Association of DNA Methylation-Based Biological Age With Health Risk Factors and Overall and Cause-Specific Mortality

Association of DNA Methylation-Based Biological Age With Health Risk Factors and Overall and... Abstract Measures of biological age based on blood DNA methylation, referred to as age acceleration (AA), have been developed. We examined whether AA was associated with health risk factors and overall and cause-specific mortality. At baseline (1990–1994), blood samples were drawn from 2,818 participants in the Melbourne Collaborative Cohort Study (Melbourne, Victoria, Australia). DNA methylation was determined using the Infinium HumanMethylation450 BeadChip array (Illumina Inc., San Diego, California). Mixed-effects models were used to examine the association of AA with health risk factors. Cox models were used to assess the association of AA with mortality. A total of 831 deaths were observed during a median 10.7 years of follow-up. Associations of AA were observed with male sex, Greek nationality (country of birth), smoking, obesity, diabetes, lower education, and meat intake. AA measures were associated with increased mortality, and this was only partly accounted for by known determinants of health (hazard ratios were attenuated by 20%–40%). Weak evidence of heterogeneity in the association was observed by sex (P = 0.06) and cause of death (P = 0.07) but not by other factors. DNA-methylation-based AA measures are associated with several major health risk factors, but these do not fully explain the association between AA and mortality. Future research should investigate what genetic and environmental factors determine AA. age acceleration, aging, biological age, cancer, DNA methylation, epigenetic clock, health risk factors, mortality Aging is associated with profound changes to DNA methylation. Globally, a decrease in methylation levels with age is observed, as well as a tendency for both hypo- and hypermethylated CpG sites to lose their initial epigenetic state and become hemimethylated (1–3). These changes are thought to play a role in the risk of disease (4–7). Recently, the concept of an “epigenetic clock” has emerged as a marker of biological aging (8). Several studies have attempted to quantify the age of the methylome using regularized regression methods to obtain a restricted set of DNA methylation measures predicting age (9–13). Two of these models have received more attention, as they predict chronological age most accurately (9, 11). Although the accurate prediction of chronological age is of interest to forensic scientists (14), the difference between predicted age (linear combination of age-associated methylation measures) and chronological age is also thought to represent an estimate of “age acceleration” (AA) and has been interpreted in epidemiologic studies as a marker of biological aging. Several studies have investigated the association between AA in blood and all-cause mortality. In one study, Marioni et al. (15) pooled data on 4 cohorts from the United Kingdom and the United States, comprising 808 deaths, and reported 16% and 9% increased mortality per 5-year increase in AA based on the Hannum et al. (9) and Horvath (11) predictors, respectively, after adjusting for several health risk factors. Similar findings were reported from an elderly German cohort (1,863 participants; 602 deaths) (16), with mortality increases of 10% and 23% per 5-year increment of AA according to the Hannum and Horvath predictors, respectively. In a Danish cohort study of 86 twin pairs, among which 55 deaths occurred, Christiansen et al. (17) reported a 35% increased risk of death per 5-year AA increment based on both Hannum’s and Horvath’s predictors. A recent pooled analysis by Chen et al. (18) showed mortality risks increased by 14% and 8%, respectively. Outcomes other than mortality have been investigated, including lung cancer (19) and Alzheimer’s disease (20). Distinct AA profiles have been observed in the blood of centenarians and their offspring (21, 22). If AA reflects biological age, it seems reasonable to expect that it is influenced by known risk factors for poor health (8). AA has been shown to be associated with obesity, most notably in liver tissue (11, 23). Cross-sectional associations with physical and cognitive ability (24), menopause (25), human immunodeficiency virus infection (26, 27), lifetime stress (28), posttraumatic stress disorder (29), and frailty in the elderly (30) have been reported. Our aim was to add to the growing literature on AA and health by 1) assessing associations between AA and known health risk factors, as well as all-cause and cause-specific mortality, and 2) examining the extent to which various lifestyle factors contribute to the relationship between AA and mortality, using a large cohort of middle-aged and older adults. METHODS Study sample We used data from studies nested within the Melbourne Collaborative Cohort Study (Melbourne, Victoria, Australia), a prospective cohort study of 41,514 healthy adult volunteers (24,469 women) aged 27–76 years (99.3% were aged 40–69 years) at baseline between 1990 and 1994 (31). DNA samples used for the present analysis were extracted from peripheral blood drawn at the time of recruitment (1990–1994). For the majority (70%) of participants, the DNA source was dried blood spots collected onto Guthrie Card Diagnostic Cellulose filter paper (Whatman plc, Kent, United Kingdom) and stored in airtight containers at room temperature. The other sources of DNA were peripheral blood mononuclear cells and buffy coats stored at −80°C for 28% and 2% participants, respectively. The study sample comprised Melbourne Collaborative Cohort Study participants selected as controls in nested case-control studies of breast, colorectal, kidney, lung, prostate, or urothelial cancer or mature B-cell malignancies (32–35). Controls had been individually matched to cases on age (they had to be free of cancer at an age within 1 year of the age at diagnosis of the corresponding case), sex, country of birth, and blood DNA source (dried blood spot, peripheral blood mononuclear cells, or buffy coat). For all but the colorectal cancer study, controls were matched to cases on year of birth. For the lung cancer study, controls were matched on smoking status at the time of blood collection. Vital status up to August 31, 2015, was ascertained through record linkage to the Victorian Registry of Births, Deaths and Marriages (via the Victorian Cancer Registry) and the National Death Index (via the Australian Institute of Health and Welfare). Cause of death, based on the underlying cause of death from the death certificate, was known for deaths occurring up to December 31, 2013. Methods relating to DNA extraction and bisulfite conversion and DNA methylation data processing are described in the Web Appendix (available at https://academic.oup.com/aje). Methylation age and AA All of the CpGs used to compute Hannum and Horvath methylation ages were available in our study after quality control. Hannum methylation age was computed using the linear combination of methylation measures at 71 age-associated CpGs (15). Horvath methylation age was calculated on the basis of methylation measures at 353 CpGs using the online calculator (11). We considered the most recent measures of AA (18, 36): 1) epigenetic AA based on the Horvath predictor—residuals resulting from a linear regression model of Horvath’s estimate of epigenetic age on chronological age, referred to as AA-Horvath; 2) the corresponding AA based on the Hannum et al. predictor (AA-Hannum); 3) intrinsic epigenetic AA (IEAA) based on the Horvath predictor—residuals resulting from a linear regression model of Horvath’s estimate of epigenetic age on chronological age and measures of blood cell counts, referred to as IEAA-Horvath; 4) the corresponding IEAA based on the Hannum et al. predictor (IEAA-Hannum); and 5) enhanced Hannum AA, defined as the AA-Hannum measure plus a weighted average of aging-associated cell counts. Results are presented in the main text for IEAA measures and in Web Tables 1–7 for other measures. Statistical analysis Correlations between AA measures were assessed using Spearman correlation coefficients. To evaluate the accuracy of the Horvath and Hannum age predictors, we computed the median absolute value of each AA measure. The reliability of methylation measures for CpG sites included in the predictors was examined using intraclass correlation coefficients computed using mixed models and based on 134 technical replicate pairs as described in previous publications (37, 38). Missing covariate data (<1% for any individual covariate) were handled using multiple imputation (39), with the R package mice (R Foundation for Statistical Computing, Vienna, Austria) (40). Health risk factors associated with AA We used linear regression models to examine the association between AA measures and known risk factors for disease, all assessed at baseline. Factors considered and adjusted for were: age; sex; country of birth (Australia/New Zealand, United Kingdom/Malta, Italy, or Greece); smoking (never smoker, former smoker who had quit ≤15 years prior, former smoker who had quit >15 years prior, current smoker of ≤20 cigarettes/day, or current smoker of >20 cigarettes/day); body mass index (BMI; calculated as weight (kg)/height (m)2 and categorized as <25, 25–29.9, 30–34.9, or ≥35); height (cm; continuous); alcohol intake (none, 1–39 g/day (males) or 1–19 g/day (females), 40–59 g/day (males) or 20–39 g/day (females), or ≥60 g/day (males) or ≥40 g/day (females)); Alternate Healthy Eating Index 2010 score (to reflect overall diet quality) (41); a physical activity score ranging from 1 to 4 and reflecting metabolic equivalents, as described in previous work (42); socioeconomic status (a score from 1 to 10 representing the relative socioeconomic disadvantage of the participant’s area of residence (43)); and education (a score from 1 to 8; 1 = primary school only and 8 = tertiary university degree or more). We also considered the following clinical variables: history of cardiovascular disease (stroke, angina, or heart attack), hypertension, diabetes, asthma, kidney stones, gallstones, arthritis, and resting heart rate (measured with the Dinamap 1846SX automated blood pressure monitor (Critikon, Tampa, Florida)). Each of these clinical factors was added individually to the previous model. Results of all models were additionally adjusted for sample type (dried blood spot, peripheral blood mononuclear cells, or buffy coat), study, and the plate on which the sample was processed (the latter 2 variables being fitted as random effects and hereafter referred to as batch effects). More detailed analyses were conducted using additional variables relating to diet, clinical factors, smoking, alcohol intake, anthropometric factors, psychological well-being, and blood measures (Web Table 5). We used the same models to assess the associations between demographic, lifestyle, and anthropometric variables and each individual CpG site included in the Hannum and Horvath predictors. In these models, smoking, BMI, and alcohol intake were modeled as continuous variables. AA and mortality We used Cox models to estimate hazard ratios for the association between AA measures and mortality. Age was used as the time scale (44), and participants were considered to be at risk from their index date (the date on which they reached the age at which their matched case in the corresponding nested case-control study was diagnosed with cancer) onward in order to avoid immortal time bias, as they had to be alive at this date. For each AA measure, the proportional hazards assumption was assessed by visual inspection of Schoenfeld residuals (45). Separate models were fitted for all-cause, cancer, cardiovascular disease, and other-cause mortality. Person-years of follow-up were calculated from the index date to the date of death, and participants were censored at the date of departure from Australia or the end of follow-up (August 31, 2015, for all-cause death and December 31, 2013, for cause-specific death), whichever came first. In cause-specific analyses, deaths from other causes were censored, which is a way to model the cause-specific hazard function, appropriately taking the competing risk of other-cause death into account (46). For comparability with previous publications, hazard ratio estimates were calculated for a 5-year increase in AA measures (15–17). For overall and cause-specific mortality analyses, 3 models were fitted in order to examine the robustness of the association between AA measures and mortality: In model 0 (unadjusted), results were adjusted for sample type and batch effects; in model 1 (minimally adjusted), we added age at blood draw, sex, and country of birth to model 0; in model 2, we added anthropometric measures, lifestyle factors, and socioeconomic variables (BMI, height, smoking, alcohol intake, diet quality, physical activity, socioeconomic status, and education) to model 1; and in model 3, we added clinical variables to model 2. Batch effects were fitted as fixed effects, and similar results were obtained when using mixed-effects Cox models (47). To investigate whether AA measures were more strongly associated with specific causes of death, heterogeneity in the associations between AA and mortality by cause of death was tested using the data duplication method (48). Interactions between AA measures and each of sex, country of birth, smoking, and BMI were examined by adding interaction terms to the model and applying a likelihood ratio test. Analyses for AA were repeated with adjustment for cell composition (as defined by the Houseman algorithm (49)). Ethics Study participants provided informed consent in accordance with the Declaration of Helsinki. The study was approved by Cancer Council Victoria’s Human Research Ethics Committee and performed in accordance with the institution’s ethical guidelines. RESULTS The correlation between the Hannum and Horvath age predictors was ρ = 0.75, and their correlation with chronological age was high (ρ = 0.76 and ρ = 0.73, respectively) (Table 1). The median AA (IEAA) was, in absolute terms, 3.9 years and 4.2 years for the Hannum and Horvath predictors, respectively (Figure 1, panels A and B, respectively). The correlation between IEAA measures was 0.56 (Table 1). The reliability of methylation measures was high for CpG sites included in Hannum’s age predictor (median intraclass correlation coefficient, 0.60; interquartile range, 0.53–0.68) and more variable for those included in Horvath’s predictor (median