Obesity is defined by health organizations as the pandemic of the 21 century. The prevalence of obesity is increasing alarmingly in both developed and developing countries . Common obesity is a complex trait caused by the interaction of genes and environment, with each gene variant producing only a minor effect . Despite the well-established fact that the genetic factors explain 40-70% of inter-individual variation in common obesity, the variation attributable to identified genetic polymorphisms to date comprise only a few percents [2, 3]. Using positional cloning strategies for identifying new susceptibility genes involved in common obesity remains a difficult task with limited success thus far . However, the study of extreme monogenic, syndromic human obesity phenotypes with Mendelian patterns has greatly increased our knowledge of the mechanisms that underlie obesity .
Bardet-Biedl syndrome (BBS, MIM 209900) is a heterogeneous autosomal recessive disorder characterized by obesity, pigmentary retinopathy, polydactyly, renal malformations, mental retardation, and hypogenitalism [6, 7]. Furthermore, the onset of BBS obesity usually begins early in life and is progressive with time emphasizing the potential relevance to common obesity . To date, 14 BBS genes were identified in 70% of the BBS patients, indicating that additional mutations in known and new BBS genes remain to be identified. A homozygous null mutation in each of 14 genes (BBS1-14) is sufficient to cause the entire BBS phenotype . Although BBS obesity is rare worldwide (<1/100,000), there is considerable interest in investigating polymorphisms in BBS genes because they could contribute to common obesity. Thus, obesity has been reported in heterozygous relatives of BBS patients (case-control study), suggesting that BBS gene defects might predispose to obesity in large cohorts .
Up to now, only a few BBS gene variations were analyzed for association with common obesity. While two previous reports found no association of polymorphisms in BBS1 and BBS6 to polygenic obesity in Norfolk and Danish populations ([10, 12], respectively), recently a significant association of polymorphisms of BBS2, BBS4, and BBS6 genes in French Caucasians with common obesity was reported in a case-control study . Neither of these studies considered the potential effect of these or other BBS genes on obesity phenotypes in a general population or tested systematically association of other BBS genes with obesity in non-BBS individuals. In correlation with the human BBS phenotype, null Bbs2 , Bbs4 , and Bbs6  mice have features of the human disorder and develop obesity. Although the molecular mechanism underlying BBS obesity is not fully understood, BBS phenotypes have been mainly attributed to a defect of the primary cilium biology. This defect results in impaired leptin receptor trafficking in the hypothalamus of some BBS models and contributes to the BBS obese phenotype of both BBS animal models and humans . Recent studies showed that BBS genes have a unique and particular pattern of expression in preadipocytes and that at this young stage primary cilia are present and are important for normal adipogenesis [20, 21]. This strongly suggests that a peripheral primary dysfunction of adipogenesis participates in the pathogenesis of obesity in BBS and potentially in common obesity.
The current study systematically investigates potential association between the tagging SNPs in all 14 BBS genes identified to date and body mass and fat in the large sample of UK adult twins.
Methods and Procedures
Data were collected about 4,465 female twins enrolled in the UK Adult Twin Registry . The twins were not selected for disease-specific studies and are representative of the UK population. Twins, both dizygotic (DZ) and monozygotic (MZ), included in this sample were females between the ages of 18 and 80 years. All subjects were clinically assessed at St Thomas' Hospital between 1996 and 2000. Phenotypic measurements were collected during clinical visits. Total fat data were obtained by DXA body composition scans (HologicQDR-2000; Vertec, Waltham, MA). Cohort included 4,465 sibling pairs and relative pairs. Each individual in the sample was assessed for following body mass-related phenotypes: body height (HT, cm), weight (WT, gr), BMI (WT/HT2 kg/m2), total body fat mass (TBF, gr), TBF/HT2 (kg/m2, analogous to calculation of BMI), and TBF/WT. Significant results were further tested in the BMI GIANT consortium population (n = 123,868, women and men) This population included the UK Twins study .
The genetic data were obtained for two subgroups of 2,462 and of 2,003 individuals, using the Infinium 610k assay (Illumina, San Diego) and the 317 assay (Illumina HAP300 chips), respectively. Genotyping was done either by the Sanger Genotyping Centre or at the Center for Inherited Diseases Research (CIDR). All participants gave written informed consent before entering the study, and the St. Thomas' Hospital research ethics committee approved the project.
