Keywords

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

The rise in the prevalence of obesity in recent decades has been spectacular: recent estimates indicate that over 500 million adults worldwide are now classed as obese [1]. While the increased prevalence of obesity is almost certainly a reflection of secular changes in environmental and lifestyle factors, including an increased intake of nutrient-dense foods coupled with reduced physical activity [2], the familial aggregation of obesity is consistent with some degree of genetic influence on body mass index (BMI) and individual predisposition to obesity. More conclusive evidence for a genetic component comes from studies that have examined the correlation of BMI between identical twins raised apart and the relationship between the BMI of adoptees and both their biological and adoptive parents [3–5]. These consistently highlight the importance of genetic factors in modulating individual susceptibility to obesity in contemporary environments. Furthermore, in controlled experiments of excessive calorie intake, consequent changes in weight and body composition were highly correlated in monozygotic twins, once again consistent with a powerful role of genetic variation in the regulation of weight [6]. Estimates for the heritability of BMI vary widely between studies, but typical figures range between 0.47 and 0.90 in twin studies and between 0.24 and 0.81 in family-based studies [7].

Other obesity related traits, including measures of fat distribution are also heritable (even after adjusting for BMI). Estimates for the heritability of waist–hip ratio (WHR), a proxy of fat distribution, range between h 2 ~ 0.31–0.70; and ~0.22–0.61 after accounting for BMI [8–11]. The heritability of WHR is higher in women and estimates of genetic correlation of WHR between men and women indicate a sex specific genetic influence on the trait [11].

The Genetics of BMI and Obesity Pre-GWAS

Genetic studies aim to find DNA sequence variants that are causally associated with the trait of interest, in the expectation that such discoveries will help to reveal fundamental mechanisms responsible for human disease. The earliest studies in this field focused on the application of family-based linkage studies to individuals and families with rare monogenic forms of obesity. The rare variants of large effects revealed by these efforts, such as those in LEP (encoding the hormone leptin, a crucial component of energy balance mechanisms) [12–14], LEPR (encoding the leptin receptor) [15], and POMC (encoding the proopiomelanocortin protein which is cleaved to form a number of key neuroendocrine messengers) [16], helped to define components of hypothalamic circuitry involved in body weight regulation in man. However, the application of linkage approaches to population-level variation in BMI and risk of common forms of obesity met with little success in terms of robust, replicated signals even in relatively well-powered meta-analysis [17]. This indicates that the genetic contribution to these traits is not dominated by the kinds of highly penetrant variants which linkage methods are best suited to detect [18].

The shift from linkage to association approaches was initially focused on the analysis of candidate genes [19], a strategy reliant on the quality of the prior biological hypotheses used to select them. One of relatively few successes from this approach was the demonstration that low frequency variants in the gene encoding the melanocortin 4 receptor (MC4R) were associated with severe, early-onset obesity [20]. These variants remain the commonest known genetic cause of morbid obesity contributing to a few percent of these cases [21]. These findings provided confirmation of the role of signalling through the hypothalamic leptin–melanocortin pathway for the maintenance of body mass in man [22]. However, the major impetus to the discovery of BMI- and obesity-associated variants has been provided by the ability to perform genome-wide scans for association.

Genome-Wide Association Studies

Genome-wide association studies (GWAS) (reviewed in [23, 24]) use dense genotyping arrays to determine how variation in genomic sequence (predominantly that due to single nucleotide polymorphisms, SNPs) associates with phenotypic traits of interest. Those traits may be categorical (e.g., obese cases and non-obese controls) or continuous (e.g., BMI or WHR). Array content and the correlation structure of variation across the genome (i.e., linkage disequilibrium) mean that GWAS to date have favored the interrogation of common variants (minor allele frequency [MAF] > 5 %). Since GWAS assay such variants across the genome, suitably powered studies enable the discovery of associated loci in an agnostic fashion, without the need for prespecified hypotheses concerning the genomic location of the association and the transcripts through which they may operate.

