Alcohol consumption and epigenetic age acceleration in young adults

Sample characteristics

Characteristics of study participants who underwent DNA methylation profiling at examination year (Y) 15 and Y20 were previously found to be largely similar to those participants who did not undergo DNA methylation profiling [19]. Table 1 presents descriptive characteristics for the 1,030 and 945 participants who underwent DNA methylation profiling by binge drinking status at Y15 and Y20, respectively. Among participants who drink, those who recently binge drank had an average of 24.6 ± 22.6 and 31.3 ± 27.5 total alcohol years at Y15 and Y20, compared to an average of 7.1 ± 8.4 and 10.8 ± 17.0 total alcohol years among those who did not recently binge drink, respectively. At both examination years, participants who recently binge drank exhibited a higher GAA compared to non-drinkers and participants who did not recently binge drink (P < 0.001), while the other EAA metrics did not show a similar trend.

Cumulative alcohol consumption on epigenetic age acceleration

Table 2 presents results for the associations between the cumulative alcohol consumption variables and EAA. After adjusting for covariates, beer years (P = 0.035 and P = 0.001) and liquor years (P = 0.003 and P < 0.001) were positively associated with GAA at Y15 and Y20, respectively. Each additional five beer years was associated with a 0.12-year [95% CI: 0.01, 0.22] and a 0.15-year [95% CI: 0.06, 0.24] higher GAA at Y15 and Y20, respectively. Each additional five liquor years was associated with a 0.32-year [95% CI: 0.11, 0.54] and a 0.33-year [95% CI: 0.16, 0.50] higher GAA at Y15 and Y20, respectively. Total alcohol years were positively associated with GAA (P = 0.005 and P < 0.001), with a 0.11-year [95% CI: 0.03, 0.19] and 0.13-year [95% CI: 0.07, 0.19] higher GAA per five alcohol years at Y15 and Y20, respectively. Wine years were not associated with GAA at either Y15 or Y20. Findings from GEE analyses yielded similar results as Y15 and Y20. IEAA (intrinsic epigenetic age acceleration), EEAA (extrinsic epigenetic age acceleration), and PAA (PhenoAge acceleration) were not associated with the cumulative alcohol consumption variables. Restricting to study participants with complete follow up yielded consistent conclusions (Supplementary Table 1). We observed similar findings when alcohol consumption was categorized into non-drinkers, low, intermediate, and high alcohol consumption (Supplementary Tables 2–4). Similar results as total alcohol years were observed using the cumulative amount of absolute alcohol consumed (Supplementary Table 5). When analyzing weekly alcohol consumption, only liquor consumption was positively associated with GAA (Supplementary Table 6). We observed correlations between beer, liquor, and total alcohol years and several GrimAge surrogate biomarkers of blood plasma proteins at Y15 and Y20, including adrenomedullin, leptin, and plasminogen activation inhibitor 1 (PAI-1) (Supplementary Figures 1, 2). Telomere length estimates were derived from DNA methylation and no cumulative alcohol variable was associated with telomere length (Supplementary Table 7).

Figure 1 displays quantile regression plots for the cumulative alcohol consumption variables at Y15 and Y20 on GAA. We plotted regression estimates for 19 quantiles ranging from 0.05 to 0.95. As displayed in the plots, beer years displayed an increasing effect on GAA across the distribution at Y15 with almost a three and a half times greater effect in the upper conditional quartiles compared to the lower conditional quartiles (0.31-year vs. 0.09-year, respectively; Figure 1A), with similar effects at Y20 (0.34-year vs. 0.10-year, respectively; Figure 1B). Liquor years demonstrated a thirty-four times greater effect in the upper conditional quartiles compared to the lower conditional quartiles at Y15 (0.68-year vs. 0.02-year, respectively; Figure 1C) and five times greater effect in the upper conditional quartiles compared to the lower conditional quartiles at Y20 (0.54-year vs. 0.11-year, respectively; Figure 1D). Wine years displayed a relatively negative effect on GAA in the lower conditional quartiles with a positive effect in the upper conditional quartiles at both Y15 and Y20. Notably, the effect estimate of wine years can be six and a half times greater in the upper conditional quartiles compared to the lower conditional quartiles of the distribution at Y15 (0.20-year gain vs. 0.03-year loss, respectively; Figure 1E) and nine times greater effect in the upper conditional quartiles compared to the lower conditional quartiles at Y20 (0.36-year gain vs. 0.04-year loss, respectively; Figure 1F). Total alcohol years demonstrated a four and a half times greater effect in the upper conditional quartiles of the distribution compared to the lower conditional quartiles (0.27-year vs. 0.06-year, respectively; Figure 1G) at Y15 and an approximately six times greater effect in the upper conditional quartiles compared to the lower conditional quartiles (0.25-year vs. 0.04-year, respectively; Figure 1H) at Y20.

