Introduction

CDKN2A and CDKN2B are adjacent cyclin-dependent kinase inhibitor genes on chromosome 9p. CDKN2A inhibits CDK4, a key regulator of pancreatic beta cell replication [1–3]. In mice, Cdkn2a overexpression leads to islet hypoplasia and diabetes [4]. Meta-analysis of genotype data from genome-wide association studies (GWASs) in northern Europeans have confirmed that single nucleotide polymorphisms (SNPs) rs10811661 and rs564398 in the CDKN2A/CDKN2B region are type 2 diabetes susceptibility variants, although the combined evidence for rs10811661 \( \left( {p = 7.8 \times 10^{ - 15} } \right) \) is far stronger than that for rs564398 \( \left( {p = 1.2 \times 10^{ - 7} } \right) \) [5–7].

The French-Canadian GWAS [8] obtained nominal association signals for proxies (r 2 ≥ 0.9) of rs10811661 (rs2383208, \( p = 2 \times 10^{ - 4} \)) and rs564398 (rs1063192, p = 0.02), but they were above the stage 1 significance threshold \( \left( {p = 5 \times 10^{ - 5} } \right) \) and therefore were not reported (Sladek et al. [8]). We sought to confirm these findings in a larger study and to further characterise the susceptibility variants at this novel locus. To this end, we performed a detailed association study of HapMap Phase II tag SNPs spanning the CDKN2A/CDKN2B locus in 3,093 French Europids.

Methods

Case–control participants

All participants were of French Europid ancestry. Individuals identified by Sladek et al. [8] to lie outside the HapMap northern and western Europe CEU ancestry cluster were excluded from the study. Type 2 diabetic participants were diabetic patients reporting a fasting blood glucose concentration ≥7.0mmol/l [9]. Normoglycaemic control participants were selected to have a fasting blood glucose concentration <7.0mmol/l [9]. The diabetic patient group was composed of: (1) 372 probands from diabetic families [10], recruited in Lille; and (2) 1,083 patients with a family history of type 2 diabetes mellitus, recruited at the Corbeil-Essonne Hospital. Control participants were composed of: (1) 353 normoglycaemic parents from type 2 diabetic families; (2) 543 participants from a prospective population-based cohort study [11]; and (3) 742 participants selected from the cohort of a large prospective study of insulin resistance in French participants [12]. In total, the case–control cohort comprised 1,455 type 2 diabetic patients (age 60 ± 12 years; BMI 29.0 ± 6.0 kg/m2; 56% men, 44% women) and 1,638 normoglycaemic participants (age 54 ± 13 years; BMI, 24.1 ± 3.3 kg/m2; 43% men, 57% women). Of these samples, 503 patients and 669 control participants were typed in common between the present study and stage 1 of an earlier GWAS [8], which comprised 694 patients and 669 control participants. Informed consent was obtained from all participants and the study was approved by the local ethics committees.

Statistical power

The case–control cohort comprised 1,455 type 2 diabetic participants and 1,638 normoglycaemic participants. At α = 0.05, this sample size provided 87 and 67% power [13] respectively, to detect the variants rs10811661 and rs564398, assuming allele frequencies of 0.80 and 0.62, a disease prevalence of 0.1, heterozygote relative risks of 1.20 and 1.12 [5–7], and a multiplicative model.

CDKN2A/CDKN2B tag SNP selection

The target region for tag SNP selection across the CDKN2A/CDKN2B locus was Chr9:21,947,752..22,009,312 (NCBI36). Using the Haploview [14] implementation of the Tagger algorithm [16], a total of 18 HapMap Phase II multimarker tagging SNPs (http://www.hapmap.org) were required to tag the region at r 2 and minor allele frequency (MAF) thresholds of 0.9 and 0.05, respectively. In addition, the GWAS-identified type 2 diabetes susceptibility SNPs rs10811661 and rs564398 [5–7] were added to the SNP set. Thus, 18 tagging SNPs, plus rs10811661 and rs564398, making a total of 20 SNPs, were tested for association with type 2 diabetes.

SNP genotyping

Genotyping was performed with the Sequenom iPLEX system [15]. SNP genotype frequencies were tested for accordance with Hardy–Weinberg equilibrium using χ 2 analysis. SNPs with a call rate <90%, a Hardy–Weinberg p < 0.05 or those that exhibited poorly defined genotype clusters were disqualified from association analysis.

