Meta-analysis identifies 13 new loci associated with waist-hip ratio and reveals sexual dimorphism in the genetic basis of fat distribution

Abstract

Waist-hip ratio (WHR) is a measure of body fat distribution and a predictor of metabolic consequences independent of overall adiposity. WHR is heritable, but few genetic variants influencing this trait have been identified. We conducted a meta-analysis of 32 genome-wide association studies for WHR adjusted for body mass index (comprising up to 77,167 participants), following up 16 loci in an additional 29 studies (comprising up to 113,636 subjects). We identified 13 new loci in or near RSPO3, VEGFA, TBX15-WARS2, NFE2L3, GRB14, DNM3-PIGC, ITPR2-SSPN, LY86, HOXC13, ADAMTS9, ZNRF3-KREMEN1, NISCH-STAB1 and CPEB4 (P = 1.9 × 10⁻⁹ to P = 1.8 × 10⁻⁴⁰) and the known signal at LYPLAL1. Seven of these loci exhibited marked sexual dimorphism, all with a stronger effect on WHR in women than men (P for sex difference = 1.9 × 10⁻³ to P = 1.2 × 10⁻¹³). These findings provide evidence for multiple loci that modulate body fat distribution independent of overall adiposity and reveal strong gene-by-sex interactions.

Central obesity and body fat distribution, as measured by waist circumference and WHR, are associated with individual risk of type 2 diabetes (T2D)^1,2 and coronary heart disease³ and with mortality from all causes⁴. These effects are independent of overall adiposity as measured by body mass index (BMI). WHR is of particular interest as a measure of body fat distribution because it integrates the adverse metabolic risk associated with increasing waist circumference with the more protective role of gluteal fat deposition with respect to diabetes, hypertension and dyslipidemia^5,6.

There is abundant evidence that body fat distribution is influenced by genetic loci distinct from those regulating BMI and overall adiposity. First, even after accounting for BMI, individual variation in WHR is heritable^7,8, with heritability estimates ranging from 22%–61%^7-10. Second, the striking abnormalities of regional fat deposition associated with lipodystrophic syndromes demonstrate that genetic variation can have dramatic effects on the development and maintenance of specific fat depots^11,12. Third, in a previous genome-wide association analysis, we identified a locus near LYPLAL1 strongly associated with WHR independent of any effects on BMI¹³, providing proof of principle for the genetic control of body fat distribution distinct from that of overall adiposity.

Within the Genetic Investigation of Anthropometric Traits (GIANT) consortium, we performed a large-scale meta-analysis of genome-wide association studies (GWAS) informative for WHR using adjustment for BMI to focus discovery toward genetic loci associated with body fat distribution rather than overall adiposity^14-16.

RESULTS

Genome-wide significant association of WHR with 14 SNPs

We conducted a two-stage study among individuals of European descent (Supplementary Table 1 and Online Methods). In the discovery stage, up to 2,850,269 imputed and genotyped SNPs were examined in 32 GWAS comprising up to 77,167 participants informative for anthropometric measures of body fat distribution. We performed a fixed-effects meta-analysis of WHR, employing study-specific linear regression adjusted for BMI and age, stratified by gender, and using an additive genetic model. After genomic control adjustment per each individual study and in the meta-analysis, these analyses revealed a substantial excess of low P values (Fig. 1a,b).

Figure 1.

Genome-wide association analyses for WHR in discovery studies. (a) Manhattan plot shows results of the WHR association meta-analysis in discovery studies (with P values on the y axis and the SNP genomic position on the x axis). Colored genomic loci indicate significant association (P < 5 × 10⁻⁸) detected previously (blue)¹³, in our GWAS stage (red) and after the meta-analysis combining GWAS data with that from the follow-up studies (orange). Two loci tested in the follow-up stage did not achieve genome-wide significance (green). (b) Quantile-quantile plot of SNPs for the discovery meta-analysis of WHR (black) and after removing SNPs within 1 Mb of either the recently reported LYPLAL1 signal (blue) or the 14 significant associations (green). The gray area represents the 95% CI around the test statistic under the null distribution.

We selected SNPs representing the top 16 independent (defined as being located >1 Mb apart) regions of association (discovery P < 1.4 × 10⁻⁶; Table 1) and evaluated them in 29 additional, independent studies (comprising up to 113,636 individuals) using a mixture of in silico data and de novo genotyping. In these follow-up studies, 14 of the 16 SNPs analyzed showed strong directionally consistent evidence for replication (P < 1.0 × 10⁻³) and ten SNPs reached genome-wide significance (P < 5.0 × 10⁻⁸). Joint analysis of the discovery and follow-up results revealed genome-wide significant associations for 14 signals (with P values between 1.9 × 10⁻⁹ and 1.8 × 10⁻⁴⁰; Table 1). Between-study heterogeneity was low (I² < 30%) for all but two signals (GRB14 and LYPLAL1; Supplementary Note), and all 14 associations remained genome-wide significant in a random-effects meta-analysis (Supplementary Table 2).

