Differential contributions of cardiovascular health-related lifestyle factors to epigenetic ageing: implications for healthy longevity

Abstract

Background

Lifestyle and cardiovascular health (CVH) are key determinants of the biological ageing process, which is closely linked to healthy ageing and longevity. Multiple CVH factors may interact and exert combined effects on the biological ageing process; however, studies exploring their collective association, particularly in Asian cohorts, remain limited. In this cross-sectional study, we investigated the relative contributions of CVH-related behaviours (diet, sleep, avoidance of nicotine, and physical activity) and clinical indicators (BMI, blood lipids, blood glucose, and blood pressure) to epigenetic age acceleration (EAA), as well as their collective associations.

Methods

We included 1940 participants from two large-scale prospective cohorts within the Korean Genome and Epidemiology Study. We assessed CVH based on the American Heart Association’s Life’s Essential 8 and measured five EAA (intrinsic EAA (Horvath DNAmAge acceleration), extrinsic EAA (Hannum DNAmAge acceleration), PhenoAge acceleration, GrimAge2 acceleration, and Dunedin Pace of Ageing Calculated from the Epigenome (DunedinPACE)). We employed quantile-based g-computation to delineate the sex-specific relative contributions of CVH to EAA.

Results

Better CVH was associated with lower EAA (ψ, the estimates of collective associations, ranged from − 4.29 to − 0.79, depending on the EAA measure). The CVH components contributing most to lower EAA varied by sex. In males, nicotine avoidance and better glucose contributed most to lower EAA, accounting for 91% and 77% of the overall associations of CVH with lower Grim2AA and DunedinPACE, respectively, whereas better blood glucose accounted for 86% and 94%. In females, physical activity and better glucose or BMI were the greatest contributors to lower EAA. Physical activity accounted for 44% of the CVH–Grim2AA association. Better glucose explained 54% and 50% of the association with Grim2AA and DunedinPACE, respectively. Better BMI contributed 46% to lower PhenoAA.

Conclusions

Simultaneous improvements in multiple CVH components were associated with lower EAA, with sex-specific differences in the most influential factors. Tailored health strategies—emphasizing smoking cessation and glucose control for males, and physical activity, glucose control, and weight management for females—may help slow ageing. These findings highlight the need for public and clinical interventions that incorporate sex differences in health behaviours and the potential biological mechanisms of ageing.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12916-025-04394-3.

Background

The rapid ageing of the global population [1] underscores the need for a greater focus on slower and healthier ageing, not merely on extending lifespan. Biological ageing has received increasing attention as it may provide a more accurate reflection of age-related physiological decline than chronological age alone. DNA methylation-based epigenetic age, derived from DNA methylation levels in leukocytes or multiple tissues, has been associated with the biological ageing process reflected by age-related diseases and life expectancy [2]. Lifetime exposures, such as health-related behaviours or lifestyle, can influence gene expression through epigenetic alterations, including DNA methylation [3]. Taken together, identifying lifestyle-related health behaviours and physical indicators that may accelerate epigenetic ageing can contribute to promoting healthy longevity.

Cardiovascular health (CVH) is a particularly crucial component of healthy ageing since it is a major determinant of mortality worldwide [4]. The American Heart Association (AHA) has suggested Life’s Essential 8, a comprehensive measure of four health behaviours and four clinical indicators designed to assess and monitor CVH [5]. Abundant evidence has shown that an ideal CVH score is associated with a lower risk for a wide range of diseases [6], including cardiovascular disease and all-cause mortality [7, 8].

Given that CVH-related behaviours complexly interplay and simultaneously influence both mortality and multi-morbidity [9], it is important to assess their joint association as well as individual association. In addition, sex is a key factor reflecting distinct patterns in lifestyle and self-management practices. For example, poor diet and smoking are more common in males, while females tend to be more sedentary [10–12]. Males are prone to undergo faster epigenetic ageing than females [13], and health behaviours may be one of the factors contributing to the sex difference in lifespan [14]. Accordingly, it is needed to elucidate the association of multiple health behaviours and clinical indicators with epigenetic age acceleration (EAA) by sex. To date, however, evidence on sex-specific CVH profiles and their differential impact on biological ageing remains scarce.

This study aimed to investigate the collective associations of health behaviours and clinical indicators with EAA and to examine the relative contribution of each component using two nationwide cohorts. We hypothesized that better health behaviour and indicators are related to lower epigenetic age and that the associations may differ by sex. To disentangle the contributions of individual health-related components in the joint association, we adapted quantile-based g-computation (QGC) [15], a method widely used to examine environmental contaminants and extendable but not yet extensively applied to multiple social factors as a mixture [16].

