits [25–28] (Table 1). For other candidate genes tested in large-scale studies or in meta-analyses and for most of the other 127 candidate genes reported in the Human Obesity Gene Map, further follow-up is required to prove or refute unambiguously their role in obesity susceptibility.

In summary, the last 15 years of candidate gene studies have only recently starting to succeed; by means of large-scale studies and meta-analyses at least five variants in four candidate genes have been found to be robustly associated with obesity-related traits.

Genome-wide linkage studies

Genome-wide linkage studies are hypothesis-generating and, through surveying the whole genome, aim to identify new, unanticipated genetic variants associated with a disease or trait of interest. Genome-wide linkage studies rely on the relatedness of study participants and test whether certain chromosomal regions co-segregate with a disease or trait across generations. A whole-genome linkage survey requires 400–600 highly polymorphic markers, genotyped at 10-cM intervals. Genome-wide linkage studies have a rather coarse resolution and typically identify broad intervals that require follow-up genotyping to pinpoint the genes that underlie the linkage signal.

Since the first genome-wide linkage study was published in 1997, the number of chromosomal loci linked to obesity-related traits has grown exponentially. The latest Human Obesity Gene Map update reported 253 loci from 61 genome-wide linkage scans, of which 15 loci have been replicated in at least three studies [5]. Yet none of these replicated loci could be narrowed down sufficiently to pinpoint the genes or variants that underlie the linkage signal. Despite substantial power, a meta-analysis of 37 genome-wide linkage studies with data on >31 000 individuals from 10 000 families of European origin could not locate a single obesity or BMI locus with convincing evidence [29]. This meta-analysis indicates that genome-wide linkage might not be an effective approach for identifying genetic variants for common obesity.

Genome-wide association studies

Similar to genome-wide linkage, the genome-wide association approach interrogates the entire genome, unconstrained by prior assumptions. It aims to identify previously unsuspected genetic loci associated with a disease or trait of interest and, thereby, to expand our understanding of the underlying physiology. Genome-wide association studies screen the whole genome at higher resolution levels than genome-wide linkage studies and are thus able to narrow-down the associated locus more accurately. Genome-wide association does not rely on familial relatedness and can therefore achieve larger sample sizes than typical family-based studies.

Two major advances, i.e. the rapid expansion of our knowledge of the human genome in concert with substantial progress in high-throughput genotyping technology, have led the way to the genome-wide association. Together, these two new advances have given rise to the production of smartly designed chips that permit interrogation of the entire genome in one single experiment.

Genome-wide association has revolutionized the field of genetic epidemiology and has already resulted in an unprecedented chain of discoveries with >300 replicated associations for >70 common diseases and traits.

The genome-wide association study design

Genome-wide association studies typically comprise two (or more) stages: a discovery stage, followed by at least one replication stage.

The discovery stage entails high-density genotyping of hundreds of thousands of genetic variants, typically single nucleotide polymorphisms (SNPs), across the genome. Each SNP is tested for association with a trait or disease of interest. Studies with large sample sizes at this stage tend to be more successful, in particular for common traits with moderate heritability, as they are better powered to detect associations of small effect size. Collective efforts have led to genome-wide meta-analyses, which combine summary statistics of a series of individual genome-wide association studies in one analysis. Because individual studies often use different genotyping platforms, meta-analyses using genotyped data only would be limited to the subset of SNPs that is common to all platforms, which is a rather inefficient use of the available data. To facilitate genome-wide meta-analyses imputation of genotypes, identical across all studies involved in the meta-analysis, is now routinely done. In brief, based on the observed haplotype structure of the genotyped SNPs and that of a reference panel (e.g. the CEU HapMap population), the genotypes of ∼2.5 million untyped HapMap SNPs are inferred for each of the individuals of each study separately using one of the publicly available imputation software programs. Imputation is not perfect and this uncertainty is accounted for in the association analyses. The genome-wide meta-analysis requires only the association summary statistics of each SNP, directly genotyped and imputed, for each of the individual genome-wide association studies to calculate the overall significance of the associations.

Associations that meet the genome-wide significance threshold are taken forward for replication in subsequent independent studies to validate the initial observation. While it has been recommended to use a nominal P-value threshold of <5.0 × 10−8[30], which corresponds to a 5% genome-wide type I error rate, more liberal thresholds have been used by early genome-wide association studies. Ideally, study design, ancestry of replication samples, and trait or disease of discovery and replication stages are the same or at least very similar. The sample size of the replication stage is preferably at least as large as that of the discovery stage to ensure sufficient power to overcome the so-called winner's curse (i.e. the fact that effect sizes at discovery stage may have been inflated). Only SNPs or loci for which the association observed at the discovery stage is confirmed at the replication stage are considered ‘true hits’.