In the remainder of this chapter, we focus on the loci which have been shown by GWAS to be associated with overall obesity or fat distribution. We distinguish between studies of traits of overall obesity (including BMI, fat percentage, and dichotomized indices of extreme obesity) and those of fat distribution (including WHR, waist circumference (WC), and measures of visceral and subcutaneous fat). In total 74 genome-wide significant loci have been associated with these traits and most of these (52 in number) are common variant loci influencing continuous obesity-related traits found in European samples. Others derive from equivalent studies in non-European samples (4 loci), and some have emerged exclusively from case–control studies in individuals selected from the extremes of the BMI distribution (11 loci) or by clinical classifications of overweight and obesity (7 loci).

Overall Obesity

Genome-Wide Association Studies of BMI

The first report from a GWA study claiming to have identified variants associated with common forms of overall obesity came in 2006 [25]. The researchers used a two-stage family-based design to identify a signal mapping close to the INSIG2 encoding insulin induced gene 2 [25]. However, this association has not been proven robust to replication in the much larger samples that have been examined in subsequent studies (see below). In fact, the association p-value observed in this study fell short of the now-widely accepted threshold (p < 5 × 10−8: based on p < 0.05 corrected for a million independent tests [26]), highlighting the value of such stringent criteria as a means of avoiding inflation of the type 1 error, and the attribution of biological significance to loci which, like INSIG2, are likely to have been false positives.

The first report of a robust genome-wide significant locus influencing BMI and risk of obesity locus came from Frayling et al. [27] in 2007, and concerned a cluster of common variants close to the FTO (“Fat mass and obesity-associated”) gene. These variants account for ~0.35 % of the phenotypic variance in BMI in Europeans [28] such that the two groups of homozygotes differ in weight by around 2.5 kg. The BMI association has now been widely replicated [28–33] and it is also clear that the same FTO variants are associated with risk of obesity at all grades of severity [27, 34].

Given that the only locus emerging from this first round of GWA studies [25, 27] had a relatively modest effect size, it was clear that larger sample sizes would be needed to extend these discoveries, both to common alleles of lesser effect, and to less frequent risk alleles. This provided the motivation for ever-larger meta-analyses efforts, which have dominated discovery efforts over the past few years. The largest of the studies published to date assembled data from almost 250,000 individuals [28]. The current count of BMI-associated loci detected in Europeans by these studies, most of them conducted under the aegis of the Genetic Investigation of ANthropometric Traits (GIANT) consortium [28–30], is 32 (Fig. 3.1).

Fig. 3.1
figure 1

Overlap of genome-wide significant loci of overall obesity (BMI), fat distribution (BMI-adjusted WHR) and BMI extremes (or clinical obesity classes) in European populations. Diagram depicts the overlap of reported GWAS loci (p < 5 × 10−8) of BMI [27–31], BMI-adjusted WHR [98], and BMI extremes or clinical obesity classes II–III [34, 38, 51–53, 55, 120]

Full size image

The first such meta-analyses uncovered common regulatory variants influencing BMI near MC4R (encoding melanocortin receptor 4) [29]: low-frequency coding variants in this same gene had previously been implicated in severe obesity [20]. Subsequently, the parallel publications from GIANT [30] and the deCODE group [31] added nine BMI loci (mapping near GNPDA2, KCTD15, MTCH2, NEGR1, SH2B1, TMEM18, BDNF, ETV5, and SEC16B) to the list. It is of note that BDNF, encoding a brain derived neurotrophic factor involved in regulation of development of neuronal circuits [35], is also involved in monogenic forms of obesity [36]. The largest meta-analysis of BMI associations added 18 further loci to the tally [28] including regions near known obesity genes such as POMC (proopiomelanocortin) [16], known to be involved in neuroendocrine regulation of weight, as well as associations in or near novel genes such as GPRC5B (G protein-coupled receptor, family C, group 5, member B), implicated in regulation of adipose inflammatory processes and progression to insulin resistance in obesity in mice [37].

