ACDC/Adiponectin Polymorphisms Are Associated With Severe Childhood and Adult Obesity

We genotyped 424 nuclear families including 634 obese children (defined as BMI >97th percentile for age and sex). Ninety-five unrelated obese children recruited in Toulouse were also genotyped (online appendix 1 [available from http://diabetes.diabetesjournals.org]). We used 534 unrelated obese children for case/control studies and 634 obesity trios (two parents and one obese child) for transmission disequilibrium test (TDT) analyses of obesity status and quantitative traits. Control set 1 was composed of 655 individuals (BMI <27 kg/m² and fasting glucose <5.6 mmol/l) recruited in Lille or through the Fleurbaix-Laventie Ville Santé (FLVS) study. We genotyped 695 unrelated morbidly obese adults (BMI ≥40 kg/m² and 75% were obese children or adolescents) recruited in Lille or Paris and compared them with age- and sex-matched control subjects (control set 2) (BMI <25 kg/m² and fasting glucose <5.6 mmol/l) selected among participants of the DESIR study. Data were also available for 224 nuclear families representative of the Northern France general population (FLVS study), from which we selected 400 trios and 197 unrelated lean children. The genetic study was approved by ethical committees of Hôtel Dieu Hospital in Paris and Centre Hospitalier Régional Universitaire in Lille. Insulin sensitivity–related phenotypes were analyzed in a subgroup of normal glucose-tolerant obese (n = 599) and FLVS (n = 427) children. Adiponectin levels were analyzed in 334 obese and in 367 FLVS children, where data were available. Homeostasis model assessment of insulin resistance was calculated as fasting insulin/[22.5e^{−ln(fasting glucose)}].

SNPs −11,377C>G and −11,391G>A were genotyped by LightCycler technology (Roche), and SNPs +45T>G and +276G>T were genotyped by Taqman technology (Applied Biosystems). Control set 2 genotypes were provided by the DESIR group. Genotyping error rates calculated from duplicate genotypes of 260 individuals were 0.01% for SNPs −11,377C>G, −11,391G>A, and +276G>T and 0.00% for SNP +45T>G. All SNPs were in Hardy-Weinberg equilibrium.

Adiponectin levels were measured in sera from 334 obese children and 367 children from the FLVS study with a commercial assay kit (LINCO Research) in P.E.S.’s laboratory.

Nuclear extracts were obtained as previously described (9) from differentiated 3T3-L1 cells (10). DNA probes were labeled with T4 kinase (Roche) using [γ-³²P] ATP. Binding reactions were performed in 20 μl of 50% glycerol, 20 mmol/l Tris/HCl, pH 7.5, 100 mmol/l KCl, 2 μg dI-dC, 1 mmol/l dithiothreitol, and 10 μg nuclear extract. Probes (50,000 cpm) were added to the binding mixture and incubated for 30 min at 20°C. The DNA-protein complexes were resolved on 5% polyacrylamide at 4°C. Gels were dried and exposed to X-ray films (Kodak). Quantification of the signal was performed with NIH Image software (available from http://rsb.info.nih.gov/nih-image/).

ACDC promoters were generated by PCR of 1.3 kb of genomic DNA from three homozygous subjects using an expand high-fidelity PCR system (Roche). PCR products were cloned into pGL3-basic (Promega). Competent bacteria were transformed with the constructs. All plasmids used for cell transfection were controlled by bidirectional sequencing.

COS-7 cells grown in Dulbecco’s modified Eagle’s media (Life Technologies) with 10% fetal bovine serum and 25 μg/ml gentamicin at 37°C and 5% CO₂ were transfected with 2 μg (TransFast reagent; Promega). Transfection efficiencies were normalized by 50 ng pRL-TK, the Renilia luciferase vector. A Dual-Luciferase Reporter Assay (Promega) was performed after 48 h incubation.

Case/control analyses used the χ² test, and P values were empirically computed with the CLUMP program (11). Allelic means for quantitative traits (corrected for BMI, age, and sex), haplotypes, allelic TDT, and quantitative TDT were analyzed by UNPHASED software (12), effects of SNPs on haplotype association by THESIAS software (13), and relative luciferase activities by SPSS 10.1.

Data are given as means (n). Analyses were performed in allelic model for traits corrected for BMI, age, and sex using the Qtphase subprogram of UNPHASED (df = 1).

