Introduction

Obesity is a major risk factor for type 2 diabetes. Genome-wide association studies (GWAS) have identified 38 loci convincingly associated with BMI in populations with European, Asian and African ancestry [1–3]. Recent studies in Europeans have demonstrated that the genetic susceptibility to obesity leads to increased risk of type 2 diabetes, through its obesity-increasing effect [4]. However, these studies did not include the most recently identified loci. Moreover, the BMI-associated loci may act differently in Asians and Europeans, since their adiposity phenotype and genetic background differ. For example, the FTO locus confers type 2 diabetes risk through its effect on BMI in Europeans, whereas in East Asians, the association remains after adjusting for BMI [5]. Therefore, it is of interest to test whether the genetic predisposition to obesity also contributes to type 2 diabetes risk through its effect on BMI in Han Chinese individuals. In this case–control study of 3,712 unrelated Han Chinese, we examined the association of a genetic risk score (GRS), based on 38 BMI-associated single-nucleotide polymorphisms (SNPs), with type 2 diabetes and glycaemic traits, and whether BMI mediated the association.

Methods

Study population

Our analyses included 3,975 individuals (1,999 type 2 diabetes cases and 1,976 controls) from four studies. Full details of study populations, design, protocols and biochemical measurements have been described previously [6]. BMI was calculated as weight/height2 (kg/m2). Indices of insulin sensitivity (HOMA-S) and beta cell function (HOMA-B) were estimated by the homeostasis model using Levy’s computer model (www.dtu.ox.ac.uk/homacalculator/download.php, 20 February 2014). The criteria for type 2 diabetes cases and non-diabetic controls are shown in the electronic supplementary material (ESM) Table 1. All studies were approved by local ethic committees, and informed written consent was obtained from all participants.

Genotyping

After excluding the samples (n = 263) who had call rate <97% and less than three SNPs with missing genotypes, 1,873 cases and 1,839 controls were included in this study (ESM Table 2). To date, 38 BMI-related loci have been identified in GWAS with p < 5 × 10−8, of which eight are monomorphic in Han Chinese individuals [1–3]. Therefore, a total of 30 BMI-associated SNPs representing independent loci were included in this study. Even though the vast majority of these SNPs were first identified in populations of Europeans, most SNPs have also shown associations with BMI in East Asian populations [1, 2]. Yet, because of differences in linkage disequilibrium structure between East Asians and other ancestry populations, the selected SNPs may not always be the ones that show the strongest association in East Asians and thus could lower the statistical power. Genotyping and imputation have been described elsewhere [6]. Briefly, DNA samples were genotyped using the Illumina Human660W-Quad BeadChip (Illumina, San Diego, CA, USA). All genetic variants passed initial quality-control criteria with a call-rate ≥95% and genotype distribution in Hardy–Weinberg equilibrium (p > 1 × 10−3). The IMPUTE software (version 2.1.2; http://mathgen.stats.ox.ac.uk/impute/impute.html) was used to impute SNPs from the phase 2 HapMap CHB+JPT (release 22) reference panel with high quality (Proper_info >0.8; ESM Table 3).

Statistical analyses

Due to the non-normal distribution, values of insulin and HOMA-S were natural log-transformed before analyses. We calculated the unweighted GRS of each individual by summing the number of BMI-increasing alleles [1–3] of the 30 SNPs. The weighted GRS was calculated by weighting each BMI-increasing allele with the ratio of its reported β-coefficient to the sum of all SNPs’ β-coefficients [1–3], and summing these values. Missing genotypes were replaced with the average BMI-increasing allele number for each SNP. Logistic and general linear regression models were applied to test the associations with type 2 diabetes risk and glycaemic traits, respectively, with adjustment for age, sex, region and BMI (where appropriate) under additive genetic model. The power calculations were performed using Quanto software (http://biostats.usc.edu/Quanto.html, version 1.2.4, May 2009) with minor allele frequencies and mean values of BMI from the current study and the effect sizes reported previously [1–3]. All p values were two-sided, with nominal significance defined as p ≤ 0.05. Bonferroni correction was used to adjust for multiple testing (i.e. five tests for the GRS and 150 tests for individual SNP analyses). As such, associations of which p < 0.01 for GRS and p < 0.0003 for single-SNPs were considered significant at a 5% α-level. Statistical analyses were performed in R version 2.15.0 (www.r-project.org/, 30 March 2012; University of Science and Technology of China, Hefei, China).

Results

The unweighted GRS was significantly associated with increased BMI in both non-diabetic controls (β [SE] 0.061 [0.022]; p = 0.0059) and the whole samples (β [SE] 0.070 [0.016], p = 1.33 × 10−5; ESM Fig. 1). Individually, 22 of the 30 SNPs showed directionally consistent associations with increased BMI in controls (p = 0.0081 for binomial test), but only LRRN6C-rs10968576 showed significant association with BMI in the whole samples after Bonferroni correction (β [SE] 0.352 [0.090]; p = 8.98 × 10−5; ESM Table 4).