The BBS1 through BBS14 genes are located on 11q13, 16q21, 3p12-q13, 15q22.3, 2q31, 20p12, 4q27, 14q32.11, 7p14, 12q21.1, 9q33.1, 4q27, 17q23, and 12q21.3 chromosomal segments, respectively. As previous studies have mapped to non-coding areas of the genes [10, 12, 13], we chose to examine all available tagging SNPs located in coding and non-coding regions of the BBS genes (Table 1). That is, for each gene, we considered genomic regions between 10 kbp upstream and 10 kbp downstream from the transcribed region. In total, 105 SNPs were selected for association analysis based on genotype data obtained in the discovery sample (610 K), using the default option of Tagger feature of the Haploview 4.2 , with r2≥ 0.8. Additional selection criteria included minor allele frequency, MAF ≥ 0.1 and correspondence of genotype frequency distributions to Hardy-Weinberg equilibrium expectations [25, 26].
Table 1. List of BBS genes, genomic location and genomic area analyzed in the study
|HGNC symbol||BBS gene||Chr||Gene Start (bp)||Gene End (bp)||-10 kbp||+10 kbp|
Statistical genetic analysis
Prior to genetic analysis, descriptive statistics were obtained for each study phenotype, using STATISTICA 7.1 package . Preliminary analysis showed highly significant correlations among all body mass variables. They were therefore subjected to principal component analysis (PCA). A first principal component derived in this analysis was used as generalized obesity phenotype. The association analysis was conducted on age-adjusted variables. For each trait, the best-fitting polynomial model of age dependence was found as described elsewhere [28, 29]. Heritability estimates for each trait were obtained using variance component analysis as repeatedly described by us previously  and implemented in the statistical package MAN .
The central component of the study included tests of genetic association between tSNPs in the selected genes and obesity phenotypes. The main results were obtained using GenABEL, a package for genome-wide association analysis between quantitative or binary traits and single-nucleiotide polymorphisms (SNPs), available at http://cran.r-project.org/web/packages/GenABEL/index.html . The SNPs were checked for genotyping quality and thresholds accepted for the study (minor allele frequency (MAF) > 0.1) using “check.marker” function in GenABEL. “polygenic” function in GenABEL was applied for each studied phenotype to determine the optimal additive polygenic model. This allowed us to factor out estimated polygenic effect and to obtain environmental residuals. The latter were used as input for “mmscore” function in GenABEL, which performs a score test for association between the trait and genetic polymorphism, in samples of related individuals . Next, the results which were less likely to have occurred by chance (smallest P-values) were tested in the replication sample (317 K). In order to account for the number of tests conducted, the results of this analysis were subjected to multiple testing correction procedure using the false discovery rate (FDR = 0.05) approach for multiple testing under dependency. For the sake of convenience, some additional statistical tests are described briefly in the Results section. Additionally, 14 most promising association results (with lowest likelihood to occur by chance) were tested in confirmation sample of the BMI GIANT consortium sample consisting of 123,868 men and women of European ancestry, using the BMI phenotype only.
Table 1 shows the list of the BBS genes, their genomic locations, and selected genomic areas analyzed in the study. Table 2 provides the basic descriptive statistics of the studied population. The data for traits are presented in the original units. There was no statistical difference between the subgroups (610 K and 317 K) in age or any of the body mass traits, thus enabling us to use both groups for the association studies. Due to the high pairwise correlations between the study phenotypes, ranging between 0.81 and 0.99 (with P < 0.001 for all), they were subjected to principal components analysis (Table 3). Only one principal component was retained in the analysis and was defined as OB_PC. It explained 87.9% of the total variability of the original traits and was used as a generalized trait of obesity in further analysis. To reduce the redundancy in data analysis, we prompted to examine only the following phenotypes: OB_PC, BMI, and TBF/HT2.
Table 2. Descriptive statistics of study population
|Heritability estimate (± SE), adjusted for age||72.58 ± 3.21%||77.34 ± 2.12%||76.53 ± 2.12%||75.16 ± 2.16%||75.37 ± 2.15%||65.92 ± 3.95%|
Table 3. Main results of principal component analysis of body mass traits in total (N = 4457) twins sample
All phenotypes displayed significant heritability estimates (Table 2), as obtained in variance component analysis and therefore provided the further argument for search of the specific genetic determinants. The main results of the association analysis conducted between each of the 105 selected SNP and three body mass and fat phenotypes are presented in Table 4. As seen, all phenotypes showed a number of nominally significant results and mostly with the same SNP. The most interesting include rs10262995 in an intronic region of BBS9 gene; rs803944 (intronic) and rs1661300 (upstream), respectively, to the BBS11 gene. In total, 13 statistically significant results, at P < 0.01 were obtained out of 315 tests. This is substantially higher than only three significant results, obtained by chance only, as expected from type I error. However, by FDR = 0.05 criteria, required P-value cutoff 0.0002 was not achieved in any of the analyses.
