Bmc Medical Genetics Association between Tcf7l2 Gene Polymorphisms and Susceptibility to Type 2 Diabetes Mellitus: a Large Human Genome Epidemiology (huge) Review and Meta-analysis

Background: Transcription factor 7-like 2 (TCF7L2) has been shown to be associated with type 2 diabetes mellitus (T2MD) in multiple ethnic groups in the past two years, but, contradictory results were reported for Chinese and Pima Indian populations. The authors then performed a large meta-analysis of 36 studies examining the association of type 2 diabetes mellitus (T2DM) with polymorphisms in the TCF7L2 gene in various ethnicities, containing rs7903146 C-toT (IVS3C>T), rs7901695 T-to-C (IVS3T>C), a rs12255372 G-toT (IVS4G>T), and rs11196205 G-to-C (IVS4G>C) polymorphisms and to evaluate the size of gene effect and the possible genetic mode of action.

in different ethnic populations when stratification of IVS3C>T and IVS4G>T polymorphisms except for Africans, where a dominant, additive genetic mode is suggested for IVS3C>T polymorphism.
Conclusion: This meta-analysis demonstrates that four variants of TCF7L2 gene are all associated with T2DM, and indicates a multiplicative genetic model for all the four polymorphisms, as well as suggests the TCF7L2 gene involved in near 1/5 of all T2MD. Potential gene-gene and geneenvironmental interactions by which common variants in the TCF7L2 gene influence the risk of T2MD need further exploration.
The polygenic basis of T2DM has been intensely investigated by large consortia worldwide, and consensus on a range of valid susceptibility genes has emerged. Such genes, for example PPARG, KCNJ11 and CPN10, appear to contain common variation that confers only 10-20% additional risk of T2DM, and only studies with large numbers of case and controls have had sufficient statistical power to examine their role robustly [72]. Recent years, some novel loci in genes such as TCF7L2, SLC30A8, HHEX, ETX2, CDKN2A/CDKN2B, IGF2BP2, CDKAL1, FTO, PPARG, and KCNJ11 have been found with susceptibility to T2DM [2,[73][74][75][76], and these has promoted the development of genetics on T2DM. From 2006 to 2008, study on TCF7L2 gene polymorphisms and its association with T2DM has become the focus in diabetes mellitus, therefore, some meaningful results have reported one by one. In March, 2006, Grant and colleagues [2] first proposed an association with TCF7L2 gene variants with T2DM. The study [2] consisted of three samples from Ice-land, Denmark and USA with strikingly similar results. Since then, this potential association has been examined in over 50 samples from around the world. Given the amount of accumulated data, it is important and urgent to perform a quantitative synthesis of these evidences. We therefore undertake a large meta-analysis of studies of the association between TCF7L2 gene polymorphisms and T2DM. To exclude heterogeneity of these studies, we stratified the data by Hardy-Weinberg equilibrium (HWE), publishing time, ethnicity, body mass index (BMI), diagnosis and age.
Although one meta-analysis [77] and several narrative reviews about TCF7L2 gene and T2DM had been reported [78][79][80], Succedent studies have now made more data available. Compared with the only one meta-analysis written by Cauchi and his colleagues [77] and published online in May, 2007, 28 more studies from 14 new papers [53,55,57,63,68,69,71,[73][74][75][76][81][82][83]were available to us. And that, this meta-analysis of the association between TCF7L2 gene and T2DM, only summarized IVS3C>T polymorphism by just analyze allele frequency [77], the authors did not generalize the association between other important variations such as IVS4G>T, IVS3T>C and IVS4G>C polymorphisms and T2DM. Further, the only meta-analysis was based on published data, namely, without stratification, and, was thus limited in the interpretations they allowed. However, using the original data, we are able to explore more deeply, searching for sources of heterogeneity. Thus, we carried out a large meta-analysis to conduct a broad comprehensive and quantitative assessment on the association between variants of TCF7L2 gene and T2DM, and then, to perform a detailed examination of sources of heterogeneity. We sought to estimate magnitude of the genetic association as well as the genetic mode of action, and to determine the extent of heterogeneity in the strength of associations between different studies.

