Genome-wide association study of body fat distribution identifies novel loci and sex-specific effects

Body mass and composition are complex traits of clinical interest due to their links to cardiovascular- and metabolic diseases. In this study, we performed genome-wide association studies (GWAS) for the distribution of body fat to the arms, legs and trunk. Proportions of fat, distributed to the different compartments, were calculated for 362,499 individuals from the UK biobank, based on segmental bioimpedance analysis (sBIA) estimates. A total of 85 body fat distribution loci were identified, using data from 116,138 participants, and replicated in an independent set of participants (N = 246,361). Out of these loci, 28 were associated with the proportion of fat in the arms, 43 with the legs and 57 with the trunk. A large degree of overlap was observed between legs and trunk loci (N=33), while arm loci overlapped to a smaller degree with leg and trunk loci (N=4 and 6, respectively). As many as 50 of the loci have not previously been associated with any adiposity-related trait. Within the novel loci we found lipid metabolism-related genes such as CILP2 and OSBPL7, as well as androgen receptor function-related genes such as ESR1, ID4 and ADAMTS17. Significant interactions between the top SNP and sex were observed for 38 loci. Our findings provide evidence for multiple loci that affect the distribution of body fat to discrete compartments of the human body, and highlight that genetic effects differ between men and women, in particular for distribution of body fat to the legs and trunk.

Developments in BIA technology has now allowed for cost-efficient segmental body composition scans that estimate of the fat content of the trunk, arms and legs, with high accuracy [19]. In this study, we utilize segmental BIA data on 362,499 participants of the UK Biobank to study the genetic determinants of body fat distribution to the trunk, arms and legs. For this purpose, we performed GWAS on the proportion of body fat distributed to these compartments. We also performed sexstratified analyses to determine sex-specific effects and interactions.

Results
We conducted a two-stage GWAS using data from the interim release of genotype data in UK Biobank as a discovery cohort. Another set of participants for which genotype data were made available as part of the second release was used for replication. After removing non-Caucasians, genetic outliers and related individuals, 116,138 and 246,361 participants remained in the discovery and replication cohort respectively. The proportions of body fat distributed to the arms -arm fat ratio (AFR), the legs -leg fat ratio (LFR) and the trunk -trunk fat ratio (TFR) were calculated by dividing the fat mass per compartment with the total body fat mass in each participant.
Women were found to have higher total fat mass compared to men (p < 2.2*10 -16 , Table 1), as well as higher amount of fat in the arms and legs (p < 2.2*10 -16 , Table 1).
Males had on average about a 12% higher proportion of their body fat located in the trunk compared with females (62.2% vs. 50.3%, p < 2.2*10 -16 , Table 1), while women had approximately 12% higher proportion of body fat located in the legs (39.7% vs. 28.1%, p < 2.2*10 -16 , Table 1). While the amount of adipose tissue in the arms was estimated to be higher in women compared to men, the relative amounts of fat in the arms were more similar. Only marginal differences were seen in basic characteristics between the discovery and replication cohorts (Table 1).

Table 1. General descriptives of UK Biobank participants * included in the analyses.
Data is presented for the discovery and replication cohorts. The cohorts were filtered for unrelated Caucasians. Mean values are presented ± standard deviations. Interestingly, we did not find a strong correlation between our ratios and other anthropometric traits. A substantial part of the variation in AFR could be explained by BMI or waist circumference (Table 2). However, anthropomorphic traits did not substantially contribute to explaining the variance in LFR or TFR (Table 2).

Discovery GWAS for body fat distribution
In the discovery GWAS, each of the three phenotypes (AFR, LFR and TFR) were analyzed in the whole discovery cohort (sex-combined) and by stratifying by sex (males and females). We first estimated the proportion of variance in body fat distribution to the arms, legs and trunk that were explained by the genotyped SNPs (N=730,616) [20]. We find that approximately 20-24% of the variance can be explained by considering all genotyped SNPs simultaneously in the sex-combined cohort. Interestingly, we also find that a higher degree of variance could be explained by genotyped SNPs in females compared to males (40-42% vs. 18-27%), and that this was consistent across all phenotypes (Table 3). : SNP heritability estimates from GCTA which represents the fraction of the phenotypic variance that can be explained by the SNPs analyzed.
A total of 25,472,837 imputed SNPs with MAF of at least 0.01% were analyzed in the discovery GWAS. A total of 11,350 associations with P< 1*10 -7 were identified to be associated with any of the phenotypes: AFR, LFR or TFR in the sex-combined and sex-stratified analyses ( Figure 1, S1-2 Fig, S1 supplementary Data). Genomic inflation factors were low for all GWAS (λ= 1.05-1.14, S3 Fig). The associations corresponded to 5,542 unique SNPs across 111 loci. Many of the regions were associated with multiple phenotypes or associated with one phenotype in more that one of the strata (males, females or sex-combined). Conditional analysis also identified secondary associations with TFR at two loci, near ACAN and ADAMTS17 on chromosome 15q26.1 and 15q26.3, respectively (S1 Table).

