Association of common variants identified by recent genome-wide association studies with obesity in Chinese children: a case-control study

We conducted an association study in two independent study groups, recruited from the urban regions of Beijing, China. The first study group, including 386 obese, 400 overweight and 151 normal-weight individuals, came from the study on Adolescent Lipids, Insulin Resistance and Candidate Genes (ALIR) in nine middle schools of Dongcheng District of Beijing. The second study group, including 319 obese, 318 overweight and 456 normal-weight individuals, was from the Comprehensive Prevention Project for Overweight and Obese Adolescents (CPOOA) with physical exercise and healthy nutrition as instruments in five elementary and middle schools of the Haidian District of Beijing. The ascertainment strategies for the two study groups have been described in detail previously [16,17]. The two studies were approved by the ethics committee of Peking University Health Science Center. Written informed consent was provided by all participants and, in the case of minors, their parents.

Anthropometric measurements, including height and weight, were determined according to standard protocols. BMI was calculated as weight in kilograms divided by the squared height in meters. We used the BMI percentile criteria to define obesity, overweight and normal-weight in children and adolescents, which were determined in a representative Chinese population [18]. According to the criteria (Table 1), the children and adolescents with an age- and gender-specific BMI ≥ 95^th percentile were defined as obese, while those with a BMI between 85^th and 95^th percentile were overweight and those with a BMI between 15^th and 85^th percentiles were normal-weight. Individuals with cardiovascular or metabolic diseases were excluded. The sex- and age-specific BMI standard deviation score (BMI-SDS) was calculated by using the growth reference data of the World Health Organization for children and adolescents aged 5–19 years [19].

The general characteristics of the study samples are shown in Table 2. The study consisted of 2 030 Chinese children, including 607 normal-weight (mean age 12.55 ± 3.04 years, mean BMI 18.77 ± 2.50 kg/m²), 718 overweight (mean age 13.26 ± 2.36 years, mean BMI 23.86 ± 2.20 kg/m²) and 705 obese (mean age 12.85 ± 2.59 years, mean BMI 28.12 ± 3.94 kg/m²).

We selected 40 obesity-related loci identified by five recent GWA studies, with one representative SNP for each locus. Firstly, we selected the 32 SNPs reported by Speliotes et al [4]. Then we selected 4 SNPs (rs121458332 of SDCCAG8, rs13278851 of TNKS/MSRA, rs9568856 of OLFM4, rs9299 of HOXB5), which were identified by two GWA studies that focused on children [5,6]. Additionally, we selected 4 SNPs (rs2206734 of CDKAL1, rs11142387 of KLF9, rs261967 of PCSK1, rs12597579 of GP2), which were associated with BMI in two GWA studies of East Asian populations [7, 8].

Fasting venous blood samples were collected. Genomic DNA was extracted from blood leukocytes by the phenol/chloroform extraction method. Sequenom’s MassARRAY system (Sequenom, San Diego, CA, USA) was applied to genotype the 40 SNPs. Primers, including a pair of amplification primers and an extension primer for each SNP, were designed with SpectroDESIGNER software (Sequenom, San Diego, CA). A multiplex polymerase chain reaction was performed, and unincorporated double stranded nucleotide triphosphate bases were dephosphorylated with shrimp alkaline phosphatase followed by primer extension. The purified primer extension reaction was spotted onto a 384-element silicon chip (SpectroCHIP, Sequenom) and analyzed in the Matrix assisted laser desorption ionization time of flight mass Spectrometry (MALDI-TOF MS, Sequenom). The resulting spectra were processed with MassArray Typer (Sequenom, San Diego, CA).

Multiplex SNP assays designs failed for 6 out of 40 SNPs (rs10968576, rs4771122, rs10150332, rs12444979, rs29941 and rs261967); these were replaced by 6 proxy SNPs (rs16912921, rs9579083, rs10145154, rs6497416, rs29942, rs261966) with strong linkage disequilibrium (r² > 0.80) in populations comparable to the discovery studies in HapMap data Release 24/phase II Nov08 (http://​hapmap.​ncbi.​nlm.​nih.​gov; Han Chinese in Beijing, China (CHB) for rs261966, Utah residents with Northern and Western European ancestry (CEU) for the other 5 SNPs).

