including reasons for exclusion can be found in Additional file 1: Table S2. The remaining 3669 methylation probes were checked for association with tobacco smoking using a linear mixed model with the LME4 package in R version 3.1.0 with Dasen normalized beta-values of the CpG sites as outcome measure [27]. We first compared current to never smokers and then performed a sensitivity analysis on the identified CpG sites comparing current to former smokers. Covariates were selected based on known association with DNA methylation and different distributions between current and never smokers in our samples. The selected covariates with fixed effects were age, sex and BMI [28–31]. Houseman estimated white blood cell proportions were used as fixed effects to correct for cell mixture distribution [32]. Array number and position on array were added in the model as covariates with random effects to correct for batch effects. We corrected for multiple testing using a robust Bonferroni-corrected p-value of 1.4 × 10−5 as the threshold for significance (0.05/3669 probes).