Search strategy and inclusion criteria
We conducted a comprehensive search of the US National Library of Medicine's PubMed, OMIM, ISI Web of Science, and Embase databases for all genetic association studies on at least one of the above four polymorphisms in the TCF7L2 gene and T2DM published from March, 2006, when the association between TCF7L2 gene and T2DM was first reported [2], through March, 2008. The search strategy was based on combination of the key words "transcription factor 7-like 2" or "TCF7L2" or "TCF7L2 gene polymorphism", and "diabetes" or "diabetes 2" or "type 2 diabetes" or "type 2 diabetes mellitus" or "T2D" or "T2DM". The references of all computer-identified publications were searched for additional studies, and the PubMed option "Related Articles" was used to search for potentially relevant papers. Reference lists in retrieved articles were also screened.
Searching was performed in duplicate by two independent reviewers (YZ and JYY). Without any language restriction (we will translate articles published in other languages into English by professional interpreters), we only selected published manuscripts (including their online supporting materials and the original data kindly sent by the authors) if they met the following criteria: 1) the publication was a population-based association study (family-based study design with linkage considerations was excluded), regardless of sample size, and 2) there were enough results for extraction of data, that is, the Cumulative meta-analysis of the odds ratios of association between rs7903146 polymorphism of TCF7L2 gene and T2DM for T allele vs. C allele Figure 4 Cumulative meta-analysis of the odds ratios of association between rs7903146 polymorphism of TCF7L2 gene and T2DM for T allele vs. C allele.
number of subjects with each allele and genotype in the T2DM and control groups. Where eligible papers had insufficient information, we contacted authors by e-mail or mail for additional information. In the case of sequential or multiple publications of analyses of the same data or overlapping data sets, the publication that reported data from the largest or most recent study was included.

Data extraction
Data were extracted independently and in duplicate by two investigators (YZ and JJY) who used the recommended guidelines for reporting on meta-analyses of observational studies [84]. The following data were extracted from the eligible studies: authors, journal and year of publication, country of origin, selection and characteristics of cases and controls, demographic data, ethnicity of the study population (Caucasian, North European, East Asian, African, Middle Eastern, American, or Pacific Asian), numbers of eligible and genotyped cases and controls, and genotype distributions in cases and controls and available subgroups thereof. Furthermore, we examined whether matching had been used; whether there was specific mention of blinding of the genotyping personnel to the clinical status of subjects; whether the genotyping method used had been validated; and whether genotype frequencies in control groups were in HWE. Any disagreement was adjudicated by a third author (YWZ).