The studies above mostly focused on the analysis of SNPs but there is some evidence that copy number variations (CNVs) may be causal at some loci. For example, in the study by Willer et al. the BMI-associated SNP at the NEGR1 (neuronal growth regulator 1) locus detected by GWAS was shown to tag a 45 kb deletion that might have stronger functional grounds for being causal [30]. More detailed studies published recently locate the causal allele at this locus to a second 8 kb deletion near NEGR1 [38]. Rare CNVs have also been implicated in syndromic forms of obesity. For example, a rare deletion in the 16p11.2 region is associated with the combination of severe obesity and mental retardation [39, 40], and duplication of the same region is associated with underweight [41].

To date, most GWAS studies have been performed in populations of European origin but studies in other ethnic groups can help to identify novel loci, to characterize the extent of aetiological overlap, and to fine-map causal variants (such as in the FTO locus [42, 43]). Two large GWAS meta-analyses of BMI in East Asian populations were recently published [32, 33]. Between them, seven of the known loci in Europeans could be replicated to genome-wide levels of significance. Furthermore, evaluating the associations in East Asians of previously reported BMI loci, an additional 11 loci, besides those genome-wide significant, were associated with BMI at lesser levels of significance (Okada et al. [32], p < 0.02; Wen et al. [33], p < 0.05), indicating considerable overlap in signals between East Asian and European populations. As in populations of European origin, the association at FTO locus explained the largest proportion of phenotypic variance (~0.2 %).

In addition, these studies identified four novel loci, mapping near CDKAL1, KLF9, PCSK1 and GP2 [32, 33]. Mutations in PCSK1 cause monogenic obesity [44] and, while a candidate study previously associated nonsynonymous variants in PCSK1 with common obesity risk in a European population [45], the PCSK1 signal in East Asians (also nominally associated with BMI in Europeans in GIANT [28]) likely represents an independent signal. Genetic variants in CDKAL1 (a CDK5 regulatory associated protein 1-like 1 with methythiotransferase function [46]), in strong LD (r 2 ~ 0.8) with the BMI GWAS SNPs in East Asians, have previously been associated with increased risk of Type 2 Diabetes (T2D) [47, 48]. The T2D risk allele(s) is associated with decreased glucose-stimulated insulin secretion [47, 49, 50]. Furthermore, the BMI-lowering allele of rs2206734 (also nominally associated with BMI in Europeans) was associated with increased risk of T2D in the same study population [32], indicating that variation near CDKAL1 may play a complex role with respect to variation in both BMI and T2D-risk.

Case–Control Studies of Dichotomized BMI

In addition to studies of the variance in BMI in population-based studies, a complementary approach treats obesity in terms of a dichotomous “case–control” analysis. A variety of different schemes for this dichotomization are possible (Table 3.1). In the largest study of this type [51], featuring case–control analyses restricted to the “tails” of the BMI distribution using data from studies previously included in GIANT meta-analyses [28], Berndt et al. found considerable overlap in the pattern of association signals seen as compared to those observed in population- or cohort-wide analyses. However, where such studies focus on cases of more extreme definitions of obesity and/or leanness (that is, individuals several standard deviations away from the population mean), there may be the opportunity to detect additional, novel, signals that may have limited impact on overall population-level variance and which are therefore difficult or impossible to detect using GWAS approaches. The rare, penetrant variants causal for monogenic and syndromic forms of obesity provide the most obvious example of this phenomenon.

Table 3.1 Novel GWAS loci identified in case–control analyses of dichotomized BMI
Full size table

Indeed, whilst several of the loci reaching genome-wide significance in dichotomous analyses focused on extreme obesity in adults overlap with those previously reported (e.g., BDNF, FTO), there are several signals that appear unique to dichotomous analyses including KCNMA1, NPC1, PTER, and HS6ST3 (Table 3.1, Fig. 3.1) [51–53]. However, most of these have appeared in a single study and have not, as yet, been replicated, even in other extreme case–control analyses. In equivalent case–control analyses in children, the more relaxed criteria adopted by Bradfield et al [54] detected many of the known adult BMI association signals but also highlighted novel signals near OLFM4 and HOXB5. In contrast, studies of children selected from the extremes of the distribution have detected signals at (or approaching) genome-wide significance near LEPR, PACS1, PRKCH, RMST, SDCCAG8, and TNKS/MSRA (Table 3.1, Fig. 3.1) [38, 55], the latter locus also detected in some studies of fat distribution [56].