Table 4. Association between the selected SNPs and body mass traits in the UK Twin sample (610K cohort)
|rs12536844||7||33510003||BBS9||0.257||0.126|| ||0.018|| |
|rs4141711||9||118511702||BBS11||0.165||0.034|| || || |
|rs12596017||16||55077515||BBS2||0.112||0.154|| || || |
|rs4637716||7||33476688||BBS9||0.478||0.904|| ||0.070|| |
|rs1419901||7||33490317||BBS9||0.274||0.243|| ||0.041|| |
In an attempt to clarify the situation, we examined the genotype and allelic distribution of the SNPs consistently significantly associated with three body mass phenotypes (Table 4), in individuals with normal body mass (BMI 20-25) and obese (BMI>30), and by twin. That is, the sample was divided into two subgroups: twin 1 and twin 2. No significant differences in genotype or allele distributions were found in this analysis (Table 5). Finally, to confirm significant findings discovered in the total 610 K sample, we tested all the significant associations presented in Table 4 using our replication sample, 317 K cohort. However, we found no significant associations in this sample. Analysis of association of the 14 SNPs with BMI GIANT consortium BMI data of men and women found no significant results (Table 6). However, it should be noted that only the BMI phenotype was analyzed; the population included men and women, and there was overlapping between the populations.
Table 5. Genotypic and allelic distribution of the BBS SNPs, associated with elevated adult body mass in 610K cohort, in weight-normal and obese individuals and in both twins separately
|BMI:20-25||455 (0.7712)||125 (0.2119)||10 (0.0169)||456 (0.7716)||130 (0.22)||5 (0.0085)|
|BMI>30||140 (0.8383)||25 (0.1497)||2 (0.012)||142 (0.8554)||24 (0.1446)||0 (0)|
|X2: 3.48783, df=2, P = 0.174840||χ2: 6.17838, df=2, P = 0.045543|
|BMI: 20-25||470 (0.7939)||11 (0.0186)||111 (0.1875)||452 (0.7661)||10 (0.0169)||128 (0.2169)|
|BMI > 30||136 (0.8144)||1 (0.006)||30 (0.1796)||139 (0.8373)||0 (0)||27 (0.1627)|
|X2: 1.41808, df=2, P =0.492119||χ2: 5.51775, df=2, P = 0.063368|
|BMI: 20-25||138 (0.2331)||169 (0.2855)||285 (0.4814)||121 (0.2047)||151 (0.2555)||319 (0.5398)|
|BMI > 30||34 (0.2036)||56 (0.3353)||77 (0.4611)||30 (0.1807)||49 (0.2952)||87 (0.5241)|
|X2: 1.70592, df=2, P = 0.426155||χ2: 1.20641, df=2, P = 0.547058|
|BMI: 20-25||260 (0.4392)||268 (0.4527)||64 (0.1081)||253 (0.4281)||264 (0.4467)||74 (0.1252)|
|BMI > 30||67 (0.4012)||82 (0.491)||18 (0.1078)||76 (0.4578)||77 (0.4639)||13 (0.0783)|
|X2: .851187, df=2, P = 0.653383||χ2: 2.82869, df=2, P = 0.243090|
|BMI: 20-25||469 (0.7922)||115 (0.1943)||8 (0.0135)||467 (0.7915)||119 (0.2017)||4 (0.0068)|
|BMI > 30||133 (0.7964)||29 (0.1737)||5 (0.0299)||125 (0.753)||37 (0.2229)||4 (0.0241)|
|X2: 2.34639, df=2, P = 0.309381||χ2: 4.19862, df=2, P = 0.122547|
|BMI: 20-25||115 (0.1943)||299 (0.5051)||178 (0.3007)||115 (0.1946)||292 (0.4941)||184 (0.3113)|
|BMI > 30||36 (0.2156)||81 (0.485)||50 (0.2994)||48 (0.2892)||75 (0.4518)||43 (0.259)|
|X2: 0.402545, df=2, P = 0.817690||χ2: 7.04284, df=2, P = 0.029561|
Table 6. Association between SNPs and BMI trait in the GAINT consortium population
Recent advances in BBS obesity research revealed defects in basic mechanisms of fat cell differentiation and potential defects in neural control of food intake related to mutations in BBS genes [17, 20, 21]. On the basis of these molecular studies, the early age of BBS obesity development and Bbs null mouse models, BBS genes were suggested as potential candidate genes contributing to common obesity. To date, 14 BBS genes have been identified.