Statistical Analysis
We used the odds ratio as the metric of choice and it was estimated for each study. In order to explore the possible association between TCF7L2 gene polymorphisms and T2MD and to avoid excessive comparisons, we calculated the odds ratio by two methods, namely, allele comparison (i.e., T allele vs. C allele in IVS3C>T polymorphism), and genotype comparison, that is, comparing the risk variant homozygotes and heterozygotes with wild homozygotes, respectively (i.e., TT vs. CC (OR 1 ) and TC vs. CC (OR 2 ) in IVS3C>T polymorphism; TT vs. GG (OR 3 ) and TG vs. GG (OR 4 ) in IVS4G>T polymorphism; CC vs. TT (OR 5 ) and CT vs. TT (OR 6 ) in IVS3T>C polymorphism; CC vs. TT (OR 7 ) and CT vs. TT (OR 8 ) in IVS4G>C polymorphism). We estimated and characterized the summary prevalence of risk allele by using only the data on controls. When we analyze genotype data for meta-analysis, cells with zero count had 0.5 been added. In addition, we calculated the Cumulative meta-analysis of the odds ratios of association between rs7903146 polymorphism of TCF7L2 gene and T2DM for T homozygotes vs. C homozygotes Figure 5 Cumulative meta-analysis of the odds ratios of association between rs7903146 polymorphism of TCF7L2 gene and T2DM for T homozygotes vs. C homozygotes.
population attributable risk (PAR) of risk allele according to method of Chang et al [71].
We first compared alleles for cases and controls in order to detect overall differences and genetic association. Allele frequencies were computed for studies reporting only genotypic data. Pooled odds ratio were computed three times, by the fixed-effects method of Mantel and Haenszel [85], by the random-effects method of DerSimonian and Laird [86], and by the Bayesian random-effects method of Warn et al [87]. Unless stated otherwise, random-effects estimates (DerSimonian and Laird) are reported here.
Our primary analysis for TCF7L2 gene polymorphisms and T2DM is based on genotype comparisons in order that magnitude of the genetic association and the genetic mode of action can be identified exactly. Once an overall gene effect was confirmed by the allele comparison, the genotype effects and genetic model were estimated by using the genetic model-free approach suggested by Minelli et al [88], where no assumptions about genetic models were required. The OR 1 to OR 8 were calculated by using multivariate meta-analysis with Bayesian method [88]. The logarithm (log) odds ratios were modeled on the base of both between and within study variations. Four separate stochastic lambda values (reflecting the genetic model), consisting of the ratios of logOR 2 versus logOR 1 for λ 1 , logOR 4 versus logOR 3 for λ 2 , logOR 6 versus logOR 5 for λ 3 , and logOR 8 versus logOR 7 for λ 4 , were calculated [88]. The parameter λ capture information about the genetic mode of action, briefly, the model is a recessive model if λ = 0, a codominant model if λ = 0.5, a dominant model if λ = 1, and homozygous or heterosis model if λ < 0 or λ > 1. Pooled odds ratio were also computed by the fixed-effects method of Mantel and Haenszel [85] and by the random-effects method of DerSimonian and Laird [86]. Unless stated otherwise, Bayesian estimates are reported here.
We tested for deviations from Hard-Weinberg equilibrium (HWE) in both case and control populations by using the exact method reported by Emigh et al [89] whether the author provide the P values of HWE or not. For above allele-based and genotype-base analyses, we Cumulative meta-analysis of the odds ratios of association between rs7903146 polymorphism of TCF7L2 gene and T2DM for heterozygotes vs. C homozygotes Figure 6 Cumulative meta-analysis of the odds ratios of association between rs7903146 polymorphism of TCF7L2 gene and T2DM for heterozygotes vs. C homozygotes.
took two methods to handle Hard-Weinberg disequilibrium. Firstly, we performed sensitivity analyses by including by including and excluding studies in Hard-Weinberg disequilibrium (case and control group were all considered). Secondly, we included all studies regardless of HWE and instead adjusted for the degree of disequilibrium by using the inbreeding coefficient (F) suggested by Trikalinos et al [90]. Briefly, the F was estimated for each study by using data in the control group. Predicted genotype frequencies were estimated [91] and used instead of the observed frequencies in the summary analysis of magnitude and genetic model.
In sensitivity analysis, we estimated between-study heterogeneity across all eligible comparisons using Cochran's Q statistic [92]. Heterogeneity was considered significant at P value less than 0.10 [92]. We also report I 2 statistic, which describes the percentage of variability in point estimates due to sample heterogeneity rather than sampling error [93,94], and can quantify heterogeneity irrespective of the number of studies [94,95]. As a guide, I 2 values less than 25 percent may be considered "low", values of 31 percent to 56 percent have been defined as indicating "low" to "moderate" heterogeneity, whereas values greater than 56 percent are considered indicative of "notable" heterogeneity, and "large" heterogeneity is stipulated for I 2 values of larger than or equal to 75 percent [94,95]. The examination of heterogeneity was performed separately for two odds ratios (OR 1 and OR 2 for IVS3C>T polymorphism, OR 3 and OR 4 for IVS4G>T polymorphism, OR 6 and OR 7 for IVS3T>C polymorphism, OR 7 and OR 8 for IVS4G>C polymorphism) in genotype-based analysis. If there was heterogeneity on at least one of these two odds ratios, the source of the heterogeneity was explored by fitting a covariable (such as age, gender, ethnicity, BMI) in a meta-regression model [96][97][98]. Possible subgroup analyses were involved in ethnicity (Caucasians, North Europeans, East Asians, Indians and other racial groups), age (< 65 years vs. ≥ 65 years), diagnosis (normal diagnosis vs. no diagnosis or false diagnosis), publishing time (2006 vs. 2007 and 2008) and BMI (normal (< 30.0 kg/m 2 ) vs. abnormal (≥ 30.0 kg/m 2 ) according to the meta-regression analysis. Logistic regression analysis was used to compare odds ratios among these possible groups. In Odds ratios of association between rs7903146 polymorphism of TCF7L2 gene and T2DM for T allele vs. C allele Figure 7 Odds ratios of association between rs7903146 polymorphism of TCF7L2 gene and T2DM for T allele vs. C allele.
addition, genetic association studies may be especially susceptibility to the selective publication of positive findings on gene association [99]. In order to investigate the publication bias among studies, funnel plots [100] and cumulative meta-analysis [101] were used.
All analyses were conducted in Stata software, version 9.0 [102], using meta, metan, metabias, metacum, and metareg commands, except for the Bayesian method of genotypebased analysis. We fit the Bayesian models by using Markov chain Monte Carlo (MCMC) methods with Bayesian framework and perform our inferences using Win-BUGS 1.4.3 [103], taking advantage of its flexibility as well as its ability to incorporate full uncertainty in all unknown parameters. Bayesian analyses yield credible intervals rather than confidence intervals; a 95% credible interval (CrI) describes a range in which the probability that the unknown quantity lies within this interval, after seeing the data, is 95%. A 'burn-in' of 10,000 iterations is carried out for models, followed by 50,000 iterations for parameter estimates. A P value less than 0.05 was consid-ered statistically significant, except for tests of heterogeneity, where a level of 0.10 was used.
between TCF7L2 gene polymorphisms and T2DM were identified by the inclusion criteria and all of them were written in English ( There was considerable diversity of ethnic groups. Eligibility criteria for T2DM patients were shown in Table 1 (Additional file 1). Controls were mainly healthy populations who were described as nondiabetic and/or, Normoglycemic, and/or normal glucose tolerance (NGT), although varying details were presented regarding the extent of testing that had been done to excluded controls with impaired glucose tolerance and diabetes (table 1, Additional file 1). Remaining 3 investigations did not report the details of controls [60,66,67]. 1 study matched for geographic region [53]; 1 study matched for age, race, and BMI [56]; 5 studies matched for race [55,67,70,83]; and one study 1 study matched for age, sex, and geographic region [58,65]. The other investigations did not provide for the details of the matching information. Detailed characteristics of each investigation, along with P values for testing HWE, are listed in the Table 2-5 (Addional file 1) for IVS3C>T polymorphism, IVS3T>C polymorphism, IVS4G>T polymorphism and IVS4G>C polymorphism, respectively.