intraclass correlation coefficient, 0.48; interquartile range, 0.22–0.61) (Web Table 8 and Web Figure 1). Table 1. Characteristics of Predictors of Age Acceleration, Melbourne Collaborative Cohort Study, Australia, 1990–1994 Age Acceleration Predictor  Horvath Predictor  Hannum Predictor  Mean (SD)  ρ  Mean (SD)  ρ  Predicted age, years  59.1 (8.0)    59.0 (10.1)    Chronological age, years  59.0 (7.6)    59.0 (7.6)    Correlation with agea    0.73    0.76  Correlation with age predictor           AAb  0.3 (7.1)  −0.01  −0.2 (6.7)  0.03   IEAAc  0.0 (6.9)  −0.06  0.0 (6.4)  −0.03  Median age differenced           AA  4.4    4.2     IEAA  4.2    3.9    Age Acceleration Predictor  Horvath Predictor  Hannum Predictor  Mean (SD)  ρ  Mean (SD)  ρ  Predicted age, years  59.1 (8.0)    59.0 (10.1)    Chronological age, years  59.0 (7.6)    59.0 (7.6)    Correlation with agea    0.73    0.76  Correlation with age predictor           AAb  0.3 (7.1)  −0.01  −0.2 (6.7)  0.03   IEAAc  0.0 (6.9)  −0.06  0.0 (6.4)  −0.03  Median age differenced           AA  4.4    4.2     IEAA  4.2    3.9    Abbreviations: AA, age acceleration; IEAA, intrinsic epigenetic age acceleration; SD, standard deviation. a Correlation between the Horvath (11) and Hannum et al. (9) age predictors: ρ = 0.75. b Correlation between the Horvath and Hannum AA measures: ρ = 0.56. c Correlation between the Horvath and Hannum IEAA measures: ρ = 0.54. d Median age difference = absolute value of AA; age and AA were measured in years. Figure 1. View largeDownload slide Distribution of intrinsic epigenetic age acceleration (IEAA) measures (years) in the Melbourne Collaborative Cohort Study, Australia, 1990–1994. A) IEAA based on the Hannum et al. predictor (9); B) IEAA based on the Horvath predictor (11). “Frequency” (y-axis) denotes the number of observations in each 5-year age group bar. Figure 1. View largeDownload slide Distribution of intrinsic epigenetic age acceleration (IEAA) measures (years) in the Melbourne Collaborative Cohort Study, Australia, 1990–1994. A) IEAA based on the Hannum et al. predictor (9); B) IEAA based on the Horvath predictor (11). “Frequency” (y-axis) denotes the number of observations in each 5-year age group bar. A total of 2,818 participants were included in the analysis of health risk factors. Higher AAs were observed in males compared with females (IEAA-Hannum: β = 1.37 years, P < 0.001; IEAA-Horvath: β = 1.23 years, P = 0.005); in Greek participants compared with those born in Australia/New-Zealand (IEAA-Horvath: β = 1.29 years, P = 0.01); in current and former smokers compared with never smokers (IEAA-Hannum: current smoker of ≤20 cigarettes/day—β = 2.12 years, P < 0.001; current smoker of >20 cigarettes/day—β = 1.26 years, P = 0.01), and in overweight (BMI ≥25) and obese (BMI ≥30) persons compared with lean individuals (BMI <25) (IEAA-Hannum: BMI 25–29.9—β = 0.69 years, P = 0.01; BMI 30–34.9—β = 1.24 years, P < 0.001; BMI ≥35—β = 1.70 years, P = 0.01). Higher education was associated with lower AA (AA-Horvath: per 1-unit increment in education score, β = −0.16 years, P = 0.04 (Web Table 1)). Most reported comorbid conditions were not associated with AA measures, except for diabetes (IEAA-Horvath: β = 1.46 years, P = 0.04) and prior history of gallstones (IEAA-Horvath: β = 1 year, P = 0.03) (Table 2). Intake of meat was associated with increased AA (IEAA-Horvath: P = 0.01), and intake of fruits, but not vegetables, was associated with decreased AA (IEAA-Hannum: P = 0.03 and P = 0.91, respectively). Most body composition measures were associated with increased AA (Web Table 5). Table 2. Characteristics of the Study Sample and Their Association With Measures of Intrinsic Epigenetic Age Acceleration (n = 2,818), Melbourne Collaborative Cohort Study, Australia, 1990–1994 Variablea  No. of Persons  %  IEAA-Hannum Predictor  IEAA-Horvath Predictor  βb  SE  P Value  βb  SE  P Value  Demographic, anthropometric, and lifestyle variables                   Male sex  1,722  61  1.37  0.4  <0.001  1.23  0.44  0.005   Country of birth                    Greece  288  10  0.32  0.45  0.48  1.29  0.50  0.01    Italy  377  13  0.54  0.40  0.18  0.76  0.44  0.09    United Kingdom/Malta  215  8  0.14  0.44  0.75  0.21  0.49  0.68   Tobacco smoking                    Former smoker                     Quit >15 years prior  545  19  −0.03  0.33  0.92  0.45  0.36  0.22     Quit ≤15 years prior  533  19  0.65  0.33  0.05  0.78  0.37  0.03    Current smoker                     ≤20 cigarettes/day  159  6  2.12  0.52  <0.001  1.33  0.58  0.02     >20 cigarettes/day  228  8  1.26  0.48  0.01  0.86  0.53  0.11   Body mass indexc                    25–29.9  1,384  49  0.69  0.27  0.01  0.52  0.30  0.09    30–34.9  451  16  1.24  0.37  <0.001  0.29  0.41  0.47    ≥35  105  4  1.70  0.64  0.01  2.23  0.71  0.002   Height, cm  167 (9.2)d  0.02  0.02  0.26  0.00  0.02  0.94   Alcohol intake                    Low  1,604  57  −0.01  0.29  0.98  0.44  0.33  0.18    Moderate  263  9  −0.20  0.45  0.66  0.42  0.50  0.40    High  154  5  0.09  0.56  0.88  0.09  0.62  0.89   Alternate Healthy Eating Index 2010 scoree  63 (10.9)d  −0.02  0.01  0.10  −0.01  0.01  0.33   Physical activity scoref  2.6 (1.0)d  0.18  0.11  0.11  0.07  0.13  0.59   Education scoreg  4.9 (1.9)d  −0.14  0.07  0.06  −0.10  0.08  0.22  Clinical variables                   History of asthma  312  11  0.52  0.37  0.16  0.37  0.41  0.36   History of hypertension  677  24  −0.02  0.27  0.93  −0.09  0.31  0.78   History of diabetes  99  4  0.80  0.64  0.21  1.46  0.71  0.04   History of arthritis  997  35  −0.30  0.25  0.23  −0.12  0.28  0.66   History of kidney stones  193  7  −0.08  0.46  0.87  0.15  0.51  0.77   History of gallstones  253  9  0.25  0.41  0.54  1.00  0.45  0.03   History of cardiovascular disease  189  7  −0.21  0.47  0.65  0.01  0.52  0.99   Resting heart rate, beats/minute  68.2 (10.0)d  0.00  0.01  0.68  0.02  0.01  0.22  Variablea  No. of Persons  %  IEAA-Hannum Predictor  IEAA-Horvath Predictor  βb  SE  P Value  βb  SE  P Value  Demographic, anthropometric, and lifestyle variables                   Male sex  1,722  61  1.37  0.4  <0.001  1.23  0.44  0.005   Country of birth                    Greece  288  10  0.32  0.45  0.48  1.29  0.50  0.01    Italy  377  13  0.54  0.40  0.18  0.76  0.44  0.09    United Kingdom/Malta  215  8  0.14  0.44  0.75  0.21  0.49  0.68   Tobacco smoking                    Former smoker                     Quit >15 years prior  545  19  −0.03  0.33  0.92  0.45  0.36  0.22     Quit ≤15 years prior  533  19  0.65  0.33  0.05  0.78  0.37  0.03    Current smoker                     ≤20 cigarettes/day  159  6  2.12  0.52  <0.001  1.33  0.58  0.02     >20 cigarettes/day  228  8  1.26  0.48  0.01  0.86  0.53  0.11   Body mass indexc                    25–29.9  1,384  49  0.69  0.27  0.01  0.52  0.30  0.09    30–34.9  451  16  1.24  0.37  <0.001  0.29  0.41  0.47    ≥35  105  4  1.70  0.64  0.01  2.23  0.71  0.002   Height, cm  167 (9.2)d  0.02  0.02  0.26  0.00  0.02  0.94   Alcohol intake                    Low  1,604  57  −0.01  0.29  0.98  0.44  0.33  0.18    Moderate  263  9  −0.20  0.45  0.66  0.42  0.50  0.40    High  154  5  0.09  0.56  0.88  0.09  0.62  0.89   Alternate Healthy Eating Index 2010 scoree  63 (10.9)d  −0.02  0.01  0.10  −0.01  0.01  0.33   Physical activity scoref  2.6 (1.0)d  0.18  0.11  0.11  0.07  0.13  0.59   Education scoreg  4.9 (1.9)d  −0.14  0.07  0.06  −0.10  0.08  0.22  Clinical variables                   History of asthma  312  11  0.52  0.37  0.16  0.37  0.41  0.36   History of hypertension  677  24  −0.02  0.27  0.93  −0.09  0.31  0.78   History of diabetes  99  4  0.80  0.64  0.21  1.46  0.71  0.04   History of arthritis  997  35  −0.30  0.25  0.23  −0.12  0.28  0.66   History of kidney stones  193  7  −0.08  0.46  0.87  0.15  0.51  0.77   History of gallstones  253  9  0.25  0.41  0.54  1.00  0.45  0.03   History of cardiovascular disease  189  7  −0.21  0.47  0.65  0.01  0.52  0.99   Resting heart rate, beats/minute  68.2 (10.0)d  0.00  0.01  0.68  0.02  0.01  0.22  Abbreviations: IEAA, intrinsic epigenetic age acceleration; SE, standard error. a Reference categories were as follows: sex—female; country of birth—Australia/New Zealand; tobacco smoking—never; body mass index—<25; alcohol intake—none; clinical variables—no history. b All estimates in the upper part of the table (demographic, lifestyle, and anthropometric variables) were mutually adjusted and further adjusted for age at blood draw, socioeconomic status (score representing the relative socioeconomic disadvantage of the participant’s area of residence, ranging from 1 to 10), sample type, and batch effects. All estimates in the lower part of the table (clinical variables) were adjusted for variables in the upper part of the table (demographic, lifestyle, and anthropometric variables) and further adjusted for socioeconomic status, sample type, and batch effects. c Weight (kg)/height (m)2. d Values are expressed as mean (standard deviation). e More details can be found in the paper by Chiuve et al. (41). f Physical activity score reflected metabolic equivalents, as described in previous work (42); possible scores ranged from 1 to 4. g Education was an ordinal questionnaire variable with possible scores ranging from 1 to 8 (1 = primary school only and 8 = tertiary university degree or more). Several individual CpG sites included in the Hannum and Horvath predictors were associated with demographic, lifestyle, and anthropometric variables (after correction for multiple testing for 424 tests, P < 1.1 × 10−4). Most of them were strongly associated with age, and associations were also observed with sex (74 CpGs), smoking (9 CpGs), and alcohol intake (6 CpGs) (Web Table 9). A total of 831 deaths were observed during follow-up (median, 10.7 years), including 692 with information on cause of death (follow-up until December 31, 2013). Unadjusted analyses (model 0) showed associations of AA measures with all-cause mortality (IEAA-Hannum: per 5-year increment in AA, hazard ratio (HR) = 1.08, 95% confidence interval (CI): 1.02, 1.14; findings were similar for other AA measures) (Table 3). After adjustment for age, sex, and country of birth (model 1), the associations for IEAA were attenuated by up to 7% (IEAA-Hannum: 7%; IEAA-Horvath: 0%) (Table 3). After adjustment for lifestyle and anthropometric and socioeconomic variables (model 2), hazard ratios were attenuated by a further 18%–37% (IEAA-Hannum: 37%; IEAA-Horvath: 18%) but not further after adjustment for reported comorbidity (model 3). In total, the association of AA with overall mortality was attenuated by 18%–45% after adjustment for known health risk factors; the remaining association was 5% increased mortality per 5-year AA increment for all AA measures. AA was more strongly associated with cancer-related death (n = 240) than with other causes of death (IEAA-Hannum: HR = 1.12 (95% CI: 1.02, 1.24); IEAA-Horvath: HR = 1.11 (95% CI: 1.01, 1.21)) (Table 4), with weak evidence of heterogeneity by cause of death (P-heterogeneity = 0.07 and P-heterogeneity = 0.14, respectively). Table 3. Associations of Age Acceleration (Intrinsic Epigenetic Age Acceleration, per 5-Year Increase) With All-Cause Mortality Risk During a Follow-up Period Extending to August 31, 2015 (n = 2,818; 831 Deaths During 31,449 Person-Years of Follow-Up), Melbourne Collaborative Cohort Study, Australia, 1990–1994 Age Acceleration Predictor  Model  Model 0a  Model 1b  Model 2c  Model 3d  HR  95% CI  P Value  HR  95% CI  P Value  HR  95% CI  P Value  HR  95% CI  P Value  IEAA-Hannum  1.08  1.02, 1.14  0.008  1.07  1.01, 1.13  0.02  1.04  0.99, 1.10  0.13  1.04  0.99, 1.11  0.12  IEAA-Horvath  1.07  1.02, 1.14  0.01  1.07  1.01, 1.12  0.01  1.06  1.00, 1.11  0.05  1.05  1.00, 1.11  0.05  Age Acceleration Predictor  Model  Model 0a  Model 1b  Model 2c  Model 3d  HR  95% CI  P Value  HR  95% CI  P Value  HR  95% CI  P Value  HR  95% CI  P Value  IEAA-Hannum  1.08  1.02, 1.14  0.008  1.07  1.01, 1.13  0.02  1.04  0.99, 1.10  0.13  1.04  0.99, 1.11  0.12  IEAA-Horvath  1.07  1.02, 1.14  0.01  1.07  1.01, 1.12  0.01  1.06  1.00, 1.11  0.05  1.05  1.00, 1.11  0.05  Abbreviations: CI, confidence interval; HR, hazard ratio; IEAA, intrinsic epigenetic age acceleration. a Model 0 (unadjusted results) adjusted for sample type and batch effects. b Model 1 (minimally adjusted results) adjusted for age at blood draw, sex, country of birth (Australia/New Zealand, United Kingdom/Malta, Italy, or Greece), sample type, and batch effects. c In model 2, the following anthropometric measures, lifestyle factors, and socioeconomic variables were added to model 1: body mass index (weight (kg)/height (m)2; <25, 25–29.9, 30–34.9, or ≥35), height (cm; continuous), smoking (never smoker; former smoker who quit ≤15 years prior; former smoker who quit >15 years prior; current smoker of ≤20 cigarettes/day; current smoker of >20 cigarettes/day), alcohol intake (1–19 g/day (females); 40–59 g/day (males) or 20–39 g/day (females); or ≥60 g/day (males) or ≥40 g/day (females)), diet quality (Alternate Healthy Eating Index 2010), physical activity (physical activity score ranging from 1 to 4 and reflecting metabolic equivalents (39)), socioeconomic status (score representing the relative socioeconomic disadvantage of the participant’s area of residence, ranging from 1 to 10 (40)), and education (score ranging from 1 to 8, with 1 = primary school only and 8 = tertiary university degree or more). d In model 3, the following clinical variables were added to model 2: history of asthma (yes/no), hypertension (yes/no), diabetes (yes/no), arthritis (yes/no), kidney stones (yes/no), gallstones (yes/no), or cardiovascular disease (yes/no) and resting heart rate (beats/minute; continuous). Table 4. Associations of Age Acceleration (Intrinsic Epigenetic Age Acceleration, per 5-Year Increase) With Cause-Specific Mortality Risk (Model 2a) During a Follow-up Period Extending to December 31, 2013 (n = 2,818; 692 Deaths During 28,022 Person-Years of Follow-up), Melbourne Collaborative Cohort Study, Australia, 1990–1994 Mortality Outcome  No. of Deaths  IEAA-Hannum Predictor  IEAA-Horvath Predictor  HR  95% CI  P Value  HR  95% CI  P Value  CVD mortality  203  1.08  0.97, 1.21  0.17  1.00  0.90, 1.12  0.93  Cancer mortality  240  1.12  1.02, 1.24  0.02  1.11  1.01, 1.21  0.03  Other-cause mortality  249  0.96  0.86, 1.06  0.37  1.06  0.97, 1.16  0.22  Mortality Outcome  No. of Deaths  IEAA-Hannum Predictor  IEAA-Horvath Predictor  HR  95% CI  P Value  HR  95% CI  P Value  CVD mortality  203  1.08  0.97, 1.21  0.17  1.00  0.90, 1.12  