Replication of top GWAS SNPs
For each phenotype and strata, the most significant SNP in each locus was taken forward for replication. These SNPs were selected independent on whether the locus was associated with multiple phenotypes/strata. Since different phenotypes/strata had Out of the 78 loci that replicated, a total of 28 loci were associated with AFR, 43 with LFR and 57 with TFR in either the sex-combined or sex-stratified analyses. There was substantial overlap between LFR and TFR loci (N= 33). AFR loci overlapped to a smaller degree with LFR (N=4 loci), and TFR (N=6 loci). Three loci in the vicinity of KRTCAP2/PKLR, ADAMTSL3 and GDF5 were associated with all three phenotypes. As many as 50 of the 78 loci did not overlap with previous GWAS loci for body adiposity-related traits and were considered novel adiposity loci (Table 4).
However, 24 of the novel adiposity loci overlap with previous GWAS for height.  Leading SNPs for each locus is presented along with the chromosome and basepair position (hg19/build 37). MAF: minor allele frequency. A1: effect allele. Results from association tests in the discovery ( 1 ) and replication cohort ( 2 ) are presented. N: number of participants with non-missing data for each SNP. β : estimated effect size (change in rank-transformed LFR) per allele. p: p-values from Z-tests for deviance of β from zero.

Sex specific effects and SNP x sex interactions
Clear contrasts could be seen between males and females with regards to the number of associations. As many as 33 loci were associated with LFR in females, but only

Discussion
We performed GWAS for body fat distribution, using segmental BIA measurements, and identified 78 loci body fat distribution to the arms (N=28), legs (N=43) and trunk (N=57). As many as 50 of the loci have not been associated with an adiposity related phenotype previously. This is probably due to the low correlation between our derived phenotypes and commonly used variables for adiposity (i.e. BMI). In contrast to previous studies, we have not addressed the total amount of fat but rather the fraction of the total body fat mass that is located in the arms (AFR), the legs (LFR), or the trunk (TFR). While most of the loci were novel, with regards to adiposity, we did see an overlap with previously reported height loci, e.g. loci near SLC12A2, ADAMTSL3 and BMP2 [21]. This is surprising since we did see a very limited covariance between height and all our analyzed phenotypes. These results suggest that there is a shared genetic contribution between height and body fat distribution. It follows by logic that genes that are involved in growth can potentially influence several different tissue types such as bone, adipose tissue and muscle.
Interestingly, some of the novel loci overlapped with regions that have previously been associated with lipid-related traits. We found that our lead SNP (rs10402308) at the at the CLIP2/PBX4-locus (associated with TFR, and LFR in women) was in strong linkage disequilibrium (LD) (r > 0.8) with several SNPs previously associated with triglycerides, and LDL cholesterol [22]. The lead SNP at the TFR-associated locus within OSBPL7, rs2074188, was also associated with higher expression of OSBPL7 in the thyroid [23]. OSBPL7 encodes oxysterol-binding protein-like protein 7, which is highly expressed in the thyroid, skeletal muscles, GI-tract, kidney and seminal vesicles (www.proteinatlas.org). Oxysterol-binding proteins encompass a family of lipid-binding proteins involved in lipid trafficking, lipid metabolism and intracellular signaling [24].
Within the novel body loci we also find several genes related to estrogen and androgen signaling. Associations were observed between TFR and variants within the estrogen receptor-encoding gene, ESR1, in females. In addition, the TFR-, and LFRassociated SNP at the ADAMTS17-locus in females, rs72755233, is a missense mutation [25] which causes a potentially deleterious threonine to isoleucine substitution at position 446 of the ADAMTS17 protein (www.ensembl.org). This gene encodes a secreted metalloproteinase that is inducible in response to estrogen and inhibits breast cancer cell growth [26]. LFR-, and TFR-associated SNPs were also observed near ID4 in women. ID4 encodes a helix-loop-helix transcription factor that is highly expressed in the thyroid gland (www.proteinatlas.org) and also regulates androgen receptor function in the prostate [27].
AFR was the phenotype that was most highly correlated with BMI. In agreement with this, the most significant AFR loci were FTO, MC4R, TMEM18, SEC16B and TFAP2B, which have previously been associated with BMI and body adiposityrelated traits [8][9][10]28]. Several TFR and LFR-associated loci have also previously been associated with anthropometric traits. In contrast to AFR, most of these loci did not overlap with previously known BMI-loci, but to a larger extent with waist-, and hip-associated loci such as MTMR11, GDF5, ZBTB38, ADAMTS10, ADAMTS17. [28]; and with height loci such as HIST1H1D, ADAMTSL3, LIN28B [29].
Comparing men to women showed that genetic effects differ between sexes for a large fraction of the loci. For trunk and legs, many effects were only detected (or significantly higher) in women. In agreement with this, a larger fraction of the variance in fat distribution to different compartments could also be attributed to the SNPs investigated in women, as compared to men. These results are consistent with previous GWAS that have revealed sexual dimorphisms in genetic loci for adiposityrelated phenotypes, such as waist-circumference, waist-to-hip ratio, and visceral fat mass [14,[30][31][32]. In our study we find evidence for 43 loci whose effects differed between the males and females, of which one overlapped with a locus (LYPLAL1) that has previously been reported to display a different effect between sexes. Our lead SNP (rs1415287) at the LYPLAL1 locus is in strong LD with rs2820443 and rs4846567 (R 2 =1.00 and 0.99), which have been associated with stronger effects on WHR and WHR adjusted for BMI in women [14,32].
One possible limitation of our study is the use of segmental BIA measurements for assessments of body adiposity in contrast to using more exact methods such as DXA or MRI. However, the relatively low cost and ease of use has allowed for assessment of body composition in almost the entire UK Biobank cohort, which enables us to perform highly powered association studies. The accuracy in reference to DXA of the body scanner used in UK Biobank, the Tanita BC-418, has previously been assessed in a European sample showing that total fat mass were accurately estimated.
However, some biases were present depending on sex and anatomical compartment [33]. This is unlikely to affect our results as we analyzed each compartment separately and also performed sex-stratified analyses in addition to the sex-combined GWAS.

