A recent study suggested that four CD36 polymorphisms (namely rs3211867, rs3211883, rs3211908, and rs1527483) were associated with an increased risk of obesity, an increased BMI and percentage of body fat in European adolescents. We first attempted to confirm these results in three independent case-control genome-wide association studies (GWAS) data totaling 3,509 subjects of French and German origin, but we were unable to find any association of these variants with early onset obesity risk. We then genotyped the four CD36 single-nucleotide polymorphisms (SNPs) in a large population-based study of 4,667 Finnish subjects and we did not replicate any of the recently reported associations with BMI. By combining all available data in a meta-analysis (N = 9,973), we found no evidence for an association of the reported four variants in CD36 with increased obesity risk or increased BMI (0.07 ≤ P values ≤ 0.93). Finally, we assessed the contribution of the full CD36 locus gene variation to obesity risk in 3,509 subjects and we did not detect any significant association with obesity after correction for multiple testing. In summary, we were unable to confirm the recently reported association of variants in CD36 with early onset obesity in populations of European ancestry.
CD36, a transmembrane glycoprotein of the class B scavenger receptor family (1,2,3), with multiple, cell-specific, functions (4), is implicated in mechanisms of angiogenesis, apoptosis, inflammatory responses, and phagocytosis and contributes to atherosclerosis (5), cardiomyopathy (6,7), arterial hypertension, Alzheimer's disease (8), malaria susceptibility (9,10), and metabolic disorders (11). CD36 is expressed in sensory cells and is involved in insect pheromone signaling (12,13) and regulates rodent fatty food preference (14,15,16,17). CD36 is a membrane-bound fatty acid transporter (18) and has also been identified in the mitochondrial fatty acid uptake system (19). In the gut, CD36 promotes absorption of long-chain fatty acids (20,21).
CD36 deficiency is known to be involved in impaired fatty acid metabolism, glucose intolerance and insulin resistance (22,23,24,25). We have previously reported that the −178 A>C CD36 variant was significantly associated with circulating adiponectin and lipid levels, which are established markers of insulin resistance (26,27).
Recently, Bokor et al. (28) reported that four variants (rs3211867, rs3211883, rs3211908, and rs1527483) in the CD36 gene were associated with adolescent obesity in a case-control study of 307 obese and 339 normal weight European adolescents and with BMI and percentage of body fat in a cross-sectional study of 1,151 European adolescents. The present study aimed to perform a replication study, using case-control and population-based designs, in order to assess the contribution of the four above-mentioned CD36 variants (rs3211867, rs3211883, rs3211908, and rs1527483) to obesity risk and BMI modulation in a large independent sample of European ancestry (N = 8,176). We then achieved a meta-analysis of six independent studies totaling 9,973 subjects. The contribution of the full CD36 locus gene variation to obesity risk was assessed in 1,858 controls and 1,651 subjects with early onset obesity, all of French or German origin.
Methods and Procedures
The study protocol was approved by all local ethics committees and informed consent was obtained from each subject before participating in the study.
Case-control studies. We extracted data from two recent genome-wide association studies (GWAS) (29,30). The first French case-control study included 685 French obese children (recruited by the CNRS UMR8090 and the Toulouse Children's Hospital) and 685 French lean children (selected from the STANISLAS study or from the Fleurbaix-Laventie Ville Santé II study). The second case-control study was performed in 479 French morbidly obese adults who developed obesity during childhood or adolescence (recruited by the CNRS UMR8090 and the Department of Nutrition of Paris Hotel Dieu Hospital) and 731 French lean adults (selected from the cohort DESIR (Data from the Epidemiological Study on the Insulin Resistance syndrome) prospective study (31)). The third German case-control study included 487 extremely obese children and adolescents, recruited in hospitals specialized for in-patient treatment of extreme obesity, and 442 healthy lean young adults who were ascertained at the University of Marburg (Germany).
The Northern Finland Birth Cohort 1986 population-based study. The Northern Finland Birth Cohort (NFBC) 1986 is a genetically homogenous cohort of Caucasian children born in the two northernmost provinces of Finland, namely Oulu and Lapland, with expected dates of birth between 1 July 1985 and 30 June 1986. The mothers were enrolled during routine visits to communal maternity health centers during the 25th week of gestation. Ninety-nine percent of eligible women and their children were included in the cohort. Obstetrical and birth data were coded from hospital records at the time of delivery. A postal questionnaire concerning the health of their child was mailed to parents when children were 7 years old. Ninety percent of the parents responded to the questionnaire. At the same time, information on the child's health was obtained from hospital records. At 16 years of age, 74% of children attended a clinical examination, including blood draw.