IVS3C>T polymorphism
The eligible studies for analysis of IVS3C>T polymorphism included a total of 33,135 cases with T2DM and 36,316 controls ( Table 2, Additional file 1). Allele and genotype data were available for all the cases and controls in the eligible studies. There was no evidence of Hard-Weinberg disequilibrium in both cases and controls.

Bias diagnostics
We did not find any evidence of publication bias of the eligible studies. Funnel plots for the comparisons most Odds ratios of association between rs7903146 polymorphism of TCF7L2 gene and T2DM for heterozygotes vs. C homozygotes Figure 9 Odds ratios of association between rs7903146 polymorphism of TCF7L2 gene and T2DM for heterozygotes vs. C homozygotes.
commonly made, that is, for T allele vs. C allele, T homozygotes (TT) vs. C homozygotes (CC), and heterozygotes (TC) vs. C homozygotes (CC) are indicated in figure 1, 2 and 3. These three plots give corrected P = 0.410 (corrected z = 0.82), corrected P = 0.379 (corrected z = 0.88), and corrected P = 0.426 (corrected z = 0.80), respectively. Figure 4, 5, and 6 indicate the cumulative metaanalysis after each study from March 2006 to February 2008, for T allele vs. C allele, T homozygotes vs. C homozygotes, and heterozygotes vs. C homozygotes. The random-effects odds ratio of T homozygotes vs. C homozygotes waved slightly among studies performed in 2006 and has changed very little from around 2.0 after the eleventh study ( figure 5). Cumulative meta-analysis of the other two comparisons by year of publication showed that the association between T2DM and TCF7L2 IVS3C>T polymorphism has remained significant and been consistent over time ( figure 4 and figure 6).  23.0% (95% CI: 9.4, 36.6)). We first contained all the 35 studies to evaluate the overall association between TCF7L2 IVS3C>T polymorphism and T2DM. With our allele-based analysis, the pooled odds ratio suggested a 1.416-fold increase in susceptibility to T2DM among persons with the T allele (random-effects OR TvC = 1.416, 95% CI: 1.346, 1.491), with a finding that was highly statistically significant (P = 0.000), but there was significant between-study heterogeneity (Q = 104.966, P = 0.000; I 2 = 67.6%) ( Table 6 [Additional file 1], figure 7).

Genotype comparisons
The genotype frequency of TCF7L2 IVS3C>T polymorphism between case and control groups is characterized in web table. The genotype effects for TT versus CC (OR 1 ) and TC versus CC (OR 2 ) were calculated for each study (data are not shown and available from the first author on request). In our primary concern, multivariate meta-analysis was conducted to estimate the pooled effects and there was an significant increased risk of T2DM among populations with both homozygous variant TT genotype (Bayesian random-effect OR 1 = 1.968, 95 percent credible interval (CrI): 1.790, 2.157) and heterozygous variant TC genotype (Bayesian random-effect OR 2 = 1.406, 95 percent CrI: 1.341, 1.476), with a "moderate" between-study heterogeneity (Q = 66.102, P = 0.001, I 2 = 48.6%, figure 8) in the former and a "notable" between-study heterogeneity (Q = 84.028, P = 0.000, I 2 = 59.5%, figure 9) in the latter. The estimated stochastic parameter λ 1 was 0.504 (95 percent CrI: 0.461, 0.551), which suggested a codominant genetic mode of action.