Genome-Wide Association Studies of Fat Percentage

BMI, although a widely used proxy of overall obesity, represents an aggregate measure of the lean and the fat mass of the individual. In an effort to better define the genetic determinants of obesity, Kilpeläinen et al. focused on body fat percentage, as a more direct measure of adiposity, generating a GWAS meta-analysis of 36,626 individuals [57]. As well as detecting FTO, these analyses recovered two loci (IRS1 and SPRY2) not previously associated with BMI. The body fat-increasing alleles at the IRS1 (insulin receptor substrate 1 signalling protein) signal are, intriguingly, associated with a healthy metabolic profile (including reduced risk of T2D [58] and unhealthy lipid profile [59]). The IRS1 locus is associated with measures of subcutaneous, but not visceral fat, indicating that the effect on fat mass at the IRS1 locus is through regulation of subcutaneous fat deposition [57, 60]. The SPRY2 locus has also been implicated in T2D risk [61, 62], though the body fat-associated SNP is not coincident with this previously reported T2D SNP. Contrary to the observations at IRS1, the body-fat increasing allele at the SPRY2 locus is associated with an adverse metabolic profile [57].

Genetic Architecture of Overall Obesity (BMI)

Despite the success in identifying a growing numbers of loci to genome-wide significance, in European populations these signals, in combination, explain no more than 1.5 % of phenotypic variance in BMI. Of the established loci, the FTO locus has the largest effect accounting for ~0.35 % of population variance [28]. These numbers fall well short of estimates of the heritability of this trait (see above). The basis for this “missing” genetic variance remains unclear, though there is no lack of possible explanations [63, 64]. At least part of the “missing” genetic variance can be attributed to the effects of additional common variants that lie below the genome-wide significance threshold. Using full GWAS data sets (not just the “proven” hits), Yang and colleagues could recover approximately 17 % of the phenotypic variance in BMI that was tagged by common variants [65]. Part of the remaining shortfall likely reflects incomplete linkage disequilibrium between the variants genotyped on GWAS arrays and those which are causally responsible for the BMI associations [66], but other mechanisms are almost certainly involved [63, 67].

There is also the possibility that the estimates of heritability against which these measures of explained variance are evaluated, are themselves inaccurate. For example, intrauterine events that lead to epigenetic modifications with long-term phenotypic impacts can lead to increased sibling resemblance, inflating heritability estimates under some designs. Similarly, estimates derived from the comparison of the phenotypic correlations observed between monozygotic and dizygotic twin pairs are based on the assumption that both types of twin are exposed to a similar degree of shared environment [68], an assumption that may not be appropriate for intrauterine exposures.

Notwithstanding the above, it seems likely that an appreciable component of the genetic variance remains unexplained, and that at least part of this will be attributable to low frequency and rare variants not well captured by GWAS studies to date. The current wave of sequencing studies should shed some light on the extent to which these variants are contributing to inherited risk.

From GWAS Associations to Potential Functional Roles in Overall Obesity

As we have seen, GWAS have powered the identification of many genetic regions associated with BMI and obesity. However, this information is of limited value unless it can be translated into improved understanding of the pathophysiology of disease, and thereby into novel clinical approaches. However, in BMI, as with most other complex traits, the regions revealed by GWAS do not lend themselves to easy biological inference. The effect sizes are modest, and most signals map to non-coding sequence, frustrating efforts to identify the “causal” transcript (that is, the specific gene that is mediating the association signal). At the same time, the extensive local correlations between common variants (that is, linkage disequilibrium) can make fine-mapping of the causal variants challenging.