0.93  Cancer mortality  240  1.12  1.02, 1.24  0.02  1.11  1.01, 1.21  0.03  Other-cause mortality  249  0.96  0.86, 1.06  0.37  1.06  0.97, 1.16  0.22  Abbreviations: CI, confidence interval; CVD, cardiovascular disease, HR, hazard ratio; IEAA, intrinsic epigenetic age acceleration. a Results were adjusted for age at blood draw, sex, country of birth (Australia/New Zealand, United Kingdom/Malta, Italy, or Greece), body mass index (weight (kg)/height (m)2; <25, 25–29.9, 30–34.9, or ≥35), height (cm; continuous), smoking (never smoker; former smoker who quit ≤15 years prior; former smoker who quit >15 years prior; current smoker of ≤20 cigarettes/day; current smoker of >20 cigarettes/day), alcohol intake (1–19 g/day (females); 40–59 g/day (males) or 20–39 g/day (females); or ≥60 g/day (males) or ≥40 g/day (females)), diet quality (Alternate Healthy Eating Index 2010), physical activity (physical activity score ranging from 1 to 4 and reflecting metabolic equivalents (39)), socioeconomic status (score representing the relative socioeconomic disadvantage of the participant’s area of residence, ranging from 1 to 10 (40)), education (score ranging from 1 to 8, with 1 = primary school only and 8 = tertiary university degree or more), sample type, and batch effects. The estimated associations of IEAA measures with overall mortality appeared stronger for men than for women (IEAA-Hannum: men—HR = 1.08 (95% CI: 1.01, 1.15); women—HR = 0.97 (95% CI: 0.88, 1.07) (P-heterogeneity = 0.06); IEAA-Horvath: men—HR = 1.09 (95% CI: 1.02, 1.16); women—HR = 0.99 (95% CI: 0.91, 1.09) (P-heterogeneity = 0.11)). No evidence of effect modification was found for smoking status, BMI, or country of birth (Table 5). When the analysis was restricted to participants for whom follow-up started less than 5 years after blood draw, the hazard ratio estimates were similar to those based on all participants. Table 5. Associations of Age Acceleration (Intrinsic Epigenetic Age Acceleration, per 5-Year Increase) With All-Cause Mortality According to Selected Health Risk Factors (Model 2a), Melbourne Collaborative Cohort Study, Australia, 1990–1994 Age Acceleration Measure  IEAA-Hannum Predictor  IEAA-Horvath Predictor  HR  95% CI  P-het  HR  95% CI  P-het  Sex               Female  0.97  0.88, 1.07  0.06  0.99  0.91, 1.09  0.11   Male  1.08  1.01, 1.15  1.09  1.02, 1.16  Smoking status               Never smoker  1.01  0.93, 1.10  0.65  1.05  0.97, 1.13  0.63   Former smoker            Quit >15 years prior  1.04  0.92, 1.17  1.04  0.93, 1.16    Quit ≤15 years prior  1.12  1.01, 1.26  1.10  0.99, 1.23   Current smoker            <20 cigarettes/day  1.01  0.84, 1.22  0.93  0.77, 1.13    ≥20 cigarettes/day  1.03  0.88, 1.21  1.10  0.94, 1.28  Body mass indexb               <25  1.03  0.93, 1.13  0.50  1.04  0.95, 1.13  0.74   25–29.9  1.08  1.00, 1.17  1.08  1.00, 1.16   ≥30  1.00  0.90, 1.12  1.03  0.93, 1.15  Country of birth               Australia/New-Zealand  1.04  0.97, 1.11  0.77  1.04  0.98, 1.11  0.44   United Kingdom  1.08  0.93, 1.27  1.09  0.93, 1.28   Italy  1.00  0.88, 1.15  1.01  0.89, 1.16   Greece  1.12  0.92, 1.37  1.21  1.00, 1.46  Time between blood draw and start of follow-up, years               <5  1.07  0.96, 1.19  0.75  1.06  0.96, 1.17  0.25   5–10  1.05  0.96, 1.16  1.11  1.02, 1.21   >10  1.02  0.94, 1.11  1.00  0.93, 1.09  Age Acceleration Measure  IEAA-Hannum Predictor  IEAA-Horvath Predictor  HR  95% CI  P-het  HR  95% CI  P-het  Sex               Female  0.97  0.88, 1.07  0.06  0.99  0.91, 1.09  0.11   Male  1.08  1.01, 1.15  1.09  1.02, 1.16  Smoking status               Never smoker  1.01  0.93, 1.10  0.65  1.05  0.97, 1.13  0.63   Former smoker            Quit >15 years prior  1.04  0.92, 1.17  1.04  0.93, 1.16    Quit ≤15 years prior  1.12  1.01, 1.26  1.10  0.99, 1.23   Current smoker            <20 cigarettes/day  1.01  0.84, 1.22  0.93  0.77, 1.13    ≥20 cigarettes/day  1.03  0.88, 1.21  1.10  0.94, 1.28  Body mass indexb               <25  1.03  0.93, 1.13  0.50  1.04  0.95, 1.13  0.74   25–29.9  1.08  1.00, 1.17  1.08  1.00, 1.16   ≥30  1.00  0.90, 1.12  1.03  0.93, 1.15  Country of birth               Australia/New-Zealand  1.04  0.97, 1.11  0.77  1.04  0.98, 1.11  0.44   United Kingdom  1.08  0.93, 1.27  1.09  0.93, 1.28   Italy  1.00  0.88, 1.15  1.01  0.89, 1.16   Greece  1.12  0.92, 1.37  1.21  1.00, 1.46  Time between blood draw and start of follow-up, years               <5  1.07  0.96, 1.19  0.75  1.06  0.96, 1.17  0.25   5–10  1.05  0.96, 1.16  1.11  1.02, 1.21   >10  1.02  0.94, 1.11  1.00  0.93, 1.09  Abbreviations: CI, confidence interval; het, heterogeneity; HR, hazard ratio; IEAA, intrinsic epigenetic age acceleration. a Results were adjusted for age at blood draw, sex, country of birth (Australia/New Zealand, United Kingdom/Malta, Italy, or Greece), body mass index (weight (kg)/height (m)2; <25, 25–29.9, 30–34.9, or ≥35), height (cm; continuous), smoking (never smoker; former smoker who quit ≤15 years prior; former smoker who quit >15 years prior; current smoker of ≤20 cigarettes/day; current smoker of >20 cigarettes/day), alcohol intake (1–19 g/day (females); 40–59 g/day (males) or 20–39 g/day (females); or ≥60 g/day (males) or ≥40 g/day (females)), diet quality (Alternate Healthy Eating Index 2010), physical activity (physical activity score ranging from 1 to 4 and reflecting metabolic equivalents (39)), socioeconomic status (score representing the relative socioeconomic disadvantage of the participant’s area of residence, ranging from 1 to 10 (40)), education (score ranging from 1 to 8, with 1 = primary school only and 8 = tertiary university degree or more), sample type, and batch effects. b Weight (kg)/height(m)2. Associations of health risk factors with AA measures and associations of AA measures with mortality were similar for other AA measures (Web Tables 2–4, 6, and 7) and were generally stronger in analyses of AA adjusted for cell composition (Web Tables 6 and 7). DISCUSSION Our study adds to the evidence regarding the use of the epigenetic clock as a marker of biological age. We observed that male sex, Greek nationality (country of birth), smoking, BMI, and diabetes were strongly associated with AA. In addition, AA measures were associated with increased risk of mortality, and these associations, albeit small, were independent of known mortality risk factors, including a large number of demographic, lifestyle, and anthropometric variables, and medical conditions. This suggests that a substantial proportion of AA is due to unidentified factors, which may include early-life exposures, health-associated genetic variants, or other unmeasured environmental factors. The main limitation of our analysis of health risk factors associated with AA was its cross-sectional nature. Future studies should investigate what factors may explain temporal changes in AA. Information on most variables was self-reported, which may have induced some measurement error. Although CpGs included in the age predictors had higher-than-average reliability, measurement error may also have decreased the accuracy of AA measures, particularly for the Horvath predictor. For the mortality analysis, we left-truncated follow-up at the age at which the matching case was diagnosed, in order to account for the procedure used to select controls. This not only created a lag time between exposure assessment and the start of follow-up (implicitly, we assumed that AA remained constant between blood draw and the index date) but also substantially reduced the follow-up time and meant that controls were selected for not developing the cancer of interest in each individual nested case-control study. However, because each cancer was rare, any selection bias would have been minimal. The Melbourne Collaborative Cohort Study has a low prevalence of unhealthy lifestyles (only 13% of participants were current smokers at baseline) and lower mortality rates than those in the general Australian population (50). Thus, it may appear surprising that the estimates from both predictors were close to chronological age (mean chronological age = 59.0 years; mean Horvath-predicted age = 59.1 years; mean Hannum-predicted age = 59.0 years). Study participants were relatively old at the start of follow-up, which may have resulted in weaker observed associations (51). However, when analyses were restricted to younger persons or to those with shorter lag times between blood sampling and the start of follow-up, the resulting hazard ratio estimates were not higher, indicating only a small influence of these limitations. In the seminal study by Marioni et al. (15), the adjusted (though not for BMI) mortality risk estimates varied by cohort, with hazard ratios ranging from 1.08 to 1.40 for AA-Hannum (with a pooled effect of HR = 1.16) and from 0.96 to 1.17 for AA-Horvath (with a pooled effect of HR = 1.09). In the study by Perna et al. (16), a substantial proportion of included deaths (52%) were for participants selected on the basis of their vital status, and a case-cohort design was employed, so their estimates may not be comparable to those of our study. The hazard ratios were 1.10 for AA-Hannum and 1.23 for AA-Horvath. These studies used AA and adjusted for cell composition, so their results are comparable to the estimates presented in Web Table 7. Chen et al. (18) included 2,734 deaths in their study and reported postadjustment hazard ratios similar to those for our model 3 (per 5-year increment in IEAA-Hannum, HR = 1.0145 = 1.07) and higher for the extrinsic measure (per 5-year increment in enhanced-Hannum, HR = 1.0295 = 1.15). While not nominally statistically significant, the estimated associations between AA and mortality were stronger among men in our study. It is unclear why this would be the case. A consistent finding was reported by Perna et al. (16), although the difference by sex was not formally tested. Effect estimates were similar for men and women in the Marioni et al. (15) study. Like our study, the study by Perna et al. (16) lacked the power to detect effect heterogeneity between causes of death. No comparable findings were reported by Marioni et al. (15), and no heterogeneity by sex was observed in the Chen et al. pooled analysis (18). We did not correct our P values for multiple testing, which may have resulted in false-positive findings. Most of the observed associations between AA and health risk factors and mortality were in the expected direction. Additional studies will be required to confirm these associations. In a recent study examining lifestyle factors associated with AA, Quach et al. (52) found that higher BMI was associated with increased AA and that higher education and fish and alcohol consumption were associated with decreased AA. BMI was also associated with IEAA, as was poultry consumption (52). Unlike this and other studies, we found a strong association of AA with smoking, which is consistent with the dramatic effects of smoking on health outcomes, but our findings did not replicate the previously reported association with consumption of poultry and fish. We also observed an association of AA and IEAA with education, which was not found in another large-scale study (53). Many molecular markers of biological aging have been proposed (54), and future research should address how methylation-defined AA measures correlate with these. An analysis of the joint effects of methylation age and telomere length on mortality using 2 cohorts showed that AA and telomere length had independent effects on mortality (55). Those analyses were limited by small sample sizes, and results were minimally adjusted (i.e., simply for age, sex, and estimated white blood cell composition). In an elderly German population, the comprehensive frailty measure that was evaluated appeared to be associated with AA-Horvath more than with telomere length (30). There was no strong correlation of AA-Horvath with telomere length. Estimates of heritability computed by Marioni et al. (15) were around 0.40 for both AA-Hannum and AA-Horvath. In another study, Levine et al. (25) estimated heritability to be 0.65 using genome-wide complex trait analysis. Li et al. (56) used twins to examine the determinants of AA measures and concluded that although genetic factors played a role, the importance of shared environmental factors might have been overlooked, such that these heritability measures were overestimates. These studies, taken together, nevertheless suggest that genetic confounders of the AA-mortality association may exist, but we could not control for them in our analysis. Hannum and Horvath measures of biological age require methylation measurement at a small number of CpG sites. The association of AA with health risk factors and with mortality suggests that it could be used to monitor an individual’s health status, although this requires formal assessment. Given the sources of possible systematic bias and measurement error encountered with the assay we used (Infinium HumanMethylation450 BeadChip array (Illumina Inc., San Diego, California)) (38, 57), one would expect that more accurate AA predictors could be derived using higher-resolution methylation measurement techniques (58). Assays for obtaining AA estimates at lower cost are also being developed (59). Tissues other than blood may offer useful alternative and potentially organ- or disease-specific AA measures (20, 23). An advantage of the Horvath AA measures is that they give an accurate prediction of age across multiple tissue types, allowing some generalizability of our findings to tissues other than blood (11). In conclusion, AA measures are associated with several established health risk factors and with mortality, independently of many of these health risk factors. In future studies, investigators should seek to identify genetic and additional environmental factors associated with AA and improve blood DNA methylation-based measures of biological age. ACKNOWLEDGMENTS Author affiliations: Cancer Epidemiology and Intelligence Division, Cancer Council Victoria, Melbourne, Victoria, Australia (Pierre-Antoine Dugué, Julie K. Bassett, Laura Baglietto, Dallas R. English, Gianluca Severi, Melissa C. Southey, Graham G. Giles, Roger L. Milne); Center for Epidemiology and Biostatistics, Melbourne School of Population and Global Health, University of Melbourne, Parkville, Victoria, Australia (Pierre-Antoine Dugué, Laura Baglietto, Daniel Schmidt, Enes Makalic, Shuai Li, Margarita Moreno-Betancur, Daniel D. Buchanan, Dallas R. English, John L. Hopper, Gianluca Severi, Melissa C. Southey, Graham G. Giles, Roger L. Milne); Genetic Epidemiology