Table 1.

Fourteen sNPs associated with WHr at genome-wide significant levels

						Discovery			Follow-up			Combined
SNP	Chr.	Position (b36)	Nearby genes	EAa	EAFb	P	β	n	P	β	n	P	β
SNPs evaluated in follow up achieving genome-wide significance
rs9491696	6	127,494,332	RSPO3	G	0.520	2.10 × 10−14	0.037	77,164	3.27 × 10−28	0.045	113,582	1.84 × 10−40	0.042
rs6905288	6	43,866,851	VEGFA	A	0.562	4.72 × 10−10	0.033	77,129	1.18 × 10−16	0.039	95,430	5.88 × 10−25	0.036
rs984222	1	119,305,366	TBX15-WARS2	G	0.365	3.81 × 10−14	0.037	77,167	1.56 × 10−12	0.031	109,623	8.69 × 10−25	0.034
rs1055144	7	25,837,634	NFE2L3	T	0.210	1.49 × 10−8	0.034	77,145	3.26 × 10−18	0.043	113,636	9.97 × 10−25	0.040
rs10195252	2	165,221,337	GRB14	T	0.599	3.23 × 10−10	0.031	77,119	3.18 × 10−16	0.036	102,449	2.09 × 10−24	0.033
rs4846567	1	217,817,340	LYPLAL1	G	0.283	2.37 × 10−12	0.037	77,167	3.15 × 10−10	0.032	91,820	6.89 × 10−21	0.034
rs1011731	1	170,613,171	DNM3-PIGC	G	0.572	1.72 × 10−10	0.031	77,094	7.47 × 10−9	0.026	92,018	9.51 × 10−18	0.028
rs718314	12	26,344,550	ITPR2-SSPN	G	0.741	2.41 × 10−8	0.031	77,167	1.49 × 10−10	0.030	107,503	1.14 × 10−17	0.030
rs1294421	6	6,688,148	LY86	G	0.387	6.31 × 10−9	0.029	77,154	2.69 × 10−10	0.028	102,189	1.75 × 10−17	0.028
rs1443512	12	52,628,951	HOXC13	A	0.239	3.33 × 10−8	0.031	77,165	2.92 × 10−10	0.030	112,353	6.38 × 10−17	0.031
rs6795735	3	64,680,405	ADAMTS9	C	0.406	2.47 × 10−7	0.025	77,162	6.75 × 10−8	0.026	84,480	9.79 × 10−14	0.025
rs4823006	22	27,781,671	ZNRF3-KREMEN1	A	0.569	4.47 × 10−8	0.027	77,086	2.41 × 10−5	0.019	93,911	1.10 × 10−11	0.023
rs6784615	3	52,481,466	NISCH-STAB1	T	0.941	3.18 × 10−7	0.052	76,859	1.56 × 10−4	0.036	109,028	3.84 × 10−10	0.043
rs6861681	5	173,295,064	CPEB4	A	0.340	1.40 × 10−6	0.026	77,164	2.13 × 10−4	0.019	85,722	1.91 × 10−9	0.022
Further sNPs evaluated in follow up but not achieving genome-wide significance in the combined analysis
rs2076529	6	32,471,933	BTNL2	C	0.570	2.22 × 10−8	0.041	34,532	0.012	0.011	92,778	3.71 × 10−7	0.020
rs7081678	10	32,030,629	ZEB1	A	0.085	5.76 × 10−7	0.045	76,270	0.094	0.013	100,527	5.57 × 10−6	0.027

P values and β coefficients (per change of WHR-increasing allele) for the association with WHR on the inverse normal transformed ranked scale in the meta-analyses of discovery studies (up to 77,167 subjects), follow-up studies (up to 113,636 subjects) and both combined (up to 190,781 subjects). Fourteen of the sixteen SNPs examined in the follow-up samples showed genome-wide significant results (P < 5 × 10⁻⁸) in the combined analysis. P values in the discovery stage were genomic control corrected per study and in the meta-analysis. Details on between-study heterogeneity are given in Supplementary Table 1c.

^aEA, effect allele (WHR-increasing allele on the forward strand).

^bEAF, effect allele frequency. Chr., chromosome.

One of these SNPs, rs4846567, is in linkage disequilibrium (LD) (r² = 0.64, D′ = 0.84; HapMap European CEU population) with the previously reported WHR-associated variant near LYPLAL1 (rs2605100)¹³. The remaining 13 loci were in or near genes not previously associated with WHR or other measures of adiposity: RSPO3, VEGFA, TBX15-WARS2, NFE2L3, GRB14, DNM3-PIGC, ITPR2-SSPN, LY86, HOXC13, ADAMTS9, ZNRF3-KREMEN1, NISCH-STAB1 and CPEB4 (Fig. 2). These 14 loci explain 1.03% of the variance in WHR (after adjustment for BMI, age and sex), with each locus contributing from 0.02% (ZNRF3-KREMEN1) to 0.14% (RSPO3) of the variance based on effect estimates in the follow-up stage.

