Introduction

Wolfram syndrome (OMIM: #222300) is a rare, progressive, neurological disorder with an autosomal-recessive mode of inheritance, which frequently manifests in childhood [1]. Diabetes insipidus and (non-autoimmune) diabetes mellitus with optic atrophy and deafness are features of the syndrome, giving rise to its alternative name: DIDMOAD.

Positional cloning studies in families with Wolfram syndrome identified linkage peaks on the short arm of chromosome 4 (4p16.1) [2] and mutations in the gene encoding wolframin (WFS1), which maps to that region, have since been shown to cause the syndrome [3].

In a recent report, four common single nucleotide polymorphisms (SNPs) (rs10010131, rs6446482, rs752854 and rs734312 [H611R]) at the WFS1 locus were shown to be convincingly associated with type 2 diabetes in six UK studies and one study of an Ashkenazi Jewish population, which together comprised 9,533 patients and 11,389 controls [4]. In that study, the summary effect estimates [odds ratios (ORs)] for the four WFS1 polymorphisms ranged from 0.90 to 0.92 and all were statistically associated.

In an accompanying report in this issue of Diabetologia [5] from the Diabetes Prevention Program (DPP), three of the WFS1 SNPs (rs10010131, rs752854, rs734312) were tested for association with incident type 2 diabetes. No statistical association was found overall, although when stratified by treatment arm, a modest effect on diabetes incidence was observed in the lifestyle intervention group.

In this study, we attempted to replicate the previously reported associations between WFS1 SNPs and risk of type 2 diabetes in a case–control study from the county of Västerbotten in northern Sweden. We then sought support for the nominal associations with diabetes identified in the north Swedish cohort by conducting an updated meta-analysis of type 2 diabetes case–control studies using published and unpublished data.

Methods

Västerbotten type 2 diabetes case–control study

We identified 1,296 adult type 2 diabetic patients through registries covering the county of Västerbotten in northern Sweden (Fig. 1); as controls we selected 1,412 non-diabetic individuals, who were group-matched for age, sex, examination date and geographic region of residence, from the Västerbotten Intervention Programme (VIP) (Table 1) [6]. Virtually all were European white individuals. Type 2 diabetes was determined using the 1999 diagnostic criteria of the World Health Organization [7]. Participants with fasting capillary glucose concentrations <7.0 mmol/l and no documented history of diabetes underwent a 75 g anhydrous oral glucose tolerance test. Accordingly, control participants were those without a documented history of diabetes and with glucose concentrations below the thresholds for type 2 diabetes [7]. Type 2 diabetes in the patient group was defined by clinical diagnoses. In a subsample of 1,013 cases, additional validation procedures were undertaken such as an independent OGTT or documented treatment with glucose-lowering drugs (Fig. 1). Analyses were repeated in this subsample to verify the genetic associations (see Results). All living participants provided written informed consent. Permission was obtained from the Local Research Ethics Committee of Umeå University, and approval for genetics investigations of this material was granted by the Swedish Data Inspection Board.

Fig. 1
figure 1

Case selection criteria for participants in the Västerbotten diabetes case–control study; aCases in whom the clinician had documented a diagnosis of type 2 diabetes on the medical chart, but where corroboration of this diagnosis through other means such as an independent OGTT or record of glucose-lowering drug treatment was absent; bBMI data were missing in eight individuals; c n varies according to genotype

Full size image
Table 1 Characteristics of participants in the Västerbotten type 2 diabetes case–control study
Full size table

Clinical measurements

The VIP survey is a population-based observational cohort study, in which, since 1985, all residents of the county of Västerbotten who are aged 40, 50 and 60 years have been invited to attend their primary healthcare centre for a clinical examination. Demographic and clinical data used in this report were collected as part of the VIP survey by trained research nurses using a protocol standardised across study centres [8]. Height and weight were measured using a calibrated wall-mounted stadiometer and scales, respectively. BMI was calculated as weight in kg/(height in m)2. Plasma glucose was assayed using fresh capillary plasma on a benchtop analyser (Reflotron; Boehringer Mannheim, Mannheim, Germany). A threshold glucose value of ≤2 mmol/l was used to exclude potentially spurious values, resulting in the exclusion of three control individuals from the dataset.

Genetic analysis within the Västerbotten study

DNA was extracted from peripheral white blood cells. Prior to genotyping, all genomic DNA samples were diluted to 4 ng/μl. SNPs were assayed at the Wellcome Trust Sanger Institute using the Taqman MGB chemistry (Applied Biosystems, Foster City, CA, USA) in accordance with the recommended protocol [9]. Genotyping success and concordance rates were >98 and 100% for all SNPs, respectively.

