Common Variants of the Novel Type 2 Diabetes Genes CDKAL1 and HHEX/IDE Are Associated With Decreased Pancreatic β-Cell Function

Table 1 summarizes the key phenotypic data for the men and women, confirming that this was a cohort of healthy subjects. The study recruitment methods and inclusion/exclusion criteria have been previously described (7). Briefly, healthy men and women of European ancestry, aged between 30 and 60 years, were recruited from 19 centers in 14 European countries. Subjects with diabetes, hypertension, or dyslipidemia were excluded (7). The analysis presented in this study is based on 1,276 subjects who had passed the eligibility criteria and had completed genotype data. Local ethics committee approval was obtained by each recruitment center, and written consent was obtained from all participants.

Subjects underwent detailed anthropometric assessment and a 75-g oral glucose tolerance test (OGTT) with blood sampling before and 30, 60, 90, and 120 min after the oral glucose load. Fat mass was determined as the difference between body weight and fat-free mass determined by bioimpedance (Tanita International Division, Yiewsley, U.K.). On a separate day within 1 month of the OGTT, subjects underwent a hyperinsulinemic-euglycemic clamp as previously reported (7). To ensure consistency across study centers, the clamp procedure was standardized and each center underwent prestudy training. Clamp data were then transferred and analyzed at the RISC coordinating center (Pisa, Italy) and quality assured against preset criteria. These were as follows: clamped glucose levels within 20% of target (fasting glucose concentration) and coefficient of variation (CV) of ≤15%, as well as avoidance of hypoglycemia (glucose <3.5 mmol/l). Insulin sensitivity was assessed as the mean glucose infusion rate over the last 40 min of the clamp, corrected for the mean plasma insulin levels achieved during the same period (M/I).

Pancreatic β-cell function was assessed using the OGTT data. The 30-min insulin response (8) was calculated as the ratio of the insulin concentration increment to the 30-min glucose concentration (30-min insulin − 0-min insulin/30-min glucose). Indexes of β-cell function parameters were derived from mathematical analysis of plasma glucose and C-peptide, using C-peptide deconvolution (9) and a model previously illustrated (10,11). In short, the model describes the relationship between glucose concentration and insulin secretion (ISR). It includes an ISR versus plasma glucose concentration dose-response curve (modulated by a potentiation factor accounting for effects due to hyperglycemia exposure and incretins) and a parameter describing early insulin release (rate sensitivity, marker of first-phase secretion). The key parameter is β-cell glucose sensitivity, i.e., the slope of dose-response curve. An OGTT-derived measure of insulin sensitivity was also calculated as previously described (12).

Samples were processed and stored locally before transfered to the central assay laboratories and analyzed as previously reported (7). Genomic DNA was extracted using a Nucleon BACC2 kit (Tepnel Life Sciences, Manchester, U.K.). Single nucleotide polymorphisms (SNPs) were selected as those best representing the recently reported type 2 diabetes associations and for which there were robust assays. All SNPs were genotyped by KBiosciences (3,6), except the Pro12Ala variant of PPARG, which was genotyped by PCR–restriction fragment–length polymorphism (13). A haplotype formed by the T-alleles of rs10757283 and rs10811661 in the CDKN2A/2B region was associated with type 2 diabetes more strongly than the individual SNPs (3). We therefore used unphased software (14) to assess the association between this haplotype and the traits under investigation.

Variables with a positive skewed distribution (M/I, 30-min insulin response, and all model parameters from the β-cell model) were log transformed to normalize distributions. These data are presented as the median (interquartile range).

As all of the novel type 2 diabetes risk alleles conformed to an additive inheritance pattern (3), linear trend analysis was first performed to test for associations between SNP genotypes and selected phenotypes after adjusting for age, sex, and recruitment center. If an SNP displayed a nonadditive inheritance pattern for a given phenotype, general linear model analysis was used to test for association. For all analyses, P ≤ 0.01 was considered statistically significant. We calculated that the cohort, for example, had 80% power at P = 0.01 to detect differences of 0.13 and 0.18 SDs per allele for minor allele frequencies of 0.44 and 0.30, respectively.