Figure 2.

Regional plots of 14 loci with genome-wide significant association. Shown is the SNP association with WHR in the meta-analysis of discovery studies for 14 loci (with −log₁₀ P values on the y axis and the SNP genomic position on the x axis). In each panel, an index SNP is denoted with a purple diamond and plotted using the P attained across discovery and follow-up data (Table 1). Estimated recombination rates are plotted in blue. SNPs are colored to reflect LD with the index SNP (pairwise r² values from HapMap CEU). Gene and microRNA annotations are from the UCSC genome browser.

Sexual dimorphism at several of the WHR loci

Given the known sexual dimorphism of WHR and the evidence from variance decomposition studies that this reflects sex-specific genetic effects¹⁷, we performed sex-specific meta-analyses for the 14 WHR-associated SNPs. These analyses included up to 108,979 women (42,735 in the discovery stage and 66,244 in the follow up) and 82,483 men (34,601 in the discovery and 47,882 in the follow up). In a joint analysis of discovery and follow-up data, 12 of the 14 SNPs reached genome-wide significance in women, but only three SNPs reached genome-wide significance in men (Table 2). At all but one locus (TBX15-WARS2), effect-size estimates were numerically greater in women. At seven of the loci (those near RSPO3, VEGFA, GRB14, LYPLAL1, HOXC13, ITPR2-SSPN and ADAMTS9), there were marked differences in sex-specific β coefficients (with P values ranging from 1.9 × 10⁻³ to 1.2 × 10⁻¹³). All loci displayed consistent patterns of sex-specific differences in both the discovery and follow-up studies (Table 2). These 14 loci explain 1.34% of the variance in WHR (after adjustment for BMI and age) in women but only 0.46% of the variance in WHR in men.

Table 2.

Evidence of sex-differences in the WHr association at seven of the 14 associated loci

		Men						Women						Sex difference
SNP	Nearby genes	Discovery		Follow up		Combined		Discovery		Follow up		Combined		Combined
		P	β	P	β	P	β	P	β	P	β	P	β	P
rs9491696	RSPO3	1.68 × 10−4	0.026	6.97 × 10−9	0.036	1.05 × 10−11	0.031	1.62 × 10−12	0.047	8.84 × 10−22	0.053	1.93 × 10−32	0.050	1.94 × 10−3
rs6905288	VEGFA	0.066	0.013	2.09 × 10−4	0.025	7.38 × 10−5	0.020	7.72 × 10−13	0.052	3.14 × 10−15	0.051	2.27 × 10−26	0.052	5.20 × 10−6
rs984222	TBX15- WARS2	3.32 × 10−9	0.041	2.43 × 10−5	0.029	9.41 × 10−13	0.035	1.21 × 10−7	0.036	1.33 × 10−8	0.033	1.02 × 10−14	0.034	0.951
rs1055144	NFE2L3	6.00 × 10−4	0.029	5.67 × 10−8	0.040	2.52 × 10−10	0.035	2.34 × 10−6	0.040	7.13 × 10−12	0.046	1.41 × 10−16	0.044	0.270
rs10195252	GRB14	0.201	0.009	0.114	0.011	0.043	0.010	6.33 × 10−15	0.053	4.95 × 10−21	0.054	3.84 × 10−34	0.054	1.41 × 10−11
rs4846567	LYPLAL1	0.191	0.010	0.982	0.000	0.358	0.005	4.84 × 10−18	0.064	8.12 × 10−17	0.055	4.95 × 10−33	0.059	1.18 × 10−13
rs1011731	DNM3- PIGC	4.88 × 10−7	0.034	1.95 × 10−3	0.022	7.81 × 10−9	0.028	2.13 × 10−5	0.028	7.03 × 10−7	0.030	6.90 × 10−11	0.029	0.855
rs718314	ITPR2- SSPN	0.177	0.010	2.02 × 10−3	0.022	1.41 × 10−3	0.017	8.29 × 10−10	0.047	4.21 × 10−9	0.038	2.41 × 10−17	0.042	4.67 × 10−4
rs1294421	LY86	4.18 × 10−3	0.020	7.00 × 10−6	0.030	1.63 × 10−7	0.025	3.44 × 10−8	0.038	7.32 × 10−6	0.026	2.40 × 10−12	0.031	0.357
rs1443512	HOXC13	0.184	0.011	9.74 × 10−4	0.024	9.45 × 10−4	0.018	1.43 × 10−9	0.048	3.09 × 10−8	0.035	6.38 × 10−16	0.040	2.23 × 10−3
rs6795735	AD- AMTS9	0.011	0.017	0.614	0.004	0.027	0.011	7.85 × 10−7	0.033	2.95 × 10−11	0.042	1.92 × 10−16	0.038	8.50 × 10−5
rs4823006	ZNRF3- KRE MEN1	6.87 × 10−3	0.019	0.094	0.012	1.94 × 10−3	0.015	6.86 × 10−8	0.037	3.81 × 10−5	0.024	3.24 × 10−11	0.030	0.032
rs6784615	NISCH- STAB1	1.51 × 10−3	0.045	0.033	0.032	1.68 × 10−4	0.039	6.23 × 10−5	0.057	1.72 × 10−3	0.039	6.01 × 10−7	0.047	0.574
rs6861681	CPEB4	1.88 × 10−3	0.023	0.045	0.015	3.03 × 10−4	0.019	2.14 × 10−4	0.027	1.58 × 10−3	0.021	1.55 × 10−6	0.024	0.555