Laboratory, Department of Pathology, University of Melbourne, Parkville, Victoria, Australia (JiHoon E. Joo, Ee Ming Wong, Melissa C. Southey); Department of Clinical and Experimental Medicine, School of Medicine, University of Pisa, Pisa, Italy (Laura Baglietto); Melbourne Bioinformatics, University of Melbourne, Victoria, Australia (Chol-Hee Jung); Centre de Recherche en Épidémiologie et Santé des Populations (INSERM U1018), Université Paris-Saclay, Université Paris-Sud, Université Versailles Saint-Quentin-en-Yvelines, Institut Gustave Roussy, Villejuif, France (Gianluca Severi); Italian Institute for Genomic Medicine, Turin, Italy (Giovanni Fiorito, Paolo Vineis, Gianluca Severi); Clinical Epidemiology and Biostatistics Unit, Murdoch Children’s Research Institute, Melbourne, Victoria, Australia (Margarita Moreno-Betancur); Colorectal Oncogenomics Group, Genetic Epidemiology Laboratory, Department of Pathology, University of Melbourne, Parkville, Victoria, Australia (Daniel D. Buchanan); Genetic Medicine and Familial Cancer Center, Royal Melbourne Hospital, Parkville, Victoria, Australia (Daniel D. Buchanan); and MRC-PHE Center for Environment and Health, Imperial College London, London, United Kingdom (Paolo Vineis). This work was supported by the National Health and Medical Research Council (NHMRC) of Australia (grants 1088405 and 1074383) and by the European Commission (H2020 grant 633666; http://www.lifepathproject.eu/). L.B. was supported by a Marie Curie International Incoming Fellowship within the European Commission 7th Framework Programme. Cohort recruitment in the Melbourne Collaborative Cohort Study was funded by VicHealth and Cancer Council Victoria. The Melbourne Collaborative Cohort Study was further supported by NHMRC grants 209057 and 396414 and by infrastructure provided by Cancer Council Victoria. The nested case-control methylation studies were supported by NHMRC grants 1011618, 1026892, 1027505,1050198, and 1043616. Cases and their vital status were ascertained through the Victorian Cancer Registry and the Australian Institute of Health and Welfare, including the National Death Index and the Australian Cancer Database. Conflict of interest: none declared. Abbreviations AA age acceleration BMI body mass index CI confidence interval HR hazard ratio IEAA intrinsic epigenetic age acceleration REFERENCES 1 Jones MJ, Goodman SJ, Kobor MS. DNA methylation and healthy human aging. Aging Cell . 2015; 14( 6): 924– 932. Google Scholar CrossRef Search ADS PubMed  2 Issa JP. Aging and epigenetic drift: a vicious cycle. J Clin Invest . 2014; 124( 1): 24– 29. Google Scholar CrossRef Search ADS PubMed  3 Jung M, Pfeifer GP. Aging and DNA methylation. BMC Biol . 2015; 13: 7. Google Scholar CrossRef Search ADS PubMed  4 Issa JP, Ottaviano YL, Celano P, et al.  . Methylation of the oestrogen receptor CpG island links ageing and neoplasia in human colon. Nat Genet . 1994; 7( 4): 536– 540. Google Scholar CrossRef Search ADS PubMed  5 Calvanese V, Lara E, Kahn A, et al.  . The role of epigenetics in aging and age-related diseases. Ageing Res Rev . 2009; 8( 4): 268– 276. Google Scholar CrossRef Search ADS PubMed  6 Christensen BC, Houseman EA, Marsit CJ, et al.  . Aging and environmental exposures alter tissue-specific DNA methylation dependent upon CpG island context. PLoS Genet . 2009; 5( 8): e1000602. Google Scholar CrossRef Search ADS PubMed  7 Teschendorff AE, West J, Beck S. Age-associated epigenetic drift: implications, and a case of epigenetic thrift? Hum Mol Genet . 2013; 22( R1): R7– R15. Google Scholar CrossRef Search ADS PubMed  8 Mather KA, Jorm AF, Parslow RA, et al.  . Is telomere length a biomarker of aging? A review. J Gerontol A Biol Sci Med Sci . 2011; 66( 2): 202– 213. Google Scholar CrossRef Search ADS PubMed  9 Hannum G, Guinney J, Zhao L, et al.  . Genome-wide methylation profiles reveal quantitative views of human aging rates. Mol Cell . 2013; 49( 2): 359– 367. Google Scholar CrossRef Search ADS PubMed  10 Bocklandt S, Lin W, Sehl ME, et al.  . Epigenetic predictor of age. PLoS One . 2011; 6( 6): e14821. Google Scholar CrossRef Search ADS PubMed  11 Horvath S. DNA methylation age of human tissues and cell types. Genome Biol . 2013; 14( 10): R115. Google Scholar CrossRef Search ADS PubMed  12 Weidner CI, Lin Q, Koch CM, et al.  . Aging of blood can be tracked by DNA methylation changes at just three CpG sites. Genome Biol . 2014; 15( 2): R24. Google Scholar CrossRef Search ADS PubMed  13 Florath I, Butterbach K, Müller H, et al.  . Cross-sectional and longitudinal changes in DNA methylation with age: an epigenome-wide analysis revealing over 60 novel age-associated CpG sites. Hum Mol Genet . 2014; 23( 5): 1186– 1201. Google Scholar CrossRef Search ADS PubMed  14 Bekaert B, Kamalandua A, Zapico SC, et al.  . Improved age determination of blood and teeth samples using a selected set of DNA methylation markers. Epigenetics . 2015; 10( 10): 922– 930. Google Scholar CrossRef Search ADS PubMed  15 Marioni RE, Shah S, McRae AF, et al.  . DNA methylation age of blood predicts all-cause mortality in later life. Genome Biol . 2015; 16: 25. Google Scholar CrossRef Search ADS PubMed  16 Perna L, Zhang Y, Mons U, et al.  . Epigenetic age acceleration predicts cancer, cardiovascular, and all-cause mortality in a German case cohort. Clin Epigenetics . 2016; 8: 64. Google Scholar CrossRef Search ADS PubMed  17 Christiansen L, Lenart A, Tan Q, et al.  . DNA methylation age is associated with mortality in a longitudinal Danish twin study. Aging Cell . 2015; 15( 1): 149– 154. Google Scholar CrossRef Search ADS PubMed  18 Chen BH, Marioni RE, Colicino E, et al.  . DNA methylation-based measures of biological age: meta-analysis predicting time to death. Aging (Albany NY) . 2016; 8( 9): 1844– 1865. Google Scholar CrossRef Search ADS PubMed  19 Levine ME, Hosgood HD, Chen B, et al.  . DNA methylation age of blood predicts future onset of lung cancer in the Women’s Health Initiative. Aging (Albany NY) . 2015; 7( 9): 690– 700. Google Scholar CrossRef Search ADS PubMed  20 Levine ME, Lu AT, Bennett DA, et al.  . Epigenetic age of the pre-frontal cortex is associated with neuritic plaques, amyloid load, and Alzheimer’s disease related cognitive functioning. Aging (Albany NY) . 2015; 7( 12): 1198– 1211. Google Scholar CrossRef Search ADS PubMed  21 Horvath S, Mah V, Lu AT, et al.  . The cerebellum ages slowly according to the epigenetic clock. Aging (Albany NY) . 2015; 7( 5): 294– 306. Google Scholar CrossRef Search ADS PubMed  22 Horvath S, Pirazzini C, Bacalini MG, et al.  . Decreased epigenetic age of PBMCs from Italian semi-supercentenarians and their offspring. Aging (Albany NY) . 2015; 7( 12): 1159– 1170. Google Scholar CrossRef Search ADS PubMed  23 Horvath S, Erhart W, Brosch M, et al.  . Obesity accelerates epigenetic aging of human liver. Proc Natl Acad Sci USA . 2014; 111( 43): 15538– 15543. Google Scholar CrossRef Search ADS PubMed  24 Marioni RE, Shah S, McRae AF, et al.  . The epigenetic clock is correlated with physical and cognitive fitness in the Lothian Birth Cohort 1936. Int J Epidemiol . 2015; 44( 4): 1388– 1396. Google Scholar CrossRef Search ADS PubMed  25 Levine ME, Lu AT, Chen BH, et al.  . Menopause accelerates biological aging. Proc Natl Acad Sci USA . 2016; 113( 33): 9327– 9332. Google Scholar CrossRef Search ADS PubMed  26 Horvath S, Levine AJ. HIV-1 infection accelerates age according to the epigenetic clock. J Infect Dis . 2015; 212( 10): 1563– 1573. Google Scholar CrossRef Search ADS PubMed  27 Gross A, Jaeger PA, Kreisberg JF, et al.  . Methylome-wide analysis of chronic HIV infection reveals five-year increase in biological age and epigenetic targeting of HLA. Mol Cell . 2016; 62( 2): 157– 168. Google Scholar CrossRef Search ADS PubMed  28 Zannas AS, Arloth J, Carrillo-Roa T, et al.  . Lifetime stress accelerates epigenetic aging in an urban, African American cohort: relevance of glucocorticoid signaling. Genome Biol . 2015; 16: 266. Google Scholar CrossRef Search ADS PubMed  29 Boks MP, van Mierlo HC, Rutten BP, et al.  . Longitudinal changes of telomere length and epigenetic age related to traumatic stress and post-traumatic stress disorder. Psychoneuroendocrinology . 2015; 51: 506– 512. Google Scholar CrossRef Search ADS PubMed  30 Breitling LP, Saum KU, Perna L, et al.  . Frailty is associated with the epigenetic clock but not with telomere length in a German cohort. Clin Epigenetics . 2016; 8: 21. Google Scholar CrossRef Search ADS PubMed  31 Milne RL, Fletcher AS, MacInnis RJ, et al.  . Cohort profile: the Melbourne Collaborative Cohort Study (Health 2020) [published online ahead of print June 21, 2017]. Int J Epidemiol . ( doi:10.1093/ije/dyx085). 32 Severi G, Southey MC, English DR, et al.  . Epigenome-wide methylation in DNA from peripheral blood as a marker of risk for breast cancer. Breast Cancer Res Treat . 2014; 148( 3): 665– 673. Google Scholar CrossRef Search ADS PubMed  33 Dugué PA, Brinkman MT, Milne RL, et al.  . Genome-wide measures of DNA methylation in peripheral blood and the risk of urothelial cell carcinoma: a prospective nested case-control study. Br J Cancer . 2016; 115( 6): 664– 673. Google Scholar CrossRef Search ADS PubMed  34 Wong Doo N, Makalic E, Joo JE, et al.  . Global measures of peripheral blood-derived DNA methylation as a risk factor in the development of mature B-cell neoplasms. Epigenomics . 2016; 8( 1): 55– 66. Google Scholar CrossRef Search ADS PubMed  35 Baglietto L, Ponzi E, Haycock P, et al.  . DNA methylation changes measured in pre-diagnostic peripheral blood samples are associated with smoking and lung cancer risk. Int J Cancer . 2017; 140( 1): 50– 61. Google Scholar CrossRef Search ADS PubMed  36 Horvath S, Gurven M, Levine ME, et al.  . An epigenetic clock analysis of race/ethnicity, sex, and coronary heart disease. Genome Biol . 2016; 17( 1): 171. Google Scholar CrossRef Search ADS PubMed  37 Dugué PA, English DR, MacInnis RJ, et al.  . The repeatability of DNA methylation measures may also affect the power of epigenome-wide association studies. Int J Epidemiol . 2015; 44( 4): 1460– 1461. Google Scholar CrossRef Search ADS PubMed  38 Dugué PA, English DR, MacInnis RJ, et al.  . Reliability of DNA methylation measures from dried blood spots and mononuclear cells using the HumanMethylation450k BeadArray. Sci Rep . 2016; 6: 30317. Google Scholar CrossRef Search ADS PubMed  39 Rubin DB. Multiple Imputation for Nonresponse in Surveys . Hoboken, NJ: John Wiley & Sons, Inc.; 2004. 40 Buuren S, Groothuis-Oudshoorn K. mice: Multivariate imputation by chained equations in R. J Stat Softw . 2011; 45( 3): 1– 67. Google Scholar CrossRef Search ADS   41 Chiuve SE, Fung TT, Rimm EB, et al.  . Alternative dietary indices both strongly predict risk of chronic disease. J Nutr . 2012; 142( 6): 1009– 1018. Google Scholar CrossRef Search ADS PubMed  42 MacInnis RJ, English DR, Hopper JL, et al.  . Body size and composition and colon cancer risk in men. Cancer Epidemiol Biomarkers Prev . 2004; 13( 4): 553– 559. Google Scholar PubMed  43 Pink B; Australian Bureau of Statistics. Socio-Economic Indexes for Areas (SEIFA), 2011. Canberra, Australian Capital Territory, Australia: Australian Bureau of Statistics; 2013. (ABS publication no. 2033.0.55.001). 44 Thiébaut AC, Bénichou J. Choice of time-scale in Cox’s model analysis of epidemiologic cohort data: a simulation study. Stat Med . 2004; 23( 24): 3803– 3820. Google Scholar CrossRef Search ADS PubMed  45 Grambsch PM, Therneau TM. Proportional hazards tests and diagnostics based on weighted residuals. Biometrika . 1994; 81( 3): 515– 526. Google Scholar CrossRef Search ADS   46 Andersen PK, Geskus RB, de Witte T, et al.  . Competing risks in epidemiology: possibilities and pitfalls. Int J Epidemiol . 2012; 41( 3): 861– 870. Google Scholar CrossRef Search ADS PubMed  47 Therneau T. Mixed Effects Cox Models. Vienna, Austria: R Foundation for Statistical Computing; 2015. https://cran.r-project.org/web/packages/coxme/vignettes/coxme.pdf. Updated June 15, 2015. Accessed June 26, 2017. 48 Lunn M, McNeil D. Applying Cox regression to competing risks. Biometrics . 1995; 51( 2): 524– 532. Google Scholar CrossRef Search ADS PubMed  49 Jaffe AE, Irizarry RA. Accounting for cellular heterogeneity is critical in epigenome-wide association studies. Genome Biol . 2014; 15( 2): R31. Google Scholar CrossRef Search ADS PubMed  50 Giles GG, English DR. The Melbourne Collaborative Cohort Study. IARC Sci Publ . 2002; 156: 69– 70. Google Scholar PubMed  51 Boonekamp JJ, Simons MJ, Hemerik L, et al.  . Telomere length behaves as biomarker of somatic redundancy rather than biological age. Aging Cell . 2013; 12( 2): 330– 332. Google Scholar CrossRef Search ADS PubMed  52 Quach A, Levine ME, Tanaka T, et al.  . Epigenetic clock analysis of diet, exercise, education, and lifestyle factors. Aging (Albany NY) . 2017; 9( 2): 419– 446. Google Scholar PubMed  53 Linnér RK, Marioni RE, Rietveld CA, et al.  . An epigenome-wide association study meta-analysis of educational attainment. Mol Psychiatry . 2017; 22( 12): 1680– 1690. Google Scholar CrossRef Search ADS PubMed  54 López-Otín C, Blasco MA, Partridge L, et al.  . The hallmarks of aging. Cell . 2013; 153( 6): 1194– 1217. Google Scholar CrossRef Search ADS PubMed  55 Marioni RE, Harris SE, Shah S, et al.  . The epigenetic clock and telomere length are independently associated with chronological age and mortality. Int J Epidemiol . 2016; 45( 2): 424– 432. Google Scholar CrossRef Search ADS   56 Li S, Wong EM, Joo JE, et al.  . Genetic and environmental causes of variation in the difference between biological age based on DNA methylation and chronological age for middle-aged women. Twin Res Hum Genet . 2015; 18( 6): 720– 726. Google Scholar CrossRef Search ADS PubMed  57 Naeem H, Wong NC, Chatterton Z, et al.  . Reducing the risk of false discovery enabling identification of biologically significant genome-wide methylation status using the HumanMethylation450 array. BMC Genomics . 2014; 15: 51. Google Scholar CrossRef Search ADS PubMed  58 Walker DL, Bhagwate AV, Baheti S, et al.  . DNA methylation profiling: comparison of genome-wide sequencing methods and the Infinium Human Methylation 450 Bead Chip. Epigenomics . 2015; 7( 8): 1287– 1302. Google Scholar CrossRef Search ADS PubMed  59 Vidal-Bralo L, Lopez-Golan Y, Gonzalez A. Simplified assay for epigenetic age estimation in whole blood of adults. Front Genet . 2016; 7: 126. Google Scholar CrossRef Search ADS PubMed  © The Author(s) 2018. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png American Journal of Epidemiology Oxford University Press

Loading next page...