Statistical analyses

To test for association of CDKN2A/CDKN2B SNPs with type 2 diabetes, χ 2 analysis of allele and genotype counts was performed. Pairwise SNP linkage disequilibrium (LD) values were calculated from the genotype data of the control cohort using Haploview [14]. Quantitative phenotypes were available for 1,539 of the normoglycaemic participants in the control cohort and were log-transformed and adjusted for age, sex and BMI, as appropriate. SNPs were tested for association with adjusted quantitative traits (BMI, fasting glucose, fasting insulin) using SPSS 14.0 (SPSS, Chicago, IL, USA) with the ANOVA test under a codominant model. Quantitative trait p values are presented uncorrected for multiple testing.

Results and discussion

The first generation GWASs for type 2 diabetes [5–7], although successful in identifying susceptibility variants in the CDKN2A/CDKN2B region, provided incomplete coverage of common SNPs [16] spanning the gene locus. In this follow-up study, we set out to achieve an efficient and comprehensive coverage [17] of common variation at the CDKN2A/CDKN2B locus, by using a multimarker tagging approach at r 2 ≥ 0.9. A total of 18 HapMap Phase II tag SNPs spanning the CDKN2A/CDKN2B locus, as well as the distal susceptibility variants rs10811661 and rs564398, were tested for association with type 2 diabetes (Table 1). The genotype counts for all SNPs are presented in the Electronic supplementary material (ESM) Table 1.

Table 1 Association of CDKN2A/CDKN2B SNPs with type 2 diabetes in French Europids
Full size table

We report replication of the strong association between the major allele of rs10811661 and type 2 diabetes (\( p = 3.8 \times 10^{ - 7} \); OR 1.43 [95% CI 1.24–1.64]), previously found in the GWASs [5–7]. In agreement with the GWASs [5–7], our data indicate that rs10811661, which is 124kb upstream of the CDKN2A/CDKN2B locus, is not in LD with any of the tag SNPs spanning the locus (Fig. 1).

Fig. 1
figure 1

Type 2 diabetes association signals and pattern of linkage disequilibrium across the CDKN2A/CDKN2B region. Plot of pairwise SNP r 2 values calculated from control genotype data. Association results (−log10 p value) and RefSeq genes are shown. The plot was drawn to scale using LocusView (T. Petryshen, A. Kirby, M. Ainscow, unpublished software). T2D, type 2 diabetes mellitus

Full size image

The other ‘confirmed’ susceptibility variant, rs564398, did not quite attain statistical significance (p = 0.053; OR 1.11 [95% CI 1.00–1.24]) in the present study. This was not unexpected as our study was underpowered to detect this variant (see Methods section). The status of rs564398 as a bona fide susceptibility variant appears to be largely driven by the Wellcome Trust Case Control Consortium study (13,965 participants, \( p = 1.13 \times 10^{ - 6} \)) since the Finland–United States Investigation of NIDDM Genetics result (4,808 participants, p = 0.039) was similar to ours, while the meta-analysed Diabetes Genetics Initiative data (13,781 participants, p = 0.5) showed up negative for this variant [5–7].

We also obtained several additional nominal association signals (p < 0.05) at the CDKN2A/CDKN2B locus (Table 1). However, only the rs3218018 result (p = 0.002) survived Bonferroni correction for multiple testing (adjusted p = 0.04). This SNP (MAF = 8%) is correlated (r 2 = 0.83) with the CDKN2B exon 1 SNP rs2069422 (MAF = 9%), which also exhibited nominal association (p = 0.006).

The six SNPs nominally associated with type 2 diabetes were also tested for association with the quantitative traits BMI, fasting glucose and fasting insulin. The only association found was with rs3731201 (Table 2). The minor allele of this variant was weakly associated with lower BMI (p = 0.017) and also showed a trend towards lower glucose levels (p = 0.06). However, it was also the minor allele that was associated with type 2 diabetes (p = 0.007). Thus, these associations are physiologically inconsistent and probably indicate that these qualitative and/or quantitative trait associations are false positives.

Table 2 Association of CDKN2A/CDKN2B SNPs with quantitative traits BMI, fasting glucose and fasting insulin
Full size table

In conclusion, our comprehensive association study of common variation spanning the CDKN2A/CDKN2B locus provides further confirmation of the strong association between the distal susceptibility variant rs10811661 and type 2 diabetes in the French population. Further genetic and functional studies are required to identify the aetiological variants at this locus and determine the cellular and physiological mechanisms by which they act to modulate type 2 diabetes susceptibility.