Statistical analysis

The data was analyzed by a linear mixed model in R version 3.1.0 (http://www.r-project.org) with family number as a random factor and sex, age, BMI, and HbA1c as fixed factors in all models. All components of the mixed model were checked for distribution normality by evaluation of histograms. Factors that did not show normal distribution were transformed by natural logarithm. Residuals from the mixed model analyses were checked for normality by qq-plots. Furthermore, all analyses were run separately for each sex and without inclusion of T2D patients. The results from the mixed models are presented as β (effect estimate) with 95 % confidence intervals and P values. For logarithmically transformed variables, β corresponds to percentage change. P values ≤0.05 were considered significant. Spearman’s correlations (r) were used to analyze associations between methylation levels. Using the SOLAR software (solar-eclipse-genetics.org), the influence of familiality (i.e., genetic and shared environmental effects combined) on DNA methylation and gene expression of HIF3A was estimated from a polygenic model as the proportion of the additive genetic variation and shared environmental effects on the total variation (the variance component approach). In the SOLAR models, familiality of HIF3A DNA methylation and gene expression was adjusted for age, sex, BMI, and HbA1c levels.