 
/lp/ou_press/association-of-dna-methylation-based-biological-age-with-health-risk-keeyPlecDn
Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.
ISSN
0002-9262
eISSN
1476-6256
D.O.I.
10.1093/aje/kwx291
Publisher site
See Article on Publisher Site

Abstract

Abstract Measures of biological age based on blood DNA methylation, referred to as age acceleration (AA), have been developed. We examined whether AA was associated with health risk factors and overall and cause-specific mortality. At baseline (1990–1994), blood samples were drawn from 2,818 participants in the Melbourne Collaborative Cohort Study (Melbourne, Victoria, Australia). DNA methylation was determined using the Infinium HumanMethylation450 BeadChip array (Illumina Inc., San Diego, California). Mixed-effects models were used to examine the association of AA with health risk factors. Cox models were used to assess the association of AA with mortality. A total of 831 deaths were observed during a median 10.7 years of follow-up. Associations of AA were observed with male sex, Greek nationality (country of birth), smoking, obesity, diabetes, lower education, and meat intake. AA measures were associated with increased mortality, and this was only partly accounted for by known determinants of health (hazard ratios were attenuated by 20%–40%). Weak evidence of heterogeneity in the association was observed by sex (P = 0.06) and cause of death (P = 0.07) but not by other factors. DNA-methylation-based AA measures are associated with several major health risk factors, but these do not fully explain the association between AA and mortality. Future research should investigate what genetic and environmental factors determine AA. age acceleration, aging, biological age, cancer, DNA methylation, epigenetic clock, health risk factors, mortality Aging is associated with profound changes to DNA methylation. Globally, a decrease in methylation levels with age is observed, as well as a tendency for both hypo- and hypermethylated CpG sites to lose their initial epigenetic state and become hemimethylated (1–3). These changes are thought to play a role in the risk of disease (4–7). Recently, the concept of an “epigenetic clock” has emerged as a marker of biological aging (8). Several studies have attempted to quantify the age of the methylome using regularized regression methods to obtain a restricted set of DNA methylation measures predicting age (9–13). Two of these models have received more attention, as they predict chronological age most accurately (9, 11). Although the accurate prediction of chronological age is of interest to forensic scientists (14), the difference between predicted age (linear combination of age-associated methylation measures) and chronological age is also thought to represent an estimate of “age acceleration” (AA) and has been interpreted in epidemiologic studies as a marker of biological aging. Several studies have investigated the association between AA in blood and all-cause mortality. In one study, Marioni et al. (15) pooled data on 4 cohorts from the United Kingdom and the United States, comprising 808 deaths, and reported 16% and 9% increased mortality per 5-year increase in AA based on the Hannum et al. (9) and Horvath (11) predictors, respectively, after adjusting for several health risk factors. Similar findings were reported from an elderly German cohort (1,863 participants; 602 deaths) (16), with mortality increases of 10% and 23% per 5-year increment of AA according to the Hannum and Horvath predictors, respectively. In a Danish cohort study of 86 twin pairs, among which 55 deaths occurred, Christiansen et al. (17) reported a 35% increased risk of death per 5-year AA increment based on both Hannum’s and Horvath’s predictors. A recent pooled analysis by Chen et al. (18) showed mortality risks increased by 14% and 8%, respectively. Outcomes other than mortality have been investigated, including lung cancer (19) and Alzheimer’s disease (20). Distinct AA profiles have been observed in the blood of centenarians and their offspring (21, 22). If AA reflects biological age, it seems reasonable to expect that it is influenced by known risk factors for poor health (8). AA has been shown to be associated with obesity, most notably in liver tissue (11, 23). Cross-sectional associations with physical and cognitive ability (24), menopause (25), human immunodeficiency virus infection (26, 27), lifetime stress (28), posttraumatic stress disorder (29), and frailty in the elderly (30) have been reported. Our aim was to add to the growing literature on AA and health by 1) assessing associations between AA and known health risk factors, as well as all-cause and cause-specific mortality, and 2) examining the extent to which various lifestyle factors contribute to the relationship between AA and mortality, using a large cohort of middle-aged and older adults. METHODS Study sample We used data from studies nested within the Melbourne Collaborative Cohort Study (Melbourne, Victoria, Australia), a prospective cohort study of 41,514 healthy adult volunteers (24,469 women) aged 27–76 years (99.3% were aged 40–69 years) at baseline between 1990 and 1994 (31). DNA samples used for the present analysis were extracted from peripheral blood drawn at the time of recruitment (1990–1994). For the majority (70%) of participants, the DNA source was dried blood spots collected onto Guthrie Card Diagnostic Cellulose filter paper (Whatman plc, Kent, United Kingdom) and stored in airtight containers at room temperature. The other sources of DNA were peripheral blood mononuclear cells and buffy coats stored at −80°C for 28% and 2% participants, respectively. The study sample comprised Melbourne Collaborative Cohort Study participants selected as controls in nested case-control studies of breast, colorectal, kidney, lung, prostate, or urothelial cancer or mature B-cell malignancies (32–35). Controls had been individually matched to cases on age (they had to be free of cancer at an age within 1 year of the age at diagnosis of the corresponding case), sex, country of birth, and blood DNA source (dried blood spot, peripheral blood mononuclear cells, or buffy coat). For all but the colorectal cancer study, controls were matched to cases on year of birth. For the lung cancer study, controls were matched on smoking status at the time of blood collection. Vital status up to August 31, 2015, was ascertained through record linkage to the Victorian Registry of Births, Deaths and Marriages (via the Victorian Cancer Registry) and the National Death Index (via the Australian Institute of Health and Welfare). Cause of death, based on the underlying cause of death from the death certificate, was known for deaths occurring up to December 31, 2013. Methods relating to DNA extraction and bisulfite conversion and DNA methylation data processing are described in the Web Appendix (available at https://academic.oup.com/aje). Methylation age and AA All of the CpGs used to compute Hannum and Horvath methylation ages were available in our study after quality control. Hannum methylation age was computed using the linear combination of methylation measures at 71 age-associated CpGs (15). Horvath methylation age was calculated on the basis of methylation measures at 353 CpGs using the online calculator (11). We considered the most recent measures of AA (18, 36): 1) epigenetic AA based on the Horvath predictor—residuals resulting from a linear regression model of Horvath’s estimate of epigenetic age on chronological age, referred to as AA-Horvath; 2) the corresponding AA based on the Hannum et al. predictor (AA-Hannum); 3) intrinsic epigenetic AA (IEAA) based on the Horvath predictor—residuals resulting from a linear regression model of Horvath’s estimate of epigenetic age on chronological age and measures of blood cell counts, referred to as IEAA-Horvath; 4) the corresponding IEAA based on the Hannum et al. predictor (IEAA-Hannum); and 5) enhanced Hannum AA, defined as the AA-Hannum measure plus a weighted average of aging-associated cell counts. Results are presented in the main text for IEAA measures and in Web Tables 1–7 for other measures. Statistical analysis Correlations between AA measures were assessed using Spearman correlation coefficients. To evaluate the accuracy of the Horvath and Hannum age predictors, we computed the median absolute value of each AA measure. The reliability of methylation measures for CpG sites included in the predictors was examined using intraclass correlation coefficients computed using mixed models and based on 134 technical replicate pairs as described in previous publications (37, 38). Missing covariate data (<1% for any individual covariate) were handled using multiple imputation (39), with the R package mice (R Foundation for Statistical Computing, Vienna, Austria) (40). Health risk factors associated with AA We used linear regression models to examine the association between AA measures and known risk factors for disease, all assessed at baseline. Factors considered and adjusted for were: age; sex; country of birth (Australia/New Zealand, United Kingdom/Malta, Italy, or Greece); smoking (never smoker, former smoker who had quit ≤15 years prior, former smoker who had quit >15 years prior, current smoker of ≤20 cigarettes/day, or current smoker of >20 cigarettes/day); body mass index (BMI; calculated as weight (kg)/height (m)2 and categorized as <25, 25–29.9, 30–34.9, or ≥35); height (cm; continuous); alcohol intake (none, 1–39 g/day (males) or 1–19 g/day (females), 40–59 g/day (males) or 20–39 g/day (females), or ≥60 g/day (males) or ≥40 g/day (females)); Alternate Healthy Eating Index 2010 score (to reflect overall diet quality) (41); a physical activity score ranging from 1 to 4 and reflecting metabolic equivalents, as described in previous work (42); socioeconomic status (a score from 1 to 10 representing the relative socioeconomic disadvantage of the participant’s area of residence (43)); and education (a score from 1 to 8; 1 = primary school only and 8 = tertiary university degree or more). We also considered the following clinical variables: history of cardiovascular disease (stroke, angina, or heart attack), hypertension, diabetes, asthma, kidney stones, gallstones, arthritis, and resting heart rate (measured with the Dinamap 1846SX automated blood pressure monitor (Critikon, Tampa, Florida)). Each of these clinical factors was added individually to the previous model. Results of all models were additionally adjusted for sample type (dried blood spot, peripheral blood mononuclear cells, or buffy coat), study, and the plate on which the sample was processed (the latter 2 variables being fitted as random effects and hereafter referred to as batch effects). More detailed analyses were conducted using additional variables relating to diet, clinical factors, smoking, alcohol intake, anthropometric factors, psychological well-being, and blood measures (Web Table 5). We used the same models to assess the associations between demographic, lifestyle, and anthropometric variables and each individual CpG site included in the Hannum and Horvath predictors. In these models, smoking, BMI, and alcohol intake were modeled as continuous variables. AA and mortality We used Cox models to estimate hazard ratios for the association between AA measures and mortality. Age was used as the time scale (44), and participants were considered to be at risk from their index date (the date on which they reached the age at which their matched case in the corresponding nested case-control study was diagnosed with cancer) onward in order to avoid immortal time bias, as they had to be alive at this date. For each AA measure, the proportional hazards assumption was assessed by visual inspection of Schoenfeld residuals (45). Separate models were fitted for all-cause, cancer, cardiovascular disease, and other-cause mortality. Person-years of follow-up were calculated from the index date to the date of death, and participants were censored at the date of departure from Australia or the end of follow-up (August 31, 2015, for all-cause death and December 31, 2013, for cause-specific death), whichever came first. In cause-specific analyses, deaths from other causes were censored, which is a way to model the cause-specific hazard function, appropriately taking the competing risk of other-cause death into account (46). For comparability with previous publications, hazard ratio estimates were calculated for a 5-year increase in AA measures (15–17). For overall and cause-specific mortality analyses, 3 models were fitted in order to examine the robustness of the association between AA measures and mortality: In model 0 (unadjusted), results were adjusted for sample type and batch effects; in model 1 (minimally adjusted), we added age at blood draw, sex, and country of birth to model 0; in model 2, we added anthropometric measures, lifestyle factors, and socioeconomic variables (BMI, height, smoking, alcohol intake, diet quality, physical activity, socioeconomic status, and education) to model 1; and in model 3, we added clinical variables to model 2. Batch effects were fitted as fixed effects, and similar results were obtained when using mixed-effects Cox models (47). To investigate whether AA measures were more strongly associated with specific causes of death, heterogeneity in the associations between AA and mortality by cause of death was tested using the data duplication method (48). Interactions between AA measures and each of sex, country of birth, smoking, and BMI were examined by adding interaction terms to the model and applying a likelihood ratio test. Analyses for AA were repeated with adjustment for cell composition (as defined by the Houseman algorithm (49)). Ethics Study participants provided informed consent in accordance with the Declaration of Helsinki. The study was approved by Cancer Council Victoria’s Human Research Ethics Committee and performed in accordance with the institution’s ethical guidelines. RESULTS The correlation between the Hannum and Horvath age predictors was ρ = 0.75, and their correlation with chronological age was high (ρ = 0.76 and ρ = 0.73, respectively) (Table 1). The median AA (IEAA) was, in absolute terms, 3.9 years and 4.2 years for the Hannum and Horvath predictors, respectively (Figure 1, panels A and B, respectively). The correlation between IEAA measures was 0.56 (Table 1). The reliability of methylation measures was high for CpG sites included in Hannum’s age predictor (median intraclass correlation coefficient, 0.60; interquartile range, 0.53–0.68) and more variable for those included in Horvath’s