The FTO locus provides an excellent example of the difficulties inherent in moving from an association signal—in this case, a comparatively strong one—to a clear mechanism of action. We have now known for more than 6 years that a cluster of highly correlated common variants in the first intron of the FTO gene is associated with BMI and obesity [27]. Epigenetic analyses have suggested that the BMI-associated haplotype may influence local methylation status [69] but fine-mapping efforts have yet to provide compelling localization of the causal variant. When it comes to defining downstream effects, we still have no convincing evidence from man that the FTO transcript itself is in any way involved. There is for example, no instance of the co-occurrence of loss of function alleles in FTO and severe obesity in humans [70, 71]. On the other hand, the adjacent gene RPGRIP1L (or FTM), which is known to be coordinately regulated with FTO via a common promoter [72], and to display a similar pattern of hypothalamic expression, has an intriguing connection to obesity through its known causal role with respect to monogenic ciliopathies [73] some of which result in marked early obesity.

In fact, the most compelling evidence implicating FTO comes from mouse models: transgenic knockdown of the murine homologue Fto results in reduced weight, and overexpression to weight gain compared to control mice [74, 75]. One possible explanation consistent with these data is that the common intronic variants within FTO identified by GWAS, exert their effects on energy balance in man through coordinate dysregulation of both FTO and RPGRIP1L.

The identification of the signal at FTO naturally prompted interest in the normal function of this transcript. In humans, FTO encodes a 2-oxoglutarate-dependent nucleic acid demethylase [76] thought to be involved in nucleic acid repair. In vitro studies have suggested a role for FTO demethylation in cellular sensing of amino acids [77], which could be relevant to regulation of appetite control in the hypothalamus. Nonetheless, it is clear that we remain some way from a complete description of how these variants influence BMI and obesity risk.

At certain other BMI GWAS loci, the situation is better understood. At four GWAS loci (near BDNF, PCSK1, POMC and MC4R) the common variant associations overlap genes in which coding mutations have been shown to be causal for monogenic or syndromic forms of obesity [20, 36, 78, 79]. In the case of three of these—PCSK1 (proprotein convertase 1), POMC (proopiomelanocortin), and MC4R (melanocortin receptor 4)—there are strong mechanistic ties to the hypothalamic leptin–melanocortin signalling pathways that regulate energy balance [80]. BDNF encodes a brain derived neurotrophic factor involved in neurogenesis and thought to be involved in food intake [81]. These GWAS signals therefore demonstrate that the neuroendocrine mechanisms documented in monogenic forms of obesity extend to population level variance in BMI and to more common forms of obesity.

At other BMI-associated GWAS, efforts to define the causal transcript are supported by additional sources of genomic data (regulatory annotations [82] or mRNA expression [83]). For example, it can be very useful through integration with mRNA and/or miRNA transcriptomic data [83–88] to demonstrate that the set of BMI-associated variants at a given locus also drives cis-expression of one of the regional transcripts. In the most recent GIANT meta-analysis [28], this approach led to positional candidates being identified at almost half the 32 BMI-associated loci.

These candidacy assignments can often be bolstered by other sources of data. Consider for example the association signal mapping close to the SH2B1 gene, encoding SH2B adapter protein 1. Cis-expression data point to SH2B1 [28], as does the high expression of this transcript in the hypothalamus [30]. The neuronal isoform of SH2B1 is involved in regulation of energy balance via effects on leptin and insulin signalling, and systemic deletion of the gene in mice results in severe leptin resistance [89].

For some loci, the data seem to point towards peripheral rather than central mechanisms of action. The BMI association on chromosome 19 lies close to the GIPR gene, encoding the gastric inhibitory polypeptide receptor, and the lead SNP is in strong LD with a missense SNP in that transcript (though the functional consequences of that mutation are not yet established). GIPR plays an important role in mediating the incretin response, which augments insulin release in response to the ingestion of food. The same locus has also been shown to associate with glucose response and insulin secretion in response to a glucose challenge [90]. Another example, mentioned earlier, is the mechanistic relationship between insulin signalling and obesity implicated by the association between IRS1 variants and fat percentage [57]. Though both central and peripheral mechanisms may be involved at IRS1, the fact that the fat percentage-associated allele is associated with improved insulin sensitivity and a healthy metabolic profile [58, 59] is consistent with enhanced insulin-mediated adipogenesis as the driver of the adiposity.

