Introduction

Despite convincing evidence from twin and family studies that type 2 diabetes has a strong genetic component, unravelling the genetic predisposition has proved frustrating. Our laboratory and many others have reported significant associations in genes identified by positional cloning efforts and/or strong candidacy, but replication of these findings has been difficult to achieve. The most widely accepted type 2 diabetes susceptibility variants, Pro12Ala in the gene encoding peroxisome proliferator-activated receptor gamma (PPARG) [1] and Glu23Lys in the beta cell potassium channel gene potassium inwardly-rectifying channel, subfamily J, member 11 (KCNJ11) [2, 3] confer only a modestly increased risk of type 2 diabetes in Europids. Recently, Grant et al. [4] identified a microsatellite in intron 3 of the transcription factor 7-like 2 gene (TCF7L2) that was strongly associated with type 2 diabetes, with a combined p value in Icelandic, Danish and US populations of 10−18 and a combined relative risk (RR) of 1.56. The single nucleotide polymorphisms (SNPs) rs12255372 and rs7903146 were in strong linkage disequilibrium with the microsatellite and showed similarly strong associations. Subsequent to this initial report, numerous published studies have examined this gene, with consistent replication among multiple cross-sectional Europid populations [5–10] and increased risk in several prospective studies [11–13]. These studies have resulted in extraordinarily low p values, odds ratios (ORs) of 1.4 to 1.9 and a risk of 1.4 per risk allele in Europid cohorts from prospective studies. Although no association was reported in African American [11] or African Caribbean [13] subsets, recently Helgason et al. [14] reported that SNP rs7903146 was associated with type 2 diabetes in families from west Africa (p = 0.0002, RR 1.45).

TCF7L2 binds to β-catenin downstream of wingless MMTV integration site family (WNT) to regulate proglucagon gene expression in gut endocrine L cells, but reportedly not in pancreatic alpha cells [15]. Other studies report that TCF7L2 is widely expressed, including in mature pancreatic beta cells as well as in peripheral and omental adipocytes [8]. The WNT pathway is important in adipogenesis [16]; thus, a broader physiological role for TCF7L2 in glucose homeostasis is plausible. Both Florez et al. [11] and Saxena et al. [6] reported reduced insulin secretion in response to oral glucose in carriers of the risk allele. Damcott et al. [10] examined 48 unrelated Europid individuals by frequently sampled IVGTT and found a significant association with a reduced insulin sensitivity index (SI) and the beta cell compensation for insulin resistance (disposition index [DI]), but only a marginal reduction in acute insulin response to glucose (AIRg). Munoz et al. [17] reported reduced AIRg and DI in the subset of 138 Europid and 118 African American women who were homozygous for the TT genotype at SNP rs12255372, albeit with statistical significance only in Europids and with no association for SNP rs7903146. In contrast, Scott et al. [7] reported no association with fasting insulin, glucose tolerance, SI or DI in Finland–US Investigation of NIDDM Genetics (FUSION) families, while Cauchi et al. [18] found only non-significant trends towards reduced fasting insulin and homeostatic estimates of insulin secretion (homeostatic model assessment [HOMA]-B).

We sought to further explore the role of TCF7L2 in type 2 diabetes and prediabetic traits by testing the association of previously reported variants in 932 individuals from Europid and African American case–control populations and in 693 members of Europid families ascertained for multiple diabetic siblings. To examine the physiological impact of TCF7L2 variants, we examined 585 non-diabetic participants who had undergone frequently sampled IVGTT, including members of families ascertained in Utah and unrelated Europid and African American participants ascertained in Arkansas. Finally, we examined TCF7L2 expression in adipose tissue, muscle tissue and transformed lymphocytes by genotype. Our results suggest ethnic differences in the risk of type 2 diabetes conferred by TCF7L2 alleles. We provide support for a role of TCF7L2 variants in insulin sensitivity and impaired glucose homeostasis in Europids, but we did not find significant evidence for genotypic effects on gene expression in three different tissues.

