Association of Transcription Factor 7-Like 2 (TCF7L2) Variants With Type 2 Diabetes in a Finnish Sample

RESEARCH DESIGN AND METHODS

We studied Finnish type 2 diabetic case subjects and normal glucose tolerant control subjects from the Finland-United States Investigation of NIDDM Genetics (FUSION) study (3,4) and from the Finrisk 2002 study, a population-based national risk factor survey in Finland (5). Diabetes was defined according to 1999 World Health Organization criteria (6) of fasting plasma glucose concentration ≥7.0 mmol/l or 2-h plasma glucose concentration ≥11.1 mmol/l, by report of diabetes medication use, or based on medical record review. Normal glucose tolerance (NGT) was defined as having fasting glucose <6.1 mmol/l and 2-h glucose <7.8 mmol/l. We selected 784 unrelated type 2 diabetic case subjects from the FUSION type 2 diabetes affected sibling pair families. We selected control subjects with NGT for the FUSION case subjects from three sources: 140 spouses of FUSION type 2 diabetic individuals, 217 subjects who had NGT by oral glucose tolerance tests at age 65 and 70 years, and 241 individuals from the Finrisk 2002 study sample. From the Finrisk 2002 study sample, we selected an additional set of 367 unrelated type 2 diabetic case subjects and 355 unrelated control subjects with NGT, approximately frequency matched for age, sex, and province of birth. Study protocols for the FUSION and Finrisk 2002 studies were approved by local ethics committees and/or institutional review boards of each participating recruitment or analysis site, and informed consent was obtained from all study participants.

RESULTS

We genotyped the five SNPs originally reported to be associated with type 2 diabetes by Grant et al. (2) in a Finnish sample of 1,151 type 2 diabetic case subjects and 953 control subjects with NGT from the FUSION (3,4) and Finrisk 2002 (5) studies (Table 1). We found significant evidence for type 2 diabetes association for all five of these SNPs (P < 0.05) with the same risk allele as the original report (2) (Table 2 and online appendix Table 1 [available at http://diabetes.diabetesjournals.org]). Our strongest evidence for type 2 diabetes association was for rs12255372, located in intron 4 (OR 1.36 [95% CI 1.15–1.61], P = 0.00026), and rs7903146, located in intron 3 (1.33 [1.14–1.56], P = 0.00042). These two SNPs are separated by 50 kb and were in moderately strong LD in our Finnish control subjects (r² = 0.70, D′ = 0.91) (online appendix Table 2). Our strongest evidence for association was for an additive model (online appendix Table 1), which was consistent with the findings of Grant et al. (2). We combined three separate controls groups, two from FUSION and one from Finrisk, for comparison to the FUSION case subjects. We tested for heterogeneity of allele frequencies among these three control groups and found no significant differences (data not shown). The SNP–type 2 diabetes association OR estimates between the Finrisk and FUSION samples did not differ more frequently than expected by chance given the number of tests performed (online appendix Table 3).

To further investigate the type 2 diabetes TCF7L2 association in this region, we genotyped 12 additional TCF7L2 SNPs that tagged all 63 SNPs in LD of r² > 0.2 or D′ > 0.9 with rs12255372 based on the CEU HapMap sample; these 12 SNPs all were within introns 3 (88 kb in length) and 4 (101 kb in length). Using this tag SNP set, we observed a region of high LD based on D′ that extended ∼56 kb from the middle of intron 3 (rs17747324) to the first part of intron 4 (rs12255372) and contained the two most strongly associated SNPs, rs7903146 and rs12255372 (online appendix Table 2). In our sample, all but 2 of the 12 additional SNPs were in lower r² with rs12255372 than the other 4 SNPs from Grant et al. (2), and the 3 most distal SNPs were in lower LD with rs12255372 than predicted by the CEU HapMap sample (online appendix Table 2). Under an additive model, each of the 12 additional SNPs was less strongly associated with type 2 diabetes (OR ≤1.26, P ≥ 0.0054) than rs12255372 or rs7903146 (Table 2).

Haplotype analysis did not reveal a risk haplotype that explained the association substantially more than any individual SNP. Within a logistic regression model framework, either of the two most strongly associated SNPs was, in our sample, sufficient to explain the observed association, since adding a second SNP did not significantly improve model fit.

Risk allele frequencies for the two SNPs with strongest type 2 diabetes associations, rs12255372 and rs7903146, were 8–12% (10–19%) lower in our Finnish control subjects (cases) than in the Icelandic, Danish, and American (U.S.) control subjects (cases) (Table 3). Allele frequencies for these SNPs did not vary significantly among the 13 historical Finnish provinces in control subjects (P ≥ 0.23, data not shown).