The discoveries

Genome-wide association studies have led to a rapid expansion in the number of loci implicated in predisposition of various polygenic diseases, including obesity. Since the introduction of the genome-wide association approach, we have witnessed three fruitful waves of discoveries based on large-scale high-density genome-wide association studies for obesity-related traits (Table 3). Most studies have tested for association with BMI as a continuous trait [31–35]. BMI is considered a good proxy-measure of adiposity in adults that is easy to obtain and available in many studies [36]. Others have examined association with risk of early-onset or adult extreme obesity [37, 38], under the assumption that morbidly obese individuals might be enriched for variants that predispose to obesity in the general population. Studies have been carried out predominantly in individuals of European ancestry, apart from one, which was based on an Indian Asian population [33].

Table 3. 
Summary of three waves of large-scale high-density genome-wide association studies for BMI and extreme obesity risk (ordered chronologically)
StudyFrayling et al. (Science 2007) [39]Scuteri et al. (PLoS Genetics 2007) [31]Hinney et al. (PLoS ONE 2007) [37]Loos et al. (Nature Genetics 2008) [32]Chambers et al. (Nature Genetics 2008) [33]Willer et al. (Nature Genetics 2009) [34]Thorleifsson et al. (Nature Genetics 2009) [35]Meyre et al. (Nature Genetics 2009) [38]
  1. BMI, body mass index; SNP, single nucleotide polymorphism; HDL, high-density lipoprotein; HOMA-IR, homeostasis model assessment of insulin resistance.

Discovery stage
 Study designCase–control studyPopulation-based studyCase–control studyMeta-analysis of 4 population-based studies and 3 disease-case seriesPopulation-based studyMeta-analysis of 10 population-based cohorts and 5 disease-case seriesMeta-analysis of 4 population-based cohortsCase–control study
 EthnicityWhite Europeans from the UKWhite Europeans from Sardinia (Italy)White Europeans from GermanyWhite Europeans from the UK and SwitzerlandIndian AsiansWhites from the UK, USA, Sweden, Finland, Italy, Switzerland and GermanyWhites from Iceland, the USA and the Netherlands, and African-AmericansWhite Europeans from France
 TraitsType 2 diabetesBMI, hip and weightEarly-onset extreme obesityBMIBMI, waist circumference and waist–hip ratio, weight, HOMA-IR, diastolic blood pressure, triglycerides, HDL-cholesterol, Type 2 diabetes and composite metabolic syndromeBMIBMI and weightEarly-onset and morbid adult obesity
 n1924 Type 2 diabetes patients; 2938 population controls4741487 extremely obese young individuals; 442 healthy lean controls16 876268432 38734 4161380 with early-onset and morbid adult obesity; 1416 age-matched normal-weight controls
 SNPs testedN/A362 129440 794344 883318 2372399 588305 846308 846 SNPs
 Significance levelN/AFDR Q-value <0.2015 most significant SNPs (corresponding to P < 3.6 × 10−6)P < 10−5P < 10−535 SNPs drawn from among the most strongly associated lociP < 10−5FDR Q-value <0.20
 SNPs for replicationN/A2111023354338
Replication stage
 Study designType 2 diabetes case-series, case- series and population-based series. Adults and childrenFamily studies. AdultsObesity family study (families with at least one extremely obese child or adolescent). Adults and childrenPopulation-based studies, case-series, obesity case–control studies, family study. Adults and childrenPopulation-based studies. AdultsPopulation-based studies, case-series, obesity case–control studies. Adults and childrenGIANT meta-analysis (Willer et al.) and population based study. AdultsFour case–control studies and 2 population-based studies
 EthnicityWhite Europeans from the UK, ItalyEuropean Americans, Hispanic Americans, and African AmericansWhite Europeans from GermanyWhite Europeans from the UK, Italy, Germany, USA, Sweden and FinlandIndian Asians and White Europeans from the UKWhites from the UK, USA, Sweden, Finland, Italy, and the NetherlandsWhites from the UK, USA, Sweden, Finland, Italy, the Netherlands and DenmarkWhite Europeans from France, Germany, Finland, and Switzerland
 TraitsBMI and obesityBMI, hip and weightExtreme obesityBMI and obesityBMI, waist circumference and waist–hip ratio, weight, HOMA-IR, diastolic blood pressure, triglycerides, HDL-cholesterol, Type 2 diabetes and composite metabolic syndromeBMI and obesityBMI and weightObesity and BMI
 n38 7593205230975 99111 95559 08237 97314 186
 Replicated loci/genesIn FTOIn FTOIn FTOIn FTO near MC4RNear MC4RIn FTO Near MC4R Near NEGR1 Near TMEM18 Near KCTD15 & CHST8 In SH2B1 Near GNPDA2, GUF1 and YPF7 In MTCH2In FTO Near MC4R Near NEGR1 Near TMEM18 Near KCTD15 & CHST8 In SH2B1 In SEC16B, near RASAL2 Near SFRS10, ETV5, DGKG In BDNF and near LGR4 and LIN7C, Near BCDIN3D, FAIM2In FTO Near MC4R In NPC1 Near MAF Near PTER