predictor (median intraclass correlation coefficient, 0.48; interquartile range, 0.22–0.61) (Web Table 8 and Web Figure 1). Table 1. Characteristics of Predictors of Age Acceleration, Melbourne Collaborative Cohort Study, Australia, 1990–1994 Age Acceleration Predictor  Horvath Predictor  Hannum Predictor  Mean (SD)  ρ  Mean (SD)  ρ  Predicted age, years  59.1 (8.0)    59.0 (10.1)    Chronological age, years  59.0 (7.6)    59.0 (7.6)    Correlation with agea    0.73    0.76  Correlation with age predictor           AAb  0.3 (7.1)  −0.01  −0.2 (6.7)  0.03   IEAAc  0.0 (6.9)  −0.06  0.0 (6.4)  −0.03  Median age differenced           AA  4.4    4.2     IEAA  4.2    3.9    Age Acceleration Predictor  Horvath Predictor  Hannum Predictor  Mean (SD)  ρ  Mean (SD)  ρ  Predicted age, years  59.1 (8.0)    59.0 (10.1)    Chronological age, years  59.0 (7.6)    59.0 (7.6)    Correlation with agea    0.73    0.76  Correlation with age predictor           AAb  0.3 (7.1)  −0.01  −0.2 (6.7)  0.03   IEAAc  0.0 (6.9)  −0.06  0.0 (6.4)  −0.03  Median age differenced           AA  4.4    4.2     IEAA  4.2    3.9    Abbreviations: AA, age acceleration; IEAA, intrinsic epigenetic age acceleration; SD, standard deviation. a Correlation between the Horvath (11) and Hannum et al. (9) age predictors: ρ = 0.75. b Correlation between the Horvath and Hannum AA measures: ρ = 0.56. c Correlation between the Horvath and Hannum IEAA measures: ρ = 0.54. d Median age difference = absolute value of AA; age and AA were measured in years. Figure 1. View largeDownload slide Distribution of intrinsic epigenetic age acceleration (IEAA) measures (years) in the Melbourne Collaborative Cohort Study, Australia, 1990–1994. A) IEAA based on the Hannum et al. predictor (9); B) IEAA based on the Horvath predictor (11). “Frequency” (y-axis) denotes the number of observations in each 5-year age group bar. Figure 1. View largeDownload slide Distribution of intrinsic epigenetic age acceleration (IEAA) measures (years) in the Melbourne Collaborative Cohort Study, Australia, 1990–1994. A) IEAA based on the Hannum et al. predictor (9); B) IEAA based on the Horvath predictor (11). “Frequency” (y-axis) denotes the number of observations in each 5-year age group bar. A total of 2,818 participants were included in the analysis of health risk factors. Higher AAs were observed in males compared with females (IEAA-Hannum: β = 1.37 years, P < 0.001; IEAA-Horvath: β = 1.23 years, P = 0.005); in Greek participants compared with those born in Australia/New-Zealand (IEAA-Horvath: β = 1.29 years, P = 0.01); in current and former smokers compared with never smokers (IEAA-Hannum: current smoker of ≤20 cigarettes/day—β = 2.12 years, P < 0.001; current smoker of >20 cigarettes/day—β = 1.26 years, P = 0.01), and in overweight (BMI ≥25) and obese (BMI ≥30) persons compared with lean individuals (BMI <25) (IEAA-Hannum: BMI 25–29.9—β = 0.69 years, P = 0.01; BMI 30–34.9—β = 1.24 years, P < 0.001; BMI ≥35—β = 1.70 years, P = 0.01). Higher education was associated with lower AA (AA-Horvath: per 1-unit increment in education score, β = −0.16 years, P = 0.04 (Web Table 1)). Most reported comorbid conditions were not associated with AA measures, except for diabetes (IEAA-Horvath: β = 1.46 years, P = 0.04) and prior history of gallstones (IEAA-Horvath: β = 1 year, P = 0.03) (Table 2). Intake of meat was associated with increased AA (IEAA-Horvath: P = 0.01), and intake of fruits, but not vegetables, was associated with decreased AA (IEAA-Hannum: P = 0.03 and P = 0.91, respectively). Most body composition measures were associated with increased AA (Web Table 5). Table 2. Characteristics of the Study Sample and Their Association With Measures of Intrinsic Epigenetic Age Acceleration (n = 2,818), Melbourne Collaborative Cohort Study, Australia, 1990–1994 Variablea  No. of Persons  %  IEAA-Hannum Predictor  IEAA-Horvath Predictor  βb  SE  P Value  βb  SE  P Value  Demographic, anthropometric, and lifestyle variables                   Male sex  1,722  61  1.37  0.4  <0.001  1.23  0.44  0.005   Country of birth                    Greece  288  10  0.32  0.45  0.48  1.29  0.50  0.01    Italy  377  13  0.54  0.40  0.18  0.76  0.44  0.09    United Kingdom/Malta  215  8  0.14  0.44  0.75  0.21  0.49  0.68   Tobacco smoking                    Former smoker                     Quit >15 years prior  545  19  −0.03  0.33  0.92  0.45  0.36  0.22     Quit ≤15 years prior  533  19  0.65  0.33  0.05  0.78  0.37  0.03    Current smoker                     ≤20 cigarettes/day  159  6  2.12  0.52  <0.001  1.33  0.58  0.02     >20 cigarettes/day  228  8  1.26  0.48  0.01  0.86  0.53  0.11   Body mass indexc                    25–29.9  1,384  49  0.69  0.27  0.01  0.52  0.30  0.09    30–34.9  451  16  1.24  0.37  <0.001  0.29  0.41  0.47    ≥35  105  4  1.70  0.64  0.01  2.23  0.71  0.002   Height, cm  167 (9.2)d  0.02  0.02  0.26  0.00  0.02  0.94   Alcohol intake                    Low  1,604  57  −0.01  0.29  0.98  0.44  0.33  0.18    Moderate  263  9  −0.20  0.45  0.66  0.42  0.50  0.40    High  154  5  0.09  0.56  0.88  0.09  0.62  0.89   Alternate Healthy Eating Index 2010 scoree  63 (10.9)d  −0.02  0.01  0.10  −0.01  0.01  0.33   Physical activity scoref  2.6 (1.0)d  0.18  0.11  0.11  0.07  0.13  0.59   Education scoreg  4.9 (1.9)d  −0.14  0.07  0.06  −0.10  0.08  0.22  Clinical variables                   History of asthma  312  11  0.52  0.37  0.16  0.37  0.41  0.36   History of hypertension  677  24  −0.02  0.27  0.93  −0.09  0.31  0.78   History of diabetes  99  4  0.80  0.64  0.21  1.46  0.71  0.04   History of arthritis  997  35  −0.30  0.25  0.23  −0.12  0.28  0.66   History of kidney stones  193  7  −0.08  0.46  0.87  0.15  0.51  0.77   History of gallstones  253  9  0.25  0.41  0.54  1.00  0.45  0.03   History of cardiovascular disease  189  7  −0.21  0.47  0.65  0.01  0.52  0.99   Resting heart rate, beats/minute  68.2 (10.0)d  0.00  0.01  0.68  0.02  0.01  0.22  Variablea  No. of Persons  %  IEAA-Hannum Predictor  IEAA-Horvath Predictor  βb  SE  P Value  βb  SE  P Value  Demographic, anthropometric, and lifestyle variables                   Male sex  1,722  61  1.37  0.4  <0.001  1.23  0.44  0.005   Country of birth                    Greece  288  10  0.32  0.45  0.48  1.29  0.50  0.01    Italy  377  13  0.54  0.40  0.18  0.76  0.44  0.09    United Kingdom/Malta  215  8  0.14  0.44  0.75  0.21  0.49  0.68   Tobacco smoking                    Former smoker                     Quit >15 years prior  545  19  −0.03  0.33  0.92  0.45  0.36  0.22     Quit ≤15 years prior  533  19  0.65  0.33  0.05  0.78  0.37  0.03    Current smoker                     ≤20 cigarettes/day  159  6  2.12  0.52  <0.001  1.33  0.58  0.02     >20 cigarettes/day  228  8  1.26  0.48  0.01  0.86  0.53  0.11   Body mass indexc                    25–29.9  1,384  49  0.69  0.27  0.01  0.52  0.30  0.09    30–34.9  451  16  1.24  0.37  <0.001  0.29  0.41  0.47    ≥35  105  4  1.70  0.64  0.01  2.23  0.71  0.002   Height, cm  167 (9.2)d  0.02  0.02  0.26  0.00  0.02  0.94   Alcohol intake                    Low  1,604  57  −0.01  0.29  0.98  0.44  0.33  0.18    Moderate  263  9  −0.20  0.45  0.66  0.42  0.50  0.40    High  154  5  0.09  0.56  0.88  0.09  0.62  0.89   Alternate Healthy Eating Index 2010 scoree  63 (10.9)d  −0.02  0.01  0.10  −0.01  0.01  0.33   Physical activity scoref  2.6 (1.0)d  0.18  0.11  0.11  0.07  0.13  0.59   Education scoreg  4.9 (1.9)d  −0.14  0.07  0.06  −0.10  0.08  0.22  Clinical variables                   History of asthma  312  11  0.52  0.37  0.16  0.37  0.41  0.36   History of hypertension  677  24  −0.02  0.27  0.93  −0.09  0.31  0.78   History of diabetes  99  4  0.80  0.64  0.21  1.46  0.71  0.04   History of arthritis  997  35  −0.30  0.25  0.23  −0.12  0.28  0.66   History of kidney stones  193  7  −0.08  0.46  0.87  0.15  0.51  0.77   History of gallstones  253  9  0.25  0.41  0.54  1.00  0.45  0.03   History of cardiovascular disease  189  7  −0.21  0.47  0.65  0.01  0.52  0.99   Resting heart rate, beats/minute  68.2 (10.0)d  0.00  0.01  0.68  0.02  0.01  0.22  Abbreviations: IEAA, intrinsic epigenetic age acceleration; SE, standard error. a Reference categories were as follows: sex—female; country of birth—Australia/New Zealand; tobacco smoking—never; body mass index—<25; alcohol intake—none; clinical variables—no history. b All estimates in the upper part of the table (demographic, lifestyle, and anthropometric variables) were mutually adjusted and further adjusted for age at blood draw, socioeconomic status (score representing the relative socioeconomic disadvantage of the participant’s area of residence, ranging from 1 to 10), sample type, and batch effects. All estimates in the lower part of the table (clinical variables) were adjusted for variables in the upper part of the table (demographic, lifestyle, and anthropometric variables) and further adjusted for socioeconomic status, sample type, and batch effects. c Weight (kg)/height (m)2. d Values are expressed as mean (standard deviation). e More details can be found in the paper by Chiuve et al. (41). f Physical activity score reflected metabolic equivalents, as described in previous work (42); possible scores ranged from 1 to 4. g Education was an ordinal questionnaire variable with possible scores ranging from 1 to 8 (1 = primary school only and 8 = tertiary university degree or more). Several individual CpG sites included in the Hannum and Horvath predictors were associated with demographic, lifestyle, and anthropometric variables (after correction for multiple testing for 424 tests, P < 1.1 × 10−4). Most of them were strongly associated with age, and associations were also observed with sex (74 CpGs), smoking (9 CpGs), and alcohol intake (6 CpGs) (Web Table 9). A total of 831 deaths were observed during follow-up (median, 10.7 years), including 692 with information on cause of death (follow-up until December 31, 2013). Unadjusted analyses (model 0) showed associations of AA measures with all-cause mortality (IEAA-Hannum: per 5-year increment in AA, hazard ratio (HR) = 1.08, 95% confidence interval (CI): 1.02, 1.14; findings were similar for other AA measures) (Table 3). After adjustment for age, sex, and country of birth (model 1), the associations for IEAA were attenuated by up to 7% (IEAA-Hannum: 7%; IEAA-Horvath: 0%) (Table 3). After adjustment for lifestyle and anthropometric and socioeconomic variables (model 2), hazard ratios were attenuated by a further 18%–37% (IEAA-Hannum: 37%; IEAA-Horvath: 18%) but not further after adjustment for reported comorbidity (model 3). In total, the association of AA with overall mortality was attenuated by 18%–45% after adjustment for known health risk factors; the remaining association was 5% increased mortality per 5-year AA increment for all AA measures. AA was more strongly associated with cancer-related death (n = 240) than with other causes of death (IEAA-Hannum: HR = 1.12 (95% CI: 1.02, 1.24); IEAA-Horvath: HR = 1.11 (95% CI: 1.01, 1.21)) (Table 4), with weak evidence of heterogeneity by cause of death (P-heterogeneity = 0.07 and P-heterogeneity = 0.14, respectively). Table 3. Associations of Age Acceleration (Intrinsic Epigenetic Age Acceleration, per 5-Year Increase) With All-Cause Mortality Risk During a Follow-up Period Extending to August 31, 2015 (n = 2,818; 831 Deaths During 31,449 Person-Years of Follow-Up), Melbourne Collaborative Cohort Study, Australia, 1990–1994 Age Acceleration Predictor  Model  Model 0a  Model 1b  Model 2c  Model 3d  HR  95% CI  P Value  HR  95% CI  P Value  HR  95% CI  P Value  HR  95% CI  P Value  IEAA-Hannum  1.08  1.02, 1.14  0.008  1.07  1.01, 1.13  0.02  1.04  0.99, 1.10  0.13  1.04  0.99, 1.11  0.12  IEAA-Horvath  1.07  1.02, 1.14  0.01  1.07  1.01, 1.12  0.01  1.06  1.00, 1.11  0.05  1.05  1.00, 1.11  0.05  Age Acceleration Predictor  Model  Model 0a  Model 1b  Model 2c  Model 3d  HR  95% CI  P Value  HR  95% CI  P Value  HR  95% CI  P Value  HR  95% CI  P Value  IEAA-Hannum  1.08  1.02, 1.14  0.008  1.07  1.01, 1.13  0.02  1.04  0.99, 1.10  0.13  1.04  0.99, 1.11  0.12  IEAA-Horvath  1.07  1.02, 1.14  0.01  1.07  1.01, 1.12  0.01  1.06  1.00, 1.11  0.05  1.05  1.00, 1.11  0.05  Abbreviations: CI, confidence interval; HR, hazard ratio; IEAA, intrinsic epigenetic age acceleration. a Model 0 (unadjusted results) adjusted for sample type and batch effects. b Model 1 (minimally adjusted results) adjusted for age at blood draw, sex, country of birth (Australia/New Zealand, United Kingdom/Malta, Italy, or Greece), sample type, and batch effects. c In model 2, the following anthropometric measures, lifestyle factors, and socioeconomic variables were added to model 1: body mass index (weight (kg)/height (m)2; <25, 25–29.9, 30–34.9, or ≥35), height (cm; continuous), smoking (never smoker; former smoker who quit ≤15 years prior; former smoker who quit >15 years prior; current smoker of ≤20 cigarettes/day; current smoker of >20 cigarettes/day), alcohol intake (1–19 g/day (females); 40–59 g/day (males) or 20–39 g/day (females); or ≥60 g/day (males) or ≥40 g/day (females)), diet quality (Alternate Healthy Eating Index 2010), physical activity (physical activity score ranging from 1 to 4 and reflecting metabolic equivalents (39)), socioeconomic status (score representing the relative socioeconomic disadvantage of the participant’s area of residence, ranging from 1 to 10 (40)), and education (score ranging from 1 to 8, with 1 = primary school only and 8 = tertiary university degree or more). d In model 3, the following clinical variables were added to model 2: history of asthma (yes/no), hypertension (yes/no), diabetes (yes/no), arthritis (yes/no), kidney stones (yes/no), gallstones (yes/no), or cardiovascular disease (yes/no) and resting heart rate (beats/minute; continuous). Table 4. Associations of Age Acceleration (Intrinsic Epigenetic Age Acceleration, per 5-Year Increase) With Cause-Specific Mortality Risk (Model 2a) During a Follow-up Period Extending to December 31, 2013 (n = 2,818; 692 Deaths During 28,022 Person-Years of Follow-up), Melbourne Collaborative Cohort Study, Australia, 1990–1994 Mortality Outcome  No. of Deaths  IEAA-Hannum Predictor  IEAA-Horvath Predictor  HR  95% CI  P Value  HR  95% CI  P Value  CVD mortality  203  1.08  0.97, 1.21  0.17  1.00  0.90, 1.12  0.93  Cancer mortality  240  1.12  1.02, 1.24  0.02  1.11  1.01, 1.21  0.03  Other-cause mortality  249  0.96  0.86, 1.06  0.37  1.06  0.97, 1.16  0.22  Mortality Outcome  No. of Deaths  IEAA-Hannum Predictor  IEAA-Horvath Predictor  HR  95% CI  P Value  HR  95% CI  P Value  CVD mortality  203  1.08  0.97, 1.21  0.17  1.00  0.90, 1.12  0.93  