Figure 1. Estimated parameters by quantile with 95% confidence limits for the effect of cumulative alcohol consumption on GAA at examination years 15 and 20. Quantile regression plots for beer (A, B), liquor (C, D), wine (E, F), and total alcohol (G, H) years at Y15 and Y20, respectively. The x-axis represents the quantile scale, and the y-axis represents the effect of alcohol on GAA for a given quantile. Results are adjusted for chronological age, sex, race, center, education, pack years of smoking, BMI, and physical activity.

Recent binge drinking on epigenetic age acceleration

Table 3 presents results for the associations between binge drinking and EAA. Recent binge drinking was positively associated with GAA at Y15 (P < 0.001) and Y20 (P < 0.001). Compared to participants who did not binge drink in the past 30 days, those who did exhibited a 1.22-year [95% CI: 0.69, 1.76] and a 1.55-year [95% CI: 0.99, 2.11] higher GAA at Y15 and Y20, respectively. Additionally, the number of days of binge drinking was associated with GAA at Y15 (P = 0.002) and Y20 (P < 0.001). Each additional day of binge drinking was associated with a 0.11-year [95% CI: 0.04, 0.18] and a 0.21-year [95% CI: 0.12, 0.29] higher GAA at Y15 and Y20, respectively. GEE yielded comparable associations as Y15 and Y20. IEAA, EEAA, and PAA were not associated with either binge drinking variable. Restricting to study participants with complete follow up yielded consistent conclusions (Supplementary Table 8). We observed similar correlations between days of recent binge drinking and several GrimAge surrogate biomarkers of blood plasma proteins at Y15 and Y20 (Supplementary Figures 1, 2). When modeling both cumulative alcohol consumption and recent binge drinking during GAA analyses, we observed recent binge drinking remained associated with GAA, while the cumulative alcohol consumption variables became less significant, although liquor years remained associated with GAA (Supplementary Table 9). Recent binge drinking was associated with telomere length derived from DNA methylation (Supplementary Table 7).

Cumulative alcohol consumption and chronological age interaction on GrimAge acceleration

Table 4 presents the interaction and stratified results for the joint associations of cumulative alcohol consumption and chronological age on GAA. Summary statistics by age quartiles are presented in Supplementary Table 10. At Y15, we observed a 0.35-year [95% CI: 0.14, 0.57] higher GAA per 5 beer years among participants in quartile 1 compared to a 0.05-year [95% CI: –0.30, 0.19] loss in GAA among those in quartile 4 (P_interaction < 0.001). Additionally, participants in quartile 1 displayed a 0.26-year [95% CI: 0.11, 0.41] higher GAA per 5 total alcohol years compared to participants in quartile 4 who experienced a 0.06-year [95% CI: –0.11, 0.22] higher GAA (P_interaction = 0.002). No statistical interactions were observed between liquor and wine years with chronological age.

At Y20, we observed a 0.29-year [95% CI: 0.12, 0.46] higher GAA in quartile 1 per 5 beer years compared to a 0.17-year [95% CI: –0.34, –0.01] loss in GAA in quartile 4 (P_interaction < 0.001). Moreover, quartile 1 experienced a 0.20-year [95% CI: 0.08, 0.31] higher GAA compared to a 0.07-year [95% CI: –0.19, 0.04] loss in GAA in quartile 4 per 5 total alcohol years (P_interaction < 0.001). Interaction and stratified results from GEE provided similar findings. Restricting to study participants with complete follow up yielded consistent conclusions (Supplementary Table 11). We observed similar results when adjusting for both cumulative alcohol consumption and recent binge drinking (Supplementary Table 12). While interactions of cumulative alcohol consumption with sex yielded primarily non-significant associations, female participants displayed higher GAA with greater consumption of beer, wine, and total alcohol compared to male participants (Supplementary Table 13). Additionally, Black participants experienced a higher GAA with greater consumption of liquor and total alcohol compared to White participants, although these interactions were non-significant (Supplementary Table 14).

Study sample

Details of the original CARDIA study design and examinations have previously been described [41]. Briefly, CARDIA is a population-based, cohort study designed to investigate the determinants of subclinical and clinical cardiovascular disease. Four centers across the United States recruited 5,115 Black and White participants ages 18 to 30 years between 1985 to 1986 and received in person examinations at 1985-1986 (baseline visit; year 0 [Y0]), 1987-88 (Y2), 1990-91 (Y5), 1992-93 (Y7), 1995-96 (Y10), 2000-01 (Y15), 2005-06 (Y20), 2010-11 (Y25), 2015-16 (Y30), and currently participating in 2020-21 (Y35).