The first wave took place in 2007 and comprised three high-density genome-wide association studies that each confirmed fat mass and obesity associated gene (FTO) as the first gene incontrovertibly associated with common obesity and related traits (Tables 3 and 4). Each of the three studies examined a different trait of interest, yet they all identified the same locus, reflecting the robustness of the observation. The first study was a genome-wide association study for Type 2 diabetes [39], identifying a cluster of SNPs in the first intron of FTO that were highly significantly associated with Type 2 diabetes. After adjusting for BMI, the association with Type 2 diabetes was completely abolished, suggesting that the FTO–Type 2 diabetes association was mediated through BMI. The replication stage, including 13 cohorts with 38 759 adults and children, firmly validated FTO as an obesity susceptibility locus as each study confirmed association with BMI and obesity risk. Virtually at the same time, the study by Scuteri et al. [31] was published, which was the first large-scale high-density genome-wide association study of BMI in >4000 Sardinians. In their discovery stage, variants in the FTO and PFKP (platelet-type phosphofructokinase) gene showed the strongest association, but only those in FTO replicated in 1496 European Americans and 839 Hispanic Americans. A third genome-wide association study examined the risk of early-onset extreme obesity by comparing 487 extremely obese young individuals and 442 healthy lean controls [37]. Of the 15 most significant SNPs that were taken forward for replication, six were located in the FTO gene, which were the only ones for which association with obesity was confirmed in 644 nuclear families with at least one obese offspring.

Table 4. 
Risk-allele frequencies and effect sizes reported for the established obesity-susceptibility loci discovered through three waves of large-scale high-density genome-wide association studied for BMI and extreme obesity risk
(Nearby) gene(s)ChromosomeFrayling et al. (Science 2007) [39]Scuteri et al. (PLoS Genetics 2007) [31]Loos et al. (Nature Genetics 2008) [32]
SNP (rs-number & position)BMI-increasing alleleSNP (rs-number & position)BMI-increasing alleleSNP (rs-number & position)BMI-increasing allele
AlleleFreq (%) (HapMap Phase III)Effect size (per risk-allele)AlleleFreq (%) (HapMap Phase III)Effect size (per risk-allele)AlleleFreq (%) (HapMap Phase III)Effect size (per risk-allele)
FTO16q12rs9939609 (52378028)ACEU: 46% CHB: 15% JPT: 19% YRI: 51%0.10 z-score or ∼0.40  kg m−2rs9930506 (52387966)GCEU: 48% CHB: 21% JPT: 23% YRI: 20%0.13 z-score or ∼0.66 kg m−2rs1121980 (52366748)ACEU: 48% CHB: 20% JPT: 22% YRI: 47%0.06 z-score or 0.26 kg m−2
Near MC4R18q21        rs17782313 (56002077)CCEU: 27% CHB: 19% JPT: 24% YRI: 31%0.05 z-score or ∼0.22 kg m−2
Near NEGR11p31            
Near TMEM182p25            
SH2B1, ATP2A116p11            
Near KCTD15 & CHST819q13            
Near GNPDA2, GUF1 and YPF74p13            
In SEC16B, near RASAL21q25            
Near SFRS10, ETV5 and DGKG3q27            
In BDNF and near LGR4 and LIN7C,11p14            
Near BCDIN3D and FAIM212q13            
Near PTER10p12            
Near MAF16q23            
Chambers et al. (Nature Genetics 2008) [33]Willer et al. (Nature Genetics 2009) [34]Thorleifsson et al. (Nature Genetics 2009) [35]Meyre et al. (Nature Genetics 2009) [38]
SNP (rs-number & position)BMI-increasing alleleSNP (rs-number & position)BMI-increasing alleleSNPBMI-increasing alleleSNPBMI-increasing allele
AlleleFreq (%) (HapMap Phase III)Effect size (per risk-allele)AlleleFreq (%) (HapMap Phase III)Effect size (per risk-allele)AlleleFreq (%) (HapMap Phase III)Effect size (per risk allele)alleleFreq (%) (HapMap Phase III)Effect size (per risk allele)
  1. Effect sizes represent the increase in body mass index (BMI) for each additional risk allele. Effect sizes were derived from replication association studies, except for Frayling et al. [39], Scuteri et al. [31] and Loos et al. [32]*Additive effect size inferred from effect sizes reported for dominant model for comparison.†Absolute BMI values inferred from z-scores assuming a SD of 4.3 kg m−2.