P values and β coefficients (per change of WHR-increasing allele in the sex-combined analysis as in Table 1) for the WHR association are given for the discovery (up to 34,601 men and 42,735 women), the follow-up (up to 47,882 men and 65,780 women) and the combined meta-analysis (up to 81,301 men and 107,429 women). Also given are the P values for testing for difference between sex-specific β coefficients in the combined meta-analysis; SNPs with P for sex difference < 3.6 × 10⁻³ (0.05/14) were considered to show a significant sex difference.

Association with other anthropometric measures

By focusing on WHR after adjustment for BMI, our goal was to detect effects on body fat distribution independent of those influencing overall adiposity. As expected, we found very little evidence that known BMI-associated variants were detected in our WHR analysis. Of the ten loci shown to be associated with BMI in previous GWAS^14,15,18, only two showed nominally significant (P < 0.05) associations for BMI-adjusted WHR in the discovery analysis (FTO, rs8050136, P = 0.03, n = 77,074; TMEM18, rs6548238, P = 3.0 × 10⁻³, n = 77,016).

We also tested the 14 WHR-associated SNPs for their effect on BMI using data from up to 242,530 participants available from the GIANT consortium (including most of the studies available for WHR association). Of the 14 WHR loci, four (near TBX15-WARS2, CPEB4, LYPLAL1 and GRB14) also showed evidence of association with BMI (4.1 × 10⁻³ ≤ P ≤ 3.2 × 10⁻⁶), with the WHR-increasing allele associated with decreased BMI (Supplementary Table 3). After adding an interaction term of SNP with BMI into the model, we observed that BMI modified the WHR association at the LY86 locus (P for interaction = 9.5 × 10⁻⁵), with a larger WHR effect among obese individuals compared to non-obese individuals (Supplementary Note).

To determine whether the WHR-associated signals exert their effects primarily through an effect on waist or hip circumference, we performed meta-analyses for these specific phenotypes in the discovery and follow-up studies (Supplementary Tables 1 and 3). Overall, we observed stronger associations for hip circumference than for waist circumference. Effect-size estimates were numerically greater for hip circumference than for waist circumference at 11 of the 14 loci, and there were nominal associations (P < 0.05) with hip circumference for 12 of the WHR-associated loci but there were only four associations with waist circumference. In both sexes, the WHR-associated loci displaying nominal association with hip circumference always featured the WHR-increasing allele associated with reduced hip circumference. In contrast, we observed sexual dimorphism in the pattern of waist circumference associations. In women, the WHR-increasing allele at all 14 loci was associated with increased waist circumference, whereas this was only true for six of these loci in men (Fig. 3). At GRB14, for example, the WHR-increasing allele was associated with increased waist circumference in women (P = 3.6 × 10⁻⁴) but with decreased waist circumference in men (P = 6.8 × 10⁻³). These differences in the relationships between waist circumference, hip circumference and WHR underlie some of the sexual dimorphism in the patterns of WHR association.

Figure 3.

Association of the 14 WHR loci with waist and hip circumference. β coefficients for waist circumference (WC, x axis) and hip circumference (HIP, y axis) in women and men derived from the joint discovery and follow-up analysis. P for WC and HIP are represented by color. In men, gray gene labels refer to those SNPs that were not significant in the male-specific WHR analysis. More details can be found in Supplementary Table 3.