Methods

Study population

This study draws on data from the Korean Genome and Epidemiology Study (KoGES), conducted by the Korea Center for Disease Control and Prevention in Korea. We used two large-scale longitudinal cohorts in KoGES: the Health Examinee (HEXA) study and the Ansan and Ansung study. Adults aged 40 years or more at baseline were recruited from the national health examinee registry, ensuring that each participant provided written informed consent. The baseline survey in HEXA was performed between 2004 and 2013, and a follow-up exam was conducted between 2012 and 2016 (a total of 2 waves). The Ansan and Ansung study began collecting data from 10,030 participants between 2001 and 2002, with 9 biennial follow-up surveys conducted, resulting in a total of 10 waves by 2020. More detailed information on Ansan and Ansung and HEXA is included in Additional file 1: sMethod 1 [17, 18]. In this cross-sectional study, the participants were drawn from the first wave of HEXA, followed up from 2004 to 2013 (N = 173,195), and the fifth wave of Ansan and Ansung, followed up from 2009 to 2010 (N = 6665). Of these, 2353 participants with available DNA methylation data (837 from HEXA and 1516 from Ansan and Ansung) were included. We excluded 372 samples with missing information on CVH components and potential confounder variables (58 from HEXA and 314 from Ansan and Ansung). We also excluded 41 participants who had major diseases (cancer and cardiovascular diseases), which may involve different physiological conditions and heterogeneous effects on epigenetic ageing, leaving 1940 participants in the final analytic dataset.

Cardiovascular health components

We assessed and defined cardiovascular health-related components using Life's Essential 8 from the AHA [5], which consists of four health behaviours (healthy diet, healthy sleep, avoidance of nicotine, and participation in physical activity), and four healthy clinical indicators (healthy weight (body mass index, BMI (kg/m²)), healthy levels of blood lipids (mg/dL), blood glucose (mg/dL or %), and blood pressure (mmHg)). Participants’ dietary pattern was assessed using the Healthy Lifestyle Score (HLS) [19, 20] based on the average consumption frequency and portion size for each food type. Due to the limited availability of diet data in the follow-up exams in Ansan and Ansung, we utilized data from the first wave for calculating the diet score of Ansan and Ansung participants (see Additional file 1: sMethod 2 for details).

Each metric was assigned a score ranging from 0 to 100 based on the AHA’s guideline (Additional file 1: Table S1) [5]. The overall CVH score was calculated as the unweighted average of the scores for the CVH components, designed to range from 0 to 100. We categorized the levels of overall CVH score into three groups using the cutoff values recommended by the AHA [5]: ideal (scores of 80 to 100), intermediate (50–79), and poor (0–49). In this study, CVH was examined in five dimensions: (i) the overall CVH score (continuous); (ii) the level of CVH score (ideal, intermediate, and poor); (iii) individual CVH components; (iv) a mixture of CVH components; (v) the relative contribution of CVH components in a mixture.

DNA methylation profiling and EAA calculation

DNA samples were collected from 2004 to 2013 in the HEXA study and from 2009 to 2010 in the Ansan and Ansung study. We measured DNA methylation and epigenetic age acceleration from 2353 blood samples using Illumina Infinium Human Methylation EPIC BeadChip (Illumina, San Diego, CA, USA), which covers more than 850 k CpG probes. We used the R packages minfi [21] to load the data and ENmix [22] to conduct quality control (QC). We filtered CpGs with low detection p-value (< 10⁻⁶), samples with > 5% low-quality CpGs, or sodium bisulfite conversion probes of extremely low intensity (< 3 SDs (standard deviation) from the mean), and outliers defined by Tukey’s method [23]. After QC, the remaining samples underwent a background correction, dye-bias adjustment, and normalization [24] to account for technical variations such as batch effects (e.g., chip or plate) and to ensure consistency in methylation measurements using the preprocessENmix function in the R ENmix package [21].

We included five EAA measures: IEAA (intrinsic epigenetic age acceleration or Horvath DNA methylation age (DNAmAge) acceleration) [25], EEAA (extrinsic epigenetic age acceleration or Hannum DNAmAge acceleration) [26], PhenoAA (PhenoAge acceleration) [27], Grim2AA (GrimAge2 acceleration) [28], and DunedinPACE (Pace of Ageing Calculated from the Epigenome) [29]. Horvath DNAmAge and Hannum DNAmAge were trained on chronological age, adjusted for or weighted by blood cell type proportions [25, 26]. PhenoAge and GrimAge2 were trained on a composite of biomarkers that contributed to time-to-death or disease-promoting factors [27, 28]. We estimated Horvath DNAmAge, Hannum DNAmAge, PhenoAge, and GrimAge2 using the online calculator (https://dnamage.clockfoundation.org/) [30]. Acceleration or deceleration of epigenetic age was calculated from the residual of a linear model of each epigenetic age on chronological age. Therefore, positive values of IEAA, EEAA, PhenoAA, and Grim2AA indicate accelerated epigenetic ageing, while negative values indicate decelerated ageing. Lastly, DunedinPACE represents the pace of biological ageing, benchmarked the average personal rate of longitudinal Change across 19 biomarkers calculated in the Dunedin Study [29]. DunedinPACE was computed using the R package DunedinPACE (https://github.com/danbelsky/DunedinPACE) [29]. Values above 1 on the DunedinPACE indicate a faster ageing rate, whereas values below 1 represent a slower pace. We expressed individual DunedinPACE values on a percentage scale, where a one-unit increase corresponds to a one percentage-point faster rate of ageing.

