Introduction

Linkage studies in Caucasian families with type 2 diabetes have suggested that one or more diabetes susceptibility gene(s) reside within human chromosome 20q12–q13 [1–3]. This region includes a major positional candidate, the transcription factor hepatocyte nuclear factor (HNF)-4α gene whose coding mutations cause MODY1 [4] and more rarely familial monogenic late-onset type 2 diabetes [5]. In both cases, defective insulin secretion was shown in diabetic and not yet diabetic mutation carriers.

The pancreatic beta cell HNF4α7 isoform was shown to be almost exclusively transcribed from a tissue-specific promoter, termed P2, and an alternative exon 1 that in humans are located 45.6 kb upstream of the previously identified P1 promoter [6]. The P2 promoter contains functional binding sites for the MODY transcription factors HNF-1α, HNF-1β and Pdx1/IPF1, and was proposed to serve as a positive cross-regulatory feedback circuit between HNF-1α and HNF-4α in human pancreatic cells [7]. As a strong candidate gene for the linkage signal observed at 20q13, the HNF-4α exons with P1 and P2 promoter sequences were screened for variants in several studies [8, 9]. However, the rare variants identified so far failed to explain the strong evidence for linkage in this region. Two reports have independently demonstrated, in Ashkenazi and Finnish populations, a convincing association with variants adjacent to the HNF-4α P2 promoter [10, 11]; crucially, these variants explain much of the evidence for linkage to 20q13 reported in both these populations. These new findings justify a reinvestigation of the genetic variation in the extended 5′-end of the HNF-4α transcription unit, since it may serve as a large target for transcriptional interference and could possibly explain the linkage to this region in the French families as well.

In the present study, we aimed to define whether the most relevant haplotype tag single nucleotide polymorphisms (htSNPs) near the HNF-4α P2 promoter, rs2144908, rs6031552 and rs2425637, which had been identified in both previous studies and shown to be associated with diabetes status in the Ashkenazi and Finnish samples, also represent a risk of type 2 diabetes in the French Caucasian population. We also evaluated whether the linkage to the 20q13 region observed in the type 2 diabetic French families could be attributed to the families carrying the at-risk allele(s).

Subjects and methods

Subjects

The French Caucasian population studied is composed of two groups of unrelated type 2 diabetic patients diagnosed with diabetes according to WHO criteria or known to be treated for diabetes. The first group consisted of probands from type 2 diabetic families recruited by the CNRS–Institut Pasteur Unit in Lille (Group 1, n=372; sex ratio: 163 men: 209 women; mean age of diagnosis 44.6±11.3 years; mean BMI 26.7±3.8 kg/m2). The second group consisted of diabetic patients examined at the Endocrinology–Diabetology department of the Corbeil-Essonne Hospital (Group 2, n=372, 206 men, 166 women; mean age of diagnosis 44.9±11.8 years; mean BMI 29.5±4.8 kg/m2). MODY and autoimmune subtypes were excluded. Two groups of normoglycaemic control subjects (fasting glucose level <5.6 mmol/l) of French Caucasian origin were analysed. The first had 320 subjects with normal glucose tolerance from the Family collection recruited in Lille (156 men, 164 women; mean age at examination 70±17 years, mean BMI 25.7±4.7 kg/m2). The second group had 366 non-obese and non-diabetic subjects from the French cohort MONICA (154 men, 212 women, mean age at examination 54.8±5.7 years, mean BMI 25.3±4.0 kg/m2).

The type 2 diabetes French families have been described by Vionnet et al. [12]. Informed written consent was obtained from all subjects before participation. The genetic study was approved by the Ethical Committee of Hotel Dieu in Paris.

Genotyping

SNP genotyping was done by an assay based on fluorogenic 5′ nuclease allowing simultaneous amplification and detection of specific SNP alleles. The chemistry of the assay includes allele-specific hybridisation probes conjugated with a fluorophore at the 5′-end, and with minor groove binder ligand and quencher moiety at the 3′-end (7900 HT SNP genotyping platform, Applied Biosystems, Foster City, CA, USA). In order to avoid genotyping errors and to confirm the association seen with one marker, we also used a second assay based on FRET technology (LightTyper and LightCycler technology, Roche, Meylan, France). SNP detection using sequence-specific FRET hybridisation probes was based on the melting-curve profile and derivative peak analysis. The principle of FRET centres on the transfer of energy from one fluorescent molecule (fluorescein) to another fluorescent molecule (LC Red 640). All genotypes were validated by a second independent reading.

Statistical analysis

Deviation of genotype frequencies from Hardy–Weinberg equilibrium was tested using the chi square test. Significance and confidence intervals for type 2 diabetes odds ratios (ORs) were calculated based on case-control chi square analysis. Additionally, we displayed a summary OR calculated with the Mantel–Haenszel method [13]. Homogeneity of the different ORs was assessed by the Breslow–Day test [14]. Both tests and estimations were performed under a fixed effect model. We considered P values of less than 0.05 statistically significant.

We assessed association and linkage with the Family Based Association Test implemented in FBAT, which defines a statistic test that reflects association between a phenotype (T) and a marker value (X). The distribution of S=∑TX (sum over all offsprings in all families of the data set) under H0 was calculated using the distribution of offspring genotype, conditional on the trait T, and on the parental genotype, or the sufficient statistics when parental genotype was not observed. The value S−E(S)/√Var(S) follows a Normal (0.1).

To test association with evidence of linkage, we split the sample of affected sib-pairs into pairs concordant for genotype categories (11, 12, 22, 1X, 2X). If a genotype is responsible for the observed linkage, we expect that the maximum likelihood score (MLS) for the set of concordant pairs, or identical by state for this genotype, which we call MLSg, is significantly higher than what is expected under H0. We then derive a p value through simulation after preserving the linkage information (i.e. the identical by descent status for each pair) and the association (the number of pairs concordant for the genotype).