Enrichment of association with metabolic traits

We evaluated the 14 WHR-associated loci for their relationships with related metabolic traits using GWAS data provided by trait-specific consortia^19-21 as well as our de novo genotyped follow-up studies. As expected given the sample overlap between this GWAS data and our WHR GWAS data as well as information on known trait correlations (Supplementary Table 4), we observed directionally consistent enrichment of associations (P < 0.05) between the 14 WHR-associated alleles and increased triglycerides, low-density lipoprotein (LDL) cholesterol, fasting insulin and homeostasis model assessment (HOMA)-derived measures of insulin resistance (binomial P from 3.2 × 10⁻⁴ to 1.8 × 10⁻⁸; Table 3 and Supplementary Table 5). For example, the WHR-increasing allele at GRB14 showed strong associations with increased triglycerides (P = 7.4 × 10⁻⁹), fasting insulin levels (P = 5.0 × 10⁻⁶) and insulin resistance (P = 1.9 × 10⁻⁶). Eleven of the 14 WHR-associated loci showed directionally consistent associations with T2D, with three of these loci (at ADAMTS9, NISCH-STAB1 and ITPR2-SSPN) reaching nominal significance (P < 0.05) (Table 3 and Supplementary Table 5). Because the association signals for correlated traits in this analysis were vulnerable to overestimation given the overlap in the GWAS samples examined, we repeated these analyses and restricted the samples included to those from our de novo genotyped follow-up studies. Although this also resulted in a lower sample size, similar patterns of enrichment were still observed (Supplementary Table 5).

Table 3.

WHR signals show enrichment of association with other traits related to metabolic disorders

Trait	Sample sizea	SNPs in concordant directionb		SNPs in concordant direction with P < 0.05c
		n	P	n	P
Triglycerides	43,826	14	6.10 × 10−5	7	1.79 × 10−8
HDL-C	45,561	13	9.16 × 10−4	4	3.20 × 10−4
LDL-C	43,889	10	0.090	1	0.298
Fasting glucose	63,849	10	0.090	1	0.298
Fasting insulin	54,883	13	9.16 × 10−4	5	1.62 × 10−5
HOMA-IR	53,625	13	9.16 × 10−4	6	6.17 × 10−7
2 h glucose	27,011	7	0.605	0	1.000
Type 2 diabetes	10,128d	11	0.029	3	4.62 × 10−3

The 14 WHR SNPs were tested for association with other traits by meta-analysis of GWAS data from previous reports¹⁹–²¹,³⁹ together with our non-overlapping de novo genotyped follow-up studies. HDL-C, high density lipoprotein cholesterol; LDL-C, low density lipoprotein cholesterol; HOMA-IR, index of insulin resistance; 2 h glucose, glucose levels 2 h after an oral glucose challenge.

^aMaximum number of subjects available for any of the 14 SNPs.

^bNumber of the 14 SNPs for which the WHR-increasing allele is associated with the trait in the concordant direction (that is, increased levels, except for HDL-C) and corresponding binomial P value to test whether this number is greater than that expected by chance and not accounting for the correlation between the traits.

^cNumber of SNPs in concordant direction that show P < 0.05 for the association with the trait and the corresponding binomial P value as in ^b.

^d4,549 cases, 5579 controls.

Pathway analysis and potential biological roles

To identify potential functional connections and pathway relationships between genes mapping at the WHR-associated loci, we focused on the 95 genes located in a 2-Mb interval centered around each of the 48 independent SNPs that attained P < 1.0 × 10⁻⁵ in the WHR discovery studies.

First, we performed a survey of the published literature using GRAIL²² to search for connectivity between the genes and specific keywords that describe these functional connections (Online Methods). Although there was no evidence after correcting for multiple testing that the connectivity between these genes was greater than chance, we identified eight genes with nominal significance (P < 0.05) for potential functional connectivity (PLXND, HOXC10, TBX15, RSPO3, HOXC4, HOXC6, KREMEN1 and HOXC11). The keywords associated with these connections included ‘vegf’, ‘homeobox’, ‘patterning’, ‘mesenchyme’, ‘embryonic’, ‘development’ and ‘angiogenesis’.

Additionally, we performed pathway analyses using the PANTHER database²³ based on the same set of 95 genes (Online Methods and Supplementary Note). This analysis generated some evidence for over-representation of ‘developmental processes’ (P = 5.8 × 10⁻⁸) and ‘mRNA transcription regulation’ (P = 2.7 × 10⁻⁶) but neither of these factors retained nominal significance after adjustment for bias (for example, due to non-random SNP coverage in relation to genes) and the number of biological processes tested (Supplementary Note and Supplementary Table 6).

Finally, we examined the described functional roles of some of the most compelling candidates based on either proximity to the signal or the other analyses described in this paper. These analyses uncovered possible genetic roles in adipocyte development (TBX15), pattern formation during embryonic development (HOXC13), angiogenesis (VEGFA, RSPO3 and STAB1), Wnt and β-catenin signaling (RSPO3 and KREMEN1), insulin signaling (ADAMTS9, GRB14 and NISCH), lipase activity (LYPLAL1), lipid biosynthesis (PIGC) and intracellular calcium signaling (ITPR2) (Supplementary Note).