Methods

Study participants

The study populations are summarised in Table 1. We tested for an association of TCF7L2 variants with type 2 diabetes in a northern European (Europid) population primarily ascertained in Utah and in an African American population ascertained primarily in Arkansas. For both studies, diabetic participants were previously diagnosed and on pharmacological therapy or had clearly diabetic glucose tolerance tests. All diabetic patients had at least one first-degree relative with type 2 diabetes. Control participants had normal glucose tolerance tests or fasting or post-meal glucose levels below 5.6 mmol/l. Controls were selected for no family history of type 2 diabetes in parents, siblings or grandparents. Metabolic effects of TCF7L2 variants were examined in three non-diabetic study populations. Non-diabetic members of northern European Utah families ascertained for at least two diabetic siblings underwent tolbutamide-modified, frequently sampled IVGTT, as described previously [19]. Europid participants and African American non-diabetic individuals ascertained in Arkansas were studied using either a tolbutamide-modified frequently sampled IVGTT (122 Europid, 69 African American) as described previously [20] or, because tolbutamide became unavailable during the study, an insulin-modified test (0.04 U/kg of insulin at 20 min) using a similar protocol [21, 22] (217 Europids, 88 African Americans) (Table 2). Arkansas participants were ascertained for variable family history of type 2 diabetes and representing a range of weight, age at study, and glucose tolerance from normal to impaired [23, 24]. Participants underwent two visits: a standard 75 g 2 h OGTT with sampling at baseline, 30 min, 60 min, 90 min and 120 min for insulin and glucose and a frequently sampled IVGTT. A subset of 144 Europid and 60 African American participants underwent determination of the maximum insulin response to arginine (AIRmax). At the conclusion of the frequently sampled IVGTT, the glucose was increased over 10 min to 23 to 25 mmol/l, which was maintained for 30 min. At 30 min of hyperglycaemia, a 5 g intravenous arginine bolus was given over 30 s. Insulin response to arginine was measured at baseline (25 mmol/l glucose) and at 2, 3, 4, 6, 8 and 10 min.

Table 1 Summary of study populations
Full size table
Table 2 Descriptive statistics of unrelated, non-diabetic participants for metabolic studies
Full size table

Family-based analyses for type 2 diabetes, lipids and glucose homeostasis traits based on 75 g OGTTs were conducted in 673 members of 68 Utah families for which both genotypic and phenotypic data were available, as summarised in Table 1 and reported previously [25, 26]. A subset of non-diabetic participants from Arkansas who had had metabolic studies also underwent fat and muscle biopsies [27], including 28 participants with impaired glucose tolerance [23]. All participants provided written informed consent under approved Institutional Review Board protocols of either the University of Utah Health Sciences Center or the University of Arkansas for Medical Sciences.

Genotyping

We typed rs7903146 and rs12255372 using pyrosequencing on a pyrosequencer (PSQ-96; Biotage, Uppsala, Sweden). Primer design was modified to permit use of a biotinylated universal primer. More than110 duplicate samples were included across all studies with 100% agreement for both SNPs. The microsatellite marker DG10S478 [4] was typed by using infrared dye labelled M13 primer on a sequencer (GR4200; LI-COR Biosciences, Lincoln, NE, USA) and scored using GeneImage IR software (version 3.5.6; Scanalytics, Fairfex, VA, USA). All primer sequences are available on request. Based on strong linkage disequilibrium between the SNPs and the dichotomised microsatellite [4], only the SNPs were typed for family and metabolic studies.

