EPIC-NL: Study design

In the EPIC-NL cohort the two Dutch contributions to the European Investigation into Cancer and Nutrition (EPIC) project are combined: the Prospect-EPIC and the MORGEN-EPIC (Monitoring Project on Risk Factors for Chronic Diseases) cohorts. Both cohorts were initiated in 1993.The study design of the combined cohort is described in detail elsewhere [8]. In brief, Prospect is a prospective cohort study among 17 357 women aged 49–70 who participated in a breast cancer screening program between 1993 and 1997. The MORGEN-project consists of 22 654 men and women aged 20–59 years recruited from three Dutch towns (Amsterdam, Doetinchem, and Maastricht). From 1993 to 1997, each year a new random sample of approximately 5000 individuals were examined.

Laboratory and genetic analyses were performed in a 6.5% random sample of the EPIC-NL study, in all incident T2D cases and in all incident CVD cases. In our study we only used the random sample. After exclusion of participants with missing blood samples (n = 157), missing values for haemoglobin A1c (HbA1C), waist circumference, high-density lipoprotein (HDL) cholesterol, systolic blood pressure, diastolic blood pressure, triglycerides or C-reactive protein (CRP) (n = 128), or with missing SNP data (n=433) the study population consisted of 1886 participants. All participants signed informed consent before study inclusion. Both studies complied with the Declaration of Helsinki. The Prospect-EPIC study was approved by the Institutional Review Board of the University Medical Center Utrecht and the MORGEN project was approved by the Medical Ethical Committee of TNO, The Netherlands.

Baseline measurements

At baseline, a physical examination was performed and non-fasting blood samples were drawn. During the physical examination, systolic and diastolic blood pressure measurements were performed twice in the supine position on the right arm using a BosoOscillomat (Bosch & Son, Jungingen, Germany) (Prospect) or on the left arm using a random zero sphygmomanometer (MORGEN). The mean of both measurements was taken. Waist circumference and height were measured to the nearest 0.5 cm. Body weight was measured with light indoor clothing without shoes on, to the nearest 100 gr.

Statistics

Distributions of genotypes were tested for deviation from hardy-weinberg equilibrium (HWE) by chi-square analyses. Triglycerides and hsCRP were log-transformed to improve normality. Participants on blood pressure medication were excluded from the analyses on blood pressure, participants on glucose lowering medication from the analysis on HbA1C, and participants with acute inflammation (hsCRP > 10 mmol/L) from the analyses on hsCRP.

SNPs were divided into 5 groups (Table 1) according to the known associations in GWAS. These groups of SNPs were associated with the corresponding phenotype using the linear regression model of Goeman’s global test [7]. Subsequently, for each group of SNPs the association with MetS was analysed using the log-linear model of Goeman’s global test, and the association with MetS-score using the linear regression model of Goeman’s global test (see Figure 1: data-analyses scheme). If one of these associations was statistically significant, we conducted additional data-analyses. First, to see whether the association with the group of SNPs was mediated by the corresponding phenotype from GWAS, we adjusted the association between MetS and the group of SNPs for this phenotype. Second, we tested if the group of SNPs was also associated with the individual MetS features or hsCRP using the linear regression model of Goeman’s global test. Third, we analysed the association of the individual SNPs in this group with MetS using log-linear models and with MetS-score using linear regression. For the individual SNPs which were significantly associated with MetS or MetS-score, we analysed associations with the individual MetS features and hsCRP using linear regression. We only conducted analyses with individual SNPs, for those SNPs which were on the group level associated with MetS or MetS-score. Consequently, only a few associations with individual SNPs were analyses. Therefore, adjustment for multiple testing is not appropriate.

All analyses were adjusted for age, sex and cohort. Significance was defined as a 2-sided P-value <0.05. The global test was calculated in R version 2.12.1 (R Foundation for Statistical Computing; http://www.r-project.org). The analyses for individual SNPs were performed with SAS version 9.2 (SAS Institute, INC., Cary, North Carolina).

Waist circumference SNPs

In our study the group of SNPs which were associated with waist circumference in GWAS (MC4R rs17782313 and FTO rs1421085) was associated with waist circumference, as well as with MetS. The association with MetS disappeared after adjustment for waist circumference, indicating that the association with MetS is driven by the association with waist circumference. Furthermore, the association with MetS seemed mainly to be driven by MC4R rs17782313. In the KORA study among 7888 adults, an association between MC4R rs2229616 (r²=1 with rs17782313) and MetS was found [30], supporting our findings. Unfortunately, in our study, data on MC4R rs17782313 were available for women only. However, since in the KORA study [30] the association between MC4R rs2229616 and MetS was not dependent on sex, we expect that this did not influence our findings. In our study the association between MC4R rs17782313 and MetS remained after adjustment for waist circumference. Furthermore, we found no association between MC4R rs17782313 and any individual MetS feature, including waist circumference. This suggests that the association between MC4R rs17782313 and MetS, is at least in part, independent of body weight. In both human and animal studies MC4R rs17782313 had an effect on insulin resistance, independent of body weight [31, 32]. Therefore, insulin resistance may explain part of the association between MC4R rs17782313 and MetS. Contrary to a meta-analysis among 12,555 Europeans, in which FTO rs9939609 (r²=1 with rs1421085) was significantly associated with MetS (OR 1.17 ; 95% CI 1.10-1.25) [33], we did not observe an association between FTO rs1421085 and MetS. This discrepancy may be explained by the weak association between FTO rs1421085 and waist circumference in our study. In our study the regression coefficient between FTO rs1421085 and waist circumference was 0.03 per SD, whereas in other studies it ranged from 0.07 per SD to 0.14 per SD [33].