    rs9939609 (52378028)ACEU: 46% CHB: 15% JPT: 19% YRI: 51%0.33  kg m−2rs8050136 (52373776)ACEU: 46% CHB: 14% JPT: 18% YRI: 46%0.080 z-score or ∼0.35  kg/m2rs1421085 (52358455)CCEU: 46% CHB: 14% JPT: 19% YRI: 7%0.080 z-score or ∼0.35  kg/m2
rs12970134 (56035730)ACEU: 28% CHB: 18% JPT: 19% YRI: 17%0.25  kg m−2rs17782313 (56002077)CCEU: 27% CHB: 19% JPT: 24% YRI: 31%0.20  kg m−2rs12970134 (56035730)ACEU: 28% CHB: 18% JPT: 19% YRI: 17%0.044 z-score or ∼0.19  kg/m2rs17782313 (56002077)CCEU: 27% CHB: 19% JPT: 24% YRI: 31%0.074 z-score or ∼0.32  kg/m2
    rs2815752 (72585028)ACEU: 64% CHB: 92% JPT: 92% YRI: 53%0.10  kg m−2rs2568958 (72537704)ACEU: 64% CHB: 92% JPT: 92% YRI: 53%0.030 z-score or ∼0.13  kg/m2    
    rs6548238 (624905)CCEU: 85% CHB: 93% JPT: 82% YRI: 90%0.26  kg m−2rs7561317 (634953)GCEU: 85% CHB: 92% JPT: 88% YRI: 79%0.043 z-score or ∼0.19  kg/m2    
    rs7498665 (28790742)GCEU: 38% CHB: 17% JPT: 13% YRI: 21%0.15  kg m−2rs8049439 (28745016)CCEU: 39% CHB: 24% JPT: 22% YRI: 41%0.034 z-score or ∼0.15  kg/m2    
    rs11084753 (39013977)GCEU: 69% CHB: 38% JPT: 33% YRI: 69%0.06  kg m−2rs29941 (39001372)CCEU: 68% CHB: 28% JPT: 26% YRI: 86%0.040 z-score or ∼0.17  kg/m2    
    rs10938397 (44877284)GCEU: 45% CHB: 25% JPT: 38% YRI: 21%0.19  kg m−2        
    rs10838738 (47619625)GCEU: 36% CHB: 32% JPT: 35% YRI: 5%0.07  kg m−2        
        rs10913469 (176180142)CCEU: 25% CHB: 24% JPT: 24% YRI: 34%0.026 z-score or ∼0.11  kg/m2    
        rs7647305 (187316984)CCEU: 80% CHB: 93% JPT: 94% YRI: 61%0.043 z-score or ∼0.19  kg/m2    
        rs4923461 (27613486)ACEU: 77% CHB: 55% JPT: 63% YRI: 86%0.043 z-score or ∼0.19  kg/m2    
        rs7138803 (48533735)ACEU: 35% CHB: 30% JPT: 34% YRI: 17%0.020 z-score or ∼0.09  kg/m2    
            rs10508503 (16339957)CCEU: 91% CHB: 100% JPT: 100% YRI: 100%0.017 z-score or ∼0.07  kg/m2
            rs1424233 (78240252)ACEU: 56% CHB: 70% JPT: 76% YRI: 65%0.030 z-score or ∼0.13  kg/m2
            rs1805081 (19394430)ACEU: 53% CHB: 77% JPT: 73% YRI: 100%0.032 z-score* or ∼0.14  kg/m2

Despite the sample sizes of the first genome-wide association studies being relatively small, they were sufficiently powered to harvest low-hanging fruit. It was clear that for a second wave of discoveries, however, collaborative efforts would be required combining individual genome-wide association studies to increase sample size and thus power to identify common variants with small effects. The Genomic Investigation of Anthropometric Traits (GIANT) consortium is such an international collaborative initiative that brings together research groups from across Europe and the USA and that specifically focuses on anthropometric traits. In their first meta-analysis, data of seven genome-wide association scans for BMI including 16 876 individuals were combined [32]. Despite a quadrupling increase in sample size compared with first-wave studies, only FTO and one new locus out of 10 loci that were taken forward for replication were unequivocally confirmed (Tables 3 and 4). The newly identified locus mapped at 188 kb downstream of MC4R, an established biological candidate, rare coding mutations of which result in monogenic forms of obesity [7] and common variants associate with a reduced risk of obesity [11, 12]. The same locus was also identified by a genome-wide association study in 2684 Indian Asians and confirmed in 11 955 individuals of Indian Asian and European ancestry [33]. Although the effect size is the same in both ethnic groups, the frequency of the risk-allele is higher in Indian Asians (36%) than in White Europeans (27%), which might in part explain why this locus could be identified with a relatively small sample of Indian Asians in the discovery stage.