Each additional BMI-increasing allele of the unweighted GRS was significantly associated with a 1.029-fold increased risk of type 2 diabetes (95% CI 1.008, 1.050; p = 0.0056), adjusted for age, sex and region, and further adjustment for BMI only slightly attenuated the association (OR 1.022; 95% CI 1.002, 1.043; p = 0.035; Table 1). Individually, BMI-increasing alleles of 21 SNPs were in the direction of associations with increased type 2 diabetes risk without adjustment for BMI (p = 0.021 for binomial test) and four were nominally associated with increased type 2 diabetes risk (ESM Table 5). However, the BMI-increasing allele of CDKAL1-rs9356744 was significantly associated with decreased risk of type 2 diabetes (OR 0.72; 95% CI 0.66, 0.95; p = 2.63 × 10−11) with or without adjustment for BMI (ESM Table 5). After removing the six SNPs that showed associations with type 2 diabetes individually, the association between the GRS and type 2 diabetes risk remained unchanged, suggesting that the association was not driven by one or a few loci.

Table 1 Association between the BMI GRS and risk of type 2 diabetes
Full size table

The GRS was nominally associated with lower HOMA-B after adjusting for BMI (β [SE] −0.876 [0.345]; p = 0.011; Table 2). Individually, eight SNPs showed nominal associations with glycaemic traits, but only the association between GALNT10-rs7708584 and HbA1c value remained significant after Bonferroni correction (ESM Table 6). Notably, further adjustment for HOMA-B abolished the association between GRS and type 2 diabetes (OR 1.012; 95% CI 0.986, 1.039; p = 0.380; Table 1) and excluding the participants who were receiving glucose-lowering treatment did not change the results (OR 1.01; 95% CI 0.93, 1.10; p = 0.79). Similar results were observed for the weighted GRS (ESM Table 7). The triangular relationship between the GRS, BMI and type 2 diabetes also suggested that the association might be mediated through impaired beta cell function (ESM Fig. 2).

Table 2 Associations of BMI GRS with glycaemic traits in non-diabetic controls
Full size table

Discussion

In this type 2 diabetes case–control study of Han Chinese, we found that each additional BMI-increasing allele in the GRS was associated with about a 3% increased risk for type 2 diabetes, independent of BMI. In the non-diabetic controls, the BMI GRS was associated with lower HOMA-B after adjusting for BMI. The association of the GRS with type 2 diabetes was abolished after controlling for HOMA-B.

Similarly, previous studies in European populations have found that each additional BMI-increasing allele increased the OR of type 2 diabetes by 3–4% [4]. However, in Chinese Hans, unlike Europeans, the effect of GRS on type 2 diabetes seemed to be mediated by HOMA-B rather than by BMI. One possible explanation for this discrepancy is that BMI might represent a somewhat different adiposity phenotype between Asians and white Europeans, since for a given BMI, East Asians and South Asians have a higher percentage of body fat than white Europeans. Moreover, due to differences in healthcare between China and western countries, it is possible that Chinese are diagnosed with type 2 diabetes at a later age. Consequently, they may have lost weight by the time of diagnosis and BMI may underestimate their adiposity level before the onset of type 2 diabetes.

Our results also showed that BMI GRS was nominally associated with decreased HOMA-B, which was independent of BMI. Moreover, analysis of the triangular relationship between the GRS, HOMA-B and type 2 diabetes, using GRS as an instrumental variable, also suggested that HOMA-B mediated, at least partially, the association between the GRS and type 2 diabetes. Although previous findings from GWAS of type 2 diabetes emphasised the importance of beta cell function in the development of type 2 diabetes [7], the exact mechanisms underlying the effects of the BMI loci on HOMA-B and type 2 diabetes remain unknown.

Individually, most SNPs showed directionally consistent effects on type 2 diabetes risk, as in Europeans [4], and only the BMI-increasing allele of CDKAL1-rs9356744 was significantly associated with decreased risk of type 2 diabetes. Consistently, a recent GWAS of BMI in East Asians also found that the BMI-decreasing allele of this SNP was associated with increased type 2 diabetes risk [1].

Our study had sufficient statistical power (≥80%) to detect associations of the GRS with BMI and type 2 diabetes risk, but it generally lacked power to replicate previously reported associations with BMI for most SNPs individually. Therefore, our primary analyses focus on the GRS analyses. Nevertheless, large-scale meta-analyses in East Asians have confirmed the single-SNP associations for most of the SNPs studied here [1, 2].

In summary, we concluded that the GRS for obesity also confers type 2 diabetes risk in Han Chinese individuals, and that its effect on type 2 diabetes is partly mediated through impaired beta cell function but is independent of BMI.