Covariates

Potential confounders included self-reported sex (male or female), chronological age (continuous), marital status (married or not), education (high school graduate or below), monthly household income (above or below median, three million Korean won), employment (manual, non-manual, or unemployed), current drinking (yes or no), menopause status (pre- or post-menopause), and number of comorbidities (based on self-reported physician diagnoses; 0, 1, or 2 or more). We constructed a comorbidity count variable that included several non-cardiometabolic diseases, such as peptic ulcer disease, cholecystitis, chronic respiratory disease, thyroid disorders, arthritis, osteoporosis, and chronic kidney disease.

Statistical analysis

Descriptive analysis

We employed descriptive analyses to compare the characteristics, CVH score, and CVH level of the participants by sex, using chi-square tests for categorical variables and Student’s t-tests for continuous variables. We also performed Spearman’s correlation tests to assess correlations within CVH components, within EAA measures (except for the IEAA-EEAA pair, which was assessed using Pearson’s test based on the normality test result), and between epigenetic age indicators and chronological age.

Linear regression models

To assess the associations of the overall CVH scores (continuous and categorical) with EAA, we employed multiple linear regression, adjusting for demographics (sex, chronological age, marital status, education level, household income, and employment) and health-related variables (current drinking and number of comorbidities). The robustness of the association was examined via sensitivity analyses controlling for demographic variables only. Next, we evaluated the association between individual CVH components and EAA, with a regression model additionally adjusted for the remaining CVH components. The slope coefficients from linear regression models represent mean changes of CVH score per increase of one SD. We also examined sex-specific associations using stratified analyses, and statistical interactions were assessed by adding a product term between overall or individual CVH and sex. Lastly, we performed further sensitivity analyses, additionally adjusting cell type composition in each regression model.

Quantile-based g-computation (QGC)

The R package qgcomp was employed to conduct the QGC analyses. We performed QGC for each mixture group (health behaviours and health indicators) by further adjusting for CVH components of each mixture in addition to demographics and health-related variables. Each continuous exposure $X_j$ was transformed into a categorical variable $X_j^q$ based on the quantile ranges of the exposure values (sextiles in this study), and then fitted into a linear model. In QGC, a weighted average of all exposures is defined as $S_i$, a mixture index, and given as ${\sum }_{j=1}^d{w}_j{X}_{\mathit{ji}}^q$ (where the weights ($w_j$) are the mean weight across bootstrap samples, representing a relative contribution of each exposure). $\psi$ represents the coefficient of mixture association, interpreted as the change in EAA per one quantile change of $S_i$ (four exposures) at the same time. The equation is as follows:

$$Y_i={\beta }_0+\psi S_i+Z+{\epsilon }_i$$

$$={\beta }_0+\psi \sum _{j=1}^4{w}_j{X}_{\mathit{ji}}^q+Z+{\epsilon }_i$$

where $Z$ is the confounder and ${\epsilon }_i$ is the error term. $\psi$ is given as ${\sum }_{j=1}^d{\beta }_j$ (where ${\beta }_j$ is the effect size of the j th exposure for the i th individual, $X_{\mathit{ji}}^q$), and the weight for the kth component is given as $w_k={\beta }_k/{\sum }_{j=1}^4{\beta }_j$, when the directions for all weights are homogeneous. When there is no directional homogeneity, the weight of each component represents a proportion of the total negative or positive beta estimate. In each direction, the sum of the weights is −1 (for negative) or 1 (for positive).

$$Y_i={\beta }_0+\sum _{j=1}^4{\beta }_j{X}_{\mathit{ji}}^q+Z+{\epsilon }_i$$

When the magnitude of contributions for the negative direction (i.e., negative association) is greater than that for the positive direction (i.e., positive association), it results in an overall negative association between the collective components and EAA (i.e., better behaviours and/or indicators are associated with lower EAA). Similar to the traditional multivariable regression, we performed QGC for all participants and additionally conducted sex-stratified analyses and interaction testing (CVH*sex).

We used R Statistical Software version 4.3.2 (R Core Team 2023) for all analyses. Two-sided P-values less than 0.05 were considered statistically significant.

Results

Participants’ characteristics

Table 1 describes the characteristics of the 1940 study participants (mean chronological age (SD): 57.51 (8.79); 1125 males and 815 females). While PhenoAA was comparable between males and females (P = 0.219), female participants tended to have lower Grim2AA and DunedinPACE (mean (SD): − 1.03 (3.05) for Grim2AA; − 5.36 (10.13) for DunedinPACE) than male participants (mean (SD): 0.56 (3.64) for Grim2AA; 0.18 (10.56) for DunedinPACE). Additional file 1: Table S2 shows the distribution of CVH scores and CVH levels. The average score was 63.9 (SD: 14.2) among total participants, with 60.4 (SD: 14.1) for males and 68.8 (SD: 13.0) for females (P < 0.001). All clinical indicators were positively correlated ($\rho$ ranged from 0.08 to 0.23, P < 0.001), and among health behaviours, nicotine avoidance was positively correlated with physical activity ($\rho$ = 0.08, P < 0.001) and healthy diet ($\rho$ = 0.21, P < 0.001) (Additional file 1: Fig. S1). Among EAA measures, Grim2AA and DunedinPACE showed relatively strong correlations ($\rho$ = 0.66, P < 0.001) (Additional file 1: Fig. S2). Epigenetic age measures were moderately or strongly correlated with each others, ranging from 0.35 to 0.84 (P < 0.001) (Additional file 1: Fig. S3).