Evaluation of copy number variants and non-synonymous changes

Both common and rare copy number variants (CNVs) have been reported to be associated with overall adiposity^14,15,24,25, but the impact of CNVs on fat distribution has not been evaluated previously. To examine the potential contribution of common CNVs to variation in WHR, we looked for evidence of association in our genome-wide association discovery meta-analysis using a set of 6,018 CNV-tagging SNPs which collectively capture >40% of common CNVs that are greater than 1 kb in length^26,27 (Online Methods and Supplementary Note).

One CNV-tagging SNP (rs1294421 in LY86) was observed among our 14 WHR-associated loci. This SNP is in strong LD (r² = 0.98) with a 2,832-bp duplication variant (CNVR2760.1)²⁷ located 12 kb from an expressed sequence tag (BC039678) and 87 kb from LY86 such that the duplication allele is associated with reduced WHR. The duplicated region consists entirely of noncoding sequence but includes part of a predicted enhancer sequence (E.5552.1)²⁸.

To identify other putatively causal variants in our associated regions, we searched for non-synonymous coding SNPs in strong LD (defined as r² > 0.7) with the most strongly associated SNPs at each locus using data from the HapMap (Build 21) and 1000 Genomes Project (April and August 2009 releases). In this search, one lead SNP (rs6784615, at the NISCH-STAB1 locus) was correlated with non-synonymous changes in two nearby genes, DNAH1 (p.Val441Leu, p.Arg1285Trp and p.Arg3809Cys) and GLYCTK (p.Leu170Val). Fine-mapping and functional studies will be required to determine whether the DNAH1 or GLYCTK SNPs or the LY86 CNV are causal for the WHR associations at these loci.

Effect of WHR associations on expression in relevant tissues

Expression quantitative trait locus (eQTL) data can implicate regional transcripts that mediate trait associations, and we therefore examined the 14 WHR-associated loci using eQTL data from human subcutaneous adipose tissue (SAT)²⁹ (two separate sample sets, n = 610 and n = 603), omental fat³⁰ (n = 740), liver³⁰ (n = 518), blood²⁹ (n = 745) and lymphocytes³¹ (n = 830) (Online Methods and Supplementary Note).

At six of the loci, the WHR-associated SNP was either the strongest SNP associated with significant (P < 1.0 × 10⁻⁵) expression of a local (within 1 Mb) gene transcript or explained the majority of the association between the most significant eQTL SNP and the gene transcript in conditional analyses (adjusted P > 0.05; Table 4). For example, the WHR-associated SNP rs1011731 (near DNM3-PIGC) was strongly associated with expression of PIGC in lymphocytes (P = 5.9 × 10⁻¹⁰); furthermore, rs1011731 is in high LD (r² = 1.00, D′ = 1.00 from the HapMap CEU population) with the SNP with the strongest effect on PIGC expression (rs991790), and this cis eQTL association was abolished by conditioning on rs1011731. These analyses therefore indicate that these two signals are coincident and that PIGC is a strong candidate for mediating the WHR association at rs1011731. We found similar evidence for coincidence of the WHR signal with expression for rs984222 (TBX15 in omental fat), rs1055144 (expressed sequence tag AA553656 in SAT), rs10195252 (GRB14 in SAT), rs4823006 (ZNRF3 in SAT and omental fat) and rs6784615 (STAB1 in blood) (Table 4). Taken together, the overlap between trait association and gene expression at these loci suggests that the WHR associations may be driven through altered expression of PIGC, TBX15, AA553656, GRB1, ZNRF3 and STAB1.

Table 4.