Gene expression

RNA was extracted from Epstein–Barr virus-transformed lymphocytes grown under standard culture conditions using either Trizol (Invitrogen, Carlsbad, CA, USA) or RNEasy (Qiagen, Valencia, CA, USA). Adipose RNA was extracted using kits (RNeasy Lipid Tissue Minikit; Qiagen). Muscle RNA was extracted using RNA kits (Ultraspec; Biotecx Laboratories, Houston, TX, USA). Reverse transcription was performed using MultiScribe Reverse Transcriptase (Applied Biosystems, Foster City, CA, USA). Because of multiple splice forms for TCF7L2 [28], we initially designed primer sets for both the 5′ (exons 1 and 2/3) and 3′ (exons 12 and 13/14) ends of the gene (Electronic supplementary material [ESM] Table 1, ESM Fig. 1). However, the 5′ primer set gave a single band in all tissues tested and detected all documented splice forms. Hence, we report only the results using this assay. Similar to findings in a previous report [8], we detected measurable levels of TCF7L2 message in muscle but at 10-fold lower levels than in adipose. Real-time quantitative RT-PCR was conducted either on an ABI 7500 (Applied Biosystems) or on a Rotorgene 2000 (Corbett Life Science, Sydney, NSW, Australia) using SYBR Green (Applied Biosystems). All assays were normalised to 18S RNA.

Laboratory measures

Insulin was measured by the General Clinical Research Center Core Laboratory using an immuno-chemiluminometric assay (MLT Assay, Cardiff, Wales, UK). Plasma glucose was measured by using a glucose oxidase method.

Statistical analysis

Allele frequencies for the case–control study were compared using the Armitage Trend and Fischer Exact tests. Hardy–Weinberg equilibrium was tested separately in diabetic and control participants using the online DeFinetti program (http://ihg.gsf.de/cgi-bin/hw/hwa1.pl). Linkage disequilibrium and haplotype associations were tested in Haploview 3.32 [29]. SI was estimated from glucose and insulin measures during the frequently sampled IVGTT using the MINMOD [30] or MONMOD Millenium programs [31]. AIRg was calculated as the mean insulin response above basal from 2 to 10 min following the glucose bolus for the Utah family sample or as the AUC for the same time for Arkansas Europid and African American samples. DI was calculated as SI × AIRg. Glucose and insulin AUCs (AUCglucose, AUCinsulin) were calculated from the trapezoidal rule. Genotypic effects on glucose homeostasis traits (SI, AIRg, DI, AUCglucose, AUCinsulin, insulinogenic index) and obesity traits (BMI, WHR, per cent fat) were tested using mixed effect, general linear models. Skewed variables were ln-transformed to normality prior to analysis. All models of glucose homeostasis traits included age and BMI as covariates and sex and genotype as fixed factors. Additionally, diagnosis and pedigree membership were included as fixed factors in analyses of family members from the Utah study. We adjusted for protocol type (tolbutamide or insulin) by including a numeric code as a fixed factor for participants studied in Arkansas [32]. Analyses of obesity traits were adjusted for age and sex. The significance of associations detected in the model was examined by comparison of marginal means using the least significant difference test. Based on the strong prior evidence and complicated correlations among analyses, we considered p < 0.05 to be significant without presenting a correction for multiple testing. Gene expression ratios were tested after ln-transformation to normality using analysis of variance or mixed effects models. All analyses were conducted in SPSS for Windows, v 12 (SPSS, Chicago, IL, USA).

Family-based analyses were conducted using measured genotype analysis. The effect of each SNP was tested using logistic regression models for type 2 diabetes and normal densities for the quantitative traits using likelihood analysis implemented in the Java version of the Pedigree Analysis Package, jPAP [33]. Significance was tested using 2 df χ 2 statistics computed as twice the natural logarithm of the ratio of the likelihood upon estimating genotype-specific penetrances or means to the likelihood of a single penetrance or mean. In addition to genotype-specific penetrances or means, the effects of sex, age, BMI and polygenic inheritance were also estimated by maximum likelihood.