Examination of heterogeneity
Meta-regression was used to explore the source of heterogeneity and it was found that ethnicity was the only covariable associated with both lnOR 1 (meta-regression beta coefficient (β) = -0.0993, P = 0.002) and lnOR 2 (β = -0.0360, P = 0.002). We then did a subgroup analysis by stratifying study samples into five ethnic groups, that is, Caucasians (11 samples), North European (6 samples), East Asian (9 samples), Indians (4 samples) and other racial descent group (5 samples), and found that this stratification can remove all the between-study heterogeneity among North Europeans, East Asians and Indians for both OR 1 and OR 2 , rather than that among Caucasians and other racial group (Table 6, Additional file 1). We then did a further subgroup analysis of Caucasians by stratifying them into French Caucasians versus other Caucasians and extracted the studies with African populations for an independent analysis, and found that such analyses can only remove heterogeneity among these three populations for OR 1 rather than OR 2 (   Although BMI was not associated with both lnOR 1 and lnOR 2 by separate meta-regression analysis, the significant association was found when we evaluate it together with ethnicity (for lnOR 1 , β = -0.0512, P = 0.038; for lnOR 2 , β = -0.0311, P = 0.004). In addition, some references reported that BMI may interact with TCF7L2 gene to increase the risk of T2MD. Therefore, we did a subgroup analysis for BMI (normal (< 30.0 kg/m 2 ) vs. abnormal (≥ 30.0 kg/m 2 ). However, such stratification cannot remove heterogeneity for both lnOR 1 (Q = 35.017, P = 0.006, I 2 = 52.1%) and lnOR 2 (Q = 28.046, P = 0.013, I 2 = 50.0%). The pooled OR 1 , OR 2 , and λ 1 for populations with normal BMI were 2.017 (95%CrI: 1.711, 2.347), 1.470 (95%CrI: 1.350, 1.609), and 0.554 (95%CrI: 0.481, 0.637), respectively. The pooled OR 1 , OR 2 , and λ 1 for populations with abnormal BMI were 1.931 (95%CrI: 1.670, 2.213), 1.337 (95%CrI: 1.250, 1.431), and 0.441 (95%CrI: 0.387, 0.500), respectively. Logistic regression analysis indicated that there was no significant difference between populations with normal and abnormal BMI for both lnOR 1 (P = 0.855) and lnOR 2 (P = 0.716).

IVS4G>T
Our searches identified 29 studies (totally 57,235 participants; 28,188 cases of T2DM, 29,047 controls) meeting our inclusion criteria. All of these studies examined the IVS4G>T polymorphism (Table 4, Additional file 1). Allele and genotype data were available for all the cases and controls in the eligible studies. There was no evidence of Hard-Weinberg disequilibrium in both cases and controls.

Bias diagnostics
There was no evidence of small study bias and publication bias for any comparisons. Funnel plots for the comparisons made for T allele vs. G allele, T homozygotes (TT) vs. G homozygotes (GG), and heterozygotes (TG) vs. G homozygotes are shown in figure 10, 11 and 12. These three funnel plots give corrected P = 0.339 (corrected z = 0.96), corrected P = 0.536 (corrected z = 0.62), and corrected P = 0.442 (corrected z = 0.77), respectively. Figure  13, 14 and 15 indicate the cumulative meta-analysis after each study from March 2006 to February 2008, for allelebased analysis and two genotype-based models of IVS4G>T polymorphism. The random-effects odds ratio of allele-based analysis was significant more than 1 and changed very little from around 1.379 after the first study ( figure 13). The random-effects odds ratio of T homozygotes vs. G homozygotes waved slightly among studies performed in 2006 and has changed very little from around 1.884 after the eleventh study ( figure 14). Cumulative meta-analysis of heterozygotes vs. G homozygotes by year of publication showed that the association between T2DM and TCF7L2 IVS4G>T polymorphism has remained significant and been consistent over time (figure 15).
Cumulative meta-analysis of the odds ratios of association between rs12255372 polymorphism of TCF7L2 gene and T2DM for T allele vs. H allele Figure 13 Cumulative meta-analysis of the odds ratios of association between rs12255372 polymorphism of TCF7L2 gene and T2DM for T allele vs. H allele.

Genotype comparisons
The genotype frequency of TCF7L2 IVS4G>T polymorphism between case and control groups is characterized in Table 4, Additional file 1. The genotype effects for TT versus GG (OR 3 ) and TG versus GG (OR 4 ) were calculated for each study (data are not shown and available from the first author on request). In our primary concern, multivariate meta-analysis was conducted to estimate the pooled effects and there was an significant increased risk of T2DM among populations with both homozygous variant TT genotype (Bayesian random-effect OR 2 = 1.885, 95 percent credible interval (CrI): 1.698, 2.088) and heterozygous variant TG genotype (Bayesian random-effect OR 4 = 1.360, 95 percent CrI: 1.291, 1.433), with a "moderate" between-study heterogeneity both in the former (Q = 46.515, P = 0.015, I 2 = 39.8%, figure 17) and in the latter (Q = 47.980, P = 0.011, I 2 = 41.6%, figure 18). The estimated stochastic parameter λ 2 was 0.486 (95 percent CrI: 0.431, 0.545), which suggested a codominant genetic mode of action.