The genotype frequencies for the eight SNPs and the Pro12Ala variant are summarized in the online appendix supplementary Table 1 (available at http://dx.doi.org/10.2337/db07-0634), and all were in Hardy-Weinberg equilibrium. Although all analyses were corrected for recruitment center, we compared genotype frequencies between subjects from north and south European recruitment centers. There were small but significant differences for rs564398, rs9939609, and rs1801282 (all P < 0.05).

Table 2 summarizes the relationships between the diabetes-associated alleles and the key phenotypic traits, namely whole-body insulin sensitivity (M/I) and two indexes of β-cell function (30-min insulin response and β-cell glucose sensitivity). The key findings are that the diabetes risk alleles for SNPs rs10946398 and rs1111875 of the CDKAL1 and HHEX/IDE gene regions, respectively, showed significant associations with the measures of β-cell function. For CDKAL1, the diabetes risk allele was associated with decreased glucose sensitivity (P = 9.86 × 10⁻⁵) and decreased 30-min insulin response (P = 0.0002). These associations remained significant after adjustment for M/I (P = 0.0005 and 0.006, respectively). For HHEX/IDE, the diabetes risk allele was associated with decreased 30-min insulin response (P = 0.0002) and a trend for decreased glucose sensitivity (P = 0.009). Again, these associations remained after adjustment for M/I (P = 0.0005 and 0.01, respectively). None of the other variants, including the CDKN2A/2B risk haplotype, showed an association with any measures of β-cell function.

We included the PPARG Pro12Ala variant as a “positive control” based on previous reports of an association between this variant and insulin sensitivity (15,16). This observation was replicated by general linear model analysis that showed a borderline significant association between the Ala allele and increased M/I (P = 0.026), which remained after correction for BMI (P = 0.006). In light of the previously reported relationship between FTO and BMI, we specifically examined the relationship between rs9939609 and three measures of adiposity (Table 2). There was an association between the diabetes susceptibility allele and increased adiposity, which was not strongly significant (BMI P = 0.022) but was consistent with the effect sizes previously reported (6). We also found an association between the allele and decreased insulin sensitivity (P = 0.023), but this was no longer evident after correction for BMI. None of the other novel diabetes susceptibility alleles or the CDKN2A/2B risk haplotype showed a significant association with whole-body insulin sensitivity. There were no significant associations between the OGTT-derived measure of insulin sensitivity and the diabetes susceptibility alleles (online supplementary Table 2).

We assessed whether any of the SNPs showed evidence of interacting with BMI to influence 30-min insulin response, β-cell glucose sensitivity, and M/I by regression analysis adjusting for the usual covariates. There was no evidence of interactive effects (all P > 0.05).

Recent GWAs have identified novel type 2 diabetes susceptibility loci within or close to the genes encoding FTO, CDKAL1, CDKN2A/CDKN2B, IGF2BP2, HHEX/IDE, and SLC30A8 (1–6). The purpose of this study was to examine the relationships between the diabetes-associated alleles and measures of β-cell function and whole-body insulin sensitivity.