Conclusions
GWAS of body fat distribution to the arms, legs and trunk revealed 50 novel adiposity loci. Our results indicate that the trunk and legs share genetic determinants of fat distribution, while distribution of fat to the arms is more independent. We also present evidence for 43 SNP-sex interactions that influence adipose tissue distribution.
Distribution of adipose tissue between the trunk, legs and arms differ between sexes, which may be due to hormonal differences. Sex hormones primarily affect cellular proliferation, differentiation and fate by binding to and activating nuclear receptors that act as transcription factors. Consequently, the observed interactions with sex are likely to represent genes that are affected by changes in sex hormone levels.

UK Biobank participants
The first release of genetic data from UK Biobank (N = 152,249) was used as a discovery cohort, and genotype data from an unrelated set of participants from the second genotype batch release (N = 326,565) as a replication cohort. Participants who self-reported as being of British descent (data field 21000) and were classified as Caucasian by principal component analysis (data field 22006) were included in the analysis. Genetic relatedness pairing was provided by the UK Biobank (Data field 22011). In total, 9,385 participants were removed due to relatedness based on kinship data (estimated genetic relationship > 0.044) and individuals with poor call rate (<95%), with high heterozygosity (Data field 22010), or with sex-errors (Data filed 22001) were also removed. After filtering, 116,138 participants were included in the discovery cohort and 246,361 in the replication cohort.

Genotyping
Genotyping in the discovery cohort had been performed on two custom-designed microarrays: referred to as UK BiLEVE and Axiom arrays, which genotyped 807,411 and 820,967 SNPs, respectively. Imputation had been performed using UK10K [34] and 1000 genomes phase 3 [35]

Phenotypic measurements
The phenotypes used in this study derive from impedance measurements produced by the Tanita BC418MA body composition analyzer. Participants were barefoot, wearing light indoor clothing, and measurements were taken with participants in the standing position. Height and weight were entered manually into the analyzer before measurement. The Tanita BC418MA uses eight electrodes: two for each foot and two for each hand. This allows for five impedance measurements: whole body, right leg, left leg, right arm and left arm. Body fat for the whole body and individual body parts had been calculated using a regression formula, that was derived from reference measurements of body composition by DXA in Japanese and Western subjects, and uses weight, age, height and impedance measurements [36] as input data. Arm and leg fat masses were averaged over both limbs. Arm, leg, and trunk fat masses were then divided by the total body fat mass to obtain the ratios of fat mass for the arms, legs and trunk, i.e. what proportion of the total fat in the body is distributed to each of theses compartments. These variables were analyzed in this study and were named: arm fat ratio (AFR), leg fat ratio (LFR), and trunk fat ratio (TFR).

Correlation
Correlations between fat distribution ratios and anthropomorphic traits were assessed by ANOVA of linear regression model fits. BMI, waist circumference, waist-to-hip ratio and height were included as the last term in generalized linear models with adipose tissue distributions (AFR, LFR and TFR) as the response variable. The reduction in residual deviance, i.e. , the reductions in the residual sum of squares as BMI, waist circumference, waist-to-hip ratio and height is is added to the formula, is presented as percentages in Table 1.