Examination of heterogeneity
Meta-regression was used to explore the source of heterogeneity and it was found that ethnicity was the only covariable associated with both lnOR 3 (β = -0.1083, P = 0.005) and lnOR 4 (β = -0.0418, P = 0.047). We then did a subgroup analysis by stratifying study samples into five racial descent groups, that is, Caucasians (9 samples), North European (4 samples), East Asian (8 samples), Indians (4 samples) and other racial descent group (4 samples), and found that this stratification can remove all the between-study heterogeneity among North Europeans, East Asians and Indians for both OR 3 and OR 4 , rather than that among Caucasians and other ethnic group ( Table 6, Additional file 1). We then extracted the studies with other Caucasians (7 studies) for an independent analysis, and found that such analysis can well-done remove the heterogeneity among these studies for both OR 3 (Q = 8.407, P = 0.210, I 2 = 28.6%) and OR 4 (Q = 4.327, P = 0.633, I 2 = 0.0%) ( Table 6, Additional file 1). We did not do a further analysis of other ethnic group because of different origin of populations in these studies.
Cumulative meta-analysis of the odds ratios of association between rs12255372 polymorphism of TCF7L2 gene and T2DM for heterozygotes vs. G homozygotes Figure 15 Cumulative meta-analysis of the odds ratios of association between rs12255372 polymorphism of TCF7L2 gene and T2DM for heterozygotes vs. G homozygotes.

Main findings
This is a large meta-analysis, including data from 25 papers involved in 36 genetic association studies with exactly 35,843 cases of T2DM and 39,123 controls, carefully avoiding the double-counting of participants in the study. The HuGE systematic review provides the most recent and comprehensive evaluation of the association between four TCF7L2 gene polymorphisms and susceptibility to T2DM. We find that near 70,000 subjects (33,135 cases of T2DM and 36,316 controls) were from 35 studies concerning the IVS3C>T polymorphism, and over 55,000 (28,188 cases of T2DM and 29,047 controls) subjects were from 29 studies investigating the IVS4G>T polymor-Odds ratios of association between rs12255372 polymorphism of TCF7L2 gene and T2DM for T allele vs. H allele Figure 16 Odds ratios of association between rs12255372 polymorphism of TCF7L2 gene and T2DM for T allele vs. H allele.
phism. These two variants were the main study focus on the association between TCF7L2 gene polymorphisms and susceptibility to T2DM.
On one hand, the results indicates notable associations between two main TCF7L2 gene polymorphisms, namely, IVS3C>T as well as IVS4G>T, and T2DM. The magnitudes of this association were moderate, however, statistically significant. Our primary analysis finds that among IVS3C>T polymorphism, TC heterozygotes carry just over a 1.4-fold increased risk of T2MD, and TT homozygous variants carry near a 2.0-fold increase in T2MD risk when compared with CC homozygotes, and that among IVS4G>T polymorphism, TG heterozygotes carry near a 1.4-fold increased risk of T2MD, and TT homozygous variants carry approximate a 1.9-fold increase in T2MD risk when compared with GG homozygotes. Significant magnitudes of genetic effect for heterozygotes of these two variants were confirmed by the analysis on the pooled odds ratios among different ethnic groups with slight differences except for East Asians where higher pooled odds ratio was observed in IVS4G>T polymorphism when compared with that of the other ethnic populations, and genetic effect sizes for homozygotes of these two variants were also strengthened by the analysis on the pooled odds ratios among different ethnic groups including Caucasians, North Europeans, Africans and Indians by showing a small variation. However, we found that there was no significant association between such two homozygotes and disease among East Asians, and therefore, conflicting results have appeared, in which studies on Chinese provided negative results [57,71], however, positive results were observed among Japanese populations [61,62,82]. Our meta-analysis on 9 studies of IVS3C>T polymorphism and 7 studies of IVS4G>T polymorphism among East Asians indicates a moderate statistical association and shows that the heterozygous variants contributed all the increased risk of T2MD among this population. This distinct disparity may suggest different mechanism of gene-disease between East Asians and other ethnic population. Moreover, although less samples and subjects were provided, the IVS3T>C and IVS4G>C polymorphisms are also significantly associated with T2DM. We then suggest that TCF7L2 is the most common susceptible gene for T2DM among various ethnic groups in the world.
Odds ratios of association between rs12255372 polymorphism of TCF7L2 gene and T2DM for T homozygotes vs. G homozy-gotes Figure 17 Odds ratios of association between rs12255372 polymorphism of TCF7L2 gene and T2DM for T homozygotes vs. G homozygotes.
On the other hand, we made no assumptions about genetic models and finds that the pooled odds ratios can clearly fit a multiplicative model and the lambda (λ) parameter can well-done indicate a codominant genetic mode of action with tight CIs among all these four TCF7L2 gene polymorphisms. We also explored the genetic mode among different ethnic populations and similar results were observed in Caucasians, North Europeans, East Asians and Indians except for Africans, where dominant genetic mode is suggested for IVS3C>T polymorphism. We cannot verify the genetic mode of action for Africans in the other three polymorphisms due to few studies. In addition, genetic mode of IVS3C>T variant among Africans also need a further verification by more studies. Even so, we nevertheless suggest a potential multiplicative genetic model for the four TCF7L2 gene polymorphisms.
The effect size of overall association between IVS3C>T variant and T2MD is near to 1.42, which is lower than that in a former meta-analysis about TCF7L2 and T2MD performed by Cauchi et al [77], and larger than that in a pooled analysis of three genome-wide association (GWA) study [73,74,76]. we supplement some studies about East Asians and Pima Indians [55,57,71,88] in which risk homozygous variant TT was scanty, and excluded several studies without genotype frequency data, which may decrease the overall effect but provide a more comprehensive understanding of the association. Even so, we cannot find any significant difference of effect size between our study with the former meta-analysis as well as the pooled GWA studies. In addition, we computed the PAR of TCF7L2 according to IVS3C>T variant and the PAR for the combined genotypes TT and TC were 16.9, 23.2, 14.1, 2.5, 17.9, 27.0 for overall, Caucasians, North Europeans, East Asians, Indians, and Africans, respectively, suggesting this gene polymorphism may contribute near 1/5 of all T2MD in the globe except for East Asians.
Our findings were based on some gene-association studies and tens of thousands participants and were robust to each of the planned sensitivity analyses used. We cannot find any evidence of publication bias and small study bias by funnel plots and cumulative meta-analysis, but, con-Odds ratios of association between rs12255372 polymorphism of TCF7L2 gene and T2DM for heterozygotes vs. G homozy-gotes Figure 18 Odds ratios of association between rs12255372 polymorphism of TCF7L2 gene and T2DM for heterozygotes vs. G homozygotes.
siderable between-study heterogeneity was found. Between-study heterogeneity may be due to differences in sample selection (e.g., in age, sex, diagnosis, sample content, etc), or in methods (e.g., genotyping method), or it may be due to real differences in populations (e.g., in race), or in interactions with other risk factors (genetic or/ and environmental factors). The results of primary analysis (genotype-based) on the four polymorphisms showed statistically significant between-study heterogeneity for both IVS3C>T polymorphism (OR 1 and OR 2 ), and IVS4G>C polymorphism (OR 3 and OR 4 ). In this Human Genome Epidemiology review, we examined five potential sources of between-study heterogeneity for genotypebased model by meta-regression and it was suggested that ethnicity was the only co-variable associated with the four ORs. In fact, more detailed manner of ethnic stratification, namely, French Caucasians, other Caucasians, North Europeans, East Asians, Indians, and Africans, can remove near all the heterogeneity of studies about both IVS3C>T and IVS4G>C polymorphisms for OR 1 to OR 4 . Study results reported by several articles showed that BMI may cooperate with the TCF7L2 gene to increase the risk of the T2DM. However, disagreement of the results was published by different researchers. Cauchi et al [59], Horikoshi et al [62] and Humphries et al [67] separated to analyze the nonobese type 2 diabetic subjects (BMI<30 Kg/m 2 ) and more higher allelic ORs or RRs were obtained in IVS3C>T and IVS4G>C polymorphisms; whereas, Miyake et al [82], Chandak et al [70], and Dahlgren et al [104] found that there was a slight increase or no change of odds ratio by adjustment of BMI. So we stratified the studied by BMI (normal vs. abnormal) and found that significant between-study heterogeneity nevertheless appeared in all the two subgroups. Both studies on IVS3C>T and IVS4G>C polymorphisms provided us for the similar results. But, we found a significant association when it was evaluated together with ethnicity. These results, on the one hand, suggested that BMI may be a possible factor to impact on the effect of TCF7L2 gene on T2DM by a indirect action; on the other hand, were difficult for us to explain the deep-seated reasons that how the BMI influence the effect of TCF7L2 gene on T2DM because we cannot gain the original data of BMI in each article. This is a general limitation of meta-analysis. So we hope the reasonable interpretation will be presented in the subsequent studies. Although sex [56,63,68,71,105], age of cases [63,68,71,105], drugs [106], constitution [64,69], and lifestyle [64,69,105] may regulate the effects of TCF7L2 gene, we cannot test the effects of age, sex, drugs and lifestyle on heterogeneity due to without sufficient related data being used. In addition, the age of the control NGT groups may be associated with the heterogeneity and the young controls may develop diabetes at a later date. However, there are also insufficient data.

