The search for genes associated with type 2 diabetes mellitus has frequently been complicated by small effect sizes and poor reproducibility between studies. The recent identification of common polymorphisms within the transcription factor 7-like 2 (TCF7L2) gene as risk factors for type 2 diabetes, however, provides fresh hope of unravelling the genetic basis of this complex disease. The association with a microsatellite marker (DG10S478) in intron 3 of TCF7L2 was first reported by Grant et al. 1 in Icelandic subjects and associations with surrogate markers were subsequently confirmed in several other populations 234567891011. Polymorphisms of the TCF7L2 gene demonstrate the strongest and most reproducible association with type 2 diabetes of any gene reported to date; recent large-scale meta-analyses by Florez 12 and Cauchi et al. 13 reveal associations between the TCF7L2 SNP rs7903146 and type 2 diabetes with p values of < 10^-80 and ~10^-140 respectively. The precise mechanisms by which this gene affects diabetes risk are not known, but it is thought to influence insulin secretion 8101114, possibly through its role in the Wnt signalling pathway 715, and may also affect insulin resistance 1011.

The South Asian population is genetically heterogeneous and is comprised of individuals originating from India, Pakistan and Bangladesh. In UK-resident South Asian populations, type 2 diabetes is around 6 times more common than in the indigenous white Caucasian population, affecting over 10% of South Asian adults 16. Despite the high risk of the disease, few studies have attempted to characterise its genetic basis in this ethnic group. Variants of TCF7L2 have been studied in two South Asian cohorts to date; a UK-resident population of South Indian origin 7 and a cohort from Pune, western India 11. The gene has not previously been investigated in any population from the north of the Indian subcontinent. In this study, therefore, we investigated the association of common polymorphisms within the TCF7L2 gene with type 2 diabetes in a well-characterized UK-resident South Asian population of Punjabi ancestry, originating predominantly from the Mirpur area of Azad Kashmir, Pakistan.

Type 2 diabetic subjects (N = 831) were recruited as part of the United Kingdom Asian Diabetes Study (UKADS), a multiple risk factor intervention trial investigating the impact of a culturally-sensitive, enhanced diabetes care package on the risk of cardiovascular disease in South Asian type 2 diabetes patients living in Birmingham and Coventry, UK. All subjects were of Punjabi ancestry, confirmed over three generations, and originated predominantly from the Mirpur area of Azad Kashmir, Pakistan. Diabetes was defined using the WHO criteria 17. Ethnically-matched normoglycaemic control subjects (N = 437) were recruited from the same geographical areas through community screening. Information concerning the relatedness between any subjects in the study was not collected. Normal glucose tolerance was defined as fasting glucose < 6 mmol/l and 2 hr glucose < 7.8 mmol/l on a 75 g oral glucose tolerance test (OGTT). Where OGTT was not feasible, normal glucose tolerance was defined as random blood glucose < 7 mmol/l. Venous blood was collected from each subject after obtaining informed consent and genomic DNA extracted using an adaptation of the Nucleon^® protocol (Nucleon Biosciences, Coatbridge, UK). The study was approved by the East Birmingham Research and Ethics Committee.

The clinical characteristics of the subjects in our study are shown in Table 1. Age of diagnosis, high density lipoprotein (HDL) cholesterol and HbA1c data were available for subjects with diabetes only. Body mass index (BMI), waist circumference and blood pressure measurements were available for both the diabetic group and 252 subjects from the control group.

The pattern and strength of pairwise LD between SNPs (Figure 1) was similar to that seen in the CEPH (CEU) HapMap samples, Utah residents with ancestry from northern and western Europe 19, and other previously published studies involving white Caucasian cohorts 5920.

Statistical power, calculated using the number of subjects successfully genotyped, was greater than 80% for all SNPs with the exception of rs12255372. As the power of the association test for this SNP was 71.1%, we would not have been able to confidently accept the hypothesis of no association if the results had not been significant. Given the results of the present study, however, this finding is largely redundant in the context of our manuscript. It is interesting to note that, despite a smaller control group, both statistical power and the significance of the associations were greater when using a control group age cut-off of 46 years (power ≥ 94% and p ≤ 4.90 × 10^-4 for all variants) compared to an age cut-off of 35 years.