We extracted genotypes of the rs3211867, rs3211883, rs3211908, and rs1527483 single-nucleotide polymorphisms (SNPs) in the three case-control studies from two recent GWAS (29,30). The French GWA genotyping used the Illumina Human CNV370-Duo array and the Illumina HAP300 array (30). The genotyping of German subjects used a Genome-Wide Human SNP Array 5.0 (29). We used the IMPUTE program to impute the genotypes for the rs3211867, rs3211883, rs3211908, and rs1527483 SNPs, as they were not directly genotyped in the DNA arrays. For the four above-mentioned SNPs, HapMap data were available. We also used data from the two GWAS (29,30) to assess the contribution of the full CD36 locus to early onset obesity. We determined the CD36 locus (5′ and 3′ intergenic and intragenic regions) as chr7:79,979,178-80,209,790 using NCBI Build 36. Our evaluation of the CD36 locus was based on RefSeq transcripts reviewed or validated by the RefSeq, SwissProt, or CCDS staff. Nineteen SNPs were genotyped for the French GWAS (31) and 65 SNPs for the German GWAS (30) at the CD36 locus and intergenic regions. Four CD36 SNPs were common between French and German GWAS.
The rs3211867, rs3211883, rs3211908, and rs1527483 SNPs were genotyped in the NFBC 1986 population-based study using an allelic discrimination assay-by-design TaqMan method on ABI 7900 (Applied Biosystems, Foster City, CA). A genotyping success rate of 94–99.8% and a concordance rate of 99.7–100% were observed from the analysis of 384 duplicated DNA samples. The genotypic distributions of the four polymorphisms were in Hardy-Weinberg equilibrium in the population-based cohort (P > 0.48).
Trained personnel performed one measurement of the weight and height during clinical examination, and then BMI (kg/m2) was calculated. In adults, obesity status was defined by BMI ≥30 kg/m2 for obese subjects and by BMI <25 kg/m2 for lean individuals. In children, leanness was defined as BMI <90th percentile, overweight as BMI ≥90th percentile and obesity as BMI ≥97th percentile.
Association between the SNPs and obesity was estimated using a logistic regression model, adjusted for gender and age. The statistical power was calculated with the QUANTO software, using the total number of case and control subjects (or the total number of subjects for population-based studies), the current prevalence of obesity in France at 15 years (11.3% as reported by the French Society of the Public Health), the effect size (either odds ratios (ORs) or the β coefficients) resulting from the first study (28) (case-control or cross-sectional) and an α-level of 0.05. Polymorphism effects on the quantitative trait of BMI were calculated by linear regression models adjusted by gender (and not by age as all individuals were of the same age). For the meta-analysis of BMI or obesity each trait analyzed separately, fixed-effect summary estimates were calculated for a dominant model (as proposed by Bokor et al. (28)) using the “rmeta” and “meta” packages of the R-Project (http:www.r-project.org). We also used Egger's regression method to test for publication bias (32). The summary statistics (linear regression β coefficient, or logarithm of the OR and s.e.) for each of the six separate studies were combined in meta-analyses using the weighted inverse normal method (http:www.spc.univ-lyon1.frlivreMAframe.html). P values of logistic and linear regression tests were transformed into Z-scores and a weight was given to each study, so that the sum of the squared weights equaled 1. The sum of these weighted Z-scores followed a normal distribution N (0,1). Associations with the binary trait, obesity, in the GWAS at the CD36 locus, were carried out using SNPTEST by testing three genetic models. A conservative Bonferroni correction (multiplication by the number of SNPs and three as there were three genetic models tested) was applied to the P values to correct for multiple comparisons. To evaluate the percentage of genetic variability tagged by the SNPs genotyped at the CD36 locus and intergenic regions (chr7:79,979,178-80,209,790 using NCBI Build 36), we downloaded genotype data from the International HapMap Project website (http:www.hapmap.org). We then selected genotype tagging SNPs (tag SNPs) that had a minimum r2 > 0.8 with untyped SNPs via Haploview (version 4.2) software. SNPs with minor allele frequency <0.05 were excluded from the analysis. R statistics (version 2.5.0) software was used for general statistical analyses.