Table 1.

Characteristics of study participants

Variables			Total (N = 1940)	Males (n = 1125)	Females (n = 815)	P-value †
Chronological age, year, mean (SD)			57.51 (8.79)	56.55 (8.86)	58.82 (8.53)	< .001
PhenoAA, year, mean (SD)			− 0.03 (5.20)	− 0.16 (5.25)	0.14 (5.13)	0.219
Grim2AA, year, mean (SD)			− 0.11 (3.50)	0.56 (3.64)	− 1.03 (3.05)	< .001
DunedinPACE, %, mean (SD)			− 2.14 (10.73)	0.18 (10.56)	− 5.36 (10.13)	< .001
Marital status, n (%)	Married		1704 (87.84)	1048 (93.16)	656 (80.49)	< .001
	Not married		236 (12.16)	77 (6.84)	159 (19.51)
Education level, n (%)	≤ high school		1487 (76.65)	751 (66.76)	736 (90.31)	< .001
	> high school		453 (23.35)	374 (33.24)	79 (9.69)
Monthly household income, n (%)	≤ median		1217 (62.73)	628 (55.82)	589 (72.27)	< .001
	> median		723 (37.27)	497 (44.18)	226 (27.73)
Employment, n (%)	Manual		793 (40.88)	540 (48.00)	253 (31.04)	< .001
	Non-manual		628 (32.37)	448 (39.82)	180 (22.09)
	Unemployed		519 (26.75)	137 (12.18)	382 (46.87)
Current drinking, n (%)	Yes		532 (27.42)	396 (35.20)	136 (16.69)	< .001
	No		1408 (72.58)	729 (64.80)	679 (83.31)
No. of comorbidities, n (%)	0		1522 (78.45)	961 (85.42)	561 (68.83)	< .001
	1		367 (18.92)	140 (12.44)	227 (27.85)
	≥ 2		51 (2.63)	24 (2.13)	27 (3.31)
HLS diet score, mean (SD)			3.45 (1.08)	3.25 (1.03)	3.73 (1.09)	< .001
Average hours of sleep, mean (SD)			6.35 (1.28)	6.41 (1.23)	6.24 (1.34)	0.029
Smoking, n (%)	Current		457 (23.56)	438 (38.93)	19 (2.33)	< .001
	Former		470 (24.23)	460 (40.89)	10 (1.23)
	Never		1013 (52.22)	227 (20.18)	786 (96.44)
Physical activity, min/week, mean (SD)			117.39 (212.80)	124.81 (232.91)	107.14 (181.05)	0.060
BMI (kg/m2), mean (SD)			24.43 (3.30)	24.49 (3.50)	24.34 (3.01)	0.330
Non-HDL cholesterol (mg/dl), mean (SD)			148.2 (34.35)	146.26 (34.25)	150.88 (34.33)	0.003
Fasting blood glucose (mg/dL), mean (SD)			107.56 (35.29)	108.25 (34.44)	106.60 (36.43)	0.314
HbA1c (%), mean (SD)			6.08 (1.29)	6.04 (1.21)	6.12 (1.37)	0.196
Systolic blood pressure (mmHg), mean (SD)			120.35 (15.76)	121.20 (14.72)	119.19 (17.03)	0.006
Diastolic blood pressure (mmHg), mean (SD)			76.54 (9.16)	77.85 (8.94)	74.74 (9.16)	< .001
Menopause status, n (%)		Pre-menopause	-	-	164 (20.12)	-
		Post-menopause	-	-	651 (79.88)
Diabetes, n (%)			59 (3.04)	39 (3.47)	20 (2.45)	0.251
Hypertension, n (%)			159 (8.2)	114 (10.13)	45 (5.52)	< .001

^† Global P values were obtained from $x^2$ tests for categorical variables and Student’s t tests for continuous variables to compare between the sexes; DunedinPACE values were expressed on a percentage scale

SD standard deviation; HLS Healthy Lifestyle Score

Associations of CVH components and overall CVH with EAA from regression analyses

Table 2 shows the associations between overall CVH and EAA from the regression analysis, including results for total participants and stratified by sex. In total participants, a one SD increase in CVH score was associated with lower EAA (β = − 0.51, 95% CI (− 0.75, − 0.27) for PhenoAA; β = − 0.94, 95% CI (− 1.09, − 0.78) for Grim2AA; β = − 2.62, 95% CI (− 3.06, − 2.18) for DunedinPACE). Participants in the ideal CVH group were more likely to have lower EAA (β = − 0.73, 95% CI (− 1.42, − 0.04) for PhenoAA; β = − 1.13, 95% CI (− 1.57, − 0.69) for Grim2AA; β = −3.56, 95% CI (− 4.80, − 2.31) for DunedinPACE), while those in the poor CVH group tended to show higher EAA (β = 0.98, 95% CI (0.32, 1.65) for PhenoAA; β = 1.80, 95% CI (1.37, 2.22) for Grim2AA; β = 5.81, 95% CI (4.61, 7.01) for DunedinPACE). These associations were consistent in males and females. Statistically significant interactions between CVH score and sex were observed in Grim2AA and DunedinPACE, and between CVH level and sex in Grim2AA (all P for interaction < 0.05). We also identified that the associations remained robust in the sensitivity analysis, including the reduced model (adjusted for demographic factors) and the extended model (additionally adjusted for cell type composition) (Additional file 1: Table S3 and Table S4), and the associations were prone to being attenuated with IEAA and EEAA (Additional file 1: Table S5).