Potential gene-gene and gene-environment interaction
T2DM seems to result from a complicated interplay of genetic and environmental factors influencing a number of intermediate traits of relevance to the diabetic phenotype (e.g., insulin secretion, insulin action, fat distribution, obesity). As a matter of fact, T2DM appears to be composed of subtypes where genetic susceptibility is strongly associated with environmental factors at one end of the spectrum, and highly genetic forms at the other end. Thus, there are several possible interactions between gene and gene or between gene and environmental factors.
Recent years, a series of new loci in some genes have been identified to contribute about 10~30% population attributable risk (PAR) of T2DM [73][74][75][76]107]. But, the overall PAR seems to far less than the cumulative effect. So there must be some potential interactions among genes. The strongest known T2DM association (random-effects OR TvC ≈1.42) was recently mapped to the transcription factor TCF7L2, a gene that is a target of the Wnt signaling pathway [50]. Whereafter, some other genes such as HHEX, IDE, DKK3 and KIF11, which are also in the Wnt signaling pathway [108][109][110][111][112][113] may have some interactions with TCF7L2 on the risk of T2DM. Cauchi et al [59]found that the expression of TCF7L2 gene was downregulated in obese subjects developing T2DM and they then brought forward a hypothesis of a potential interaction between TCF7L2 and Calpain-10 by a supporting material [114] that suggested that the Calpain system was involved in the constitutive regulation of β-catenin signaling functions. Wnt signaling has recently been shown to regulate pancreatic β-cell proliferation and the author suggested a possible interation between TCF7L2 gene and β-catenin gene [115]. However, there is not any reported publication about the interaction between TCF7L2 other related genes. These hypothetical interactions and their specific effects on T2DM, as well as functional analyses, will be required to further elucidate the role of variation in TCF7L2 in the pathogenesis of T2MD and very large samples are needed.
A variety of environmental factors can be implicated in the phenotype of T2DM, such as obesity, hypertension, bad lifestyle (smoking, drinking, high-energy diet), short of exercise, malnutrition, and some drugs [116]. All the above environmental factors may interact with the TCF7L2 gene to influence T2DM. So far, few specific geneenvironment interactions have been described for TCF7L2 gene polymorphisms. Because obesity is a major determinant of development of T2DM, most patients with T2DM are obese when they develop diabetes, and obesity aggravates the insulin resistance. BMI and waist circumstance may become the main focus on attention of interactions, particularly in studies of IVS3C>T and IVS4G>C polymorphism among T2DM. Cauchi and colleagues [117] reported that IVS3C>T risk allele is more prevalent in nonobese than in obese diabetics. Humphries et al [67] found a statistically significant interaction between IVS3C>T as well as IVS4G>C genotype and BMI. Wang et al [118] replicated the result of Humphries et al. In succession, Helgason et al [47] discovered that two haplotypes, that is, HapA and HapB T2D had a interaction with BMI, and that the former was associated with increased BMI, whereas the latter was associated decreased BMI. Florez et al [106], observed that another index of obesity, waist circumstance, showed a nominally significant interation with both IVS3C>T and IVS4G>T in spite of no interaction between BMI and genotype. In addition, plasma triglyceride (TG) and C-reactive protein (CRP) have also been found to interact with genotype of both IVS3C>T and IVS4G>T [67]. These early findings suggestive of geneenvironment interactions with the different TCF7L2 gene polymorphisms should be interpreted extremely cautiously, however, and much larger and more detailed studies are looked forward to substantiating such putative interactions with appropriate power and rigor. Complementarily, Smoking and drinking were important risk factors to influence on T2DM and they can also interact with genetic factor to increase the risk of this disease [116]. We found a higher proportion of current or/and past smokers and drinkers in the study populations. But, it was regretted for us that there was no any report about the interactions between TCF7L2 gene polymorphisms and smoking or/and drinking in all the open publications. This is a meaningful and valuable domain to explore.

