We used data from the FHS to study associations between variants in TCF7L2 and quantitative diabetes- and obesity-related traits. The FHS is a community-based, multi-generational, longitudinal study of cardiovascular disease and its risk factors, including diabetes. The FHS is comprised of the Original Cohort, Offspring Cohort, and Third Generation Studies. Participants described in this analysis consist of 2,512 individuals from the FHS Offspring Cohort. In this analysis, our quantitative traits of interest are from Offspring examination 7 (1998–2001) where phenotypic data from a 75 g OGTT are available for all Offspring without diagnosed diabetes, and where proinsulin and insulin levels were obtained for proinsulin/insulin ratios. The study was approved by Boston University’s Institutional Review Board and written informed consent, including consent for genetic analyses, was obtained for all study participants. The demographic characteristics of the FHS study population are presented in Table 1.
An extensive array of diabetes- and obesity-related quantitative traits have been collected in the FHS. Diabetes quantitative traits measured in this study include: HbA1c, fasting plasma glucose (FPG), fasting insulin, insulin resistance by HOMA (HOMA-IR), proinsulin/insulin ratio, Gutt’s 0–120 min insulin sensitivity index (Gutt’s ISI) and the time-averaged mean FPG level over examinations 3 to 7, spanning 16 years (mean FPG). Individuals being treated for diabetes with oral hypoglycaemic or insulin therapy were excluded from the analysis of the glycaemic traits; however, participants with diet-controlled diabetes were not excluded. Data on mean FPG and Gutt’s ISI for the polymorphism rs7903146, using different association tests, have been presented previously in the context of a 100 K genome-wide association scan . Laboratory methods for all quantitative traits have been described previously .
Obesity traits measured included: mean BMI 35–50 (which is the mean BMI between ages 35 and 50 years) and mean BMI 50–65 (which is the mean BMI between ages 50 and 65 years), and BMI, waist circumference, subcutaneous adipose tissue (SAT) and visceral adipose tissue (VAT) from examination 7. SAT and VAT were performed by multi-detector computed tomography assessment with adipose compartments delineated by manually tracing the abdominal wall as described previously , and were available in approximately one half of the participants for whom we had anthropomorphic measurements.
We used the 2003 American Diabetes Association clinical criteria to define diabetes, where a case was defined as use of oral hypoglycaemic or insulin therapy, or a FPG ≥ 7.0 mmol/l at the index examination and FPG ≥ 7.0 mmol/l on at least one prior examination .
We selected four SNPs in TCF7L2 that have been previously reported to be involved in type 2 diabetes and/or increased BMI: rs12255372, rs7903146, rs10885406 and rs7924080. As suggested by Helgason et al. , the haplotype designated HapA was tagged by two different methods: either by the combination of two SNPs, rs10885406 A and rs7903146 C, or by rs7924080 T alone. As shown below, they yielded essentially the same results.
Genotyping was performed by allele-specific primer extension of multiplex amplified products with detection by matrix-assisted laser desorption ionisation-time of flight mass spectroscopy on an iPLEX Sequenom platform (Sequenom, San Diego, CA, USA). Average genotyping call rates were 98%, and the average consensus rate based on 254 duplicate samples was 99%.
The analysis of the quantitative traits was carried out with linear mixed-effects models that tested the association between Studentised residuals of the traits and the SNPs using an additive genetic coding. The following traits were log-transformed: HbA1c, FPG, fasting insulin, HOMA-IR, mean FPG, proinsulin, mean BMI 35–50 and mean BMI 50–65. Traits that were not transformed were Gutt’s ISI, waist circumference, SAT and VAT. Studentised residuals were generated for men and women separately, and two adjustment models were used: the first with age and age2, and the second with age, age2 and BMI (kg/m2). The adjustment for both age and age2 allowed for a non-linear relationship between the trait and age. The covariates for each trait were taken from the examination at which the trait was measured, except for the mean FPG and mean BMI traits, in which case they were obtained over the same time period. The SOLAR software package was used to implement the linear mixed-effect models to account for within-family correlation of the traits . The models included within-family random effects with a covariance structure determined by the degree of relatedness between each relative pair. For the association analysis we used a linear mixed-effects model to account for the relatedness between individuals, and for the survival analysis we used a frailty term in our model to account for the correlation (in diabetes survival) between family members.
An additional analysis was carried out to determine the association between the SNPs and the ratio of proinsulin to insulin. The participant-specific insulin/pro-insulin ratio was log-transformed and Studentised residuals were obtained using examination 7 covariates, and the analysis with linear mixed-effect models was performed as described above.
In order to distinguish whether a higher proinsulin/insulin ratio indicates an intrinsic defect in insulin processing or the hypersecretion of proinsulin by beta cells that cannot keep up with insulin demand, we assessed whether the association between rs7903146 and log(insulin/proinsulin) was jointly modified by BMI and HOMA-IR. If a higher proinsulin/insulin ratio stems from an intrinsic defect in insulin processing, the association should be present in both insulin-resistant and insulin-sensitive individuals; conversely, if it is due to a system that cannot meet the demands induced by high insulin resistance, the association should be much stronger in insulin-resistant than in insulin-sensitive individuals. To distinguish between these hypotheses, the sample was divided into tertiles of BMI and HOMA-IR: individuals in the highest tertile of both BMI and HOMA-IR were placed in a high BMI/high HOMA-IR group (n = 425), while individuals in the lowest tertile of both BMI and HOMA-IR were placed in a low BMI/low HOMA-IR group (n = 415). A linear mixed-effect model was run with log(insulin/proinsulin) as the outcome to test for interaction between BMI/HOMA-IR group and genotype at rs7903146. Using QUANTO software, with 840 participants included in the interaction analysis, we have 80% power to detect a genotype × group interaction explaining 0.9% of the variance in log proinsulin ratio .
We used the two-step approach proposed by Zaykin et al.  to test the association of our obesity-related quantitative traits with a haplotype, denoted HapA, defined by the A-C alleles of SNPs rs10885406 and rs7903146. First, haplotype frequencies were obtained for our sample using the haplo.stats software implemented in the software package R . This software assumes that the sample is unrelated; although the haplotype frequencies observed in the whole sample were very similar to the haplotype frequencies observed in an unrelated subset of the sample. Individual dosages of the HapA haplotype, varying between 0 and 2, were obtained using posterior probabilities from the maximum likelihood estimation procedure. Linear mixed-effect models were run to assess the association between HapA and obesity-related quantitative traits. We tested the association between obesity (defined as BMI ≥ 30 kg/m2) and HapA using logistic regression generalised estimating equations, in which we used pedigree as a cluster and an independence correlation structure.