Table 2.

Associations between overall CVH and EAA from regression analysis

		EAA	Total (N = 1940) β (95% CI)	Males (n = 1125) β (95% CI)	Females (n = 815) β (95% CI)	Interaction P-value
CVH score †		PhenoAA	− 0.51 (− 0.75, − 0.27)**	− 0.48 (− 0.79, − 0.17)*	− 0.62 (− 0.99, − 0.25)**	0.668
		Grim2AA	− 0.94 (− 1.09, − 0.78)**	− 1.14 (− 1.34, − 0.93)**	− 0.67 (− 0.89, − 0.46)**	0.002*
		DunedinPACE	− 2.62 (− 3.06, − 2.18)**	− 3.11 (− 3.67, − 2.55)**	− 1.79 (− 2.46, − 1.11)**	0.005*
CVH level ‡	Ideal	PhenoAA	− 0.73 (− 1.42, − 0.04)*	− 0.46 (− 1.55, 0.64)	− 1.14 (− 2.04, − 0.24)*	0.738
	Poor		0.98 (0.32, 1.65)*	1.00 (0.23, 1.77)*	1.24 (− 0.17, 2.65)
	Ideal	Grim2AA	− 1.13 (− 1.57, − 0.69)**	− 1.72 (− 2.45, − 0.99)**	− 0.95 (− 1.48, − 0.42)**	0.045*
	Poor		1.80 (1.37, 2.22)**	1.94 (1.42, 2.45)**	1.54 (0.71, 2.36)**
	Ideal	DunedinPACE	− 3.56 (− 4.80, − 2.31)**	− 4.03 (− 5.99, − 2.07)**	− 3.52 (− 5.15, − 1.88)**	0.291
	Poor		5.81 (4.61, 7.01)**	6.34 (4.96, 7.72)**	4.63 (2.07, 7.18)**

Adjusted for chronological age, marital status, education level, household income, employment, current drinking, and no. of comorbidities (sex was additionally adjusted in total participants; menopause status was additionally adjusted in females)

^† β represents the Change in EAA for each 1-SD (standard deviation) increase in CVH score; participants (SD): total (14.2); males (14.1); females (13.0)

^‡ Reference = Intermediate

Interaction P-values derived from interaction analysis between CVH and sex

Stratum-specific P-values: * P < .05; ** P < .001

The associations of individual CVH components with EAA are described in Fig. 1. Among health behaviours, nicotine avoidance and physical activity were associated with lower EAA (nicotine avoidance: β = − 0.50, 95% CI (− 0.90, − 0.11) for PhenoAA; β = − 1.60, 95% CI (− 1.85, − 1.36) for Grim2AA; β = − 3.30, 95% CI (− 4.02, − 2.58) for DunedinPACE; physical activity: β = − 0.79, 95% CI (− 1.36, − 0.23) for DunedinPACE). Among clinical indicators, better blood glucose was associated with lower epigenetic age consistently across the EAA measures (β = − 0.56, 95% CI (− 0.81, − 0.32) for PhenoAA; β = − 0.85, 95% CI (− 1.00, − 0.69) for Grim2AA; β = − 2.48, 95% CI (− 2.92, − 2.04) for DunedinPACE), while better BMI and blood pressure showed associations with specific EAA measures (BMI for DunedinPACE; blood pressure for PhenoAA, respectively). A healthy diet, healthy sleep, and better blood lipids were not associated with EAA. We also identified that higher scores of nicotine avoidance, better BMI, blood glucose, and blood pressure were associated with lower EAA in both sexes (Additional file 1: Table S6). Physical activity was associated with lower EAA only among females (β = − 0.40, 95% CI (− 0.69, − 0.12) for Grim2AA; β = − 1.06, 95% CI (− 1.99, − 0.12) for DunedinPACE). The associations of individual CVH components with EAA remained robust in the sensitivity analysis with the model additionally adjusted for cell type composition (Additional file 1: Table S7).

Fig. 1.