In a linkage genome scan of 737 FUSION type 2 diabetic families, we observed modest evidence for type 2 diabetes linkage on chromosome 10q (logarithm of odds = 0.61) at 131.5 cM on the FUSION linkage map near microsatellite marker D10S1237, ∼1 Mb distal to TCF7L2 (12). Using GIST, we found evidence for an association between the presence of the risk allele in the FUSION type 2 diabetic case subjects and increased allele sharing identity by descent within FUSION type 2 diabetes sibships for the four most strongly associated SNPs by case/control analysis, rs7903146, rs7901695, rs17747224, and rs12255372 (P = 0.036, 0.033, 0.055, and 0.064, respectively), suggesting that any of these SNPs may partially explain the observed linkage signal.

One of the underlying goals of type 2 diabetes genetic research is to identify individuals at higher or lower risk for disease based on the presence or absence of risk variants from multiple genes. Of SNPs that have been tested for type 2 diabetes association by multiple groups, PPARG P12A and KCNJ11 E23K have been shown to be associated with type 2 diabetes in multiple studies (14–16,17) including FUSION (7) (Willer et al., personal communication). We examined the combined risk of type 2 diabetes from variants in PPARG, KCNJ11, and TCF7L2 in our Finnish sample. Within our sample, the ORs from additive models for PPARG P12A (1.27 [95% CI 1.07–1.50]), KCNJ11 E23K (1.23 [1.08–1.40]), and the TCF7L2 rs12255372 allele T (1.35 [1.16–1.38]) did not differ significantly, and thus, we analyzed the data in terms of the total number of risk alleles for these three SNPs (0–6 risk alleles). Each risk allele increased the odds of type 2 diabetes, which approximates the increase in risk, by a factor of 1.26 (95% CI 1.15–1.38). For example, compared with individuals with the median number of three risk alleles, individuals with one risk allele have 0.63-fold–higher risk of type 2 diabetes, and individuals with five risk alleles have 1.59-fold–higher risk of type 2 diabetes (Table 4). When we do not assume equal allele effects but instead calculate the OR for each number of risk alleles compared with the reference count of three risk alleles, we obtain similar OR estimates (Table 4).

DISCUSSION

We found type 2 diabetes association in Finns with the five type 2 diabetes–associated TCF7L2 SNPs identified by Grant et al. (2). Our two most strongly associated SNPs had the strongest evidence of association in the combined Icelandic, Danish, and American (U.S.) samples. The ORs based on our Finnish sample were consistently lower than those reported by Grant et al. (2), but there was no evidence of heterogeneity in the ORs between our Finnish sample and the original Icelandic, Danish, and American (U.S.) samples for any of the five SNPs (P = 0.21−0.45) (Table 2).

We assayed 12 additional TCF7L2 SNPs chosen to tag all known SNPs in LD with rs12255372 and did not find evidence of a more strongly associated variant. However, we have not assayed SNPs that cover the remainder of the gene, and other TCF7L2 variants that increase of risk of type 2 diabetes may exist.

The risk allele frequencies for our two most strongly associated SNPs were substantially lower in Finnish case and control subjects than in the Icelandic, Danish, or American (U.S.) case and control subjects of Grant et al. (2), suggesting that there are underlying differences in the population allele frequencies rather than differences due to case or control sampling criteria.

We found evidence from GIST (13) that the associated variants partially explain the excess allele sharing identity by descent in the region of our modest linkage signal (logarithm of odds = 0.61) (12) on chromosome 10q, again suggesting that a variant or variants that increase the risk of type 2 diabetes are present in this region. However, in the Icelandic sample, these SNPs did not explain the observed linkage signal (2).

We found additive effects on the risk of type 2 diabetes for each additional risk allele from TCF7L2, PPARG (P12A), and KCNJ11 (E23K). Hansen et al. (18) observed additive effects for the number of risk alleles from PPARG (P12A) and KCNJ11 (E23K), and Hattersley et al. (19) observed additive effects for these alleles in combination with alleles from additional risk loci. These SNPs likely represent only a small proportion of the total set of genetic variants associated with the risk of type 2 diabetes; thus, an individual with multiple risk alleles from this particular set of three variants could still have a low risk of type 2 diabetes compared with others in the population. These OR estimates also may be biased relative to those that would be observed in the general population because a large proportion of our case subjects are from type 2 diabetes affected sibpair families and because we have excluded individuals with impaired glucose tolerance or impaired fasting glucose from our control group.

We tested the TCF7L2 SNPs for quantitative trait association, including weight-related traits, blood pressure, fasting levels of insulin, glucose, lipids, and free fatty acids in FUSION case subjects; 2-h oral glucose tolerance test in FUSION control subjects; insulin and glucose in control subjects; and glucose effectiveness, acute insulin response, and insulin sensitivity and disposition index in spouse control subjects (3,4). We found no significant SNP trait associations after taking into account the number of tests. The most significant result was observed in the FUSION spouse group with lower disposition index in individuals with the G (putative risk) allele of rs11196192 (P = 0.0020) (data not shown).

In summary, we have confirmed the association of variants in TCF7L2 with type 2 diabetes observed in the Icelandic, Danish, and American (U.S.) samples of Grant et al. (2). TCF7L2 joins a growing list of transcription factors that are involved in the growth, development, and metabolism of type 2 diabetes and contain genetic variants that increase the risk of type 2 diabetes.