Introduction

Type 2 diabetes (T2D) is a complex disorder characterized by hyperglycemia that results from impaired pancreatic β-cell function and decreased action of insulin on its target tissues. T2D is one of the most common polygenic disorders, in which each genetic variation has a partial and additive effect.1 Over 50 susceptibility loci for T2D have been identified through genome-wide association studies (GWAS),2, 3, 4 most of which were conducted in European populations. KCNQ1, UBE2E2 and C2CD4A/B are among the few T2D loci discovered in non-European populations.5, 6, 7

Recently, new genetic loci were found to be associated with fasting glucose homeostasis in a meta-analysis of 21 GWAS of European non-diabetic subjects (the Meta-Analyses of Glucose and Insulin-related traits Consortium: MAGIC).4 The contribution of these new loci should be investigated further, because it is well known that there are significant differences in the frequencies of some genetic variations among different ethnic groups. Possibly because of the modest size of the effect of most variations, however, it is hard to replicate the results of studies that have analyzed large samples in other populations. In fact, only a handful of loci (CDKAL1, CDKN2A/B, IGF2BP2, TCF7L2, SLC30A8, HHEX, KCNJ11 and IRS-1) have been confirmed to be associated with T2D in Japanese populations.6, 8, 9, 10

In the meantime, arguments against the common disease–common variant hypothesis have been proposed, suggesting that rare variants with a large effect have a substantial role in the genetic architecture of common diseases. So far, we only have limited examples of rare variants conferring a markedly increased risk of common diseases. It has also recently been proposed that a very large number of common variants with effects that cannot be detected by studies employing a conventional sample size may be involved in the genetic architecture of common diseases, which is the so-called ‘infinitesimal’ hypothesis.11

Even if individual small effects are not detectable by a single SNP (single-nucleotide polymorphism) analysis, we may be able to detect their cumulative effect by combined SNP analysis.

Accordingly, the aim of the present study was to determine whether the loci identified in the MAGIC study comprise part of the genetic component of diabetes-related traits and T2D in a Japanese population by using combined SNP analysis.

Materials and methods

Cases and controls

Written informed consent was obtained from all the participants. We enrolled 1200 subjects with T2D aged 65.5±9.5 years (body mass index (BMI): 24.3±3.7 kg m−2 (mean±s.d.)) who attended the University of Tokyo Hospital (Tokyo, Japan). The control group consisted of 855 persons aged 69.5±6.8 years (BMI: 24.5±3.2 kg m−2) who participated in an annual health check conducted by the Hiroshima Atomic Bomb Casualty Council Health Management Center (Hiroshima, Japan). We also enrolled 722 subjects with T2D aged 64.9±11.1 years (BMI: 24.6±3.9 kg m−2) and 763 controls aged 66.7±10.6 years (BMI: 22.7±3.3 kg m−2) who attended the University of Toyama Hospital (Toyama, Japan) and 710 patients aged 60.9±10.2 years (BMI: 24.1±3.7 kg m−2) and 432 controls aged 75.2±8.0 years (BMI: 21.5±3.6 kg m−2) from the University of Kobe (Kobe, Japan).

Diabetes was diagnosed according to the World Health Organization criteria.12 Exclusion criteria included positivity for antibody to glutamic acid decarboxylase or diabetes secondary to (i) liver dysfunction, (ii) steroids or other drugs that might elevate glucose levels, (iii) malignancy or (iv) any monogenic disorder known to cause diabetes. Inclusion criteria for the control group were as follows: (i) HbA1c<6.2%, (ii) fasting plasma glucose <7.0 mmol l−1 and (iii) postprandial plasma glucose <11.1 mmol l−1. HbA1c was estimated according to the method of the Japan Diabetes Society and converted to the corresponding National Glycohemoglobin Standardization Program (NGSP) value.

Genotyping

We genotyped SNPs that showed a significant association with diabetes-related traits or phenotypes in the MAGIC study. SNPs with a minor allele frequency of less than 0.05 in the Japanese population (HapMap JPT) were excluded from this study. We also excluded SNPs for which the association with T2D has already been reported in samples from Tokyo or Toyama.7, 13, 14, 15 As a result, a total of nine SNPs (rs340874, rs4607517, rs2191349, rs7034200, rs10885122, rs174550, rs11605924, rs10830963 and rs35767) were genotyped.