For the third wave of discoveries, the GIANT consortium doubled its discovery stage sample size to 32 387 adults of European ancestry from 15 cohorts [34]. Of the 35 loci that were taken forward for follow-up in an independent series of 59 082 individuals, eight loci were firmly replicated. These include the previously established FTO and near-MC4R loci and six new loci, i.e. near neuronal growth regulator 1 (NEGR1), near transmembrane protein 18 (TMEM18), in SH2B adaptor protein 1 (SH2B1), near potassium channel tetramerization domain containing 15 (KCTD15), near glucosamine-6-phosphate deaminase 2 (GNPDA2), and in mitochondrial carrier homologue 2 (MTCH2). In parallel with the analyses of the GIANT consortium, deCODE genetics performed a meta-analysis of four genome-wide association studies for BMI, including 30 232 individuals of European descent and 1160 African-Americans [35]. A total of 43 SNPs in 19 chromosomal regions were taken forward for replication genotyping in 5586 Danish individuals and for confirmation in discovery stage data of the GIANT consortium. Besides the FTO and near-MC4R loci, eight additional loci reach genome-wide significance (Tables 3 and 4). Of these, four loci (near NEGR1, near TMEM18, in SH2B1, near KCTD15) had also been identified by the GIANT consortium, whereas four loci were new, i.e. in SEC16 homologue B (SEC16B), between ets variant gene 5 (ETV5) and diacylglycerol kinase (DGKG), in BDNF, and between BCDIN3 domain containing (BCDIN3D) and Fas apoptotic inhibitory molecule 2 (FAIM2). Variation in the HLA-B associated transcript 2 (BAT2) gene was consistently associated with weight, but not BMI, suggesting that this locus might contribute to overall size rather than adiposity. While the studies by the GIANT consortium and deCODE genetics focused on BMI as the main outcome, a third genome-wide association study examined association with the risk of early-onset and morbid adult obesity in 1380 cases and 1416 controls [38]. A total of 38 highly significant markers were taken forward for genotyping in 14 186 adults and children to test for replication with BMI and obesity risk. In addition to FTO and near-MC4R, three new markers were identified; in Niemann–Pick disease, type C1 (NPC1), near v-maf musculoaponeurotic fibrosarcoma oncogene homologue (MAF) and near phosphotriesterase related (PTER) (Tables 3 and 4).

For the FTO, near-MC4R and the BDNF loci is has been suggested that there might be two signals that each independently associate with BMI [32, 35]. However, further follow-up studies will be required to confirm this observation.

In this review, I have focused on large-scale high-density genome-wide association studies. However, other genome-wide association studies have been published for which findings were less robust. These studies were usually based on smaller samples, using genotyping platform with lower resolution, or did not include a replication stage. For example, the very first genome-wide association study for BMI, published in 2006, was performed in 694 individuals from the Framingham Heart Study using a 100 K SNP chip [40]. A SNP (rs7566605) 10 kb upstream of the INSIG2 was identified and replicated in four separate samples. However, results from subsequent large-scale studies have been inconclusive [41–45]. The 100 K genome-wide association study in the Framingham Heart Study was extended to a larger sample (n= 1341), but no replication data were available to confirm potential hits [46]. A genome-wide association study in 1000 White Americans using a 500 K SNP chip found replicated evidence for common variants in the catenin (cadherin-associated protein), β-like 1 (CTNNBL1) gene to be associated with BMI [47]. However, subsequent studies reported inconsistent support for this observation [48, 49] and also recent large-scale genome-wide association meta-analyses did not validate this locus [34, 38].

Taken together, three waves of high-density multistage genome-wide association analyses have so far discovered 15 new loci consistently associated with obesity-related traits. It should be noted, however, that the exact identity of the susceptibility variants at each locus is often uncertain. Pinpointing the causal variants will be a prime task before research towards novel biology and new therapies can take place.

Implications for public health

A major challenge is the translation of this new knowledge into public health and clinical practice. The flurry of discoveries has raised hopes of the development of genetic risk profiles that, based on the obesity-susceptibility variants, would predict early in life who would be at risk to develop obesity.

Effect sizes, risk and predication  Despite highly significant associations and consistent and repeated replication, each of the recently identified loci have only small effects on BMI (Table 4) and obesity risk (Figure 1). As discovery is the first aim of genome-wide association studies, significance of association and replication has been the main focus. For a better understanding of the importance of the susceptibility loci towards variation in obesity risk and BMI at the population level, effect size and prevalence of the risk-alleles need to be taken into account. Table 4 and Figure 1 summarize these key parameters for each of the 15 established obesity susceptibility loci for BMI and obesity risk, respectively. It should be noted that discoveries are often made thanks to the ‘winner's curse’ effect that typically overestimates effect sizes. True effect sizes are more likely to be derived from the replication stage.

Figure 1.