Three key findings have emerged from this study. First, the type 2 diabetes–associated alleles of the SNPs located within the CDKAL1 and HHEX/IDE gene regions were associated with impaired β-cell function. Both these SNPs showed clear associations with increased type 2 diabetes risk in GWA studies of subjects of European ancestry, with odds ratios (ORs) of 1.12 and 1.13 for CDKAL1 and HHEX/IDE, respectively (3–5). For each gene, the type 2 diabetes susceptibility allele was associated with decreased 30-min insulin response and decreased β-cell glucose sensitivity. These relationships remained after correction for whole-body insulin sensitivity (M/I), in keeping with primary defects of β-cell function rather than secondary changes in response to altered insulin sensitivity. It is worth noting, however, that these are not direct measures of insulin secretion but provide indirect indexes of β-cell function. The measure of β-cell glucose sensitivity represents the mean slope of the dose-response relationship between insulin secretion determined from C-peptide kinetics and plasma glucose concentration. Decreased β-cell glucose sensitivity therefore indicates decreased insulin secretion over the glucose concentration range achieved during the OGTT. We recently investigated predictors of progression to abnormal glucose tolerance in a high-risk nondiabetic cohort and found that decreased glucose sensitivity was the strongest independent predictor when all the other model parameters and the early insulin response were included (10). Our results support those of Steinthorsdottir et al. (2) who showed that variation within CDKAL1 was similarly associated with type 2 diabetes and decreased early insulin response. As previously reported (2–5), the role of CDKAL1 in pancreatic β-cell function remains unknown, although we found that it is strongly expressed in human adult pancreatic islets relative to other tissues (3).

The susceptibility allele at rs1111875 within the HHEX/IDE gene region was shown to be associated with type 2 diabetes in both our own and other GWA studies (1–5). HHEX is highly expressed in pancreatic islet tissue (3), and failure of ventral pancreas development is a feature of HHEX-null mice (17). Conditional knockdown of HHEX in pancreatic β-cells will help to define its role in β-cell differentiation and function. It is important to note, however, that the association signal is a region of extended linkage disequilibrium that includes other genes, including IDE and KIF11, that encode the insulin-degrading enzyme and kinesin-interacting factor, respectively (3).

We have found no relationship between the indexes of β-cell function and the susceptibility alleles in the other novel gene regions (CDKN2A/CDKN2B, IGF2BP2, and SLC30A8 genes). This is somewhat surprising, as some of the risk alleles had ORs close to those of CDKAL1 and HHEX/IDE (3). Other possible explanations are that these genes do not impact upon the phenotypes that we have examined or, perhaps, that the effect of the risk alleles is strongly age dependent, as ours is a relatively young cohort (Table 1).

The second finding is that we confirmed that the FTO susceptibility allele is associated with increased adiposity, and this was consistent across all three measures of adiposity (BMI, waist-to-hip ratio, and fat mass). The FTO susceptibility allele was also weakly associated with decreased insulin sensitivity, but this was explained through the increased adiposity.

The third finding is that none of the other novel susceptibility alleles were associated with altered whole-body insulin sensitivity. This is despite the fact that the RISC cohort is one of the largest collections of healthy subjects, all of European ancestry, to be assessed for whole-body insulin sensitivity using the gold standard hyperinsulinemic-euglycemic clamp technique. We acknowledge, however, that the exclusion criteria may have served to compress the distribution of insulin sensitivity across the cohort and limited the scope to detect associations between this phenotype and the diabetes susceptibility alleles. Nonetheless, we were able to show that homozygous Ala allele carriers of the PPARG Pro12Ala variant have greater whole-body clamp-derived insulin sensitivity compared with Pro allele carriers, in keeping with the findings of a similar study that assessed whole-body insulin sensitivity by hyperinsulinemic-euglycemic clamp (16). This would suggest, therefore, that if any of the other novel diabetes susceptibility variants do influence insulin sensitivity, then their effects must be comparatively weak and will require a larger cohort to be detected.

In conclusion, we have shown that the CDKAL1 and HHEX/IDE susceptibility alleles are associated with decreased pancreatic β-cell function and have confirmed the relationship between variation in FTO and increased adiposity. However, none of the other novel diabetes susceptibility alleles were associated with altered whole-body insulin sensitivity.

The RISC Study is supported by European Union Grant QLG1-CT-2001-01252 and AstraZeneca. L.P. is the recipient of a joint Biotechnology and Biological Sciences Research Council and Unilever U.K., Ltd. case PhD studentship.