In this study, we analyzed the association between all 14 BBS genes identified to date and body-weight- and fat-related phenotypes. We report here that of 105 selected tagging SNPs in 14 BBS genes, three SNPs in two BBS genes (BBS9 and BBS11) showed evidence of association with elevated body mass.
Marker rs10262995 (BBS9) and rs803944 and rs1661300 (BBS11) showed nominally significant associations with all the studied phenotypes: OB_PC, BMI, and TBF/HT2.
The P-values for the strongest association with OB_PC, BMI, and TBF/HT2 were as follows: rs10262995- 0.002, 0.001, 0.004, rs803944-, 0.003, 0.004, 0.005, and rs1661300- 0.002, 0.003, 0.004, respectively. However, none of the associations subsisted multiple-testing correction, and none were supported in further analysis in this cohort in our replication sample and in BMI GIANT data.
One of the potential reasons that we were unable to replicate the significant results observed in the first sample is the power of association analysis, which is crucially depends on sample size. This is the main reason for many consortia studies that often use tens of thousands of individuals. The simulation studies also showed that the significance of association signal depends, in addition on MAF and putative effect of the SNP. For example, Korostishevsky et al.  showed that if the marker effect <1%, then power of the analysis in the sample of the moderate size, even implementing such a sensitive method as the pedigree-based disequilibrium test, decreases <80%.
We did not find association between any other BBS genes and elevated body mass, which is in accordance with other studies, including Anderson et al. , who failed to identify association between BBS6 coding variants and common obesity in a Danish obese population, and Fan et al. , who did not find association between BBS1 and common obesity in Newfoundland population.
The variants in BBS9 and BBS11, nominally associated with elevated weight in our study, were located in non-coding regions of the genes. This is in line with the latest BBS-obesity association study that found BBS2, BBS4, and BBS6 non-coding sequence variations to be significantly more prevalent in obese compared with lean individuals . The biological significance of non-coding sequence variations remains largely unpredictable and difficult to recognize.
Although it has been previously suggested that heterozygous carriers of a BBS mutation are predisposed to obesity , only few BBS gene variations were analyzed for association with common obesity [10, 11]. Benzinou and colleagues  found that BBS mutations increase the risk of common obesity: one of the BBS2 SNP was associated with common adult obesity, and BBS4 and BBS6 variants were associated with common early-onset childhood obesity and common adult morbid obesity. Our results did not confirm association between BBS2, BBS4, and BBS6 coding and non-coding regions and elevated body mass. Although we did not have data for the specific SNP variants found in the study by Benzinou et al.  (BBS6 rs221667 and rs6108572, BBS2 rs4784675, and BBS4 rs7178130), the use of polymorphisms included in the 610K and the 317K SNP arrays produced a good coverage of the same genomic segments; particularly, as we studied, similar to Benzinou et al., both the coding and non-coding regions. Still, we did not obtain significant association results with body mass and fat for SNPs residing in the same genomic segments. It should be noted that Benzinou et al. studied different population than ours; whereas we studied only adult women, their study included men and women, and both children and adults, obese versus normal weight with a different experimental design (control versus obese).
It has been suggested that the sum of function-altering variants in BBS genes is associated with a more obese phenotype . We therefore tested whether the number of rare alleles in the BBS genes within the individual could be associated with the study OB-related phenotypes. However, we found only negligible (∼0.020) and statistically non-significant (P > 0.20) correlations. There could be a number of reasons for not detecting such a correlation. First, there is no evidence that rare alleles in general population, as the one examined in the present study, are “function-altering variants” and not just the regular population polymorphisms. Second, reviewing of the numerical values of the estimated regression coefficients shows that the effect of the rare allele may be positive or negative, depending on SNP (although as a rule statistically non-significant) and therefore their cumulative effected is expected to be negligible.
Despite negative results, our study has several advantages. For the first time, all 14 BBS genes were systematically examined for association with the variety of obesity-related phenotypes. Thus, for example, we employed PC analysis to extract useful information from multiple obesity phenotypes, which were highly correlated. This type of analysis is widely used to decrease the likelihood of type I error rate and is specifically relevant to obesity research as more typical obesity phenotypes might not distribute normally [32, 33].
To summarize, studying the association between all 14 BBS genes identified to date, we found only weak evidence for three variants related to common elevated body weight; however, none of the associations subsisted multiple-testing correction. The results suggest that common variations in 14 BBS genes are unlikely to have a major effect on risk of common elevated body mass and fat in Europeans Caucasians. However, studies of other SNPs, further away from the transcribed regions, deep re-sequencing to find function-altering variants [since common allelic variation may not necessarily alter gene function markedly, ] as well as studies of other populations might still shed light on possible association of BBS genes with common obesity.