For several other BMI-associated loci such as TMEM160-ZC3H4 [28], there are few clues on the biological relevance in obesity, and any one of several transcripts could be responsible. One way of leveraging the combination of genetic and prior biological data to make provisional mechanistic inference in such situations is to perform pathway-based analyses (reviewed in Wang et al. [91]), which test for enrichment of GWAS loci for transcripts that have been mapped to defined biological processes or pathways. Applied to BMI GWAS data, these analyses have tended to support the evidence for broad neuroendocrine involvement, whilst also highlighting processes that are more difficult to assimilate within the current knowledge base (e.g., platelet-derived growth factor signalling) [28].

Fat Distribution

The clinical consequences of adipose tissue excess depend not only on its quantity but also its distribution, with the accumulation of visceral (abdominal) fat leading to particularly adverse metabolic and cardiovascular effects [92, 93]. After accounting for overall obesity (as measured by BMI), fat distribution (commonly measured by WHR) shows substantial residual heritability (h 2 ~ 0.22–0.61) consistent with mechanisms of genetic control distinct from those influencing overall energy balance and BMI [9, 10]. The distinct genetic regulation of patterns of fat distribution is also supported by rare monogenic syndromes of selective adipose tissue loss (collectively, the lipodystrophies) [94]. Given the checkered history of efforts to target neuronal pathways related to overall obesity in the search for effective, safe treatments for obesity, there is considerable interest in defining the mechanisms responsible for individual variation in patterns of fat distribution, and in particular, in identifying peripheral (rather than central) targets for therapeutic intervention.

Genome-Wide Association Studies of WHR and WC

Initial efforts to map variants influencing fat distribution focused on the standard clinical traits, WHR and WC. In the first GWAS for WHR, Lindgren et al. discovered an association to a genetic variant on chromosome 1 (close to the LYPLAL1 gene encoding lysophospholipase-like 1) associated with WHR in women exclusively: this effect was independent of BMI [56]. Studies of WC generated their strongest signals at previously reported BMI loci such as FTO and MC4R, reflecting the strong correlation between these traits [27, 56, 95, 96]. With the possible exception of the association near TFAP2B, at which adjustment with BMI seems to increase the magnitude of the effect on central obesity [97], other WC-associated loci identified by GWAS (MSRA, NRXN3) are likely to reflect a primary association with BMI [28].

Given these strong trait correlations, more recent fat distribution GWAS efforts have adopted the approach of adjusting WHR (or WC) for BMI before performing the association analyses, thereby seeking to emphasize those signals that influence patterns of relative fat deposition independent of the overall obesity component. In the largest analysis to date, involving data from around 190,000 subjects, Heid et al. [98] used this approach to identify 13 novel loci for BMI-adjusted WHR as well as to replicate the signal near LYPLAL1. As might have been expected given the adjustment for BMI, the loci identified by this endeavor were completely distinct from those previously reported to influence overall obesity (Figs. 3.1 and 3.2). In line with the metabolic consequences of visceral fat accumulation, these fat distribution associated variants are also enriched for association with related metabolic traits including fasting insulin, lipids and indices of insulin resistance [98]. The obvious gender dimorphism of WHR prompted efforts to evaluate these signals in terms of their potential for different effects in males and females. Half of the 14 loci showed evidence of gender-specific effects: in each case, the effect was stronger in women [98].