Power considerations

Based on a Europid case and control population of 192 individuals each, a control allele frequency of 0.25 and an OR of 1.5, we had 70% power to detect an association at p < 0.05 or 86% power to detect an OR of 1.6. Among African Americans, we had 85% power to detect an OR of 1.5 with a control allele frequency of 0.30 in 186 control and 369 diabetic participants at p < 0.05 or 95% power at an OR of 1.6. For the quantitative trait study in Europids, we considered a simplified model, in which carriers and rare homozygotes were grouped and traits were ln-transformed for comparison by Student’s t test. Notably, we used general linear regression with genotype for the actual analyses, but power analysis of that model was technically problematic. For SI, we had 84% power to detect a 10% difference in ln SI (mean±SD; 3.3 ± 0.9 to 3.0 ± 0.9) or a difference on the linear scale of 2.7 (95%CI 0.46–15.8) to 2.0 (95%CI 0.34–11.7). For AIRg, we had 92% power to detect a 5% fall in lnAIRg (mean±SD 6.0 ± 0.8 to 5.7 ± 0.8) or a difference on the linear scale from 403 (95%CI 84–1,935) pmol/l to 299 (96% CI 62–1,434) pmol/l.

Results

Associations of TCF7L2 with type 2 diabetes mellitus

The now well-replicated association of the T allele at rs7903146 was replicated in Europids (OR 1.72, p = 0.0008), with lesser associations at SNP rs12255732 and the microsatellite DG10S478 (OR 1.46, p = 0.02). In contrast, none of these markers showed any evidence for an association with type 2 diabetes in African American participants (p > 0.5). Results are summarised in ESM Table 2. Whereas the three markers were in strong linkage disequilibrium in Europids (r 2 > 0.7), little linkage disequilibrium was observed in African Americans (r 2 < 0.3) (ESM Tables 3 and 4). No significant association with BMI was seen for any SNP in Europids (data not shown), but the rs12255372 TT genotype was nominally associated with reduced BMI in African Americans (GG 31.1 ± 16.2, GT 30.9 ± 14.0, TT 28.4 ± 7.0 kg/m2; p = 0.03).

Family-based studies of TCF7L2 SNPs in Europids

Both phenotypic and genotype data for SNPs rs7903146 and rs12255372 were available in 673 members of 68 families. Although neither SNP showed excess transmission from parents to affected offspring using methods described previously [33–35], logistic regression methods showed a significant association of the TT genotype, for each of the two SNPs, with diabetes penetrance and with increased 60 min and 90 min glucose levels, but not with reduced insulin levels or altered HOMA estimates of insulin sensitivity (Table 3; data not shown). Among the subset of 122 non-diabetic family members who underwent frequently sampled IVGTT, neither SNP altered SI, AIRg or DI (data not shown).

Table 3 Measured genotype analysis in families for type 2 diabetes and glucose homeostasis traits
Full size table

Metabolic studies of TCF7L2 SNPs in unrelated, non-diabetic participants

The characteristics of 159 non-diabetic African American and 344 non-diabetic Europid participants are shown in Table 2. The populations were similar for age and BMI, but African Americans had significantly greater insulin secretion (AIRg, DI, insulinogenic index, AUCinsulin) and more favourable blood pressure and lipid profiles than Europids. Given the differences in linkage disequilibrium, metabolic profiles and association with type 2 diabetes, we analysed Europid and African American populations separately (Table 4, ESM Table 5). Risk alleles for both rs12255372 and rs7903146 were associated with reduced SI in Europids (p < 0.007), but not in African Americans. Consistent with reduced SI in risk allele carriers, AUCinsulin was increased. In contrast, no measure of insulin secretion (AIRg, DI, AIRmax, insulinogenic index) differed significantly by genotype. AUCglucose showed a non-significant trend towards increasing values with heterozygosity and homozygosity for the risk allele (Table 4), but neither 2 h glucose nor glucose tolerance status (normal vs impaired glucose tolerance) differed by genotype at either marker (p > 0.7; data not shown). Among African Americans, AUCinsulin and the insulinogenic index were reduced in heterozygotes compared with common homozygous individuals, but not in the small number of minor allele homozygotes (Table 4, ESM Table 5). Neither SNP was significantly associated with BMI, waist circumference, WHR or per cent fat in either population (data not shown).