Our results provide important confirmation that variants of the TCF7L2 gene are strongly associated with type 2 diabetes in populations of South Asian origin, with similar effect sizes to those seen in white Caucasian populations. Two of the variants (rs7903146 and rs12255372) have been studied previously in South Asian cohorts 711. Allele frequencies for rs7903146 in our Punjabi population were similar to those reported in these studies, whereas the minor allele of rs12255372 appeared to be slightly more common in our cohort compared with these Indian populations.

Our data support the multiplicative model of inheritance for the TCF7L2 SNPs proposed by Grant et al. 1, except for the rs12255372 variant. This may be due to the minor departure from Hardy-Weinberg equilibrium observed in the control group for this SNP. All samples were re-genotyped for rs12255372 using the same technique but with fresh reagents. To rule out the possibility that the observed excess of minor allele homozygotes in the control group was due to systematical genotyping error, all subjects in this group previously genotyped as TT were genotyped by RFLP, using primers designed with the web-based tool SNP Cutter 21 and the enzyme Tsp509 (details available upon request). The rationale behind this latter method was that any individual falsely genotyped as TT by the TaqMan method would be genotyped as CT or CC by the RFLP method. Neither method revealed any inconsistencies with the initial genotyping. It seems likely, therefore, that the minor departure from Hardy-Weinberg equilibrium is purely due to chance sampling. It is worth noting that Humphries et al. 7 found no statistically significant evidence for a multiplicative model of inheritance for this SNP in a South Indian population.

A recent paper 22 has shown that a TCF7L2 haplotype (HapB_T2D), characterised by the presence of the rs7903146 T allele, is associated with decreased BMI in subjects with type 2 diabetes. In this study we observed a similar trend, replicated for the minor allele of all variants, although this did not reach statistical significance. Unlike Helgason et al., however, we observed this trend within both the case and control groups. More data will be needed to investigate this relationship further.

Data concerning the relatedness between subjects in the cohort were not available in this study. It is therefore possible that some individuals are genetically related to others in the cohort, leading to non-independence of genotypes. Such cryptic relatedness potentially increases the likelihood of obtaining a false positive result in tests of genetic association, and as such must be considered as a possible limitation of our study. Given the documented reproducibility of the association between TCF7L2 and type 2 diabetes, however, we do not believe that cryptic relatedness has affected the results presented in this manuscript.

Our study highlights the importance of implementing minimum age thresholds when selecting control subjects for genetic studies of type 2 diabetes. Not implementing these thresholds may produce an artificially low OR, as young individuals in the control group may develop type 2 diabetes later in life and are therefore not true controls. In our study this resulted in reduced statistical power (increasing the likelihood of false negative results), a factor which could become increasingly important when studying risk factors with smaller effect sizes. The effect that a control group minimum age threshold has on an association test will depend on the prevalence of disease in the studied population. The marked effect reported in this manuscript is probably due to the relatively high prevalence of type 2 diabetes in UK-resident South Asian populations, as young control subjects have a greater chance of later developing the disease. In other populations, such as those of white Caucasian origin, these thresholds may have a smaller effect.

In conclusion, our results add to the rapidly expanding body of evidence that implicates TCF7L2 as an important risk factor for type 2 diabetes in multiple ethnic groups.

Prof. A. H. Barnett has received honoraria and fees for advisory work, together with research funding from all the companies acknowledged in the relevant section of this manuscript. He and his family hold no stocks or shares in any of the companies. None of the other authors declare any competing interests.

SDR participated in the genotyping of samples and drafting of the manuscript, and performed statistical analysis. SB conceived of the study and participated in the genotyping and drafting of the manuscript. ACB performed DNA extraction. MAK participated in the drafting of the manuscript. All authors contributed to the intellectual content of the paper, and have read and approved the final manuscript.