Fig. 3.2
figure 2

Effect sizes for BMI in GIANT meta-analyses vs. BMI-adjusted WHR in GIANT meta-analyses for genome-wide significant BMI and BMI-adjusted WHR loci. In the scatterplot, data for BMI in GIANT meta-analyses [28] are shown on the X-axis and data for BMI-adjusted WHR in GIANT meta-analyses [98] on the Y-axis. The points are colored according to if they represent loci associated with BMI (red) or BMI-adjusted WHR (blue)

Full size image

In a complementary approach to studies of WHR in population-wide analysis, Berndt et al. restricted analysis to the “tails” of the WHR distribution (upper and lower 5th percentiles) and analyzed WHR in terms of dichotomous “case–control” analyses [51]. This analysis demonstrated a similar pattern of association signals as that of previous population-wide analysis [98], indicating that WHR at the “tails” of the distribution has a similar genetic architecture as that of the full distribution.

Genome-Wide Association Studies of Abdominal Fat Distribution

The use of imprecise, but widely available, clinical measures such as WHR facilitates large meta-analysis, but there is much to be gained by complementary analyses in smaller numbers of more carefully phenotyped subjects. In a recent study, more direct measurements of the extent of abdominal subcutaneous and visceral adiposity were obtained by computed tomography (CT) [60]. This analysis was able to demonstrate that the fat distribution association signal near LYPLAL1 [51, 56, 98] could also be detected using CT (as the ratio between subcutaneous and visceral fat area). It also highlighted a signal near THNSL2 that was associated with visceral adiposity in women: this survived adjustment for BMI, and has not previously been associated to obesity traits [60].

Genetic Architecture of Fat Distribution

Combined, the 14 loci for BMI-adjusted WHR uncovered by GWAS account for approximately 1 % of variance in this trait (1.34 % in women; 0.46 % in men) [98]. Using methods analogous to those for BMI described above [65], Vattikuti et al. showed that ~13 % of the overall variance in WHR could be explained by common GWAS SNPs, and thus estimated that 46 % of heritability in WHR may be captured by common variants [99]. A similar range of explanation for the missing genetic variance is possible as for overall obesity [63, 64, 67], and ongoing sequence-based efforts will help to define the extent to which this deficit can be plugged by the contribution of low frequency and rare variants.

From GWAS Associations to Potential Functional Roles in Fat Distribution

As with BMI, progress towards characterization of the mechanisms operating at each of these loci has been patchy. Expression-QTL mapping in adipose tissue, blood, and other tissues has identified promising candidate transcripts at six of the loci (AA553656, GRB14, PIGC, STAB1, TBX15, and ZNRF3) [98].

For several of these transcripts, the genetic data integrates well with the corpus of existing biological data. For example, GRB14, encoding a growth factor receptor-binding protein, is known to acts as a negative regulator of insulin receptor signalling [100, 101]. The WHR-associated variant shows directionally consistent associations with triglyceride and insulin levels [98] and other (statistically independent) variants at the same locus influence BMI-adjusted insulin and HDL-cholesterol levels [102, 103]. TBX15 encodes a mesodermal developmental transcription factor and has been indicated in adipocyte differentiation and triglyceride accumulation [104]. This transcript is also differentially expressed between visceral and subcutaneous adipose tissue, and there is evidence that visceral adipose tissue expression is negatively correlated with BMI [105].

The most consistent signal for fat distribution maps to the LYPLAL1 locus [51, 56, 60, 98]. As might be expected variants at this locus are associated with a range of related metabolic and anthropometric traits including adiponectin [106], fasting insulin adjusted for BMI [103] and height [107]. So far, there is limited evidence to demonstrate that the signal is mediated through the LYPLAL1 transcript and the region contains several other potential candidates. However, expression of this gene is induced in subcutaneous fat following obesity [108] and its presumed function as a lysophospholipase is consistent with a causal role.

As with the GWAS loci associated with BMI, pathway enrichment approaches have been applied across the 14 WHR-associated loci. Though the enrichment signals were relatively weak, they highlighted developmental processes and mRNA transcript regulation [98]. The known functions of some of the stronger positional candidates—such as angiogenesis (VEGFA), adipocyte differentiation (GRB14) and developmental function (TBX15, HOXC13)—seem to point towards peripheral mechanisms.