Additional case–control studies for meta-analysis

To increase statistical accuracy, we added relevant data from recent genome-wide association (GWA) scans of type 2 diabetes. We contacted the relevant investigators of these GWA scans [10–13] and requested summary statistics (OR and 95% CI) for SNPs at WFS1 that were correlated in HapMap at an r 2 of 1.0 with SNP rs10010131, the SNP showing the strongest statistical association in the study described by Sandhu et al. [4]; the correlated SNPs are therefore direct proxies for rs10010131.

Statistical analysis

Statistical analyses were conducted using SAS software 9.1 (SAS Institute, Cary, NC, USA). Hardy–Weinberg equilibrium (HWE) was assessed using the likelihood ratio test with 1 df. Linkage disequilibrium (LD), expressed as r 2, was calculated using Haploview 4.0 (available from http://www.broad.mit.edu/mpg/haploview, last accessed in November 2007). Power calculations were performed using Quanto 1.1.1 (available from http://hydra.usc.edu/gxe, last accessed in November 2007). Conditional logistic regression models were fitted to assess the associations between each of the WFS1 genotypes and type 2 diabetes. Models were adjusted for age, sex and BMI. Glucose variables were log-transformed to correct skewness; anti-logged means and 95% CIs are presented for the respective results. For the descriptive statistics, central tendency and variance are reported as means and SD, respectively. A value of p < 0.05 was considered statistically significant. Meta-analysis of studies was performed using STATA version 8.2 (StataCorp, College Station, TX, USA) and a fixed effects model and inverse-variance-weighted averages of log(ORs) to obtain a combined estimate of the overall OR. Between-study heterogeneity was assessed using the χ 2 statistic.

Results

WFS1 genotype associations in the Västerbotten case–control study

Table 1 shows participant characteristics for the Västerbotten case–control study; patients and controls were generally overweight or obese, middle-aged adults. The ratio of men to women was higher in the patient group.

All SNPs were consistent with HWE (p > 0.1). Minor allele frequencies in control participants for each SNP were: rs10010131 = 0.43; rs6446482 = 0.44; rs752854 = 0.35; and rs734312 = 0.48. The LD between SNPs in control participants here was generally lower than the values reported in other populations [4], ranging from 0.46 to 0.97 (Fig. 2).

Fig. 2
figure 2

Linkage disequilibrium (r 2) between WFS1 genotypes in the control sample of the Västerbotten type 2 diabetes case–control study (n = 1,412)

Full size image

In logistic regression analyses modelling log additive genetic effects, borderline statistically significant associations were observed between SNPs rs10010131, rs6446482 and rs734312 and type 2 diabetes (Table 2); the magnitude and direction of these associations are consistent with the initial study [4]. SNP rs752854 was statistically associated with type 2 diabetes, with the minor allele conveying a protective effect (OR 0.85, 95% CI 0.75–0.96, p = 0.010; Table 2).

Table 2 Effect estimates for each of the WFS1 SNPs in relation to glucose levels or type 2 diabetes in the Vasterbotten type 2 diabetes case–control study
Full size table

All analyses were re-run in the subset of individuals with validated type 2 diabetes diagnoses (n = 1,013). In these models, the point estimates for each of the SNPs were consistent with the full dataset, although the CIs were slightly wider for rs10010131, rs6446482 and rs752854, reflecting the reduced sample size (results not shown).

Results of the meta-analysis of WFS1 genotypes with regard to type 2 diabetes risk

For the meta-analysis, we included data from three of the first five type 2 diabetes GWA scans; a fourth was included in the original study [4]. The characteristics of each of the additional studies included in the meta-analysis have been reported previously [10–12]. We based these effect estimates on SNPs that were highly correlated with SNP rs10010131, which showed the strongest signal in the original report. For Sladek et al. [10], this was SNP rs4416547, typed in 686 type 2 diabetes patients and 669 controls. For the Diabetes Genetics Initiative (DGI) study [11], it was SNP rs10012946, typed in 1,464 patients and 1,467 controls. SNP rs10010131 was available for analysis from the candidate gene panels from the Finland–United States Investigation of Non-Insulin-Dependent Diabetes Mellitus Genetics (FUSION), which comprised 1,160 patients and 1,172 controls [12].

In a meta-analysis of the Västerbotten study and the three additional studies, which comprised a total of 4,606 patients and 4,720 controls, we found strong evidence for a statistical association between variation at WFS1 and type 2 diabetes (OR 0.87, 95% CI 0.82–0.93, p = 4.5 × 10−5; Fig. 3a), providing clear evidence of independent replication. The magnitude of association was highly consistent among studies, with no material heterogeneity among studies (p = 0.17 with 3 df). Finally, we conducted an updated meta-analysis of all 11 published and previously unpublished material. The unpublished data were obtained by contacting the authors of recently published genome-wide scans in which the relevant variants were genotyped (C. Dina, P. Froguel, K. L. Mohlke, C. Willer) [10–12]. As expected, this analysis showed robust evidence for statistical association (p = 5.4 × 10−11; p for heterogeneity = 0.42 with 10 df; Fig. 3b).