Table 4 Metabolic studies for SNP rs7903146 in unrelated, non-diabetic individuals from Arkansas
Full size table

TCF7L2 gene expression

We examined adipose tissue in 74 individuals for whom we had both TCF7L2 expression and genotype (13 male, 61 female; 66 Europid, eight African American). Analyses were adjusted for age, sex, ethnicity and BMI using mixed effects regression models. Expression levels did not differ by genotype for either SNP (Table 5 for rs7903146), nor did levels influence SI or AIRg (p > 0.2; data not shown). Glucose tolerance status was not a determinant of gene expression (p > 0.3; data not shown). Conclusions were not altered when analyses were restricted to Europid samples. However, in exploratory analyses, significant interaction was found between genotype and BMI to determine expression levels for SNP rs12255732 in all samples (p = 0.003) and for rs7903146 in Europid samples (p = 0.03).

Table 5 Gene expression levels in subcutaneous adipocytes and muscle from non-diabetic individuals by rs7903146 genotype
Full size table

Muscle tissue was examined from 64 individuals for whom genotypes were available (58 Europid, six African American; eight male, 56 female). Data for rs7903146 are presented in Table 5. A non-significant trend towards lower expression in TT homozygotes was observed for both SNPs (p > 0.25); again results did not differ if only Europid samples were included. Muscle TCF7L2 expression was approximately tenfold lower than in adipose. Expression levels did not correlate with and were not significant determinants of SI or AIRg (r 2 < 0.01, p > 0.3); again, they were not altered by glucose tolerance status (p > 0.2; data not shown). Among the 38 individuals for whom both fat and muscle tissue were available, expression levels showed no significant correlation between tissues (r 2 = 0.015, p = 0.6).

We next examined transformed lymphocyte cDNA from transformed lymphocytes from 25 Europid (10 non-diabetic, 15 diabetic) and 30 African American (10 non-diabetic, 20 diabetic) participants. African Americans had lower expression levels (0.94 ± 0.58) than Europids (1.39 ± 1.38; p = 0.03), a difference not observed in the smaller number of adipose and muscle samples available from African American participants. Regardless of whether African American and Europid samples were considered separately or together, expression levels did not differ by diabetic status or by genotype (Table 6). In an exploratory analysis, rs7903146 interacted with diabetic status to reduce TCF7L2 expression in non-diabetic TT homozygotes relative to common CC homozygotes, whereas levels were increased in CT and TT diabetic participants (Fig. 1).

Table 6 TCF7L2 expression studies of transformed lymphocytes in diabetic and non-diabetic participants of Europid and African American ancestry
Full size table
Fig. 1
figure 1

Expression was evaluated in a mixed effects model with ethnicity and diagnosis (diabetic or non-diabetic) as co-factors. Expression ratio is for the 5′ primer set normalised to 18S RNA. Number of samples: CC (7 control, 13 diabetic); CT (8 control, 13 diabetic); TT (9 control, 4 diabetic). diamonds, control; triangles, diabetic

Full size image

Discussion

TCF7L2 is the best replicated and strongest genetic risk factor identified to date for type 2 diabetes in Europids, with SNP rs7903146 showing the most consistent and strongest association [6]. No coding variant explains the TCF7L2 association with type 2 diabetes [4]. Our study replicates this association in Europids, in both a case–control study and in family-based analyses, with an OR for rs7903146 of 1.7, which is comparable with many previous studies [8]. Furthermore, we found significantly higher post-challenge plasma glucose levels in non-diabetic carriers of the TCF7L2 risk alleles, both in members of high-risk families and in unrelated individuals. In contrast, we found no association in a similarly ascertained African American cohort. Few studies of TCF7L2 in African Americans have been published. In the relatively small African American subset of the Diabetes Prevention Program [11] and in 385 Afro-Caribbean participants from the UK [13], TCF7L2 was not associated with type 2 diabetes mellitus. Because the minor (T) allele frequency at rs7903146 is even higher in African Americans than in Europids, allele frequency cannot explain this discrepancy. In contrast, Helgason et al. [14] did replicate the association of rs7903146 in families from West Africa (RR 1.45, p = 0.002 after correcting for relatedness and ancestry). Differences in linkage disequilibrium between populations of African and European ancestry are unlikely to explain this discrepancy, given that rs7903146 appears to be the causative SNP [14]. Using our African American sample size, an OR of 1.5 (close to the 1.45 estimated in West Africans) and the control minor allele frequency of 0.30, we had 85% power to detect an association at p < 0.05. Were the OR only 1.3, our power would fall to 52%. Alternatively, gene–gene or gene–environment interactions could easily explain the differences among populations [36], particularly given the different lifestyles of West African and Arkansas African American individuals.