Associations between individual CVH components and EAA from regression analysis. Adjusted for sex, chronological age, marital status, education level, household income, employment, current drinking, no. of comorbidities, and four components of health behaviours or clinical indicators. β represents the Change in EAA for each 1-SD (standard deviation) increase in CVH score; components (SD): healthy diet (36.9); healthy sleep (27.6); nicotine avoidance (39.5); physical activity (45.8); better BMI (21.7); better blood lipids (28.9); better blood glucose (32.4); better blood pressure (30.7). * P< .05; ** P< .001

Collective association of CVH components with EAA and their relative contribution to EAA

Collective health behaviours and clinical indicators were associated with lower epigenetic age. Among the individual components, nicotine avoidance, physical activity, healthy diet, and better blood glucose were likely to contribute to lower EAA, consistently across all EAA measures (Fig. 2). A one-quantile increase in health behaviour scores was associated with an average of 1.75 years or 4.29% lower epigenetic age ($\psi$= − 1.75, 95% CI (− 2.34, − 1.16) for Grim2AA; $\psi$= − 4.29, 95% CI (− 5.98, − 2.61) for DunedinPACE). Among individual health behaviours, nicotine avoidance displayed the greatest contribution to the overall association between collective health behaviours and lower epigenetic age (weights: − 0.68 for Grim2AA and − 0.56 for DunedinPACE, respectively). A one-quantile increase of clinical indicator scores was associated with an average of 0.79, 0.82 years or 4.14% lower EAA ($\psi$= − 0.79, 95% CI (− 1.36, − 0.23) for PhenoAA; $\psi$= − 0.82, 95% CI (− 1.18, − 0.46) for Grim2AA; $\psi$= − 4.14, 95% CI (− 5.16, − 3.12) for DunedinPACE). Better blood glucose accounted for 67% to 100% negative contribution to EAA (weight: − 0.67 for DunedinPACE; − 0.73 for PhenoAA; − 1.00 for Grim2AA, respectively), representing the greatest extent. More detailed QGC results displaying all components and directions are presented in Additional file 1: Fig. S4.

Fig. 2.

Collective association and relative contributions of CVH components to EAA in total participants. $\psi$ represents the mixture association. (Models for health behaviours): Adjusted for sex, chronological age, marital status, education level, household income, employment, current drinking, no. of comorbidities, and four health behaviours. (Models for clinical indicators): Adjusted for sex, chronological age, marital status, education level, household income, employment, current drinking, no. of comorbidities, and four clinical indicators

Collective health behaviours and clinical indicators were associated with lower epigenetic age across all EAA measures in both sexes (Fig. 3). Among males, a one-quantile increase in health behaviour scores was associated with an average of 2.06 years lower Grim2AA ($\psi$= − 2.06, 95% CI (− 2.79, − 1.32)), and 4.90% slower DunedinPACE ($\psi$= − 4.90, 95% CI (− 6.97, − 2.84)), respectively. Better clinical indicator scores were associated with an average of 0.64 years lower Grim2AA ($\psi$= − 0.64, 95% CI (− 1.15, − 0.14)), and 3.58% slower DunedinPACE ($\psi$= − 3.58, 95% CI (− 4.95, − 2.21)). Among females, better status of health behaviours and clinical indicators was associated with lower epigenetic age (health behaviour: $\psi$ = − 1.14, 95% CI (− 2.08, − 0.19) for Grim2AA; clinical indicator: $\psi$ = − 1.87, 95% CI (− 2.79, − 0.96) for PhenoAA; $\psi$ = − 1.36, 95% CI (− 1.90, − 0.83) for Grim2AA; $\psi$ = − 5.46, 95 CI (− 7.08, − 3.84) for DunedinPACE). In both males and females, the CVH components contributing to the lower epigenetic age were generally consistent; however, the greatest component and its relative contribution varied in magnitude. In males, nicotine avoidance had the greatest negative contribution to EAA (weight ranged from − 0.77 to − 0.91), followed by physical activity (from − 0.04 to − 0.16), and healthy diet (from − 0.03 to − 0.07). Among clinical indicators, better blood glucose accounted for the largest negative contribution, ranging from 86 to 94%. In females, physical activity had 44% greatest contribution to lower EAA (weight: − 0.44 for Grim2AA), followed by nicotine avoidance (weight: − 0.32) and healthy diet (weight: − 0.24). Among clinical indicators, better blood glucose contributed most to the association with lower Grim2AA and DunedinPACE (weights: − 0.54 for Grim2AA; − 0.50 for DunedinPACE), and better BMI contributed most to lower PhenoAA with 46% contribution (weights: − 0.46). Sex modified the associations between nicotine avoidance and Grim2AA, between better BMI and all EAA, and between better blood glucose and Grim2AA (all P < 0.05).

Fig. 3.

Collective association and relative contributions of CVH components to EAA by sex. Blue: negative contribution of health components to association; Grey: positive contribution of health components to association. $\psi$ represents the mixture association. ^a Adjusted for chronological age, marital status, education level, household income, employment, current drinking, no. of comorbidities, and four health behaviours (menopause status was additionally adjusted in females). ^b Adjusted for chronological age, marital status, education level, household income, employment, current drinking, no. of comorbidities, and four clinical indicators (menopause status was additionally adjusted in females). Interactions between CVH components and sex on EAA were statistically significant in nicotine avoidance*sex on Grim2AA, better BMI*sex on all EAA, and better blood glucose*sex on Grim2AA (all P < .05) (values are not shown in the figure)