Expression quantitative trait locus analysis for 11 of the 14 WHR signals

				WHR SNP association with transcript (P)				Peak SNP association with transcript (P)
WHR SNP	Tissue	Gene	Effecta	Unadj.	Adj. for peak SNP	Transcript peak SNPb	LD (r2)c	Unadj.	Adj. for WHR SNP
rs9491696	SAT-D	RSPO3	+	1.10 × 10−7	0.03	rs1936795	0.26	2.20 × 10−13	7.40 × 10−8
rs984222	Omental	TBX15	+	7.90 × 10−10	1.00	rs984222	1.00	7.90 × 10−10	1.00
	Omental	WARS2	+	5.11 × 10−36	0.03	rs10802075	0.27	1.31 × 10−163	1.33 × 10−88
	Subcutaneous fat	WARS2	+	1.67 × 10−25	0.01	rs10802075	0.22	3.88 × 10−110	1.01 × 10−63
	Lymphocytes	WARS2	−	4.30 × 10−18	5.47 × 10−5	rs2645305	0.27	5.57 × 10−40	6.88 × 10−26
	Liver	WARS2	+	2.57 × 10−17	0.07	rs1057990	0.26	6.69 × 10−59	1.97 × 10−32
	SAT-D	WARS2	+	1.10 × 10−18	0.51	rs1057990	0.26	5.80 × 10−130	5.80 × 10−100
	Blood	WARS2	+	6.10 × 10−17	0.11	rs1057990	0.26	6.30 × 10−75	1.10 × 10−54
rs1055144	SAT-D	AA553656 d	−	1.20 × 10−11	0.96	rs7798002	0.95	7.20 × 10−12	0.32
	SAT-M	AA553656 d	−	2.46 × 10−7	0.65	rs1451385	0.77	5.93 × 10−8	0.38
rs10195252	SAT-D	GRB14	+	4.40 × 10−11	1.00	rs10195252	1.00	4.40 × 10−11	1.00
	SAT-M	GRB14	+	5.51 × 10−6	1.00	rs10184004	1.00	5.51 × 10−6	1.00
	Omental	GRB14	+	1.02 × 10−13	1.00	rs10195252	1.00	1.02 × 10−13	1.00
	SAT-M	SLC38A11	−	3.93 × 10−6	0.66	rs10184126	0.18	7.76 × 10−44	8.57 × 10−34
	SAT-D	SLC38A11	−	3.70 × 10−9	0.35	rs10184126	0.18	2.40 × 10−94	7.40 × 10−82
rs1011731	Blood	C1orf105	+	3.80 × 10−16	0.20	rs2157451	0.28	1.30 × 10−33	8.20 × 10−18
	Lymphocytes	PIGC	−	5.87 × 10−10	1.00	rs991790	1.00	5.65 × 10−10	1.00
rs718314	Lymphocytes	ITPR2	+	1.79 × 10−9	0.98	rs7976877	0.45	2.21 × 10−18	1.91 × 10−6
	Blood	ITPR2	−	2.40 × 10−9	0.20	rs2570	0.41	2.40 × 10−37	1.80 × 10−28
rs1294421	SAT-M	BC039678	−	2.43 × 10−7	0.38	rs1294404	0.64	1.89 × 10−16	3.42 × 10−4
	Omental	BC039678	−	1.09 × 10−6	0.33	rs912056	0.71	8.28 × 10−17	4.26 × 10−5
rs6795735	SAT-D	ADAMTS9	−	1.50 × 10−6	0.04	rs7372321	0.11	1.10 × 10−9	2.30 × 10−5
	Omental	AK022320	−	7.99 × 10−15	0.64	rs4521216	0.02	5.15 × 10−42	1.49 × 10−19
	SAT-D	AK022320	−	2.24 × 10−10	0.98	rs4521216	0.02	9.62 × 10−37	7.58 × 10−19
rs4823006	SAT-D	ZNRF3	−	2.40 × 10−8	0.63	rs3178915	0.81	6.70 × 10−11	8.90 × 10−4
	SAT-M	ZNRF3	−	1.08 × 10−18	0.93	rs6005975	0.79	1.59 × 10−19	0.50
	Omental	ZNRF3	−	9.13 × 10−18	0.98	rs6005975	0.79	6.07 × 10−21	0.27
rs6784615	Blood	STAB1	+	2.80 × 10−9	0.32	rs9846089	0.83	9.40 × 10−10	0.08
rs6861681	Lymphocytes	CPEB4	+	3.79 × 10−22	0.89	rs7705502	0.87	4.95 × 10−29	2.00 × 10−3
	Blood	HMP19	+	1.60 × 10−16	0.97	rs10516107	0.83	1.10 × 10−21	4.30 × 10−6

Association between the 14 WHR SNPs and expression of transcripts located within 1 Mb of the WHR SNP in two sets of abdominal subcutaneous adipose tissue (SAT-D from deCODE and SAT-M from Massachusetts General Hospital), omental fat, liver, lymphocytes and blood (Supplementary Note). Results are given if the unadjusted WHR SNP association showed P < 1.00 × 10⁻⁵. Findings are highlighted in bold font where the WHR SNP was the transcript peak SNP or where the WHR signal and the cis eQTL signal were considered coincident (that is, the transcript peak SNP was highly correlated with the WHR SNP, r² > 0.7 and the transcript peak association disappeared by adjusting on the WHR SNP, P > 0.05); see also Online Methods. Unadj., unadjusted; Adj., adjusted.

^aEffect direction for the WHR-increasing allele.

^bSNP with the strongest association with the transcript in the region (transcript peak SNP).

^cCorrelation (HapMap CEU, build 36) between the WHR SNP and the transcript peak SNP.

^dThe transcript labeled AA553656 was detected as Contig27623_RC and corresponds to chromosome 7 locations 25,854,143–25,854,203 (HapMap build 36).