SNP genotyping was performed by using the TaqMan assay (Applied Biosystems, Foster City, CA, USA). The success rates of the assay was >95%.

Statistical analysis

In the non-diabetic subjects, the genotype distributions of all SNPs were in Hardy–Weinberg equilibrium (P⩾0.05). Using the non-diabetic subjects, we performed a quantitative trait analysis to assess the association between each SNP and indexes of glucose homeostasis (fasting plasma glucose, fasting insulin, HbA1c, HOMA-β and HOMA-IR). Differences of the genotype distribution between the diabetic and non-diabetic groups were also assessed by logistic regression analysis with an additive model. We performed the above analyses separately on groups from the University of Tokyo, University of Toyama and University of Kobe, and then performed joint analyses.

We analyzed the combined effect of multiple SNPs. Risk alleles were selected from the MAGIC study and we calculated how many risk alleles there were in each sample for the nine SNPs analyzed in this study. Samples with 12 or more risk alleles were classified as the high-risk group, samples with 9–11 risk alleles were classified as the medium-risk group and samples with 8 or fewer risk alleles were classified as the low-risk group. Differences in the distribution of the cases and controls between the risk groups were analyzed. Using the control samples, we analyzed differences of diabetes-related traits between the risk groups. After excluding two SNPs (PROX1 and FADS1) that were significantly associated with T2D in the present study, we also analyzed the combined influence of the remaining seven SNPs. In this analysis, the high-risk group was defined as having 10 or more risk alleles, the medium-risk group had 7–9 risk alleles and the low-risk group had 6 or fewer risk alleles. We assessed the differences in the prevalence of T2D by using a logistic regression model and diabetes-related traits by using a linear regression model. We calculated odds ratios per one risk-group increase for T2D and effect sizes per 1 risk-group increase for diabetes-related traits. Effect sizes and s.e. were calculated for groups from the University of Tokyo, University of Toyama and University of Kobe, after which combined analysis was performed.

As dependent variables in the linear regression models, we used log-transformed data for fasting insulin, HOMA-IR and HOMA-β, whereas untransformed data were used for fasting plasma glucose and HbA1c.

Multiple corrections were done by the Benjamini–Hochberg method.16 A false discovery rate (FDR) of 0.05 was employed as the significance threshold.

These analyses were performed with R software (http://www.R-project.org).

Results

We found that 2 SNPs (rs340874 and rs174550) were significantly associated with T2D (Table 1). In the case of rs340874 (PROX1), a minor allele was associated with an increased risk of T2D (P=0.0078, FDR-adjusted P=0.035, OR: 1.12, 95% CI: 1.03–1.22). For rs174550 (FADS1), a major allele was associated with an increased risk of T2D (P=0.0071, FDR-adjusted P=0.035, OR: 1.12, 95% CI: 1.03–1.22). In addition, rs2191349 (DGKB-TMEM195) was nominally associated with T2D (P=0.035, FDR-adjusted P=0.11, OR: 1.10, 95% CI: 1.01–1.21). No associations were observed between T2D and the remaining loci in our study population.

Table 1 Association of the SNPs with type 2 diabetes
Full size table

In the non-diabetic subjects, there was a significant association with HbA1c for rs4607517 (near GCK, P=0.00056, FDR-adjusted P=0.0050) (Table 2). However, none of the loci were significantly associated with fasting plasma glucose, fasting plasma insulin, HOMA-IR or HOMA-β (Supplementary Tables 1–4).

Table 2 Association of the SNPs with HbA1c
Full size table

The higher risk groups had an increased risk of T2D (P=0.0017, OR: 1.17, 95% CI: 1.06–1.29) in the combined analysis (Table 3). It is noteworthy that the higher risk groups had an increased risk of T2D even after excluding the two SNPs (near PROX1 and at FADS1) that were significantly associated with T2D in the individual SNP analyses (P=0.011, OR: 1.14, 95% CI: 1.03–1.27) (Table 3). The higher risk groups had higher fasting plasma glucose levels and HbA1c levels (P=0.018 and P=0.013, respectively) (Table 4). The higher risk groups had a lower HOMA-β (P=0.033), but no differences of HOMA-IR or fasting plasma insulin levels were found between the risk groups.