Phenotypic characteristics of the studied populations are described in Table 1. We first analyzed data from the GWAS (29,30) and we imputed the four previously reported CD36 polymorphisms (rs3211867, rs3211883, rs3211908, and rs1527483) (28). The imputation quality scores were all over 0.9 for the four variants in each of the three GWA data sets (imputation quality score is 1 if the SNP is perfectly imputed). The first case-control study used 685 obese and 685 lean French children. A second analysis was performed in 487 extremely obese young German subjects and 442 German lean young adult controls. A third case-control study used 479 French adults with a history of overweight during childhood or adolescence and 731 French lean adults. We tested the impact of the above-mentioned CD36 polymorphisms (rs3211867, rs3211883, rs3211908, and rs1527483) on obesity in these three independent case-control studies. We noted that some of them were underpowered: statistical power of 35.8–49.4% to detect the previously reported OR (28) for the rs3211867, statistical power of 24.8–34.3% for the rs3211883, statistical power of 57.8–74.8% for the rs3211908, and statistical power of 35.3–48.7% for the rs1527483. As shown in Table 2, no nominally significant (P < 0.05) association was observed between any of the four CD36 SNPs and the risk of obesity under the dominant model (the model used in the initial study) nor in any of the three individual studies. Recessive and additive models were tested in the three case-control studies but they did not show any association with the risk of obesity (data not shown).
Table 1. Description of the populations used for case-control studies
Table 2. Genotypic distributions and odds ratios (ORs) for the CD36 polymorphisms and obesity among case-control studies
To assess the contribution of the above-mentioned CD36 polymorphisms (rs3211867, rs3211883, rs3211908, and rs1527483) to BMI variation and obesity predisposition in a general population, we genotyped the NFBC 1986 cohort (N = 4,667). Clinical characteristics of adolescents in this study are presented in Table 1.
We first tested the association between the four CD36 SNPs and BMI in the participants, aged 16 years (Table 3). The NFBC 1986 study had a statistical power of 97.3, 98.1, 99.1, and 99.2% for the rs3211908, rs1527483, rs3211883, and rs3211867, respectively to detect the effects previously reported in the HELENA cross-sectional study (28). BMI was slightly lower in subjects carrying the minor rs3211908 allele (P = 0.047 under a dominant model) but the direction of the allele effect differed from the HELENA study (28). As shown in Table 3, we observed no significant difference between BMI at 16-year and rs1527483, rs3211867, and rs3211883 SNPs under a dominant model.
Table 3. Genotype distributions according to the four CD36 polymorphisms in the NFBC 1986 study
We then performed an association study by testing the impact of the four CD36 SNPs on overweight (BMI ≥90th percentile) or obesity (BMI ≥97th percentile) risk in the NFBC 86 cohort at 16 years of age. Statistical power was between 17.7 and 43.1% to detect the effects with obesity previously reported (28). None of these SNPs was associated with obesity risk. However, we found a weak association between two polymorphisms (rs1527483 and rs3211908) and risk for overweight (OR = 0.72 (0.54–0.97), P = 0.03, and OR = 0.67 (0.48–0.93), P = 0.02, respectively under a dominant model) but the direction of the association was inconsistent compared to the initial study (28). Results are summarized in Table 3.
To increase the statistical power, we combined the three case-control studies (N = 3,509) in a meta-analysis and we observed a nominal effect (OR = 0.82 (0.68–0.99), P = 0.04) for the rs3211883 but in a direction inconsistent with the previous study (28) and with a statistical power of 70.1%. No effect was detected for the three other CD36 SNPs (rs3211867, rs3211908, and rs1527483) with obesity risk under a dominant model despite a statistical power over 80% (87.0, 87.5, and 98.8% for the rs1527483, rs3211867, and rs3211908, respectively). Neither between-study heterogeneity (0.43 < P < 0.92) nor publication bias (0.48 < P < 0.95) was significant for these four meta-analyses (Figure 1).
We also combined the HELENA cross-sectional study (28) and the NFBC 1986 population-based study in a meta-analysis and we showed no significant association between the same four SNPs and BMI in the 5,818 European adolescents (data not shown).
We performed a meta-analysis for the four CD36 SNPs by pooling our current results with the two studies from Bokor et al. (28) in six independent studies (N = 9,973) and found that none of these SNPs was significantly associated with obesity under a dominant model (P values of 0.72, 0.66, 0.07, 0.93 for rs3211867, rs3211883, rs3211908, rs1527483, respectively).