As shown in Additional file 1: Table S1, the call rates for 40 SNPs were above 95.0 %. We exclude one monomorphic (rs6497416 of GPRC5B), one triallelic (rs4836133 of ZNF608) and six rare variants with minor allele frequency below 1 % (in all genotyped individuals) from the subsequent analyses resulting in 32 SNPs. In the normal-weight group, 31 of the 32 SNPs showed no evidence for deviations from Hardy-Weinberg equilibrium (HWE; all P > 0.05). For one SNP (rs7138803) near FAIM2 we observed some evidence for a deviation from HWE (P = 0.01) but a double-checking of the genotype data revealed no obvious genotyping artifacts.

The genotype data of the normal-weight group were tested for deviations from Hardy–Weinberg equilibrium using χ² tests (see above). F-statistics (F_ST), a metric representing the effect of population subdivision, was calculated according to the following formula, F_ST  = (P₁-P₂)²/ ((P₁ + P₂)*(2-(P₁ + P₂))), where P₁ is the allele frequency estimate in the population of the discovery study and P₂ is allele frequency estimate based on the total sample of our study [20,21]. A F _ST value ≥ 0.10 indicates large genetic differentiation [22].

Logistic regression was performed to examine the effect of each SNP on risk of obesity or overweight (categorical variable). Linear regression was performed to examine the effect of each SNP allele on BMI-SDS variability. Both logistic regression and linear regression were carried out under a (log)-additive genetic model with adjustment for age, age square, sex and study group (ALIR and CPOOA). For each of the 6 proxy SNPs, the allele which was correlated with the effect allele of the original SNP in the discovery study was defined as the effect allele, while the effect alleles of other SNPs were the same as the discovery studies, for comparing our results with the published data [4-8].

To identify cumulative effects of these SNPs, we created a genetic risk score (GRS) for each individual by summing up the number of effect alleles of the SNPs. We did not weight the risk alleles on the basis of their individual effect sizes because no well-accepted effect sizes were available for each of the SNPs, and it has been shown that weighting of risk alleles may have only limited effects [23]. Again logistic regression was used to calculate odds ratio (OR) of the GRS-32 from all 32 SNPs that met our minor-allele frequency cut-off (see above) for the risk of obesity or overweight. Linear regression was performed to examine the effect of GRS-32 on BMI-SDS variability.

SPSS 18.0 software was used for the above statistical analyses (SPSS, Chicago, IL). In addition to effect sizes estimates (i.e. per allele odds ratios (OR) and 95 % confidence intervals (95 % CI)), we reported nominal two-sided P-values. We applied a nominal significance level of α = 0.05 (two-sided). Adjustment was made for multiple testing using Bonferroni correction for 32 SNPs, i.e. resulting in α_BF = 0.05/32 = 0.00156 (two-sided). Difference in effect size of each SNP between our study and the discovery study was examined by testing heterogeneity with MANTRA software, which was developed by Morris AP [24] for trans-ethnic meta-analysis of genome-wide association studies. P(heterogeneity) is the posterior probability of heterogeneity in allelic effects, which is derived from transethnic meta-analysis. If P(heterogeneity) > 50 %, there is the evidence of heterogeneity in allelic effects between the present and discovery studies [24]. The receiver operating characteristic (ROC) curves comparing normal-weight and obese children were produced by logistic regression, and the areas under the curve (AUC_ROC) from different models were compared by MedCalc software. Based on the published minor allele frequencies (see Table 3) and applying a (log)-additive genetic model, a sample size of 705 obese cases and 607 normal-weight controls has a comparisons-wise power ranging between 0.94-0.99 for a true allelic OR of 1.5 or 0.33-0.64 for a true allelic OR of 1.2 (α = 0.05; two-sided). Accounting for multiplicity these numbers changed to 0.63-0.98 or 0.05-0.20, respectively (α_BF = 0.00156; two-sided). Similarly, analyzing BMI-SDS in a sample of 2 030 children, leads to a comparisons-wise power ranging between 0.51-0.89 for a true allelic β of 0.10 (in units of BMI-SDS) or 0.17-0.36 for a true allelic β of 0.05 (α = 0.05; two-sided). Accounting for multiplicity these numbers changed to 0.12-0.51 or 0.02-0.06, respectively (α_BF = 0.00156; two-sided). These power calculations were performed using Quanto software (University of Southern California, Los Angeles, CA).