Study populations

PAGE involves several studies, described briefly below and in greater detail on the PAGE website (https://www.pagestudy.org). All studies were approved by Institutional Review Boards at their respective sites, and all participants provided informed consent.

Causal Variants across the Life Course (CALiCo) is a consortium of six demographically diverse population based studies and a central laboratory, and includes approximately 58,000 men and women ranging in age from adolescence to older adulthood. Three CALiCo studies participated in the present analysis: Atherosclerosis Risk in Communities Study (ARIC) (N = 13,383) [4], Cardiovascular Health Study (CHS) (N = 4,509) [5], and Strong Heart Cohort Study (SHCS) (N = 1,714) [6]. In addition to the studies involved in the CALiCo consortium, PAGE includes three other large studies. The Multiethnic Cohort (MEC) is a population-based prospective cohort study of over 215,000 men and women in Hawaii and California aged 45–75 at baseline (1993–1996) and primarily of five ancestries [7]. Participants eligible for the present study were controls in nested case–control studies of breast, colorectal, or prostate cancer or for biomarker studies, and who had glucose and/or insulin measurements (N=942). This analysis also included data from the Epidemiologic Architecture for Genes Linked to Environment (EAGLE) study. EAGLE accesses the genetic component of three National Health and Nutrition Examination Surveys (NHANES): NHANES III (phase 2 collected between 1991 and 1994), NHANES 1999–2000, and NHANES 2001-2002 [8–10]. Overall, 7,719 NHANES participants aged 18 and older were included in these analyses. Finally, the Women’s Health Initiative (WHI) is a multifaceted clinical trial and cohort study investigating post-menopausal women’s health in the U.S [11]. Out of the 161,808 women enrolled in WHI, 8,312 were selected and included in the present study. Except for the Women’s Health Initiative, all studies recruited men and women. All studies collected self-identified racial/ethnic group via questionnaire. In the current analysis, we included "East Asians" defined as MEC participants who identified themselves as of sole or mixed Japanese descent, and WHI participants of Japanese, Chinese, Filipino, Vietnamese, and/or Korean ancestry. Fasting glucose and insulin concentrations were measured using standard assays, at laboratories specific to each PAGE site.

At all PAGE sites, we excluded underweight (BMI<18.5 kg/m²) and extremely overweight (BMI>60 kg/m²) individuals with the assumption that these extremes could be attributable to data coding errors, an underlying illness or possibly to a familial syndrome and hence, a rare mutation. We excluded individuals self-reporting that they have ever been diagnosed with diabetes, or who report taking diabetes medications. In addition, to mirror typical exclusion criteria of other studies of glucose homeostasis, we also excluded individuals with fasting glucose concentrations consistent with diabetes (i.e., ≥126 mg/dl or ≥7.0 mmol/L ), regardless of self-reported diabetes status.

After applying the above exclusion criteria, a total of 36,579 participants were selected from the PAGE consortium for analysis.

SNP selection and genotyping

Ten SNPs in 8 genetic regions were selected for genotyping based on prior GWAS findings of positive association with glucose or insulin concentrations, and exceeding a genome-wide significance of p <5 × 10^-8 in studies published through 2010 [12–14]. Nine SNPs were previously associated with glucose, and 2 were associated with insulin, with 1 of these SNPs associated with both quantitative traits (rs780094/GCKR). In the glucose analysis, we included an additional GWAS finding for type 2 diabetes (rs7903146/TCF7L2) that had been subsequently associated with fasting glucose concentrations [15]. Each PAGE site prioritized which SNPs to genotype based on investigator interests, genotyping platforms, and resources, resulting in heterogeneity of available glucose or insulin SNPs across racial/ethnic groups. Ten SNPs were genotyped in European Americans and African Americans, 4 were genotyped in Hispanics and in American Indians, and 2 were genotyped in East Asians.