Table 3 Type 2 diabetes in the risk groups
Full size table
Table 4 Diabetes-related traits in the risk groups.
Full size table

Discussion

In the present study, we analyzed nine SNPs that were previously shown to be associated with diabetes-related traits by the MAGIC study. We found that a SNP of FADS1 and a SNP near PROX1 were significantly associated with an increased risk of T2D in a Japanese population. We also identified an association between a SNP of GCK and elevation of HbA1c in a non-diabetic Japanese population.

To our knowledge, this is the first study to identify the association of T2D with a SNP at FADS1. In the MAGIC study, FADS1 was associated with fasting plasma glucose levels within the physiological range, but not with pathological glucose levels.4 However, this was not the case in the present study because FADS1 was associated with susceptibility to T2D in our Japanese population.

The present study is the first to identify a SNP near PROX1 related to susceptibility to T2D in a Japanese population. The contribution of a SNP near PROX1 to the pathogenesis of T2D has now been demonstrated across different ethnicities because the risk allele in the present study was directionally consistent with those found in other populations.4, 17

In the combined SNP analysis, the higher risk groups had an increased risk of T2D. This suggests that these SNPs had an additive effect on the risk of T2D. Unexpectedly, an increased risk of T2D was also found even after we excluded the two SNPs that were significantly associated with T2D. Subjects with multiple risk alleles in the MAGIC study had a higher fasting plasma glucose, higher HbA1c and lower HOMA-β, even though few or none of their SNPs were significantly associated with diabetes-related traits in the individual analyses. These results suggest that SNPs with an effect that is too small to detect in individual SNP analyses have a role in the genetic architecture of T2D and support the infinitesimal hypothesis. It seems that we are able to detect their cumulative effect by combined SNP analysis.

In this study, correction for multiple testing was done by the Benjamini–Hochberg method. This controls the FDR and is less conservative than methods that control the family-wise error rate (FWER), such as Bonferroni’s correction. The seven SNPs with no significant association in the individual analyses had a significant influence on T2D in the combined analysis, which suggests that methods controlling the FWER may be too conservative and methods controlling the FDR may be preferable in such studies. Similarly, the standard GWAS threshold of 5 × 10−8 may be too strict. It could be useful to calculate a genetic risk score by using all SNPs beyond the nominal statistical thresholds in GWAS and determine whether the genetic score predicts the phenotype in another sample, but further investigation of this point is needed.

The higher risk groups had lower HOMA-β, but there were no differences of HOMA-IR or fasting plasma insulin levels between the risk groups. This was consistent with the results of the MAGIC study, which showed that most of the nine SNPs were associated with HOMA-β and only two SNPs were associated with HOMA-IR or fasting insulin.

We could not replicate all of the MAGIC study findings. One possible reason is that the linkage disequilibrium (LD) relationships of disease-susceptibility alleles may differ between populations. Because the LD block in the CRY2 regions is much larger in CEU than in JPT according to the HapMap database, it is possible that the causative SNPs under investigation are in different LD blocks in the Japanese population but are in the same LD in an European population. The fact that Japanese subjects with T2D tend to be leaner and less hyperinsulinemic than Europeans may be one of the reasons why SNPs associated with fasting insulin levels or HOMA-IR were not replicated.

Because our sample size did not have enough power to detect susceptibility genes with modest effect sizes, it cannot be denied that the other loci investigated in the current study could have some role in predisposing Japanese to T2D. In the quantitative analysis of the association between each SNP and diabetes-related traits except for HbA1c, only 1344 samples were included, which is one of the reasons we could not replicate the associations between SNPs and diabetes-related traits, except for that between GCK and HbA1c.

In conclusion, the present study was the first to demonstrate that a SNP of FADS1 contributes to susceptibility to T2D. We also identified a significant association between T2D and a SNP near PROX1. Furthermore, our results indicated that analysis of SNPs with a small contribution to disease risk could capture more of the genetic background of T2D than assessment of only the significant SNPs.