Cancer mortality  240  1.12  1.02, 1.24  0.02  1.11  1.01, 1.21  0.03  Other-cause mortality  249  0.96  0.86, 1.06  0.37  1.06  0.97, 1.16  0.22  Abbreviations: CI, confidence interval; CVD, cardiovascular disease, HR, hazard ratio; IEAA, intrinsic epigenetic age acceleration. a Results were adjusted for age at blood draw, sex, country of birth (Australia/New Zealand, United Kingdom/Malta, Italy, or Greece), body mass index (weight (kg)/height (m)2; <25, 25–29.9, 30–34.9, or ≥35), height (cm; continuous), smoking (never smoker; former smoker who quit ≤15 years prior; former smoker who quit >15 years prior; current smoker of ≤20 cigarettes/day; current smoker of >20 cigarettes/day), alcohol intake (1–19 g/day (females); 40–59 g/day (males) or 20–39 g/day (females); or ≥60 g/day (males) or ≥40 g/day (females)), diet quality (Alternate Healthy Eating Index 2010), physical activity (physical activity score ranging from 1 to 4 and reflecting metabolic equivalents (39)), socioeconomic status (score representing the relative socioeconomic disadvantage of the participant’s area of residence, ranging from 1 to 10 (40)), education (score ranging from 1 to 8, with 1 = primary school only and 8 = tertiary university degree or more), sample type, and batch effects. The estimated associations of IEAA measures with overall mortality appeared stronger for men than for women (IEAA-Hannum: men—HR = 1.08 (95% CI: 1.01, 1.15); women—HR = 0.97 (95% CI: 0.88, 1.07) (P-heterogeneity = 0.06); IEAA-Horvath: men—HR = 1.09 (95% CI: 1.02, 1.16); women—HR = 0.99 (95% CI: 0.91, 1.09) (P-heterogeneity = 0.11)). No evidence of effect modification was found for smoking status, BMI, or country of birth (Table 5). When the analysis was restricted to participants for whom follow-up started less than 5 years after blood draw, the hazard ratio estimates were similar to those based on all participants. Table 5. Associations of Age Acceleration (Intrinsic Epigenetic Age Acceleration, per 5-Year Increase) With All-Cause Mortality According to Selected Health Risk Factors (Model 2a), Melbourne Collaborative Cohort Study, Australia, 1990–1994 Age Acceleration Measure  IEAA-Hannum Predictor  IEAA-Horvath Predictor  HR  95% CI  P-het  HR  95% CI  P-het  Sex               Female  0.97  0.88, 1.07  0.06  0.99  0.91, 1.09  0.11   Male  1.08  1.01, 1.15  1.09  1.02, 1.16  Smoking status               Never smoker  1.01  0.93, 1.10  0.65  1.05  0.97, 1.13  0.63   Former smoker            Quit >15 years prior  1.04  0.92, 1.17  1.04  0.93, 1.16    Quit ≤15 years prior  1.12  1.01, 1.26  1.10  0.99, 1.23   Current smoker            <20 cigarettes/day  1.01  0.84, 1.22  0.93  0.77, 1.13    ≥20 cigarettes/day  1.03  0.88, 1.21  1.10  0.94, 1.28  Body mass indexb               <25  1.03  0.93, 1.13  0.50  1.04  0.95, 1.13  0.74   25–29.9  1.08  1.00, 1.17  1.08  1.00, 1.16   ≥30  1.00  0.90, 1.12  1.03  0.93, 1.15  Country of birth               Australia/New-Zealand  1.04  0.97, 1.11  0.77  1.04  0.98, 1.11  0.44   United Kingdom  1.08  0.93, 1.27  1.09  0.93, 1.28   Italy  1.00  0.88, 1.15  1.01  0.89, 1.16   Greece  1.12  0.92, 1.37  1.21  1.00, 1.46  Time between blood draw and start of follow-up, years               <5  1.07  0.96, 1.19  0.75  1.06  0.96, 1.17  0.25   5–10  1.05  0.96, 1.16  1.11  1.02, 1.21   >10  1.02  0.94, 1.11  1.00  0.93, 1.09  Age Acceleration Measure  IEAA-Hannum Predictor  IEAA-Horvath Predictor  HR  95% CI  P-het  HR  95% CI  P-het  Sex               Female  0.97  0.88, 1.07  0.06  0.99  0.91, 1.09  0.11   Male  1.08  1.01, 1.15  1.09  1.02, 1.16  Smoking status               Never smoker  1.01  0.93, 1.10  0.65  1.05  0.97, 1.13  0.63   Former smoker            Quit >15 years prior  1.04  0.92, 1.17  1.04  0.93, 1.16    Quit ≤15 years prior  1.12  1.01, 1.26  1.10  0.99, 1.23   Current smoker            <20 cigarettes/day  1.01  0.84, 1.22  0.93  0.77, 1.13    ≥20 cigarettes/day  1.03  0.88, 1.21  1.10  0.94, 1.28  Body mass indexb               <25  1.03  0.93, 1.13  0.50  1.04  0.95, 1.13  0.74   25–29.9  1.08  1.00, 1.17  1.08  1.00, 1.16   ≥30  1.00  0.90, 1.12  1.03  0.93, 1.15  Country of birth               Australia/New-Zealand  1.04  0.97, 1.11  0.77  1.04  0.98, 1.11  0.44   United Kingdom  1.08  0.93, 1.27  1.09  0.93, 1.28   Italy  1.00  0.88, 1.15  1.01  0.89, 1.16   Greece  1.12  0.92, 1.37  1.21  1.00, 1.46  Time between blood draw and start of follow-up, years               <5  1.07  0.96, 1.19  0.75  1.06  0.96, 1.17  0.25   5–10  1.05  0.96, 1.16  1.11  1.02, 1.21   >10  1.02  0.94, 1.11  1.00  0.93, 1.09  Abbreviations: CI, confidence interval; het, heterogeneity; HR, hazard ratio; IEAA, intrinsic epigenetic age acceleration. a Results were adjusted for age at blood draw, sex, country of birth (Australia/New Zealand, United Kingdom/Malta, Italy, or Greece), body mass index (weight (kg)/height (m)2; <25, 25–29.9, 30–34.9, or ≥35), height (cm; continuous), smoking (never smoker; former smoker who quit ≤15 years prior; former smoker who quit >15 years prior; current smoker of ≤20 cigarettes/day; current smoker of >20 cigarettes/day), alcohol intake (1–19 g/day (females); 40–59 g/day (males) or 20–39 g/day (females); or ≥60 g/day (males) or ≥40 g/day (females)), diet quality (Alternate Healthy Eating Index 2010), physical activity (physical activity score ranging from 1 to 4 and reflecting metabolic equivalents (39)), socioeconomic status (score representing the relative socioeconomic disadvantage of the participant’s area of residence, ranging from 1 to 10 (40)), education (score ranging from 1 to 8, with 1 = primary school only and 8 = tertiary university degree or more), sample type, and batch effects. b Weight (kg)/height(m)2. Associations of health risk factors with AA measures and associations of AA measures with mortality were similar for other AA measures (Web Tables 2–4, 6, and 7) and were generally stronger in analyses of AA adjusted for cell composition (Web Tables 6 and 7). DISCUSSION Our study adds to the evidence regarding the use of the epigenetic clock as a marker of biological age. We observed that male sex, Greek nationality (country of birth), smoking, BMI, and diabetes were strongly associated with AA. In addition, AA measures were associated with increased risk of mortality, and these associations, albeit small, were independent of known mortality risk factors, including a large number of demographic, lifestyle, and anthropometric variables, and medical conditions. This suggests that a substantial proportion of AA is due to unidentified factors, which may include early-life exposures, health-associated genetic variants, or other unmeasured environmental factors. The main limitation of our analysis of health risk factors associated with AA was its cross-sectional nature. Future studies should investigate what factors may explain temporal changes in AA. Information on most variables was self-reported, which may have induced some measurement error. Although CpGs included in the age predictors had higher-than-average reliability, measurement error may also have decreased the accuracy of AA measures, particularly for the Horvath predictor. For the mortality analysis, we left-truncated follow-up at the age at which the matching case was diagnosed, in order to account for the procedure used to select controls. This not only created a lag time between exposure assessment and the start of follow-up (implicitly, we assumed that AA remained constant between blood draw and the index date) but also substantially reduced the follow-up time and meant that controls were selected for not developing the cancer of interest in each individual nested case-control study. However, because each cancer was rare, any selection bias would have been minimal. The Melbourne Collaborative Cohort Study has a low prevalence of unhealthy lifestyles (only 13% of participants were current smokers at baseline) and lower mortality rates than those in the general Australian population (50). Thus, it may appear surprising that the estimates from both predictors were close to chronological age (mean chronological age = 59.0 years; mean Horvath-predicted age = 59.1 years; mean Hannum-predicted age = 59.0 years). Study participants were relatively old at the start of follow-up, which may have resulted in weaker observed associations (51). However, when analyses were restricted to younger persons or to those with shorter lag times between blood sampling and the start of follow-up, the resulting hazard ratio estimates were not higher, indicating only a small influence of these limitations. In the seminal study by Marioni et al. (15), the adjusted (though not for BMI) mortality risk estimates varied by cohort, with hazard ratios ranging from 1.08 to 1.40 for AA-Hannum (with a pooled effect of HR = 1.16) and from 0.96 to 1.17 for AA-Horvath (with a pooled effect of HR = 1.09). In the study by Perna et al. (16), a substantial proportion of included deaths (52%) were for participants selected on the basis of their vital status, and a case-cohort design was employed, so their estimates may not be comparable to those of our study. The hazard ratios were 1.10 for AA-Hannum and 1.23 for AA-Horvath. These studies used AA and adjusted for cell composition, so their results are comparable to the estimates presented in Web Table 7. Chen et al. (18) included 2,734 deaths in their study and reported postadjustment hazard ratios similar to those for our model 3 (per 5-year increment in IEAA-Hannum, HR = 1.0145 = 1.07) and higher for the extrinsic measure (per 5-year increment in enhanced-Hannum, HR = 1.0295 = 1.15). While not nominally statistically significant, the estimated associations between AA and mortality were stronger among men in our study. It is unclear why this would be the case. A consistent finding was reported by Perna et al. (16), although the difference by sex was not formally tested. Effect estimates were similar for men and women in the Marioni et al. (15) study. Like our study, the study by Perna et al. (16) lacked the power to detect effect heterogeneity between causes of death. No comparable findings were reported by Marioni et al. (15), and no heterogeneity by sex was observed in the Chen et al. pooled analysis (18). We did not correct our P values for multiple testing, which may have resulted in false-positive findings. Most of the observed associations between AA and health risk factors and mortality were in the expected direction. Additional studies will be required to confirm these associations. In a recent study examining lifestyle factors associated with AA, Quach et al. (52) found that higher BMI was associated with increased AA and that higher education and fish and alcohol consumption were associated with decreased AA. BMI was also associated with IEAA, as was poultry consumption (52). Unlike this and other studies, we found a strong association of AA with smoking, which is consistent with the dramatic effects of smoking on health outcomes, but our findings did not replicate the previously reported association with consumption of poultry and fish. We also observed an association of AA and IEAA with education, which was not found in another large-scale study (53). Many molecular markers of biological aging have been proposed (54), and future research should address how methylation-defined AA measures correlate with these. An analysis of the joint effects of methylation age and telomere length on mortality using 2 cohorts showed that AA and telomere length had independent effects on mortality (55). Those analyses were limited by small sample sizes, and results were minimally adjusted (i.e., simply for age, sex, and estimated white blood cell composition). In an elderly German population, the comprehensive frailty measure that was evaluated appeared to be associated with AA-Horvath more than with telomere length (30). There was no strong correlation of AA-Horvath with telomere length. Estimates of heritability computed by Marioni et al. (15) were around 0.40 for both AA-Hannum and AA-Horvath. In another study, Levine et al. (25) estimated heritability to be 0.65 using genome-wide complex trait analysis. Li et al. (56) used twins to examine the determinants of AA measures and concluded that although genetic factors played a role, the importance of shared environmental factors might have been overlooked, such that these heritability measures were overestimates. These studies, taken together, nevertheless suggest that genetic confounders of the AA-mortality association may exist, but we could not control for them in our analysis. Hannum and Horvath measures of biological age require methylation measurement at a small number of CpG sites. The association of AA with health risk factors and with mortality suggests that it could be used to monitor an individual’s health status, although this requires formal assessment. Given the sources of possible systematic bias and measurement error encountered with the assay we used (Infinium HumanMethylation450 BeadChip array (Illumina Inc., San Diego, California)) (38, 57), one would expect that more accurate AA predictors could be derived using higher-resolution methylation measurement techniques (58). Assays for obtaining AA estimates at lower cost are also being developed (59). Tissues other than blood may offer useful alternative and potentially organ- or disease-specific AA measures (20, 23). An advantage of the Horvath AA measures is that they give an accurate prediction of age across multiple tissue types, allowing some generalizability of our findings to tissues other than blood (11). In conclusion, AA measures are associated with several established health risk factors and with mortality, independently of many of these health risk factors. In future studies, investigators should seek to identify genetic and additional environmental factors associated with AA and improve blood DNA methylation-based measures of biological age. ACKNOWLEDGMENTS Author affiliations: Cancer Epidemiology and Intelligence Division, Cancer Council Victoria, Melbourne, Victoria, Australia (Pierre-Antoine Dugué, Julie K. Bassett, Laura Baglietto, Dallas R. English, Gianluca Severi, Melissa C. Southey, Graham G. Giles, Roger L. Milne); Center for Epidemiology and Biostatistics, Melbourne School of Population and Global Health, University of Melbourne, Parkville, Victoria, Australia (Pierre-Antoine Dugué, Laura Baglietto, Daniel Schmidt, Enes Makalic, Shuai Li, Margarita Moreno-Betancur, Daniel D. Buchanan, Dallas R. English, John L. Hopper, Gianluca Severi, Melissa C. Southey, Graham G. Giles, Roger L. Milne); Genetic Epidemiology Laboratory, Department of Pathology, University