The National Heart, Lung and Blood Institute (NHLBI) Pooled Cohorts Study12 harmonised data from four prospective population-based studies that evaluated non-cigarette tobacco use: Atherosclerosis Risk in Communities Study (ARIC)13; Coronary Artery Risk Development in Young Adults (CARDIA)14; Multi-Ethnic Study of Atherosclerosis (MESA)15 and Strong Heart Study (SHS).16 Design features of the cohorts are described in online supplemental table 1. Participants with comprehensive smoking assessments and DNA methylation measurements were included in the analyses.

The NHLBI Pooled Cohorts Study was approved by NHLBI and the institutional review boards of all collaborating institutions. The study complies with the Declaration of Helsinki. All participants provided written informed consent.

DNA was extracted from venous blood and treated with bisulfite conversion. DNA methylation profiling was performed in a random subset of participants in ARIC, CARDIA and MESA, and in study participants with available covariate data in SHS. Inclusion criteria for DNA methylation profiling are further described in online supplemental figure 1. Methylation levels were measured using the Illumina Infinium HumanMethylation450K BeadChip (Illumina., San Diego, California, USA) in ARIC and MESA and the Methylation EPIC BeadChip (Illumina) in CARDIA and SHS. Each cohort performed quality control, statistical preprocessing, data normalisation and batch correction according to validated protocols.18 19 DNA methylation measurements produced by the two platforms are proven to be highly correlated, substantiating the application of combined data in population-based research.20

Prior research has shown that cg05575921, a CpG site contained in the AHRR gene, is highly associated with cigarette smoking,6 has promising potential for clinical use21 and has a methylation set point that is not affected by ethnicity-specific variation.22 The present study thereby evaluated associations of tobacco use patterns with cg05575921 methylation, which was measured in peripheral blood leucocyte DNA a median of 2.8 years after smoking habits were assessed. AHRR methylation was quantified using methylation M-values and β -values. The M-value is the log2 ratio of the amount of methylated probe versus unmethylated probe and was used to derive all reported p values because M-values can be analysed parametrically.23 The β -value represents per cent methylation (range: 0–1) and was used to derive all reported effect estimates to aid in the biological interpretation of results. Additional details on methylation assays and quality control procedures are reported in online supplemental methods.

Lung function was measured using water-seal, dry-rolling seal or flow-sensing spirometers in accordance with American Thoracic Society guidelines.12 Spirometry was quality controlled using 2005 criteria.24 All cohorts measured prebronchodilator forced expiratory volume in 1 s (FEV1) and forced vital capacity (FVC). Respiratory symptom questionnaires were used to evaluate the presence of respiratory symptoms including chronic cough, sputum production, wheezing and shortness of breath in ARIC and CARDIA.25 All-cause mortality was defined as death from any cause. Vital status was ascertained through regular interviews with family members and/or linkage with the National Death Index.26

Age, sex, race/ethnicity and education were self-reported. Height and weight were measured using standard methods. Medications were self-reported or assessed via medication inventories. Hypertension was defined by blood pressure (≥140/90 mm Hg) or use of antihypertensive medications. Diabetes was self-reported or defined by fasting blood glucose (≥126 mg/dL) or the use of relevant medications. Beyond the exclusion of 50 participants with incomplete smoking assessments, 313 participants (3.7%) were missing data for one or more covariates (online supplemental figure 2).

Associations of tobacco use patterns with AHRR methylation were estimated using linear regression. Tobacco use was categorised as never use (reference group), exclusive use of non-cigarette tobacco products, exclusive use of cigarettes or dual use of non-cigarette tobacco products and cigarettes. Three models were used to evaluate associations of tobacco use patterns with AHRR methylation. Model 1 did not adjust for additional parameters. Model 2 adjusted for a priori confounders including age, sex, race/ethnicity, education (high school or less vs some college or more), study cohort and years from smoking assessment to AHRR methylation measurement.27 Model 3 further adjusted for cigarette smoking status, cigarette pack-years, hypertension, diabetes and body mass index (BMI). Linearity, homoscedasticity and normality were assessed using residual plot analyses. No violations of linearity assumptions were observed. Reported p values are from models evaluating methylation M-values because M-values can be analysed parametrically.23 Reported effect estimates are from models evaluating methylation β -values to aid in the biological interpretation of results.

To determine whether non-cigarette tobacco use further altered AHRR methylation among participants who smoked cigarettes, current and former cigarette smokers were stratified by cigarette smoking pack-years (<5 pack-years, ≥5 and <15 pack-years, ≥15 and <25 pack-years or ≥25 pack-years). Non-cigarette tobacco exposure was categorised as never use (reference group) or ever use. Three different linear regression models were sequentially adjusted for a priori confounders as described above.

To evaluate associations of AHRR methylation with smoking-related health outcomes among participants with distinct tobacco use patterns, participants were stratified by smoking exposures (never use, exclusive non-cigarette tobacco use, exclusive cigarette use or dual non-cigarette tobacco and cigarette use). Within each stratum, associations of AHRR methylation with key clinical outcomes were tested. Linear mixed models were used to test associations of AHRR methylation with lung function. Spirometry assessments collected at the time of DNA methylation assays and in subsequent visits were included in the analyses. Linearity assumptions were confirmed using residual plot analyses. Participants with one spirometry assessment were included in linear mixed models to increase model precision.11 Unstructured variance–covariance matrices and random intercepts were used to account for within-subject correlation and to optimise model fit.28 All models were adjusted for a priori confounders including time-varying age, sex, race/ethnicity, education, study cohort, height and years between AHRR methylation assessment and spirometry. Analyses of participants reporting dual use of non-cigarette tobacco products and cigarettes were additionally adjusted for cigarette smoking status and cigarette pack-years. To assess the longitudinal association of AHRR methylation with a rate of change in lung function, the two-way multiplicative interaction term between AHRR methylation level and time from methylation assay to spirometry evaluation was tested. Because non-cigarette tobacco use was more common in male participants, sensitivity analyses were restricted to male participants.

To evaluate associations of AHRR methylation with respiratory symptom burden, participants were stratified by distinct tobacco use patterns. Logistic regression was used to test associations of AHRR methylation with the presence of respiratory symptoms. Models were adjusted for a priori confounders including age, sex, race/ethnicity, education, study cohort, hypertension, diabetes and BMI. Analyses of participants reporting dual use of non-cigarette tobacco products and cigarettes were additionally adjusted for cigarette smoking status and cigarette pack-years.

To evaluate associations of AHRR methylation with all-cause mortality, participants were stratified by distinct tobacco use patterns. Kaplan-Meier curves were used to compare survival for participants in the lowest quartile of AHRR methylation (β value <0.75) to participants in the remaining quartiles of AHRR methylation (β-value ≥0.75). The log-rank test was used to compare survival between the two groups within each stratum. Associations of AHRR methylation with all-cause mortality were also analysed using Cox proportional hazards models. The proportional hazards assumption was confirmed using the scaled Schoenfeld residual test.29 Models were adjusted for a priori confounders including age, sex, race/ethnicity, education, study cohort, hypertension, diabetes and BMI. Analyses of participants reporting dual use of non-cigarette tobacco products and cigarettes were additionally adjusted for cigarette smoking status and cigarette pack-years. Because non-cigarette tobacco use was more common in male participants, sensitivity analyses were restricted to male participants.

Statistical analyses were performed in R (V.4.0.3) using the lme4 (linear regression models) and survival packages (Cox models). A two-tailed alpha of 0.05 was considered statistically significant for all models.