Effect sizes [odds ratios (OR)] for risk of obesity reported for the established obesity-susceptibility loci discovered through three waves of large-scale high-density genome-wide association studies for body mass index (BMI) and extreme obesity risk. Effect sizes represent the increased odds of obesity for each additional risk-allele. ORs for PTER and NPC1 were inferred from the OR reported for the dominant model. Frayling et al. (2007) (inline image); Loos et al. (2008) (inline image); Willer et al. (2009) (inline image); Thorleifsson et al. (2009) (○); Meyre et al. (2009) (inline image)

Up to now, genetic variation in the FTO gene still has the largest, yet small, effect on BMI in individuals of European descent. Each risk-allele increases BMI by 0.26–0.66 kg m−2 (Table 4), which is equivalent to ∼0.84–2.1 kg in body weight for a person 1.8 m tall. Accordingly, the risk of obesity increases by 1.25–1.32 odds, for each additional risk-allele (Figure 1). Although results were initially inconsistent in East Asian populations [50, 51], there is increasing evidence that effect sizes are similar to those observed for the European population [52–58]. Genetic variation in FTO does not seem to affect BMI in an African population [59], in African-Americans [31, 60], in South Asian Indians [61] or in Oceanic populations [62].

The effect sizes for the other susceptibility loci identified through genome-wide association studies for BMI are smaller than for FTO, i.e. effect sizes for BMI range between 0.06 and 0.26 kg m−2 (Table 4) and for risk of obesity between 1.03 and 1.19 odds (Figure 1). The smallest effects were observed for the MTCH2, KCTD15 and FAIM2 loci, whereas effects were more pronounced for the near-MC4R, TMEM18 and BDNF loci. Of interest, the three loci identified through a genome-wide association study for extreme early onset and adult obesity were not identified by studies on BMI. Their effect sizes for common obesity risk seem in general more pronounced [∼1.20 to ∼1.40 odds ratio (OR)] than of those discovered through genome-wide association studies of BMI. This observation suggests that differences in study design at the discovery stage might lead to the identification of a different series of susceptibility variants. The locus near MC4R has a similar effect on BMI in Indian Asians and Europeans [33]. So far, no data on effect sizes in different ethnicities of the other 13 obesity-susceptibility loci have been reported.

The prevalence of risk-alleles in the population is another important determinant of their contribution toward public health and varies between 27 and 91% for the 15 loci in Europeans (Table 4 and Figure 1). For example, the frequency of the FTO risk-allele is high in Europeans; 63% carry at least one risk-allele and 16% are homozygous. As such, the population attributable risk (PAR) of the FTO locus for obesity was estimated at ∼20% [39], implying that from a population perspective, 20% of obese cases could be prevented if the negative effects of the FTO risk-allele were eliminated. However, it is well-recognized that the PAR does not indicate how many individuals need treatment to obtain a 20% reduction in obesity prevalence, nor does it provide clinically useful risk predictions. More valuable risk estimates for clinical practice can be derived from prospective cohort studies. A 9-year follow-up study that examined the influence of the FTO and near-MC4R loci found that each additional risk-allele increases the incidence of obesity by 24% [63]. Such information is currently lacking for the more recently identified loci.

In view of applying genetic profiles to identify high-risk individuals, data on the combined influence of all established loci will be needed. In the paper by Willer et al. [34], SNPs of eight obesity susceptibility loci were genotyped in 20 000 individuals of the population-based EPIC-Norfolk cohort. The average BMI of individuals carrying 13 or more risk-alleles (<2% of the population) was 1.46 kg m−2 (or ∼3.7–4.7 kg in body weight) higher compared with those carrying three or fewer risk-alleles (<2% of the population). Although this is a respectable difference due to genetic variation, it should be noted that this compares only the two extreme 1–2% of the population. Overall, the eight variants explained only a sobering 0.84% of the variance in BMI [34]. Furthermore, the area under the curve of the receiver–operating characteristic curve was only 0.548, indicating that the eight loci combined have a rather limited predictive value [34].

Thus, despite overwhelming significances and repeated replications, the explained variance and predictive value of the currently identified obesity susceptibility loci is too low to be clinically useful.

Influence in childhood and adolescence  Only one genome-wide association study for the risk of early-onset extreme obesity has been reported so far, confirming FTO as an obesity susceptibility locus [37]. Collaborative efforts to increase the sample size and thus the power will be needed to allow examination of whether genome-wide association scans in children and adolescents identifies different loci from those identified in adults. It is thought that early-onset obesity is a genetically different trait from common adult obesity.