These enrichment patterns, when compared with those seen for BMI, seem consistent with the hypothesis that overall obesity is primarily defined by variation at genes involved in central neuroendocrine regulation, whereas fat distribution is largely influenced by variation at genes, which control peripheral aspects of adipose function and development.

Challenges for the Present and for the Future

Whilst there is no doubt that GWAS studies have accelerated our understanding of the genetics and biology of obesity, there remains much to do. At most of the loci discovered, we have yet to identify the causal variant (or variants) or to define with certainty which regional transcript is responsible for mediating the association effect. The accumulation of transethnic association data [109–111] combined with the growing use of next-generation sequencing to generate reference sets for imputation [112] and to interrogate phenotypically selected individuals (e.g., the morbidly obese) should help to address the former. The latter depends in part on the generation of improved annotations (particularly those from relevant tissues) that connect non-coding variation to transcript regulation, and on the development of appropriate functional assays. As always, the ability to refine the phenotypic consequences of allelic differences at variants of interest in human subjects (for example through imaging of fat tissues) will play a crucial role in defining a mechanistic understanding of these traits.

Missing Genetic Variance

As we have seen the loci identified by GWAS loci explain a surprisingly small proportion of phenotypic variance, far less than appears to be the case for other “similar” quantitative traits, such as height and lipids. Approaches that combine effects across the entire GWAS dataset, rather than considering only those signals reaching genome-wide significance, do a better job of recovering variation (indicating a long “polygenic” tail of common variant susceptibility) but still leave a substantial component of estimated heritability unexplained [65]. To discover further genome-wide significant associations to common genetic variants with increasingly smaller effects would require even larger studies than to date. The latest wave of grand meta-analyses of BMI and BMI-adjusted WHR (involving over 320,000 and 210,000 European samples respectively), currently underway, promise to reveal some of these common variant signals, as do the studies emerging from analyses in a variety of non-European samples.

It has been suggested that rare (MAF < 1 %) or low frequency (MAF 1–5 %) variants beyond the range of the historical GWAS approach, may contribute to this missing genetic variance [24, 64, 113, 114]. The rapidly decreasing cost and increasing accuracy of next-generation sequencing are bringing variants in this class under the microscope for the first time [115].

It is clear that individual risk of obesity reflects the integration of genetic and non-genetic factors including variation in food availability and extent of physical exercise [116]. Indeed, these may directly interact such that variant effects are modulated by these lifestyle factors: under some circumstances these interactions may contribute to the missing “genetic” variance [67]. The detection of such interactions at the genome scale requires massive sample sizes, unless the interaction terms are substantial. Nevertheless, there are several examples now emerging of interaction effects at obesity loci: these include an interaction between FTO and exercise [117] as well as sex-specific effects reported for WHR [56, 98] and visceral adiposity [60].

Risk Prediction, Intervention and Medication

One might hope that improved knowledge about the genetics of obesity would help to generate predictive models. These might be used to identify individuals at highest future risk of obesity who could be targeted for early intervention, and/or define genetic markers related to treatment outcome that can be used to guide therapeutic choices. However, the common variants so far identified by GWAS have too weak an effect, even in combination, to have value in this respect. Indeed, genetic risk factors are currently outperformed by traditional risk factors [118] including present BMI (a good predictor of future obesity risk [119]).

Instead, the most valuable translational benefits are likely to accrue from the biological knowledge, which grows from the genetics. Currently, there are few effective pharmaceutical treatments for obesity, and the most successful clinical intervention requires radical (bariatric) surgery. The clinical burden of obesity urgently requires the identification of novel validated therapeutic targets based around a better understanding of underlying mechanisms. The wider behavioral effects of drugs acting on central processes such as appetite may continue to prove problematic in this respect and efforts to target peripheral mechanisms of fat distribution, and thereby ameliorate the adverse metabolic consequences of obesity may prove more productive.