Results

Based on the linkage disequilibrium (LD) pattern and haplotype block structure defined in the two Ashkenazi and FUSION studies, we tested three htSNPs distributed into separate blocks to capture the genetic variability from the alternative P2 promoter to sequences proximal to the P1 promoter of HNF-4α. All three SNPs, rs2144908, rs6031552 and rs2425637, were found to be associated with diabetes in the FUSION study, whereas association was identified only with rs2144908 in the Ashkenazi samples [10, 11].

All the markers tested conform to Hardy–Weinberg equilibrium (see Table 1, genotype distributions).

Table 1 Comparison of allele frequencies for htSNPs rs1884614, rs2144908, rs6031552 and rs2425637 in the French, FUSION and Ashkenazi type 2 diabetic patients and control subjects
Full size table

A case-control analysis was carried out in 744 diabetic subjects comprising two independent unrelated type 2 diabetic sample sets from the French population and in 686 normoglycaemic control subjects of the same origin. Table 1 shows the comparison of allele frequencies between type 2 diabetic patients and control subjects for the three variants tested in the French study population.

No significant association with type 2 diabetes was seen for two of the three SNPs (rs6031552 and rs2425637) in contrast to earlier reports. With regard to SNP rs2144908, located 1.3 kb downstream of the beta cell promoter P2, we observed in the French samples a trend towards increased prevalence of the minor A allele in control subjects (p=0.022, OR 0.80, 95% CI 0.66–0.97; multiple test p value determined through permutation, p=0.065). This was in contrast to the Ashkenazi and FUSION studies, where the A allele occurred at a higher frequency in diabetic patients (Table 1). In order to avoid a technical problem and to confirm the genotype raw data, we used a second assay (see the genotyping procedure) and obtained a total concordance of genotypes between the two methods, allowing further reassurance that genotyping was correct.

Moreover, an additional SNP, rs1884614, that showed association with type 2 diabetes and strong LD with rs2144908 in both of the previous studies, was tested in all the samples of our study, and found to be also in tight LD with rs2144908 (D′=0.986, r 2=0.951), but showing a less significant p value in the case-control test (p=0.058, Table 1). A different haplotype structure could also explain such divergent findings, but we observed that the pairwise D′ and r 2 values between the four SNPs tested in our study (data available from authors) are similar to those reported in the previous cohorts where association was found [10, 11]. The differences in OR values for rs2144908 between our study and the Finnish/Ashkenazi studies are statistically significant (Interaction Test: chi square=18.6, p=1.61×10−5). Under the fixed effect model, it is unlikely that our observation is a random variation of an OR of 1.35.

Although the structure of the data does not clearly separate the different case and control groups either for ethnicity or BMI, we estimated the Mantel–Haenszel summary OR combining the two D1 vs C1 and D2 vs C2 studies, and found a less significant result for rs2144908: p=0.087, OR 0.789, 95% CI 0.602–1.034, 99% CI 0.553–1.126.

In our previous linkage studies based on two selections of French diabetic families, we found moderate significance of linkage to diabetes with a locus near ADA-HNF4α on chromosome 20q13 (MLS=1.72 and 1.81) [1, 12]. The highest peak near the marker D20S213 lies 3.5 Mb apart from the genomic location of the HNF4α gene (from UCSC Genome Browser, July 2003). FBAT was then used to detect a possible association between diabetes and the htSNPs, but no linkage disequilibrium between any of the three SNPs and the disease was found in the family sample. We also investigated the impact of the htSNPs genotypes on the positive MLS, and all the three MLSg scores were above 5 (Table 2). Nevertheless, it appears that after simulation the increase in lod score was not significantly higher than expected in a sample of pairs that were identical by state. The best p value obtained was p=0.037 for SNP rs2425637 (Table 2), but this result may be spurious given that the case-control association test was not significant.

Table 2 Family-based association test (FBAT) and association with the evidence of linkage
Full size table

Discussion

Taken together, our findings in the French population are different from those of the Ashkenazi and FUSION studies, and do not appear to be explained by different LD structures between populations. Two of the previously associated SNPs did not confer an increased risk of type 2 diabetes in French Caucasians. The case of SNP rs2144908, located in a >10 kb haplotype block spanning the P2 region of HNF-4α, is different: here we found in French Caucasians that the minor A allele, in contrast to the Ashkenazi and Finnish populations, could protect against diabetes. This result is moderated by the Mantel–Haenszel test. Such inverse allelic association between populations was already detected for variants in the glycogen synthase gene on chromosome 19 [15], and more recently for one obesity-associated allele of the SLC6A14 gene on chromosome Xq24, showing an opposite effect in a replication study [16].

Our opposite result could be due to environmental factors and/or different genetic backgrounds, or may be artefactual, but given the A allele frequency (0.20) for rs2144908 we had 70% power at p=0.01 to detect an association with an OR of 1.35 or greater. Nevertheless, if we consider an OR of 1.10, which is the lowest estimation in the 95% confidence interval of the previous studies, the power of our study declines to 7%. As the power calculation shows that our sample is not tailored to detect a small genetic effect and as we note that the multiple test does not reach significance, we cannot rule out a false negative association result. Thus, rather than a true association in an opposite direction, our finding may reflect a genetic effect of smaller magnitude than reported in the original studies. A large meta-analysis involving several well-documented and powered studies will determine whether there is a consistent association between particular SNPs upstream of HNF-4α and type 2 diabetes in several ethnic groups. Furthermore, as it appears that the region immediately upstream of the P2 promoter of HNF-4α might influence type 2 diabetes susceptibility, additional studies will be needed to identify the true at-risk SNP(s) and the diabetogenic effect.