Introduction

Type 2 diabetes is a complex disease; multiple genes are involved in its onset and development. To date, a number of genes have been reported to be associated with type 2 diabetes. Most of the genes are investigated because they are presumed to be relevant to the pathogenesis of type 2 diabetes based on the function of the gene. However, because the whole picture of the pathogenesis of type 2 diabetes is still to be clarified, this ‘candidate-gene approach’ is limited in power to detect novel disease-susceptibility genes.

Due to the recent development of single nucleotide polymorphism (SNP) typing technology and accumulation of the information on the linkage disequilibrium (LD) of the human genome, whole-genome association studies have now become a feasible way of searching for a novel disease-susceptibility gene across the whole genome without prior information as to the function of the gene. Indeed, since the original discovery of the powerful type 2 diabetes gene, transcription factor 7-like 2 (T cell specific, HMG-box) (TCF7L2), by a joint group from Iceland, the USA and Denmark [1], many groups have confirmed the significant association between type 2 diabetes and this gene in several populations [2–4]. Following the detection of TCF7L2, a number of groups, including French-Canadian [5], Icelandic [6], UK [7, 8], Finnish [9] and Finnish–Swedish [10], have described new type 2 diabetes-susceptibility genes obtained by whole-genome association study. These genes are solute carrier family 30 (zinc transporter), member 8 (SLC30A8), haematopoietically expressed homeobox (HHEX), exostoses (multiple) 2 (EXT2), CDK5 regulatory subunit associated protein 1-like 1 (CDKAL1), fat mass and obesity associated (FTO), cyclin-dependent kinase inhibitor 2A (melanoma, p16, inhibits CDK4) (CDKN2A), cyclin-dependent kinase inhibitor 2B (p15, inhibits CDK4) (CDNK2B) and insulin-like growth factor 2 mRNA binding protein 2 (IGF2BP2), which have not been well-documented and thus not thought to be good candidates for type 2 diabetes genes.

It is well known that frequencies of genetic variations are different among ethnic groups, leading to differences in the effect and importance of the same susceptibility gene according to the different ethnic groups. It is especially important to investigate this in populations in Eastern Asia, including Japan and China, where the number of individuals with type 2 diabetes is increasing rapidly. Therefore, we have investigated whether the confirmed type 2 diabetes variants in these genes in Europid populations are also associated with type 2 diabetes in participants from the Japanese population.

Methods

Participants

We performed an association study in 864 Japanese type 2 diabetes individuals and 864 Japanese non-diabetic individuals. The clinical characteristics are described in Electronic supplementary material (ESM) Table 1. The diabetic patients were randomly recruited from among those attending the outpatient clinic of the Department of Metabolic Diseases, Graduate School of Medicine, University of Tokyo (Tokyo, Japan), and the non-diabetic individuals from among those undergoing annual health check-ups at the Hiroshima Atomic Bomb Casualty Council Health Management Center (Hiroshima, Japan) [3]. Details of the inclusion and exclusion criteria of our case–control samples are described in the ESM. In order to evaluate the possibilities of population stratification in our case–control samples, we chose 50 SNPs from the database of International HapMap project that are not in LD with each other and genotyped them in 382 Tokyo cases and 382 Hiroshima controls. Then, we investigated whether there was inflation of the number of SNPs with p < 0.05 for association between the 50 SNPs and type 2 diabetes. As described in ESM Table 2, there was no evidence of inflation of the number of SNPs with p < 0.05. The number of SNPs with p < 0.05 was 2, whereas the expected number of SNPs with p < 0.05 was 2.35 (47 × 0.05, three SNPs were not variants in the Japanese population). All the participants gave informed consent, and the Ethics Committee of the University of Tokyo approved this study.