Insulin resistance SNPs

We found an association between insulin resistance SNPs and MetS and MetS-score that remained after adjustment for HbA1c, indicating that this association was not driven by HbA1C. However, since HbA1C is not an optimal marker of insulin resistance (r² between HOMA-IR – HbA1C ≈0.50 [34]), we can not rule out that insulin resistance mediates this association. Out of the group of five insulin resistance SNPs, IRS1 rs2943634 was the only SNP significantly associated with MetS and MetS score. It was also associated with HbA1C, triglycerides and HDL cholesterol. Accordingly an IRS1 knock-out mouse model displayed a MetS like phenotype with insulin resistance, increased blood pressure, increased triglycerides, decreased HDL cholesterol and decreased LPL activity [35]. Furthermore, in human studies IRS1 rs2943634 has been associate with glucose related [15] and lipid traits [36]. In contrast, in a study among 1126 non-Hispanic whites, 898 non-Hispanic blacks and 906 Mexican Americans, IRS1 rs7578326 (r² with rs2943634=0.82) was not associated with MetS, neither in the overall population, nor in specific ethnic groups [37]. However, as the number of Caucasian participants and MetS prevalence were lower than in our study, the former study may have been underpowered in Caucasian. Besides IRS1 rs2943634 the group of insulin resistance SNPs consisted of PPARG rs1801282, GCKR rs780094, GCK rs1799884, and IGF1 rs35767. In line with other studies, none of these SNPs were associated with MetS in our data [4, 37, 38]. This may be explained by the relatively weak effect on HOMA-IR of PPARG rs1801282, GCK rs1799884, and IGF1 rs35767 [14, 15] or by pleiotropic effects of PPARG rs1801282 and GCKR rs780094. The 12Pro allele of PPARG rs1801282 has opposite effects on insulin resistance and BMI in Caucasian subjects [39, 40], whereas GCKR rs780094 has opposite effects on insulin resistance and lipid levels [27]. These opposite effects may result in a zero association with MetS, as observed by for example Passaro et al. [40].

Lipid SNPs

We did not observe an association between groups of SNPs known for their association with triglycerides or HDL cholesterol and MetS. On the contrary, in a GWAS [5] and a systematic review of genetic association studies [4], the majority of SNPs associated with MetS was involved in lipid metabolism. The lack of an association with MetS may be a power issue, because the association between lipid SNPs and lipid levels was relatively weak in EPIC-NL. Subgroup analyses revealed that these weak associations were consistent for all lipid SNPs and could not be explained by medication use, sex or a difference between the MORGEN and Prospect study. Furthermore, it is unlikely that the non-fasting state of our samples gives an explanation, as in a GWAS, the association with lipid levels was independent of the fasting state for most SNPs [41].

Inflammation SNPs

We found no significant association between a group of inflammation SNPs and MetS. Accordingly, in a study among 4286 British women, a CRP haplotype was not associated with the individual features of MetS [42]. Furthermore, Rafiq et al. could not detect an association between T2D, an endpoint of MetS and 8 SNPs known to alter circulating levels of inflammatory proteins, which were located in the IL-18, IL1RN, IL6R, MIF, PAII and CRP genes [43]. Overall, this evidence may suggest that genetic variants in inflammatory genes do not play a causal role in MetS development. However, for several reasons it cannot be ruled out that some SNPs in inflammatory pathways are causally related to MetS. First, for some inflammatory proteins SNP-MetS associations have not been investigated yet. Second, as the global test gives a combined result for all SNPs, the global test may be not significant despite the presence of an association between one of the single SNPs and MetS.

In this study we have explored the biomarkers involved in MetS development, by studying SNPs related to these biomarkers. Advantage of this approach is that according to the principles of Mendelian randomization the associations we investigated are neither affected by reverse causality nor by socioeconomic and behavioural confounders [44]. Furthermore, as all participants were Caucasian, it is unlikely that our study results have been affected by population stratification. However, for some SNPs we measured, like IRS1 rs2943634 [45], allele frequencies are very heterogeneous among different populations. Therefore, our results warrant replication in other study populations. Sixteen SNPs which were associated in GWAS with MetS related traits were not on the IBC CVD array. Inclusion of the three waist circumference SNPs, which were not on the array, would probably have increased the possibility to find associations with several MetS features. As the global test of inflammation SNPs on hsCRP was already highly significant (P=7.3*10^-6), inclusion of additional SNPs, which were absent on the IBC CVD array, would not have changed our results for the global test, but may have revealed additional individual SNPs. The total number of lipid SNPs in our study was relatively large and relatively few lipid SNPs were missing. Therefore we believe that inclusion of additional lipid SNPs would not have changed our results considerably on the group level. The IBC CVD array covered all insulin resistance SNPs discovered in GWAS. However, up till now for only three SNPs a genome wide association with HOMA-IR has been found and replicated. To increase power we also included those SNPs associated with glucose related traits in GWAS, which were also associated with HOMA-IR (P≤0.05). However, as the association between the group of insulin resistance SNPs and HbA1C was just significant, the power to detect associations with MetS and its features was still low.

In conclusion, we found that SNPs associated with waist circumference or insulin resistance in GWAS were also associated with MetS. These results are in line with the hypotheses that weight regulation and insulin metabolism are causative factors for MetS.

Individual SNPs for which we found an association with MetS were MC4R rs17782312 which is involved in weight regulation and IRS1 rs2943634 which is involved in insulin resistance.

Inflammatory biomarkers

HNF1A – rs1183910[20]

ARL15 – rs4311394 [18]

APOE, APOC1, APOCII – rs4420638 [47]

CDH13 – rs3865188 [48]