Discussion

This study examined the relative contributions of CVH components, incorporating behavioural and clinical factors, to biological ageing, as well as their joint associations. CVH components were collectively associated with lower EAA, and nicotine avoidance and better blood glucose made the greatest contributions to lower EAA within each mixture of health behaviours and clinical indicators in total participants, respectively. We also observed that the relative contribution of each component to EAA differed by sex. In males, nicotine avoidance and better blood glucose were the most significant contributors to lower epigenetic ageing measured by Grim2AA and DunedinPACE. In contrast, among females, physical activity, better BMI, and blood glucose played a greater role in lower EAA—specifically, physical activity for Grim2AA, better BMI for PhenoAA, and better blood glucose for both Grim2AA and DunedinPACE. Sex differences in the key components influencing biological ageing remain understudied; we anticipate our findings may aid in exploring the mechanisms underlying sex differences in epigenetic ageing [31], proposing that health behaviours could be a contributing factor to lifespan discrepancies by sex. Also, previous studies have suggested that practicing two or more positive health behaviours may more efficiently promote healthy ageing and greater overall health benefits [32]. Our findings implicate the role of aggregated CVH components in slowing epigenetic ageing, in addition to the prior observations regarding higher CVH scores and delayed biological ageing [33].

Several studies support our findings on the key factors, such as nicotine avoidance and better glucose levels, in lowering EAA [34, 35]. Nicotine exposure has been linked to premature death and the development of pulmonary disease and cancers, with epigenetic age acceleration as a potential mediator [36]. In addition, smoking can accelerate epigenetic ageing by affecting immune cells [37], which in turn promotes inflammation, oxidative stress [38, 39], and clonal hematopoiesis [40]. Smoking-induced changes in DNA methylation are common across multiple tissues (e.g., whole blood, buccal, and lung epithelial cells) [41], and our findings support previous evidence on blood DNA methylation-based epigenetic ageing. Not only active smoking but also secondhand smoke, which is included in our analysis as nicotine exposure, may accelerate biological ageing [42]. Compared to males, the weaker contribution of nicotine avoidance in females may be attributed to a smaller proportion of current smokers [43]. Elevated fasting glucose and HbA1c levels are known to induce overexpression of growth factors, inflammation, and prolonged oxidative stress, contributing to both micro- and macrovascular complications as well as alteration of epigenetic mechanisms within target cells—vascular smooth muscle, retinal, and cardiac cells [44, 45]. In our study, an association between diabetic risk and systemic ageing as estimated from blood-based measures was observed, potentially mediated by vascular complications. Since DNA methylation patterns may differ by cell type or organ (e.g., kidney, blood vessels, and eyes) [44], further investigation is warranted in future studies. The importance of optimal levels of BMI and physical activity in females may be related to insulin resistance. Insulin resistance is partially responsible for the effect of high BMI (obesity) on epigenetic ageing [46], and particularly, menopausal females tend to be less sensitive to insulin as the protective role of endogenous estrogen decreases [47]. In addition, estrogen deficiency contributes to the accumulation of body fat, which potentially increases BMI and shortens telomere length [48, 49], and physical activity is an effective method of regulating BMI [50]. In our study, after adjusting for menopausal status—which is closely related to estrogen decline—BMI showed a comparable contribution to that of blood glucose, and was the strongest contributor to the lower PhenoAA. This suggests that factors beyond menopause, such as other causes of estrogen decline or insulin resistance in females (e.g., psychological stress and hypothalamic–pituitary–adrenal (HPA) axis dysregulation [51, 52]), may be involved.

Both in the regression and QGC analysis among total participants, there were differences in EAA measures that showed associations with the cardiovascular health components. PhenoAge, Grim2Age, and DunedinPACE were developed using glucose or HbA1c, and all three acceleration measures showed associations with blood glucose. BMI was used as a predictor only in developing DunedinPACE, and in regression analyses, DunedinPACE was solely associated with better BMI, unlike other measures. Nicotine avoidance was associated with lower EAA across all EAA measures in regression analysis, including PhenoAge and DunedinPACE, which were not directly trained on smoking, whereas Grim2Age was developed using smoking pack-years. These associations were also observed for DunedinPACE in the QGC analysis. These findings suggest that nicotine avoidance operates through multifaceted biological mechanisms that contribute to lower epigenetic ageing. The superior performance of PhenoAA, Grim2AA, and DunedinPACE in our findings may be attributed to their incorporation of cardiovascular-related physiological parameters [28, 29]. Among these EAA measures, Grim2AA and DunedinPACE exhibited more consistent associations than PhenoAA in our study. This may be partially explained by the fact that both clocks were developed using a wider range of clinical and laboratory parameters, including lifestyle factors and progressive multi-system deterioration [28, 29, 53], as well as their relatively high correlation with each other.

Our results did not emphasize the roles of healthy sleep and a healthy diet in lower EAA, despite prior evidence suggesting their importance in delaying biological ageing [54, 55]. This discrepancy highlights the need for a more comprehensive evaluation of sleep health, including factors beyond sleep duration, such as sleep quality and disorders. In previous research, high scores on diet assessments such as the Healthy Eating Index or Mediterranean-style Eating Pattern for Americans, which were recommended for use by the AHA, have been associated with lower EAA [55, 56]. We cannot rule out the possibility that the inconsistent findings between the current study and previous research are due to the different tools used for dietary assessment.