Alcohol consumption measurements

Alcohol consumption was assessed at Y0 and at each follow up examination. Participants were asked “Did you drink any alcoholic beverages in the past year?” and if answered yes, participants were asked “How many drinks of [alcohol type] do you usually have per week?” for beer (a 12-ounce glass, can, or bottle), liquor (a shot or 1.5-ounces), and wine (a 5-ounce glass). Binge drinking was also obtained by asking participants “During the past 30 days, on how many days did you have five or more drinks on the same occasion?” We considered six alcohol variables at Y15 and Y20: four continuous alcohol variables capturing cumulative consumption from Y0 to Y15 and from Y0 to Y20, i.e., beer, liquor, wine, and total alcohol (the sum of the individual alcohol types); one binary variable indicating whether a participant binge drank in the past 30 days at Y15 and Y20; and one continuous variable indicating the number of days a participant binge drank during those 30 days at Y15 and Y20. In order to calculate cumulative alcohol consumption from all available CARDIA data from Y0 to Y15 and Y20 separately, we assumed weekly consumption of alcohol measured at each CARDIA exam was consistent in-between study examinations, i.e., weekly alcohol consumption represented consumption throughout the year and between examinations. For each weekly alcohol variable, we estimated ‘alcohol years’ by multiplying each alcohol variable by 52 weeks and the number of years between visits and dividing by 365 days, where an alcohol year is equivalent to 365 days of alcohol consumption. We then estimated alcohol years for each alcohol type separately and total alcohol by summing across examinations from Y0 to Y15 and Y20.

DNA methylation profiling

Details of methylation profiling and quality control have previously been described [19, 42–44]. Briefly, a randomly selected subset of 1,200 study participants with available whole blood repeatedly collected at both Y15 and Y20 (a total of 2,400 samples) underwent DNA methylation profiling using the Illumina MethylationEPIC Beadchip. The R package ENmix was used to perform data preprocessing and quality control using default parameter settings [45]. Low quality methylation measurements were defined as having less than 3 beads or a detection P < 1 × 10⁻⁶. A total of 87 samples with low quality methylation measurements >5% or bisulfite conversion process of extremely low intensity (defined as less than 3 × standard deviation of the intensity across samples below the mean intensity) and 6,209 CpG sites with a detection rate <95% were excluded from further analysis. Furthermore, 95 samples were identified as extreme outliers via the average total intensity value [intensity of the unmethylated signal (U) + intensity of the methylated signal (M)] or β value [M/(U + M + 100)] across all CpG sites and Tukey’s method [46]. Application of a model-based background correction was conducted using ENmix and dye bias correction was performed using RELIC [47]. M and U intensities for Infinium I or II probes were quantile-normalized separately, respectively. Low quality methylation measurements and β value outliers were set to missing. After applying these quality control measures, the final methylation dataset for epigenetic age calculation included 1,042 and 957 samples at Y15 and Y20, respectively. Supplementary Figure 3 presents a graphical timeline of the examinations with phenotype and blood draw collection.

Epigenetic age calculation

Four epigenetic ages were calculated at Y15 and Y20. Horvath’s age, intrinsic epigenetic age acceleration (IEAA), was calculated with 353 CpGs and is associated with cell-intrinsic aging [7]. Hannum’s age, extrinsic epigenetic age acceleration (EEAA), was estimated from 71 CpGs and is associated with immunological aging [8]. Levine’s age, PhenoAge acceleration (PAA), was calculated with 513 CpGs and is associated with comorbidities and physical functionality [9]. Lu’s age, GrimAge acceleration (GAA), was estimated from 1,030 CpGs and is associated with lifespan [10]. A publicly available online calculator was used to calculate the DNA-methylation epigenetic age estimates (https://dnamage.genetics.ucla.edu/new). EAA for each metric was estimated as the residuals from a linear model of chronological age and the top eight technical principal components, capturing technical biases and batch effect, on epigenetic age as previously described [42].

Statistical analysis

Multiple linear regression and generalized estimating equations (GEE) were used to examine the associations between the cumulative and binge drinking alcohol variables (independent variables) and each EAA (outcome variables) at Y15 and Y20. Each cumulative alcohol variable was rescaled to units of five drink years (i.e., divided by 5) to aid in the interpretation of beta coefficients. Drinkers were further categorized into low, intermediate, or high consumption via tertiles and compared to non-drinkers for each cumulative alcohol variable. Quantile regression was additionally performed to evaluate the conditional quantiles of each cumulative alcohol variable on EAA. This approach provides a more comprehensive view of the covariate effects on variables under study [48]. Due to previously observed differences in alcohol consumption by demographic characteristics [5, 6], interaction and stratified analyses were conducted by chronological age, sex, and race with the cumulative alcohol variables. Age was categorized into quartiles during stratified analyses: quartile 1 (n_{Y15[32≤age≤37]} = 242, n_{Y20[37≤age≤42]} = 223), quartile 2 (n_{Y15[38≤age≤40]} = 246, n_{Y20[43≤age≤45]} = 224), quartile 3 (n_{Y15[41≤age≤43]} = 294, n_{Y20[46≤age≤48]} = 271), and quartile 4 (n_{Y15[44≤age≤49]} = 248, n_{Y20[49≤age≤54]} = 227). Sensitivity analyses using study participants with complete follow up were performed. All models were adjusted for chronological age, sex, race, center, education, pack years of smoking, body mass index, and physical activity. Pack years of smoking was estimated by converting the number of daily smoked cigarettes measured at each examination to packs of cigarettes, multiplying by the number of years between examinations, and summing the products from Y0 to Y15 and Y20, separately. Associations were declared significant if P ≤ 0.05. All statistical analyses were performed using SAS 9.4.