Loci identified through genome-wide association studies in adults have been tested for association with BMI and risk of obesity in children. For the FTO locus, similar effect sizes were observed in childhood that persist into adolescence [39, 60, 64, 65]. Although no association was observed with birth weight [39, 66–68], some found that already at the age of 2 weeks FTO SNPs were associated with increased weight and ponderal index [67]. The effect size of the near-MC4R locus on childhood BMI was twice as large as in adults and of similar size as the effect size observed for the FTO locus [32]. No associations were observed with body size at birth or during the first 3 years of life [32]. Association between the near-MC4R locus and childhood obesity (OR ∼1.4) was confirmed in European American children and adolescents, which is more than twice the effect size for adult obesity risk, whereas no such associations were observed in African-American children [69]. Of the more recently identified loci, significant associations with BMI were observed for the TMEM18, GNPDA2 and KCTD15 loci in a population-based sample of 11-year-old children [34]. The TMEM18 and GNPDA2 loci as well as the NEGR1, NPC1 and PTER loci were significantly associated with extreme childhood obesity [34, 38]. No associations with childhood BMI or obesity risk were observed for the MAF, SH2B1 or MTCH2 loci, while childhood data for other loci were not available [34, 38].

Gene–lifestyle interaction  It is well-recognized that a westernized lifestyle is the major driving force behind the obesity epidemic. However, genetically susceptible individuals will gain more weight in this obesogenic environment than those who are genetically protected. Recent studies have reported on such a FTO–environment interaction, showing that the association between FTO variants and BMI is more pronounced in individuals who are sedentary, whereas the association is attenuated in those who have a physically active lifestyle [63, 66, 70]. These observations suggest that genetic susceptibility towards obesity induced by variation in FTO can be overcome, at least in part, by adopting a physically active lifestyle. Yet, lifestyle intervention studies in individuals with increased risk for Type 2 diabetes could not confirm interaction with FTO in relation to weight loss [71–73]. There are multiple factors that can explain the discrepancy between the results from cross-sectional studies and intervention trials, including study design, study population, outcome, and lifestyle factors studied. For the more recently identified obesity susceptibility loci, no interactions with lifestyle have been reported so far, apart from one study in 5807 Danish that reported absence of interaction between the near-MC4R locus and physical activity in relation to BMI [74].

As gene–environment interaction studies require large sample sizes [75, 76], large-scale studies, ideally with objective measurement of lifestyle, using a prospective (intervention) design, will be needed to test the hypothesis that living a healthy lifestyle can overcome a genetic susceptibility to obesity. Meta-analysis of gene–lifestyle data is hampered by heterogeneity in measurement of lifestyle factors and outcomes. Also, publication bias might cause inaccurate results as negative interaction results often remain unpublished. Yet gene–lifestyle interaction data might hold important public health messages. They may allow identifying environments in which the currently established susceptibility loci explain a larger part of the variance in BMI and obesity risk. Furthermore, gene–lifestyle interaction studies may help remove the deterministic picture of obesity susceptibility genes, as they could show that a healthy lifestyle can overcome a genetic predisposition to become obese.

Implications for the aetiology of obesity

It is anticipated that the newly identified genetic loci will shed new light on the complex physiology governing the regulation of energy balance. This has raised expectations for the development of more effective preventive and therapeutic interventions.

New insights based on animal and humans studies have begun to accumulate, in particular on the biology of the FTO gene. FTO is a member of the nonhaeme dioxygenase superfamily that encodes a 2-oxoglutarate-dependent nucleic acid demethylase and catalyses the demethylation of 3-methylthymine in single-stranded DNA [77, 78]. Studies in rodents indicated that Fto mRNA expression is abundant in the brain, particularly in the hypothalamic nuclei governing energy balance, and dependent on the energy state [77, 79, 80]. Studies in humans have supported a central neuronal role for FTO as its risk-alleles have been associated with increased appetite and energy intake, and reduced satiety [81–85]. A recent study in mice, however, has challenged the neuronal hypothesis and has provided strong evidence for a peripheral role of FTO[86]. Fischer et al. [86] found that the loss of Fto in mice leads to a significant reduction in adipose tissue and lean body mass, which is a consequence of increased energy expenditure and systemic sympathetic activation. Interestingly, spontaneous locomotor activity is decreased, whereas relative food intake is increased. Thus, these experiments suggest that FTO might influence body composition through control of energy expenditure [86]. A peripheral role for FTO was also proposed by a study in healthy women showing that carriers of the risk-allele had reduced lipolytic activity, independent of BMI [87]. However, other studies in humans could not confirm association with resting or physical activity energy expenditure [65, 70, 83–85, 88]. Adding to the complexity is the observation that the Ftm gene, that lies in opposite orientation to Fto, has a similar hypothalamic expression pattern as Fto suggesting that the gene might be co-regulated, so that it is not completely clear yet which of these two genes (or both) functionally relevant [79].

The near-MC4R locus, identified in the second wave of genome-wide association studies, is located at 188 kb downstream of MC4R, which is an obvious candidate gene given its role in monogenic early-onset obesity. Although MC4R is the gene nearest to the association signal and the phenotypic associations are consistent with effects mediated through MC4R function, it has not yet been firmly established whether this locus indeed reflects MC4R function.