RNA expression of gluteal and abdominal fat tissue

To determine whether genes within the WHR-associated loci showed evidence of differential transcription in distinct fat depots, we compared expression levels in gluteal or abdominal SAT in 49 individuals. We focused on the 15 genes with the strongest credentials for causal involvement (on the basis of proximity to the lead SNP and/or other biological or functional data; Table 1) for which expression data were available. Five of these genes (RSPO3, TBX15, ITPR2, WARS2 and STAB1) were differentially expressed between the two tissues (using an F test, corrected for false discovery rate across the 15 expressed genes, P < 0.05; Supplementary Table 7). This supports the hypothesis that, at some loci at least, the association with WHR reflects depot-specific differences in expression patterns.

DISCUSSION

Overall, our findings demonstrate that the genetic regulation of body fat distribution involves loci and processes that are largely distinct from those that influence BMI and risk of obesity. This finding is consistent with the evidence that WHR displays substantial heritability even after adjustment for BMI. The loci that emerged from this study display no overlap with those shown to be associated with BMI either in previous reports^14-16 or in the expanded meta-analysis recently completed by the GIANT consortium³².

Another point of distinction between our findings and those for BMI relates to the evidence for sexual dimorphism that we observed at several of the WHR-associated loci. Sex differences in the regulation of body fat distribution have long been acknowledged without a clear understanding of the underlying molecular mechanisms. These differences become apparent during puberty and are generally attributed to the influence of sex hormones³³. Consistent with our findings, variance decomposition studies have shown that the genetic contribution to the overall variance in WHR, waist and hip circumference is greater in women¹⁷. Although there is some evidence for loci with differential sex effects influencing lipids³⁴, uric acid levels³⁵ and risk of schizophrenia³⁶, we are unaware of prior reports indicating such strong enrichment of female-specific associations for any other phenotype, including BMI³².

The primary objective of genetic discovery efforts is to characterize the specific mechanisms involved in regulating the trait of interest. Despite the considerable challenges associated with moving from common variant association signals to defining causal alleles and pathways, we have identified strong candidates at several of the loci. For example, the cis eQTL data implicate GRB14 as a compelling candidate for the WHR association on chromosome 2, and we were able to show that the same GRB14 variants are also associated with triglyceride and insulin levels, consistent with previous association of this locus with high-density lipoprotein (HDL) cholesterol³⁷. These inferences about the role of GRB14 are supported by evidence that Grb14-deficient mice exhibit improved glucose homeostasis despite lower circulating insulin levels, as well as enhanced insulin signaling in liver and skeletal muscle³⁸. The signal near ADAMTS9 overlaps a previously-reported T2D locus³⁹, and the lead SNP for WHR in our study is identical to the SNP displaying the strongest T2D association in a previous expanded T2D meta-analysis⁴⁰. Given evidence that ADAMTS9 T2D risk alleles are associated with insulin resistance in peripheral tissues⁴¹, these findings are consistent with a primary effect of ADAMTS9 variants on body fat distribution. At the chromosome 6 locus, VEGFA is the most apparent biological candidate given the presumed role of VEGFA as a mediator of adipogenesis⁴² and evidence that serum levels of VEGFA are correlated with obesity^43,44. Finally, at the TBX15-WARS2 locus, TBX15 emerges as the strongest candidate based on the cis eQTL data in omental fat, marked depot-specific differences in adipose tissue expression in mice and humans and associations between TBX15 expression in visceral fat and WHR^45,46.

Our efforts to use pathway- and literature-mining approaches to look for functional enrichment of the genes mapping to associated regions met with only limited success but did provide some support for over-representation of developmental processes. Developmental genes have been implicated in fat accumulation and distribution^45,46, and recent evidence supports a link between developmental genes, including HOXC13 (ref. ⁴⁷) and TBX15 (refs. ^45,48), and body fat distribution. Developmental genes may in part determine the adipocyte-specific expression patterns that have been observed in different fat depots⁴⁵. Taken together, our findings point to a set of genes influencing body fat distribution that have their principal effects in adipose tissue. This is in contrast to the predominantly central (hypothalamic) processes that are involved in the regulation of BMI and overall adiposity⁴⁹.

By providing new insights into the regulation of body fat distribution, the present study raises a number of issues for future investigation. From the genetic perspective, re-sequencing, dense-array genotyping and fine-mapping approaches will be required to characterize causal variants at the loci we have identified and to support further discoveries that may account for the substantial proportion of genetic variance unexplained by our findings. From the clinical perspective, it will be important to explore the relationship of these variants to more refined measures of body fat distribution derived from detailed imaging studies, to use the variants identified to characterize the causal relationships between body fat distribution and related metabolic and cardiovascular traits and to explore population differences in patterns of body fat distribution. Efforts to tackle overall obesity through therapeutic or lifestyle-based modulation of overall energy balance have proved extremely challenging to implement, and the manipulation of processes associated with more beneficial patterns of fat distribution offers an alternative perspective for future drug discovery.

METHODS

Methods and any associated references are available in the online version of the paper at http://www.nature.com/naturegenetics/.