Case/control analysis in 534 obese children and 655 control adults (control set 1) showed significant association between childhood obesity and the −11,377C allele (odds ratio 1.24 [1.03–1.50], P = 0.025) (Table 1). Similar results were obtained for −11,377C in 695 morbidly obese adults and in 695 age- and sex-matched control subjects (control set 2) (1.22 [1.03–1.45], P = 0.022). Pooled analysis of 2,579 French Caucasian subjects (1,229 morbidly obese and 1,350 control subjects) confirmed that the −11,377C allele is a risk factor for severe forms of obesity in our populations (1.23 [1.08–1.39], P = 0.001). Allele +276T was also associated with severe obesity (1.19 [1.05–1.36], P = 0.006), while −11,391G>A and +45T>G did not show evidence of association in our study. We genotyped SNPs −11,377C>G and +276G>T in a limited subset of 197 unrelated lean children from the north of France and obtained similar frequencies for the −11,377C and +276T alleles as seen in our control adults (0.74 vs. 0.73 and 0.28 vs. 0.26, respectively), which rules out a significant bias due to age.

We analyzed haplotypes and identified an “at-risk” haplotype, including the −11,377C and +276T alleles (1_(C)-1_(G)-1_(T)-2_(T)), more frequent in obese than in control subjects (0.20 vs. 0.17, respectively; odds ratio 1.24, P = 0.009, Table 2). In contrast, the 2_(G)-1_(G)-1_(T)-1_(G) haplotype was more frequent in control than in obese subjects (0.23 vs. 0.20, respectively; 0.92, P = 0.007). The effect of each SNP on haplotype associations was assessed using THESIAS software and showed that both associated SNPs contribute to the at-risk haplotype association. Using regression analyses, we obtained similar results (data not shown).

We also tested in our childhood obesity population familial association using TDT in 634 trios. We did not observe distortion of transmission for individual SNPs. However, haplotype TDT showed a trend in overtransmission of the obesity risk haplotype 1_(C)-1_(G)-1_(T)-2_(T) in obese trios (56.7%, P = 0.097, Table 3).

We assessed ACDC SNP effect on BMI, insulin sensitivity, and adiponectinemia in obese children and in children of similar age and representative of the general population (FLVS children). Single SNP analysis showed that allele −11,377C was associated with lower fasting insulin (P = 0.018) and lower homeostasis model assessment of insulin resistance (P = 0.013) in obese children and in trend of association with higher adiponectinemia in FLVS children (P = 0.087) (appendix 2). Allele +276T was associated with higher adiponectin levels in obese children (P = 0.03), but this finding was not observed in FLVS children (P = 0.303). Allele −11,391A was associated with higher adiponectinemia in obese children (P = 0.005), and this finding was strongly replicated in FLVS children (P = 0.00007).

Using quantitative family-based analysis, haplotype 1_(C)-2_(A)-1_(T)-2_(T), including −11,391A and obesity risk alleles −11,377C and +276T, was associated with higher adiponectin levels in obese (P = 0.008, Table 3) but not in FLVS (P = 0.49) children. Obesity risk haplotype (1_(C)-1_(G)-1_(T)-2_(T)) did not show evidence of association with adiponectin levels in obese children (P = 0.38). However, this haplotype was associated with higher adiponectinemia in FVLS children (P = 0.03).

We investigated functional proprieties of the ACDC promoter sequences where SNPs −11,377C>G and −11,391G>A are located. We performed electrophoretic mobiilty shift assays (EMSAs) using nuclear extracts from in vitro–differentiated adipocytes. Promoter wild-type probe (−11,377C and −11,391G) was shifted by a nuclear factor (Fig. 1B). Increasing amounts of wild-type unlabeled probe gradually unhooked the labeled probe showing specific interaction between DNA and nuclear proteins. To examine the effect of SNPs −11,377C>G and −11,391G>A on binding affinity, we performed EMSAs using probes that differed only on SNP position followed by autoradiogram quantification. The band shift with the −11,377G allele probe was 2.5-fold less intense than with the wild-type probe, which suggests decreased interaction with ACDC promoter in the presence of this allele. No band shift was observed for the −11,391A probe, indicating that this allele totally impedes the DNA/adipocyte nuclear protein interaction observed for the wild-type probe. We assessed the transcriptional activity of the ACDC promoter according to SNPs −11,377C>G and −11,391G>A through luciferase tests (Fig. 1C). The −11,377G allele had lower transcriptional activity than the wild-type probe, which includes the −11,377C obesity-associated allele, but this effect was not significant. In contrast, the −11,391A promoter had a 2.5-fold increase in transcriptional activity (P = 0.002), which was in accordance with the association between this allele and higher adiponectin levels observed in children. Genomatix analyses (available from http://www.genomatix.de/) failed to predict any potential transcription factor binding sites, suggesting that −11,377C>G and −11,391G>A may locate in a novel cis element recognized by putative nuclear repressors.