Selection of SNPs genotyped in this study

We selected all the SNPs in genes that were repeatedly confirmed as type 2 diabetes-susceptibility variants in the recent six association studies [5–10]. SNPs that reside in a gene which were reported only in a single study were not included in this study, e.g. SNPs in LOC387761, chromosome 4 open reading frame 32 (FLJ39370, also known as C4orf32) and protein kinase N2 (PKN2). SNPs that we had previously reported were also excluded, e.g. SNP rs7903146 in TCF7L2 and rs1801282 in peroxisome proliferator-activated receptor gamma (PPARG). As for CDKAL1, SNP rs7754840 was not included as it was only 216 bp downstream of rs10946398, and these were in LD with each other. SNP rs5219 of potassium inwardly-rectifying channel, subfamily J, member 11 (KCNJ11) was also excluded because it was within 1 kb of rs5215. All together, 16 SNPs were investigated in this study.

Genotyping SNPs

Among the 16 SNPs, FTO rs9939609 was genotyped by direct sequencing, performed with a BigDye Terminator (Applied Biosystems, Foster City, CA, USA) and resolved using an ABI 3700 automated DNA sequencer (Applied Biosystems). The rest of the SNPs were genotyped using Taqman SNP genotyping assays by means of an ABI 7900HT (Applied Biosystems) according to the manufacturer’s protocol. Genotyping success rates ranged between 97.7% (rs4402960) and 100% (rs9939609). Concordance rate, based on duplicate comparisons in 192 control participants and 192 type 2 diabetes patients, was 100%. Samples with ambiguous base calling were genotyped twice. All the SNPs were in accord with Hardy–Weinberg equilibrium in type 2 diabetes participants (p > 0.09) and control participants (p > 0.17).

Evaluating LD

The r 2 values between SNPs were estimated using a software package based on the expectation-maximisation algorithm, SNPAlyze version 3.5.0 (Dynacom, Tokyo, Japan). The LD structure was plotted using GOLD software (http://www.sph.umich.edu/csg/abecasis/GOLD/).

Biological measurements

Insulin resistance and pancreatic beta cell function were quantified using homeostasis model assessment (HOMA-IR and HOMA-beta, respectively); HOMA-IR = fasting insulin (pmol/l) × glucose (mmol/l)/22.5 × 6 (correction for μU/l in original formula) and HOMA-beta = fasting insulin (pmol/l) × 20/glucose (mmol/l) − 3.5 × 6 (correction for μU/l in original formula) as widely used [11]. Data were expressed as means±SD. Details for the methods of other biological measurements are described in the ESM.

Statistical analysis

The proportions of genotypes or alleles were compared between type 2 diabetic and non-diabetic participants using a multiple logistic regression analysis adjusted for age and sex. Differences in HOMA index according to genotypes were determined by analysis of covariates in non-diabetic participants, after adjustment for age and sex. The statistical analyses were performed using JMP for Windows version 6.00 software (SAS Institute, Cary, NC, USA). Values of p were corrected by Bonferroni adjustment and p < 0.0031 (0.05 divided by 16, the total number of SNPs studied), was considered significant. The allele-specific odds ratio (OR) was assessed by counting the number of risk alleles for each individual, and we used the number of risk alleles to predict case/control status using logistic regression.

Results

The association results of the type 2 diabetes-susceptibility variants in the present study are shown in Table 1. Among the 16 SNPs from eight genes, all three SNPs, rs5015480 [OR = 1.46 (95% CI 1.20–1.77), p = 2.0 × 10−4], rs7923837 [OR = 1.40 (95% CI 1.17–1.68), p = 2.0 × 10−4] and rs1111875 [OR = 1.30 (95% CI 1.11–1.52), p = 0.0013] in HHEX were significantly associated with type 2 diabetes in the Japanese population after adjustment for multiple testing. The OR for SNP rs5015480 was the highest in the present study. The risk alleles were all minor alleles and the associations were in the same direction as the previous studies [5, 8–10]. We determined the LD pattern in this region and found that this region was separated into two LD blocks in our samples; one containing insulin degrading enzyme (IDE) and kinesin family member 11 (KIF11), and the other containing HHEX (ESM Fig. 1). All three SNPs showed nominal association with BMI (rs5015480 TT vs TC vs CC, 24.0 ± 0.15 vs 23.3 ± 0.24 vs 22.1 ± 0.92, p = 0.011; rs7923837 AA vs AG vs GG, 24.0 ± 0.16 vs 23.3 ± 0.23 vs 23.9 ± 0.70, p = 0.037; rs1111875 TT vs TC vs CC, 24.1 ± 0.17 vs 23.4 ± 0.20 vs 23.3 ± 0.49, p = 0.042). When we stratified the participants into obese (BMI ≥ 25 kg/m2) and non-obese (BMI  < 25 kg/m2) participants, each SNP showed a higher OR in the obese participants [OR = 1.91 (95% CI 1.34–2.75), p = 5.0 × 10−4; OR = 1.58 (95% CI 1.15–2.17), p = 0.0049; OR = 1.51 (95% CI 1.16–1.98), p = 0.0025, respectively] (ESM Table 3) than in non-obese participants. The associations were negative or only nominally significant in participants with BMI  < 25 kg/m2 (p = 0.029, p = 0.011 and p = 0.080, respectively).