Fig. 3
figure 3

a Association between WFS1 SNPs and the risk of type 2 diabetes in previously unpublished studies. b Association between WFS1 SNPs and the risk of type 2 diabetes in all currently available studies. Summary data for SNP rs10010131 were used in the meta-analysis, with the exception of Sladek et al. [10] and DGI [11], where rs4416547 and rs10012946 were substituted as proxy markers. These SNPs are perfectly correlated (r 2 = 1.0) with rs10010131 in HapMap samples

Full size image

Discussion

In this study we assessed the effects of WFS1 polymorphisms on type 2 diabetes in populations from the northern and southern regions of Sweden, northern and western Finland, and France. The purpose of this investigation was to replicate the WFS1 genotype associations with type 2 diabetes that have been reported recently in several UK studies and one of an Ashkenazi Jewish population [4].

WFS1 SNP rs752854 was statistically associated with type 2 diabetes in our northern Swedish study from Västerbotten, the direction and magnitude of the association being consistent with the previous report [4]. Although the remaining three SNPs were not statistically associated with type 2 diabetes, the effects are similar in direction and magnitude to those reported previously. In the original report, rs10010131 was the variant most strongly associated with diabetes, whereas in the Västerbotten cohort, the associated variant was rs752854. Both variants are non-coding. Therefore, it is probable that they tag the true functional locus and the difference in statistical associations between studies is attributable to different genetic substructures of the populations. This possibility is supported by the lower pair-wise LD between SNPs in the Västerbotten case–control study compared with the studies included in the original report. For example, LD in the original study populations ranged between r 2 = 0.75–0.98 for pair-wise comparisons of the four associated SNPs, with the LD between the two SNPs showing the strongest statistical associations with type 2 diabetes being correlated at r 2 = 0.98. In this report, the LD between these two SNPs was similar (r 2 = 0.97), but for the remaining pair-wise comparisons the r 2 value ranged from 0.46–0.97 (Fig. 2). The differences in LD structure across the WFS1 loci within the population of Västerbotten and in the populations from elsewhere that we studied may reflect different admixture patterns. Given the very high genotyping success and concordance rates for the Västerbotten cohort, it is unlikely that genotyping error explains these differences. It is, however, also possible that these differences in LD structure are attributable to statistical fluctuations.

Given the relatively small magnitude of these associations and the nominally significant p values in the Västerbotten study, it is likely that our failure to reproduce the associations for two of the WFS1 SNPs is attributable to insufficient statistical power, rather than the absence of a true effect. In the original report on WFS1 genotypes and type 2 diabetes risk, Sandhu et al. [4] reported ORs of approximately 0.90 to −0.92 per copy of the minor alleles at each locus. We calculated the sample size that would be required to detect these associations. At a level of statistical association of p = 0.05 and a power of 80%, more than 3,000 case–control pairs would be required to detect an association of this magnitude, thus highlighting the need for large studies when seeking to detect associations where the effects on disease are modest.

Wolframin is a 100 kDa transmembrane protein that is expressed in neurons and pancreatic beta cells. The proposed functions of wolframin include the regulation of membrane trafficking, protein processing and calcium homeostasis in the endoplasmic reticulum of neurons and pancreatic beta cells [14, 15]. Disruption of these processes is believed to cause the progressive pancreatic beta cell loss and neuronal degeneration observed in Wolfram syndrome [16].

In the present study, we were unable to assess the effects of the WFS1 genotypes on measures of insulin sensitivity or secretion. However, given the known mechanisms through which wolframin functions on pancreatic beta cells, it is possible that defects in insulin production underlie the genetic associations reported here and elsewhere with type 2 diabetes. In the DPP, nominal evidence of association between WFS1 genotypes and measures of insulin secretion was observed [5]. Future studies focusing on this phenotype may help determine the mechanisms through which common variants at the WFS1 gene and type 2 diabetes are related.

In conclusion, we have replicated the association between SNPs at the WFS1 locus and risk of type 2 diabetes in a Swedish case–control study. Although we only found evidence for a statistical association with SNP rs752854, the direction and magnitude of the associations for the other three SNPs are consistent with previous reports. Furthermore, by undertaking a meta-analysis of additional data from European individuals, collectively comprising up to 12,979 cases and 14,937 controls, we have been able to robustly confirm the association of a second WFS1 locus, rs10010131, with risk of type 2 diabetes.