Association Analysis of Type 2 Diabetes Loci in Type 1 Diabetes

As described in our previous report (14), two European-descent samples were studied. The first consisted of 514 type 1 diabetic subjects and 2,027 control subjects (representing the addition of 969 healthy control subjects to the set described by Hakonarson et al. [14] in order to increase statistical power) and 483 complete type 1 diabetes family trios (affected child and both parents). The average age at onset of the type 1 diabetic children was mean ± SD 7.89 ± 4.05. The median age was 8 years, with lower and upper quartiles at 4.6 and 11 years, respectively. All patients were diagnosed under the age of 18 years and treated with insulin since diagnosis, and none have stopped treatment for any reason. Ethnic backgrounds were of mixed European descent. All samples were genotyped on the Illumina Infinium II HumanHap550 array (Illumina, San Diego, CA). The Research Ethics Board of the Montreal Children's Hospital, the Research Ethics Board of the Children's Hospital of Philadelphia, and other participating centers approved the study, and written informed consent was obtained from all subjects.

As shown in Table 1, 13 type 2 diabetes–associated single nucleotide polymorphisms (SNPs) from the 10 type 2 diabetes loci were selected for the type 1 diabetes analysis. They represent either the SNP originally reported as type 2 diabetes associated or a near-perfect proxy. The PPARG SNP rs1801282 is located at the first coding exon of PPAR-γ2 and causes the amino acid change Pro12Ala. The two CDKAL1 SNPs rs4712523 and rs7756992 have an r² = 0.747. The two CDKN2A/B SNPs rs1412829 and rs2383208 have an r² = 0.002. The two IDE/HHEX SNPs rs1111875 and rs7923837 have an r² = 0.744. The type 2 diabetes association of KCNJ11 was found from a non-synonymous SNP Glu23Lys (rs5219) at first. The rs5219 has an r² = 0.9 with rs5215 (8) and therefore is well tagged by rs5215 in GWA studies (8,10). All of the 13 SNPs have a genotyping success rate ≥98.7% and no Mendelian error in the 483 family trios. Only the WFS1 SNP rs10012946 showed borderline nominal significance of divergence from Hardy-Weinberg equilibrium in the control group, which did not reach the significance threshold adjusted for 13 tests (α = 0.00385).

For the case-control cohort, the potential population stratification was corrected using the Eigenstrat algorithm (19) implemented in the Eigensoft version 2.0 software (http://genepath.med.harvard.edu/∼reich/Software.htm). By the principal components analysis of population structure (19), 42 case and 130 control subjects were identified and removed as outliers. Therefore, 472 type 1 diabetic case and 1,897 control subjects were analyzed for genetic association. For the family cohort, the transmission disequilibrium test was performed using the Haploview software (www.broad.mit.edu/personal/jcbarret/haploview). For a joint analysis of the two cohorts, we combined the two z scores weighted by the sample sizes. According to the statistical power calculation for a case-control study with unequal sample sizes proposed by Fleiss et al. (20), the case-control cohort of 472 type 1 diabetic case and 1,897 control subjects has the statistical power equivalent to 756 case vs. 756 control subjects. The family cohort of 483 complete type 1 diabetic trios has the statistical power equivalent to 483 case vs. 483 control subjects. The joint Z score was calculated as:

where Z₁ is of the case-control cohort and Z₂ is of the family cohort. Each Z score is equivalent to the square root of the respective χ² value. A protective or undertransmitted minor allele corresponds to a negative Z score, whereas a risk or overtransmitted minor allele corresponds to a positive Z score.

As shown by our analysis (Table 2), none of the 13 SNPs from the 10 type 2 diabetes loci show statistically significant association. These 13 SNPs have a minor allele frequency range from 0.116 to 0.397. The statistical power of this study to detect an association from each SNP is shown in Fig. 1. Our study had sufficient power to detect an association with OR ≥1.20 for each SNP with different allele frequency. To further increase statistical power, we performed a combined analysis using the publicly available Wellcome Trust Case-Control Consortium (WTCCC) data (supplementary Table 1 [available in an online appendix at http://dx.doi.org/10,2337/db08-0270]). The WTCCC tested 2,000 type 1 diabetic case and 3,000 control subjects for 500 k SNPs (Affymetrix GeneChip) (8). As shown by the association analysis (Table 3), 12 of the 13 SNPs did not show statistical significance in either the WTCCC data alone or the combined analysis with our dataset. SNP rs1412829 in the CDKN2A/B locus met the significance threshold of α = 0.00385 in the WTCCC data (P = 0.002) (OR 0.879 [95% CI 0.810–0.954]) but not in the combined analysis (P = 0.039) (0.929 [0.867–0.995]). CDKN2A and CDKN2B encode two specific inhibitors of cyclin-dependent kinase 4 (CDK4), i.e., p16^INK4a and p15^INK4b, respectively. CDK5 and CDK4 play important roles in β-cell function and proliferation (10), and, as such, the locus is a reasonable functional candidate. Study of much larger cohorts will be needed to evaluate the possibility of a very weak effect in type 1 diabetes.

Our study suggests that the type 2 diabetes loci do not play any obvious role in type 1 diabetes genetic susceptibility. These known type 2 diabetes genes are mainly involved in two mechanisms, i.e., pancreatic β-cell function and peripheral insulin sensitivity. To explore whether these genes may promote the early onset of type 1 diabetes by impairing insulin secretion or insulin sensitivity, we also investigated the age-at-onset difference of different genotypes for each type 2 diabetes SNP marker. As shown by the one-way ANOVA test of age at onset of three genotypes for each SNP (Table 2), no SNP has an obvious effect on the type 1 diabetes age at onset. Unlike type 2 diabetes, type 1 diabetes typically has an acute onset that can be reliably defined.

Both type 1 and type 2 diabetes are complex diseases. With the rapid technological development of functional genomics, distinct molecular mechanisms of the two diseases are being recognized, establishing the basis of different approaches for developing novel preventive or therapeutic strategies for type 1 and type 2 diabetes. In addition, this study highlights the importance of differentiation diagnosis of adult-onset type 1 diabetes from type 2 diabetes. Because type 1 diabetes does not share common genetic susceptibility with type 2 diabetes, it is important to manage different treatment for adult type 1 diabetic patients. Some issues remain for further studies on genetic mechanisms of type 1 and type 2 diabetes. The type 1 diabetes association of CDKN2A/B needs to be confirmed by an independent study with a large sample size. Assuming a multiplicative effects model, an OR of 0.929, and a minor allele frequency of 0.452, a study with 5,790 case and 5,790 control subjects has 80% statistical power to replicate the association at α = 0.05. Both our study and the WTCCC study focused on pediatric-onset type 1 diabetes, and the possibility remains that type 2 diabetes loci may have some effect in adult-onset cases. Finally, the involvement of type 1 diabetes loci in type 2 diabetes genetics needs further investigation, the testing of which will require accurate phenotyping within the clinical spectrum of type 2 diabetes. For example, it will be interesting to study all type 1 diabetes loci in the subset of insulin-resistant, non–insulin-treated, adult-onset cases that are positive for islet autoantibodies (21).

This work was funded by the Children's Hospital of Philadelphia, the Juvenile Diabetes Research Foundation International, and Genome Canada. H.Q.Q. is supported by a fellowship from the Canadian Institutes of Health Research.

We thank all the patients, their parents, and the healthy control subjects for their participation in the study.