DNA extraction and genotyping methods followed standard protocols. Each PAGE site employed different genotyping platforms, with similar quality control criteria. CALiCo sites used TaqMan, the Illumina 370CNV BeadChip, the Affymetrix Genome-Wide Human SNP Array 6.0, and the Illumina HumanCVD BeadChip. A portion of CHS genotype data was obtained from a previous GWAS. EAGLE used Sequenom’s iPLEX® Gold coupled with MassARRAY MALDI-TOF MS detection and Illumina’s BeadXpress with a custom GoldenGate genotyping assay. MEC used Applied Biosystems OpenArray and TaqMan. WHI used Illumina BeadXpress with the Veracode GoldenGate genotyping assay. All sites used internal and blinded external controls, and excluded genotypes deviating from Hardy-Weinberg expectations (p-value < 0.001) or with low concordance (typically, <95% - 99%). In addition to site-specific quality control, all PAGE study sites genotyped 360 DNA samples from the International HapMap Project and submitted these data to the PAGE Coordinating Center for concordance checks [16]. Additional details on data collection, specimen processing, and genotyping are found in the Additional file 1: Supplementary Methods.

Statistical analysis

In order to maximize comparability with prior studies of glucose homeostasis, we converted insulin and glucose concentrations into units commonly reported in the literature. Thus, we investigated continuous fasting glucose (mmol/L) and natural log transformed fasting insulin (pmol/L). The association between each SNP and its related quantitative trait was estimated using linear regression with robust standard errors (SEs) [17]. SNP genotype was coded assuming an additive genetic model (i.e., 0, 1, or 2 copies of the coded allele). For ease in interpreting the results, we coded the allele that was associated with an increased insulin or glucose concentration in the prior GWAS. All analyses were stratified by self-identified racial/ethnic group, and adjusted for covariates known to be associated with insulin and/or glucose concentrations: smoking (current vs. former/never; smoking increases insulin resistance) [18], continuous BMI (obesity is associated with insulin resistance) [19], sex (insulin metabolism differs by sex) [20], and continuous age (insulin metabolism varies by age) [21]. Analyses were performed for each of the 6 participating PAGE studies separately and study-specific results (effect sizes and robust SEs) were combined with fixed-effects meta-analysis using R.

Based on our hypothesis that GWAS-identified glucose and insulin SNPs are associated with glucose and/or insulin concentrations across all race/ethnicities, we did not adjust for multiple testing. We labeled meta-analysis results as "replicating" (for EA) or "generalizing" (for other racial/ethnic groups) if the beta was in the same direction as the original GWAS, and was statistically significant (i.e., p < 0.05). All aggregate results will be available via dbGaP (http://www.ncbi.nlm.nih.gov/gap) at a future date.

Approximately 13% of the overall WHI study cohort was selected to contribute to PAGE. This selection was non-random, and was enriched for subjects with certain incident health conditions (e.g., cardiovascular disease and stroke), non-European American race/ethnicity, and BMI>40. Therefore, analyses of WHI data incorporated inverse probability weighting to account for this sampling strategy.

We only reported results if the meta-analysis sample size was > 400. For each racial/ethnic group, we estimated the statistical power to detect the GWAS-reported effect sizes for each SNP using Quanto (http://hydra.usc.edu/gxe/), assuming the same effect size as reported in the prior GWAS, an additive genetic model and a two-sided test of association at p = 0.05. Power calculations were based on allele frequencies specific to each racial/ethnic group. We evaluated I² as a measure of heterogeneity [22], to describe the presence or absence of excess variation across the PAGE study sites.

Glucose SNPs

Insulin SNPs

Table 3 lists meta-analysis results for EA and AA, for 2 SNPs (rs780094/GCKR and rs35767/IGF1) associated insulin in prior GWAS. The association between rs780094/GCKR and (ln) insulin replicated in EA (p-value = 1.3 × 10^-10), but did not generalize to AA (p = 0.12). This association was also significant in Hispanics (p = 0.01) and American Indians (p = 4.8 × 10^-4), but not East Asians (p = 0.39). The association between rs35767/IGF1 and insulin was not significant in EA or AA.

Evidence for heterogeneity

Overall, we observed little evidence of heterogeneity across studies. In EA, rs7903146/TCF7L2 had a statistically significant p-value for heterogeneity in the association with glucose (χ² = 18.79, p = 0.001), For this association, the site-specific betas for WHI, CHS, EAGLE, ARIC, and MEC were 0.03, 0.03, 0.02, 0.14, and 0.12 mmol/L, respectively. In American Indians, rs560887/G6PC2 had a statistically significant I² p-value for heterogeneity in the association with glucose in American Indians (χ² = 7.31, p = 0.03). For this association, the data came from 3 sites of SHCS (Arizona, Oklahoma, and South Dakota); and the betas were 0.42, -0.01, and 0.17 mmol/L. For each of these SNPs, results of random effects models are presented in Table 3.