Epigenetic signatures of starting and stopping smoking

BACKGROUND: Multiple studies have made robust associations between differential DNA methylation and exposure to cigarette smoke. But whether a DNA methylation phenotype is established immediately upon exposure, or only after prolonged exposure is less well–established. Here, we assess DNA methylation patterns from peripheral blood samples in current smokers in response to dose and duration of exposure, along with the effects of smoking cessation on DNA methylation in former smokers. METHODS: Dimensionality reduction was applied to DNA methylation data at 90 previously identified smoking–associated CpG sites for over 4900 individuals in the Generation Scotland cohort. K–means clustering was performed to identify clusters associated with current and never smoker status based on these methylation patterns. Cluster assignments were assessed with respect to duration of exposure in current smokers (years as a smoker), time since smoking cessation in former smokers (years), and dose (cigarettes per day). FINDINGS: Two clusters were specified, corresponding to never smokers (97·5% of whom were assigned to Cluster 1) and current smokers (81·1% of whom were assigned to Cluster 2). The exposure time point from which >50% of current smokers were assigned to the smoker–enriched cluster varied between 5 and 9 years in heavier smokers and between 15 and 19 years in lighter smokers. Low–dose former smokers were more likely to be assigned to the never smoker–enriched cluster in the first year following cessation. In contrast, a period of at least two years was required before the majority of former high–dose smokers were assigned to the never smoker–enriched cluster. INTERPRETATION: Our findings suggest that smoking–associated DNA methylation changes are a result of prolonged exposure to cigarette smoke, and can be reversed following cessation. The length of time in which these signatures are established and recovered is dose dependent. Should DNA methylation–based signatures of smoking status be predictive of smoking–related health outcomes, our findings may provide an additional criterion on which to stratify risk.