Table 1 Association results for confirmed type 2 diabetes-susceptibility variants in the Japanese population
Full size table

Among the two SNPs in the FTO gene, only rs8050136 showed a nominal significance in terms of the association with type 2 diabetes [OR = 1.22 (95% CI 1.03–1.46), p = 0.025]. rs8050136 was not associated with BMI (CC 23.8 ± 0.16 vs CA 23.7 ± 0.23 vs AA 23.8 ± 0.49, p = 0.99), and the nominal association between rs8050136 and type 2 diabetes was not strengthened after correction for BMI [OR = 1.22 (95% CI 1.03–1.46), p = 0.025 adjusted for age, sex and BMI]. The other SNP in the FTO gene, rs9939609, was identified as a type 2 diabetes-susceptibility variant that predisposes to diabetes through an effect on BMI in the UK population [7]. However, we could not detect any significant association between rs9939609 and type 2 diabetes or with BMI (TT 24.1 ± 0.17 vs TA 24.6 ± 0.22 vs AA 24.3 ± 0.62, p = 0.17). No other clinical parameters related to type 2 diabetes such as HbA1c, fasting glucose, fasting insulin, HOMA-IR, and HOMA-beta were associated with rs9939609.

SNPs in CDKAL1, CDKN2B and SLC30A8 genes showed nominally significant association with type 2 diabetes. As for EXT2, IGF2BP2 and KCNJ11, we could not detect any SNPs that were significantly associated with type 2 diabetes, but the associations were in the same direction as the previous reports in the European populations.

When we tested the SNPs for quantitative trait association in non-diabetic participants, we found several SNPs to be nominally associated with pancreatic beta cell function as estimated by HOMA-beta. Participants with the risk allele of rs7756992 in CDKAL1 (p = 0.023), rs10811661 in CDKN2B (p = 0.0083), rs11037909 (p = 0.0353) and rs3740878 (p = 0.022) in EXT2, and rs1470579 (p = 0.0446) and rs4402960 (p = 0.0096) in IGF2BP2 had lower HOMA-beta values (Fig. 1). However, when taking the multiple comparisons into account, none of the SNPs reached significance. As for HOMA-IR, none of the SNPs was associated with HOMA-IR.

Fig. 1
figure 1

Association between variants in CDKAL1 rs7756992 (*p = 0.023) (a), CDKN2B rs10811661 († p = 0.0083) (b), EXT2 rs3740878 (‡ p = 0.022) (c), and IGF2BP2 rs4402960 (§ p = 0.0096) (d) with pancreatic beta cell function as estimated by HOMA-beta index. Associations between SNPs representing the four genes and the indices of pancreatic beta cell function are shown. Associations were assessed in non-diabetic participants