We performed an EMSA on the intronic sequence harboring SNP +276G>T and obtained a specific interaction with adipocyte nuclear factors, but the +276T allele had no effect on this interaction (data not shown).

ACDC SNPs have been reported as risk factors for type 2 diabetes in several populations. This study is the first to report associations of the adiponectin gene SNPs with both childhood and morbid adult obesity. The −11,377C and +276T allelic frequencies were very similar in both obese children and morbidly obese adults, suggesting, as already seen for SNPs in the GAD2 (14) and ENPP1 (15) genes, that morbid adult and childhood obesity may share part of their genetic background. Although familial association results were not strong and need further confirmation, they pointed out the absence of hidden stratification.

Previously, the alternative alleles −11,377G and +276G were associated with higher risk for type 2 diabetes (6,7), which suggests that severe obesity risk alleles may decrease the risk of the etiology of type 2 diabetes. The physiological mechanisms behind this genetic finding could be through the effect of the ACDC SNPs on adiponectin levels. Our results, and data from other studies, are in favor of the association of obesity risk alleles −11,377C and +276GT with higher adiponectinemia (7,16–18). This finding is surprising, as decreased adiponectin levels were described in obese subjects. However, adiponectin levels in obese insulin-sensitive subjects can be similar to lean insulin-sensitive subjects and significantly higher than in insulin-resistant patients independent of their obesity status (3). The physiological contribution of adiponectin to weight gain may be mediated by its insulin-sensitizing action in adipose tissue. Insulin signaling in the adipocyte is important for lipid storage, and adipose tissue selective insulin receptor knockout mice are protected from obesity (19). In 3T3-L1 preadipocytes, adiponectin overexpression accelerates cell proliferation and differentiation, while in mature adipocytes autocrine adiponectin increases glucose uptake and favors lipid accumulation (20). Transgenic overexpression of adiponectin in the physiological range induced morbid obesity without insulin resistance in ob/ob mice (J.-Y. Kim, P.E.S., unpublished data). As long-term exposure to insulin resistance would limit energy storage and weight gain (21,22), higher adiponectin-induced insulin sensitivity may increase lipid storage in adipose tissue instead of organs like liver and muscle. Consequently, hyperadiponectinemia not only protects from insulin resistance, and eventually from type 2 diabetes, but also may favor additional weight gain and may predispose to obesity. Treatments in type 2 diabetic patients with peroxisome proliferator–activated receptor γ agonists improve insulin sensitivity but stimulate body fat gain (23). Several studies have reported that peroxisome proliferator–activated receptor γ or its agonists increase adiponectin expression and secretion (24,25), suggesting a putative role of adiponectin in peroxisome proliferator–activated receptor γ–induced fat gain.

Our functional experiments did not show evidence of −11,377C having a role in the modulation of ACDC transcriptional activity. As −11,377C is the wild-type allele, its presence corresponds to the basal transcriptional activity of the promoter. Our experimental design may not have enough power to detect a decrease in activity. We note that we cannot exclude linkage disequilibrium between this SNP and others at the ACDC locus. In contrast, we showed that the −11,391A allele is associated with higher adiponectinemia, probably through enhanced ACDC transcriptional activity. Our results are in agreement with recent data showing that a deletion of the human promoter region where −11,391G>A is located increases ACDC transcriptional activity (26). The lack of association for this rather infrequent SNP with severe forms of obesity is probably due to insufficient power and will require investigation in larger populations. Our EMSA experiments did not show evidence of the effect of the +276G allele. The 3′ untranslated region SNP +2019delA is in linkage disequilibrium with +276G>T in our population (D′ = 0.97, data not shown). This SNP was previously associated with adiponectin levels and explained a 3q27 quantitative trait locus linked to adiponectinemia, suggesting that it may be functionally relevant (18).

In summary, our results suggest that the adiponectin variants predispose to severe childhood and adult obesity, but further investigations are required to determine the physiological mechanisms behind the process.

This work was partly supported by the association 200 Familles pour Vaincre le Diabète et l’Obésité, the Association Française des Diabétiques, an Association Française de Recherche et d’Études sur l’Obésité/Institut Roche de l’Obésité Research Prize, and by the British Medical Research Council.

We thank F. Vasseur, C. Lecoeur, C Wachter, and S. Gaget for statistics and bioinformatics assistance; M.A. Charles and M. Tauber for supplying DNA; and K.J. Ward, M. Ghoussaini, and P. Boutin for manuscript improvements. We thank the DESIR Study Group for supplying genotypes of the control set 2. We are indebted to all subjects who participated in this study.