This study is original in its approach, framing multiple health-related components affecting biological ageing as a whole and quantifying their relative contributions; an approach that has been rarely explored in previous research in this area of study [16]. We anticipate that our findings on the relative contributions of health factors to EAA may provide evidence for prioritizing public health interventions to promote slower biological ageing. Based on our results, we suggest that CVH of the AHA may serve as a useful indicator for assessing both healthy ageing and cardiovascular health. It is also noticeable that the sex differences in the relative importance of health behaviours to lower EAA were newly reported in this study.

However, this study is not without Limitations. Beyond the quantitative measurement of health behaviours, the qualitative aspects of these components, such as sleep quality, should also be considered. Due to the data Limitations in the Ansan and Ansung study, dietary components could not be obtained from the same wave as the other seven factors, suggesting the possibility that a potential association between a healthy diet and EAA might not have been captured due to the longer time lag. Also, data for health behaviours were obtained from self-reported questionnaires, which may raise a concern of recall bias; however, we believe the bias is non-differential in association with EAA measures used in our analysis. Although the follow-up rate of the fifth wave of the Ansan and Ansung study was 67.2%, previous reports indicate no substantial differences in most baseline characteristics between responders and non-responders, except for smoking and fasting glucose levels [57]. Smoking prevalence and fasting glucose levels were comparable to those reported in nationally representative Korean data collected during a similar period [58, 59]. Nevertheless, the potential for selection bias cannot be entirely ruled out. We also cannot exclude the potential for reverse causality, as biological ageing may influence clinical indicators related to CVH. However, we sought to minimize the potential for this direction of association—where ageing leads to deterioration in serious clinical conditions—by excluding individuals with major comorbidities such as cardiovascular disease or cancer. The sex differences in our results suggest the role of underlying biological mechanisms, such as sex hormones, which may intricately interact with health behaviours. Further research is needed to elucidate the biological mechanisms linking different health behaviours by sex and EAA. Finally, although the current study utilized data from multiple, population-based large-scale cohorts, further research incorporating diverse cultural backgrounds is warranted to generalize our findings and to compare the effects of different health-related behaviours on epigenetic age deceleration.

Conclusions

In summary, our findings suggest that nicotine avoidance, physical activity, and better blood glucose levels or BMI may contribute to slower biological ageing. We suggest practicing multiple healthy behaviours, as a collective effort to improve health may also help with healthy ageing. Within the collective behavioural efforts, smoking cessation in males and engaging in physical activity in females potentially offer the most significant advantages for improved health outcomes. Based on these results, public health strategies aimed at slowing ageing should consider sex differences in health behaviours and potential underlying biological mechanisms. While the underlying mechanisms remain to be elucidated, we suggest developing effective preventive policies for slow ageing and monitoring their impact using epigenetic markers to extend healthy lifespan and reduce the healthcare burden.

Abbreviations

AHA

American Heart Association

BMI

Body mass index

CI

Confidence interval

CVH

Cardiovascular health

DNAmAge

DNA methylation age

DunedinPACE

Pace of Aging Computed from the Epigenome

EAA

Epigenetic age acceleration

EEAA

Extrinsic Epigenetic Age Acceleration

Grim2AA

GrimAge2 Acceleration

HEXA

Health Examinee

HLS

Healthy Lifestyle Score

HPA

Hypothalamic–pituitary–adrenal

IEAA

Intrinsic Epigenetic Age Acceleration

IRB

Institute Review Board

KoGES

Korean Genome and Epidemiology Study

PhenoAA

PhenoAge Acceleration

QGC

Quantile-based g-computation

QC

Quality control

SD

Standard deviation

Authors’ contributions

DE.L, K.K contributed to the study concept, design, and interpretation of results. DE.L, YS.P, HM.J, B.K, Y.K, and K.K contributed to data collection and curation. DE.L contributed to drafting the manuscript and statistical analysis. B.K, Y.K, and K.K obtained funding. Y.K, S.C, and K.K contributed to supervision, administrative, technical, or material support. All authors read and approved the final manuscript.

Funding

This work was supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science and Technology (Grant number: RS-2024–00333941 (Dr. K Kim)); National Institute of Health, Republic of Korea (Grant number: 2024-NI-005–00 (Dr. B Kim and Dr. Y Kim)).

Data availability

All data needed to evaluate the conclusions in the paper are present in the paper and/or the Additional Materials. Data from KoGES are available upon reasonable request (https://www.nih.go.kr/ko/main/contents.do.menuNo=300566), and full information on KoGES can be accessed through the Korea Biobank Project (https://www.kdca.go.kr/contents.es.mid=a30326000000). A data dictionary defining each variable will be made available upon reasonable request to the corresponding author.

Declarations

Ethics approval and consent to participate

All individuals voluntarily participated in this study and provided informed consent upon enrollment in the Korean Genome and Epidemiology Study (KoGES) survey. The protocol of this study was reviewed and approved by the Institute Review Board at National Institute of Health in Korea (No. KDCA-2024–02-07). This study was exempted from the review by the Institute Review Board of Sungkyunkwan University (IRB No. SKKU-06–029).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.