Full size image

Discussion

Recent reports have revealed novel type 2 diabetes-susceptibility genes, such as SLC30A8, CDKAL1, CDKN2A/2B, IGF2BP2, EXT2, HHEX and FTO, in the European population [5–10]. In this study, we confirmed the significant association of HHEX with type 2 diabetes in the Japanese population. The OR values of the three SNPs genotyped in HHEX (1.20–1.46) were higher than those of the European population (1.20). However, there was a noticeable difference in the frequencies of the risk alleles; 0.24–0.33 in the Japanese populations and 0.65 in Europids, which was also observed for the TCF7L2 gene variants in our previous report [3]. We observed a nominal inverse association between the SNPs in HHEX and BMI in the control participants (p = 0.011–0.042). Adjustment for BMI did not strengthen the association between SNPs in HHEX and type 2 diabetes. Haplotype analysis of the HHEX gene did not strengthen the association, either (p = 0.0013). However, when we performed a stratified association study, the ORs grew higher in the group of obese participants. This may indicate that the risk alleles of HHEX do not predispose to diabetes through an effect on BMI, but it may predispose to diabetes under obesity-inducing circumstances. Sladek et al. [5] reported that multiple SNPs within an extended region containing IDE, KIF11 and HHEX were associated with type 2 diabetes due to preserved LD in this region. In our samples, LD in this region was not relatively preserved and HHEX represents one block of strong LD (ESM Fig. 1), suggesting that an SNP in HHEX is the functional SNP conferring type 2 diabetes susceptibility. HHEX has been reported to be essential for beta cell function and development and is a target of the wingless-type MMTV integration site family (WNT) signalling pathway, as is TCF7L2 [5].

FTO, CDKAL1, CDKN2B and SLC30A8 showed nominal association with type 2 diabetes. FTO was identified as a type 2 diabetes-susceptibility variant that predisposes to diabetes through an effect on BMI in the UK population [7]. As to possible reasons for not detecting association between BMI and FTO SNPs in the Japanese population, this may be due to the fact that our samples were from a less obese Japanese population compared with Europids. Indeed, mean BMI of participants with the risk genotype in UK non-diabetic controls is 25.4–27.1 kg/m2, whereas that of Japanese non-diabetic controls is 24.3 kg/m2. The Japanese population has similar prevalence of type 2 diabetes to Europids, whereas the prevalence of obesity in the Japanese population is much lower, indicating that the Japanese population has a predisposition to type 2 diabetes under a lower burden of obesity compared with the Europid. Therefore, it is possible that the present study could not detect the effect of the FTO gene on BMI in the Japanese population, a less obese population than the Europid.

SNP rs7756992 in CDKAL1 is the SNP that is reportedly associated with type 2 diabetes in Han Chinese individuals from Hong Kong [6]. The minor allele frequency (MAF) of rs7756992 in the Han Chinese was higher than that in the European-origin population and the MAF in the Japanese was very similar to that of the Han Chinese. We detected a nominally significant association between this SNP and Japanese type 2 diabetes [OR = 1.20 (95% CI 1.03–1.39), p = 0.017]. Given that CDKAL1 and CDKN2B showed nominal association with impaired pancreatic beta cell function as estimated by HOMA-beta index in the present study, and that they may well have some role in transduction of glucose toxicity or regenerative capacity of pancreatic beta cells [6], they could be good candidate genes for Japanese type 2 diabetes. EXT2 and IGF2BP2 also showed nominal association with impaired pancreatic beta cell function, but we could not detect any significant association with type 2 diabetes.

The absence of significant association is often attributed to lack of power. The present study had 80% power to detect an OR of 1.20, when the frequency of a risk allele was 30%. However, the power to detect an OR of 1.12, the smallest OR in the whole-genome association studies [8–10], was 50%. Also, the characterisation of the control sample lacks OGTT data. These facts should be regarded as limitations of the present study.

It has been reported that life-style alteration can reduce the risk of type 2 diabetes, even in individuals carrying the type 2 diabetes-susceptibility variant of TCF7L2 [2]. Frequencies of risk alleles of three SNPs in HHEX were lower in our samples than in European populations but effects of those SNPs on susceptibility to diabetes were stronger. Therefore, genotyping SNPs of HHEX in an individual may be effective for a personalised preventive medicine in Asian populations.

In conclusion, this is the first report to describe a significant association between type 2 diabetes and HHEX in the Asian population. Further replication of these studies in Eastern Asian populations who share a quite similar genetic background are awaited.