of Melbourne, Parkville, Victoria, Australia (JiHoon E. Joo, Ee Ming Wong, Melissa C. Southey); Department of Clinical and Experimental Medicine, School of Medicine, University of Pisa, Pisa, Italy (Laura Baglietto); Melbourne Bioinformatics, University of Melbourne, Victoria, Australia (Chol-Hee Jung); Centre de Recherche en Épidémiologie et Santé des Populations (INSERM U1018), Université Paris-Saclay, Université Paris-Sud, Université Versailles Saint-Quentin-en-Yvelines, Institut Gustave Roussy, Villejuif, France (Gianluca Severi); Italian Institute for Genomic Medicine, Turin, Italy (Giovanni Fiorito, Paolo Vineis, Gianluca Severi); Clinical Epidemiology and Biostatistics Unit, Murdoch Children’s Research Institute, Melbourne, Victoria, Australia (Margarita Moreno-Betancur); Colorectal Oncogenomics Group, Genetic Epidemiology Laboratory, Department of Pathology, University of Melbourne, Parkville, Victoria, Australia (Daniel D. Buchanan); Genetic Medicine and Familial Cancer Center, Royal Melbourne Hospital, Parkville, Victoria, Australia (Daniel D. Buchanan); and MRC-PHE Center for Environment and Health, Imperial College London, London, United Kingdom (Paolo Vineis). This work was supported by the National Health and Medical Research Council (NHMRC) of Australia (grants 1088405 and 1074383) and by the European Commission (H2020 grant 633666; http://www.lifepathproject.eu/). L.B. was supported by a Marie Curie International Incoming Fellowship within the European Commission 7th Framework Programme. Cohort recruitment in the Melbourne Collaborative Cohort Study was funded by VicHealth and Cancer Council Victoria. The Melbourne Collaborative Cohort Study was further supported by NHMRC grants 209057 and 396414 and by infrastructure provided by Cancer Council Victoria. The nested case-control methylation studies were supported by NHMRC grants 1011618, 1026892, 1027505,1050198, and 1043616. Cases and their vital status were ascertained through the Victorian Cancer Registry and the Australian Institute of Health and Welfare, including the National Death Index and the Australian Cancer Database. Conflict of interest: none declared. Abbreviations AA age acceleration BMI body mass index CI confidence interval HR hazard ratio IEAA intrinsic epigenetic age acceleration REFERENCES 1 Jones MJ, Goodman SJ, Kobor MS. DNA methylation and healthy human aging. Aging Cell . 2015; 14( 6): 924– 932. Google Scholar CrossRef Search ADS PubMed  2 Issa JP. Aging and epigenetic drift: a vicious cycle. J Clin Invest . 2014; 124( 1): 24– 29. Google Scholar CrossRef Search ADS PubMed  3 Jung M, Pfeifer GP. Aging and DNA methylation. BMC Biol . 2015; 13: 7. Google Scholar CrossRef Search ADS PubMed  4 Issa JP, Ottaviano YL, Celano P, et al.  . Methylation of the oestrogen receptor CpG island links ageing and neoplasia in human colon. Nat Genet . 1994; 7( 4): 536– 540. Google Scholar CrossRef Search ADS PubMed  5 Calvanese V, Lara E, Kahn A, et al.  . The role of epigenetics in aging and age-related diseases. Ageing Res Rev . 2009; 8( 4): 268– 276. Google Scholar CrossRef Search ADS PubMed  6 Christensen BC, Houseman EA, Marsit CJ, et al.  . Aging and environmental exposures alter tissue-specific DNA methylation dependent upon CpG island context. PLoS Genet . 2009; 5( 8): e1000602. Google Scholar CrossRef Search ADS PubMed  7 Teschendorff AE, West J, Beck S. Age-associated epigenetic drift: implications, and a case of epigenetic thrift? Hum Mol Genet . 2013; 22( R1): R7– R15. Google Scholar CrossRef Search ADS PubMed  8 Mather KA, Jorm AF, Parslow RA, et al.  . Is telomere length a biomarker of aging? A review. J Gerontol A Biol Sci Med Sci . 2011; 66( 2): 202– 213. Google Scholar CrossRef Search ADS PubMed  9 Hannum G, Guinney J, Zhao L, et al.  . Genome-wide methylation profiles reveal quantitative views of human aging rates. Mol Cell . 2013; 49( 2): 359– 367. Google Scholar CrossRef Search ADS PubMed  10 Bocklandt S, Lin W, Sehl ME, et al.  . Epigenetic predictor of age. PLoS One . 2011; 6( 6): e14821. Google Scholar CrossRef Search ADS PubMed  11 Horvath S. DNA methylation age of human tissues and cell types. Genome Biol . 2013; 14( 10): R115. Google Scholar CrossRef Search ADS PubMed  12 Weidner CI, Lin Q, Koch CM, et al.  . Aging of blood can be tracked by DNA methylation changes at just three CpG sites. Genome Biol . 2014; 15( 2): R24. Google Scholar CrossRef Search ADS PubMed  13 Florath I, Butterbach K, Müller H, et al.  . Cross-sectional and longitudinal changes in DNA methylation with age: an epigenome-wide analysis revealing over 60 novel age-associated CpG sites. Hum Mol Genet . 2014; 23( 5): 1186– 1201. Google Scholar CrossRef Search ADS PubMed  14 Bekaert B, Kamalandua A, Zapico SC, et al.  . Improved age determination of blood and teeth samples using a selected set of DNA methylation markers. Epigenetics . 2015; 10( 10): 922– 930. Google Scholar CrossRef Search ADS PubMed  15 Marioni RE, Shah S, McRae AF, et al.  . DNA methylation age of blood predicts all-cause mortality in later life. Genome Biol . 2015; 16: 25. Google Scholar CrossRef Search ADS PubMed  16 Perna L, Zhang Y, Mons U, et al.  . Epigenetic age acceleration predicts cancer, cardiovascular, and all-cause mortality in a German case cohort. Clin Epigenetics . 2016; 8: 64. Google Scholar CrossRef Search ADS PubMed  17 Christiansen L, Lenart A, Tan Q, et al.  . DNA methylation age is associated with mortality in a longitudinal Danish twin study. Aging Cell . 2015; 15( 1): 149– 154. Google Scholar CrossRef Search ADS PubMed  18 Chen BH, Marioni RE, Colicino E, et al.  . DNA methylation-based measures of biological age: meta-analysis predicting time to death. Aging (Albany NY) . 2016; 8( 9): 1844– 1865. Google Scholar CrossRef Search ADS PubMed  19 Levine ME, Hosgood HD, Chen B, et al.  . DNA methylation age of blood predicts future onset of lung cancer in the Women’s Health Initiative. Aging (Albany NY) . 2015; 7( 9): 690– 700. Google Scholar CrossRef Search ADS PubMed  20 Levine ME, Lu AT, Bennett DA, et al.  . Epigenetic age of the pre-frontal cortex is associated with neuritic plaques, amyloid load, and Alzheimer’s disease related cognitive functioning. Aging (Albany NY) . 2015; 7( 12): 1198– 1211. Google Scholar CrossRef Search ADS PubMed  21 Horvath S, Mah V, Lu AT, et al.  . The cerebellum ages slowly according to the epigenetic clock. Aging (Albany NY) . 2015; 7( 5): 294– 306. Google Scholar CrossRef Search ADS PubMed  22 Horvath S, Pirazzini C, Bacalini MG, et al.  . Decreased epigenetic age of PBMCs from Italian semi-supercentenarians and their offspring. Aging (Albany NY) . 2015; 7( 12): 1159– 1170. Google Scholar CrossRef Search ADS PubMed  23 Horvath S, Erhart W, Brosch M, et al.  . Obesity accelerates epigenetic aging of human liver. Proc Natl Acad Sci USA . 2014; 111( 43): 15538– 15543. Google Scholar CrossRef Search ADS PubMed  24 Marioni RE, Shah S, McRae AF, et al.  . The epigenetic clock is correlated with physical and cognitive fitness in the Lothian Birth Cohort 1936. Int J Epidemiol . 2015; 44( 4): 1388– 1396. Google Scholar CrossRef Search ADS PubMed  25 Levine ME, Lu AT, Chen BH, et al.  . Menopause accelerates biological aging. Proc Natl Acad Sci USA . 2016; 113( 33): 9327– 9332. Google Scholar CrossRef Search ADS PubMed  26 Horvath S, Levine AJ. HIV-1 infection accelerates age according to the epigenetic clock. J Infect Dis . 2015; 212( 10): 1563– 1573. Google Scholar CrossRef Search ADS PubMed  27 Gross A, Jaeger PA, Kreisberg JF, et al.  . Methylome-wide analysis of chronic HIV infection reveals five-year increase in biological age and epigenetic targeting of HLA. Mol Cell . 2016; 62( 2): 157– 168. Google Scholar CrossRef Search ADS PubMed  28 Zannas AS, Arloth J, Carrillo-Roa T, et al.  . Lifetime stress accelerates epigenetic aging in an urban, African American cohort: relevance of glucocorticoid signaling. Genome Biol . 2015; 16: 266. Google Scholar CrossRef Search ADS PubMed  29 Boks MP, van Mierlo HC, Rutten BP, et al.  . Longitudinal changes of telomere length and epigenetic age related to traumatic stress and post-traumatic stress disorder. Psychoneuroendocrinology . 2015; 51: 506– 512. Google Scholar CrossRef Search ADS PubMed  30 Breitling LP, Saum KU, Perna L, et al.  . Frailty is associated with the epigenetic clock but not with telomere length in a German cohort. Clin Epigenetics . 2016; 8: 21. Google Scholar CrossRef Search ADS PubMed  31 Milne RL, Fletcher AS, MacInnis RJ, et al.  . Cohort profile: the Melbourne Collaborative Cohort Study (Health 2020) [published online ahead of print June 21, 2017]. Int J Epidemiol . ( doi:10.1093/ije/dyx085). 32 Severi G, Southey MC, English DR, et al.  . Epigenome-wide methylation in DNA from peripheral blood as a marker of risk for breast cancer. Breast Cancer Res Treat . 2014; 148( 3): 665– 673. Google Scholar CrossRef Search ADS PubMed  33 Dugué PA, Brinkman MT, Milne RL, et al.  . Genome-wide measures of DNA methylation in peripheral blood and the risk of urothelial cell carcinoma: a prospective nested case-control study. Br J Cancer . 2016; 115( 6): 664– 673. Google Scholar CrossRef Search ADS PubMed  34 Wong Doo N, Makalic E, Joo JE, et al.  . Global measures of peripheral blood-derived DNA methylation as a risk factor in the development of mature B-cell neoplasms. Epigenomics . 2016; 8( 1): 55– 66. Google Scholar CrossRef Search ADS PubMed  35 Baglietto L, Ponzi E, Haycock P, et al.  . DNA methylation changes measured in pre-diagnostic peripheral blood samples are associated with smoking and lung cancer risk. Int J Cancer . 2017; 140( 1): 50– 61. Google Scholar CrossRef Search ADS PubMed  36 Horvath S, Gurven M, Levine ME, et al.  . An epigenetic clock analysis of race/ethnicity, sex, and coronary heart disease. Genome Biol . 2016; 17( 1): 171. Google Scholar CrossRef Search ADS PubMed  37 Dugué PA, English DR, MacInnis RJ, et al.  . The repeatability of DNA methylation measures may also affect the power of epigenome-wide association studies. Int J Epidemiol . 2015; 44( 4): 1460– 1461. Google Scholar CrossRef Search ADS PubMed  38 Dugué PA, English DR, MacInnis RJ, et al.  . Reliability of DNA methylation measures from dried blood spots and mononuclear cells using the HumanMethylation450k BeadArray. Sci Rep . 2016; 6: 30317. Google Scholar CrossRef Search ADS PubMed  39 Rubin DB. Multiple Imputation for Nonresponse in Surveys . Hoboken, NJ: John Wiley & Sons, Inc.; 2004. 40 Buuren S, Groothuis-Oudshoorn K. mice: Multivariate imputation by chained equations in R. J Stat Softw . 2011; 45( 3): 1– 67. Google Scholar CrossRef Search ADS   41 Chiuve SE, Fung TT, Rimm EB, et al.  . Alternative dietary indices both strongly predict risk of chronic disease. J Nutr . 2012; 142( 6): 1009– 1018. Google Scholar CrossRef Search ADS PubMed  42 MacInnis RJ, English DR, Hopper JL, et al.  . Body size and composition and colon cancer risk in men. Cancer Epidemiol Biomarkers Prev . 2004; 13( 4): 553– 559. Google Scholar PubMed  43 Pink B; Australian Bureau of Statistics. Socio-Economic Indexes for Areas (SEIFA), 2011. Canberra, Australian Capital Territory, Australia: Australian Bureau of Statistics; 2013. (ABS publication no. 2033.0.55.001). 44 Thiébaut AC, Bénichou J. Choice of time-scale in Cox’s model analysis of epidemiologic cohort data: a simulation study. Stat Med . 2004; 23( 24): 3803– 3820. Google Scholar CrossRef Search ADS PubMed  45 Grambsch PM, Therneau TM. Proportional hazards tests and diagnostics based on weighted residuals. Biometrika . 1994; 81( 3): 515– 526. Google Scholar CrossRef Search ADS   46 Andersen PK, Geskus RB, de Witte T, et al.  . Competing risks in epidemiology: possibilities and pitfalls. Int J Epidemiol . 2012; 41( 3): 861– 870. Google Scholar CrossRef Search ADS PubMed  47 Therneau T. Mixed Effects Cox Models. Vienna, Austria: R Foundation for Statistical Computing; 2015. https://cran.r-project.org/web/packages/coxme/vignettes/coxme.pdf. Updated June 15, 2015. Accessed June 26, 2017. 48 Lunn M, McNeil D. Applying Cox regression to competing risks. Biometrics . 1995; 51( 2): 524– 532. Google Scholar CrossRef Search ADS PubMed  49 Jaffe AE, Irizarry RA. Accounting for cellular heterogeneity is critical in epigenome-wide association studies. Genome Biol . 2014; 15( 2): R31. Google Scholar CrossRef Search ADS PubMed  50 Giles GG, English DR. The Melbourne Collaborative Cohort Study. IARC Sci Publ . 2002; 156: 69– 70. Google Scholar PubMed  51 Boonekamp JJ, Simons MJ, Hemerik L, et al.  . Telomere length behaves as biomarker of somatic redundancy rather than biological age. Aging Cell . 2013; 12( 2): 330– 332. Google Scholar CrossRef Search ADS PubMed  52 Quach A, Levine ME, Tanaka T, et al.  . Epigenetic clock analysis of diet, exercise, education, and lifestyle factors. Aging (Albany NY) . 2017; 9( 2): 419– 446. Google Scholar PubMed  53 Linnér RK, Marioni RE, Rietveld CA, et al.  . An epigenome-wide association study meta-analysis of educational attainment. Mol Psychiatry . 2017; 22( 12): 1680– 1690. Google Scholar CrossRef Search ADS PubMed  54 López-Otín C, Blasco MA, Partridge L, et al.  . The hallmarks of aging. Cell . 2013; 153( 6): 1194– 1217. Google Scholar CrossRef Search ADS PubMed  55 Marioni RE, Harris SE, Shah S, et al.  . The epigenetic clock and telomere length are independently associated with chronological age and mortality. Int J Epidemiol . 2016; 45( 2): 424– 432. Google Scholar CrossRef Search ADS   56 Li S, Wong EM, Joo JE, et al.  . Genetic and environmental causes of variation in the difference between biological age based on DNA methylation and chronological age for middle-aged women. Twin Res Hum Genet . 2015; 18( 6): 720– 726. Google Scholar CrossRef Search ADS PubMed  57 Naeem H, Wong NC, Chatterton Z, et al.  . Reducing the risk of false discovery enabling identification of biologically significant genome-wide methylation status using the HumanMethylation450 array. BMC Genomics . 2014; 15: 51. Google Scholar CrossRef Search ADS PubMed  58 Walker DL, Bhagwate AV, Baheti S, et al.  . DNA methylation profiling: comparison of genome-wide sequencing methods and the Infinium Human Methylation 450 Bead Chip. Epigenomics . 2015; 7( 8): 1287– 1302. Google Scholar CrossRef Search ADS PubMed  59 Vidal-Bralo L, Lopez-Golan Y, Gonzalez A. Simplified assay for epigenetic age estimation in whole blood of adults. Front Genet . 2016; 7: 126. Google Scholar CrossRef Search ADS PubMed  © The Author(s) 2018. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.

Journal

American Journal of EpidemiologyOxford University Press

Published: Mar 1, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off