Full CD36 locus exploration
We then explored the entire CD36 locus (5′ and 3′ intergenic and intragenic regions) by evaluating the risk on obesity of the CD36 SNPs genotyped on the chips of the three GWAS data sets described above. Nineteen SNPs were genotyped for the French GWAS (30) and 65 SNPs for the German GWAS (29), at the CD36 locus and intergenic regions. These SNPs represented 86.5 and 93.6% of the genetic variability at the CD36 locus and intergenic regions, respectively for the French and German GWAS. For each polymorphism, we performed association analyses with obesity in the same case-control studies described above: French obese and lean children, French morbidly obese adults with a history of overweight during childhood or adolescence and lean adults and German extremely obese and lean young subjects. The three genetic models were tested.
The lists of CD36 SNPs genotyped for the two GWAS and the results of association analyses are presented in Supplementary Tables S1 and S2 online. None of the SNPs genotyped at the CD36 locus and intergenic regions remained significant after correction for multiple testing (P value ≤ 8.77 × 10−4 for the French studies and P value ≤ 2.56 × 10−4 for the German study). They did not show any significant association with childhood obesity or adult obesity, with onset of overweight during childhood or adolescence, either in French or in German populations using all possible association models: additive, dominant, and recessive and after adjustment by gender and age.
Recently, Bokor et al. (28) proposed that four CD36 SNPs may increase the risk of obesity in European adolescents. Our data, supported by case-control and population-based study designs, did not confirm the initial association of these polymorphisms with obesity.
Bokor et al. claimed that the lack of association of CD36 SNPs and obesity in previous GWAS (29,30) of children and adolescents with extreme obesity was due to the fact that no CD36 SNP was present on the chips nor was able to be imputed from other SNPs. In our analysis, 19 SNPs were genotyped in the French GWAS (30) and 65 SNPs in the German GWAS (29). These SNPs captured a large part of the common genetic variability at the CD36 locus and intergenic regions (86.5 and 93.6%, respectively for the French and German GWAS) but no association with obesity was detected. In addition, the imputation quality scores were excellent for the four variants from the GWAS data sets and described by Bokor et al.
The absence of replication in our study is unlikely to be due to a lack of statistical power. While the cohorts from the GWAS had modest sample sizes and were underpowered when analyzed alone, we reached a statistical power higher than 80% for rs3211867, rs3211908, and rs1527483 except for rs3211883 (70.1%) by combining the three case-control studies (N = 3,509) in a meta-analysis. Moreover, the NFBC 1986 study allowed us to validate our results knowing that we had a statistical power higher than 97% for the four studied SNPs. These results enabled us to definitely exclude association of rs3211867, rs3211908, rs3211883, and rs1527483 with obesity.
To explain the lack of association with childhood obesity at the CD36 locus in GWAS, Bokor et al. underlined that children with extreme obesity may suffer from different genetic defects than adolescents with mild obesity (28). In the current study, we studied cohorts of children as well as adolescents, and we tested a large range of obesity-related parameters (BMI, overweight, mild, or extreme obesity) but we were unable to replicate the association described in the initial study.
The initial study seems very likely to be a false positive result. In fact, by combining the six case-control studies in a meta-analysis we observed that the effects of our studies were in the opposite direction to the effects reported in the initial study, especially for the rs1527483, rs3211867, and rs3211883. The results from the initial study are typical of a “Winner's Curse” effect, amplified by modest sample sizes.
In conclusion, we found no effect of the CD36 variants on obesity risk in young populations of European ancestry using case-control, population-based, and meta-analysis study designs. Our results underline the importance of multiple replication, large-scale meta-analysis studies and the use of stringent statistic thresholds in genetic association reports before drawing conclusions about a true association signal.
We are indebted to all participants in this study. We thank M. Deweirder and F. Allegaert for DNA bank management. We thank S. Gaget, S. Gallina, and C. Cavalcanti-Proença for their technical support. We also thank J. Dallongeville for providing us additional information about the study recently published by Bokor et al. The German GWAS was supported by grants from the German Ministry of Education and Research (01GS0820 (NGFNplus) and 01KU0903). This work was supported by a grant from the “Programme National de Recherche en Alimentation et nutrition humaine.”