Limitation of this meta-analysis
The lack of information especially the genotype data from some articles was the main limitation, we have done our possible to contact exhaustibly to the authors of publication that did not provide the original data and obtained full data on about 75 percent of subjects involved in various ethnic populations. Next, the quality of diagnosis, match, and genotyping varied a lot among studies, but the summary results were not changed when these questions were discussed carefully. There was considerable heterogeneity in IVS3C>T and IVS4G>T polymorphisms in our initial results. However, we discovered the main sources of the heterogeneity and were able to remove it finally. A question we cannot resolve is that how the four snps discussed here in strong LD will lead to differences in T2DM association among diverse populations. Besides, since environment factors and genes, such as obesity, smoking, drinking, and hypertension as well as plasma TG, and gene related to the Wnt signaling pathway are strong candidates for a potential interaction with TCF7L2 gene polymorphisms, the lack of data available to us on most of those environmental factors (except for obesity) and genes was a limitation, which we hope will be demonstrated by the following studies.

Conclusions and recommendation for future researches
This large meta-analysis summarizes the strong evidence for an association between TCF7L2 gene and T2DM both overall and in Caucasians, North Europeans, East Asians, Indians, and Africans, and suggested a potential multiplicative genetic model for all the four polymorphisms of TCF7L2 gene among different ethnic populations except for Africans, where additive genetic mode is suggested for IVS3C>T polymorphism. Our study results also suggest that IVS3C>T and IVS4G>T variants of TCF7L2 gene can be taken as reference loci for exploring T2DM susceptibility. We found and removed the main source of heterogeneity; however, we cannot find any evidence of bias. Furthermore, we estimated the potential gene-gene and gene-environmental interactions by which common variants in the TCF7L2 gene influence risk of T2DM. These two domains are just the recommendation for future researches.