TCF7L2 binds to the promoter of the proglucagon gene in gut endocrine L cells [15], thus potentially acting through glucagon-like peptide 1 to affect insulin secretion. Indeed, some but not all studies have shown impaired insulin secretion to oral glucose in individuals with the rs7903126 TT genotype [6, 11, 18]. In contrast, we did not find evidence for such an effect. Although we found nominal evidence for reduced insulin secretion in African American carriers of the risk alleles, insulin secretion was higher among Europid individuals with risk alleles, consistent with reduced insulin sensitivity. Whereas we did observe the trend towards lower DI reported in smaller studies of Damcott et al. [10] and Munoz et al. [17]), this trend was driven by reduced SI. However, post-challenge plasma glucose was increased with the risk allele among non-diabetic members of high-risk families (Table 3), with a similar trend among non-diabetic, unrelated individuals (Table 4). Nonetheless, neither 2 h glucose as a quantitative trait nor glucose tolerance (2 h glucose higher than 7.8 mmol/l) as a dichotomous trait was associated with TCF7L2 genotype (p > 0.7).

TCF7L2, which is widely expressed, acts downstream of WNT to complex with β-catenin, which in turn activates transcription of a wide variety of genes, including those implicated in cell proliferation and neoplasia [37, 38]. The WNT pathway also plays a role in adipogenesis [39]; hence TCF7L2, which is well expressed in adipose, may play a role in insulin action and insulin sensitivity. Damcott et al. [10] first reported reduced SI in non-diabetic Europid participants with TCF7L2 risk alleles. Our data confirm that association in a much larger Europid population, with a 45% reduction in SI among TT homozygotes. In contrast, Munoz et al. found no association with SI in 138 Europid women [17]; we found no association in 122 non-diabetic members of Utah families (the current study); and Scott et al. [7] found no association with SI in members of Finnish families. Munoz et al. found a non-significant trend towards reduced SI and reduced AIRg among 118 African American women, which was not observed in our study. Differences in ascertainment, relatively small samples for published studies, particularly when considering the rare allele homozygote frequency, and possible gene–gene or gene–environment interactions may explain these differences. Neither the previously published studies nor we attempted to correct for multiple testing (two SNPs, several metabolic traits, with both the SNPs and traits showing correlation). Hence, some of the nominal p values (0.01 < p < 0.05) in our studies and published reports may be spurious.

Non-coding variants of TCF7L2 may influence glucose homeostasis by changing transcription (regulation) or splicing and may do so in a tissue-specific fashion. To our knowledge, this is the first study to examine the role of ethnicity and genotype in TCF7L2 gene expression or the role of muscle and adipocyte TCF7L2 expression in insulin sensitivity. We found no genotypic effects on TCF7L2 expression levels in adipose or muscle, nor any correlation of TCF7L2 mRNA levels with SI or BMI. In contrast, we did find a trend towards genotype determination of gene expression in transformed lymphocytes, where expression was decreased in African American and Europid non-diabetic individuals, but increased in diabetic individuals with the TT risk genotype. As this study was small (24 non-diabetic individuals, 30 diabetic individuals) and included two ethnic groups that themselves show differences in expression, this unexpected finding requires further study. Because individuals for whom transformed lymphocytes were available were not the individuals who had undergone adipose and muscle biopsies, we were unable to examine the correlation in gene expression between these tissues. Given the multiple splice forms of TCF7L2, the lack of genotype effect on total TCF7L2 expression measured in this study and the lack of correlation of adipocyte or muscle TCF7L2 expression with SI, we believe that tissue-specific splicing deserves further study.