For most of the 13 loci discovered during the third wave of genome-wide association studies, the physiological role in relation to obesity risk is not or poorly understood [34, 35, 38]. In fact, many of these loci harbour multiple genes, sometimes located within a recombination interval with high linkage disequilibrium, which hampers pinpointing the causal variants. Yet two loci contain genes with a strong candidacy. SH2B1 is implicated in leptin signalling and Sh2b1-null mice are obese [89]. Although the SH2B1 variant with the most significant association is a nonsynonymous SNP (Thr484Ala), it is in strong linkage disequilibrium with variants in neighbouring genes [34]. Thus it remains to be confirmed whether the signal is indeed representing SH2B1 function. Also, the BDNF locus is a strong prior candidate and one of the SNPs in this locus is the nonsynonymous Val66Met that has previously been examined in candidate gene studies of eating behaviour and BMI (Table 1) [20].

Of interest is that many of the currently identified obesity susceptibility loci locate near genes that are highly expressed in the brain and hypothalamus, favouring a role for the nervous system in body weight control [34, 35].

Comprehensive resequencing and fine mapping will be required to identify unambiguously the causal variant before physiologists can start exploring the functional relevance of the locus in relation to the risk of obesity, which will be key for translation into clinical practice.

Ways ahead

Despite the enormous success of genome-wide association studies, the established loci in combination explain <2% of interindividual variation in BMI. Given that the heritability of BMI is estimated at 40–70%, one should conclude that many more susceptibility loci remain to be uncovered. Various approaches have been proposed to identify more genetic loci, to pinpoint the causal variants, and to explore the physiological mechanisms and pathways that underlie the observed association [90].

One approach is to initiate a fourth wave of genome-wide association studies that further increases the sample size of the discovery stage. For example, the GIANT consortium has extended its meta-analyses, including >100 000 individuals with genome-wide association data. This will further improve power to uncover common variants with even smaller effect sizes. Alternatively, less stringent significance levels for taking variants forward for replication could identify true loci that are currently hidden amid false-positive results.

As power to reveal new loci might vary across populations because of differences in effect sizes and allele frequencies, the study of ethnicities other than White European might provide new gene discovery opportunities. For example, Chambers et al. needed only a relatively small discovery sample of Indian Asians to identify the near-MC4R because the frequency of the risk-allele was substantially higher than in Europeans [32, 33].

So far, most studies have used BMI as a simple and inexpensive proxy-measure of adiposity, which is easy to collect in large samples. More accurate measures of adiposity might further improve power, yet these are often more expensive and harder to collect. While the power of a study will depend on the balance between sample size and accuracy of measurement, emphasis on one or the other might yield different loci. Genome-wide association studies for body fat percentage, waist circumference, extreme obesity risk, for mediating traits that underlie obesity, such as food intake and energy expenditure, might reveal new obesity-susceptibility loci that are currently hidden in studies on BMI.

Risk-allele frequencies of the newly identified obesity susceptibility loci vary between 27 and 91% (Figure 1) in European populations. It is unlikely that within this range of allele frequencies there are variants with larger effect sizes than those already observed, as one would expect that at least the most recent genome-wide association studies would have had sufficient power to identify those. However, less frequent variants with larger effects may have remained undetected, as the current genome-wide scans have only limited potential to capture rarer variants. The 1000 Genomes Project, which aims also to catalogue variants of lower frequency, might lead to new genome-wide genotyping chips that more fully capture the genetic variation in humans.

The contribution of copy number variants to the predisposition of obesity has so far been unexplored. Further advances in technology and analytical modelling will be required before analyses of copy number variants can be implemented on a larger scale. Yet the observation that the NEGR1 locus might represent a copy number variant indicates the potential importance of this source of variation [34].

Besides aiming to identify more susceptibility loci, follow-up of the established loci in molecular and physiological studies will be key to determine the mechanisms through which the loci confer obesity. A major challenge before loci can be passed on to physiologists is pinpointing the causal variant or gene. This will require high-throughput sequencing of the region of interest in extreme cases and control or in different ethnicities. It is only when the causal locus is identified and its modes of action are fully understood that this knowledge can be translated into mainstream healthcare and clinical practice.


With genome-wide association, we have entered a new era of gene discovery for common obesity. While over the past 15 years candidate gene studies have identified a handful of genetic variants convincingly associated with obesity-related traits, genome-wide association studies have changed the pace of discoveries with the discovery of at least 15 loci in less than 3 years.

Recent progress in obesity genetics has already provided valuable new insights into pathophysiological mechanisms and pathways that underlie obesity development. Although these advances have raised hopes for genetic risk profiling and therapeutic intervention, the implementation of such approaches in mainstream healthcare remains distant, as we stand to learn much more about the causal variants and their functional implications.

Expectations are high but many challenges remain. Among the latter, translating new advances on the genetic predisposition to obesity into useful guidelines for prevention and treatment will be the most demanding.

Competing interests

None to declare.