Epigenetic signatures of cardiometabolic risk in men: accelerated aging and differential methylation replicated across cohorts

Abstract

Background

Men exhibit greater susceptibility to cardiovascular diseases and metabolic disorders, with an earlier onset and more aggressive progression, potentially driven by epigenetic modifications, particularly DNA methylation. Our goal was to comprehensively characterize the epigenetic landscape of a broad cardiometabolic burden in a cohort composed exclusively of men.

Results

We generated novel DNA methylation profiles from whole blood samples of men with cardiometabolic disturbances (hypertension, ischemic heart disease, obesity, dyslipidemia) and age-matched healthy controls. Cases demonstrated significant epigenetic age acceleration, most pronounced for second-generation clocks (GrimAge, GrimAge2) and pace of aging measures (DunedinPACE), accompanied by shortened epigenetic telomere length (DNAmTL). Notably, none of the 19 evaluated first-generation epigenetic clocks exhibited sensitivity to the studied diseases. Epigenome-wide association analysis identified differentially methylated positions (DMPs), predominantly hypomethylated in cases compared to controls. Gene set enrichment analysis of genes annotated to these DMPs revealed nine distinct biological pathway clusters that reflect the multifactorial processes associated with cardiometabolic burden, including chronic inflammation, GPCR signaling dysregulation, metabolic disturbances, mitochondrial dysfunction, vascular remodeling, and renal electrolyte regulation. Key findings, including GrimAge acceleration, DunedinPACE elevation, DNAmTL shortening, and enrichment of inflammatory and GPCR pathways, were replicated in an independent cohort of men with atherosclerosis.

Conclusions

Men with cardiometabolic disturbances exhibit accelerated epigenetic aging and distinct DNA methylation signatures associated with cardiometabolic burden. Analysis of a broad battery of epigenetic clock models revealed that only the second-generation (GrimAge) and third-generation (DunedinPACE) models demonstrated a pronounced sensitivity to the uncomplicated diseases evaluated. In contrast, the first-generation models, trained to predict chronological age, failed to detect significant differences between the groups, suggesting limited applicability in these pathologies. The concordance of results across original and independent replication cohorts underscores the fundamental nature of these epigenetic alterations. Our findings suggest candidate biomarkers measurable in minimally invasive blood samples that may assist in early risk stratification and monitoring of disease progression in men, warranting further prospective evaluation to facilitate clinical translation.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s13148-026-02195-w.

Introduction

Cardiovascular diseases (CVD) encompass a broad spectrum of pathological conditions affecting the heart and vasculature. Accounting for approximately one-third of all deaths globally, CVD represents one of the most significant challenges to public health [1–3]. While the age-standardized death rate has declined, the absolute number of CVD-related deaths continues to rise, driven by population growth and aging [1]. Affecting both developed and developing nations, CVD remains a persistent challenge for researchers and clinicians [2, 4]. Sex differences critically shape CVD epidemiology; men experience heart disease on average 7–10 years earlier than women, with a 2% cumulative incidence reaching 10 years earlier (45–50 years in men vs. 55–60 years in women) [5]. Men are more likely than women to develop CVD, including coronary heart disease, myocardial infarction, peripheral artery disease, and atherosclerosis [6]. Higher systolic blood pressure observed in young men compared to young women may predispose them to isolated hypertension [7], which in turn contributes to other cardiovascular pathologies [8]. Recent studies indicate that men exhibit a less favorable cardiovascular risk profile than women and face higher CVD-related mortality [9, 10]. Key modifiable risk factors include smoking, elevated BMI, systolic hypertension, dyslipidemia, and diabetes [3, 9, 11, 12], though their relative impact varies across regional and socioeconomic factors [13]. Despite advances in understanding modifiable risk factors, the mechanisms driving accelerated CVD development in men remain poorly understood. This underscores the need to investigate non-traditional pathophysiological pathways, particularly epigenetic regulatory mechanisms.

Epigenetic mechanisms – including DNA methylation, post-translational histone modifications, and non-coding RNA regulation – play a significant role in regulating the transcription of genes governing cardiac muscle function and vascular structure, thereby contributing to CVD pathogenesis [14, 15]. Previous studies have shown that these mechanisms influence the progression of coronary heart disease, atherosclerosis, vascular calcification, and hypertension, as well as other cardiovascular and cardiometabolic conditions [14–17]. Unhealthy behaviors, lifestyle factors (e.g., sleep, diet, stress), medical history, and adverse environmental exposures can shape the human epigenome, which in turn may compromise cardiovascular function [18–21]. Monitoring DNA methylation – the most studied epigenetic modification [22, 23] – along with related biomarkers can offer insights into how risk factors influence CVD development and progression, potentially enabling effective health monitoring [24]. Systematic investigations of sex-specific epigenetic signatures in CVD remain scarce, particularly for men [25].

Biological aging and its deviation from chronological age are commonly assessed using epigenetic clock models. These models are typically trained on chronological age or survival outcomes, using methylation levels at specific CpG sites. They have evolved through three successive generations. First-generation clocks, including Hannum DNAmAge [26] and Horvath DNAmAge [27], estimate epigenetic age based on age-associated CpGs. Second-generation clocks, such as DNAmPhenoAge [28] and GrimAge [29], incorporate clinical biomarkers and mortality risk factors to estimate biological age and predict mortality. Third-generation clocks, exemplified by DunedinPACE [30], estimate the pace of aging by leveraging longitudinal data.

Epigenetic clock models are frequently studied for their associations with various diseases, including CVD. Accelerated epigenetic aging has been observed using GrimAge [31–40], Horvath DNAmAge [32, 39], DunedinPACE [37] in patients with a range of cardiovascular conditions, including stroke, heart failure, atherosclerosis, atrial fibrillation, and elevated cardiovascular risk. However, most studies focus on a narrow range of epigenetic models, typically limited to first- and second-generation clocks. Furthermore, these studies predominantly analyze mixed-sex cohorts with statistical adjustment for sex, rather than explicitly evaluating sex-specific effects [32, 35]. To date, the relationship between epigenetic age and cardiovascular health in post-menopausal women has only been investigated in one study [41], while similar investigations specifically targeting men remain scarce and represent a knowledge gap.

EpiScores is another promising DNA methylation-based tool for assessment of physiological status, containing over 100 regression models that predict dimensionless levels of plasma proteins or metabolic risk factors [42]. It has been demonstrated that EpiScores can be associated with an increased risk of morbidity and mortality, particularly with regard to CVD and ischemic heart disease [42–44]. However, EpiScores studies have been implemented for mixed cohorts adjusted for sex (like for epigenetic clocks), while no studies focused on the male population have been published to date.

To identify specific CpG sites and genes associated with CVD, an epigenome-wide association study (EWAS) is typically performed. This approach facilitates understanding of the mechanisms underlying CVD development and enables the identification of disease-associated biomarkers [45]. Gene set enrichment analysis (GSEA) helps identify systematic dysregulation of biological and molecular pathways specific to the pathologies under investigation. Recent EWAS have identified numerous differentially methylated positions (DMPs) associated with myocardial infarction, acute coronary syndrome, and CVD risk; however, these findings largely lack specificity for the male population [46–49].

The primary objective of this study is to address the identified gap by providing a multifaceted analysis of epigenetic aging in cardiometabolic disturbances specifically in men. The objectives of this research include: (1) assessing epigenetic age acceleration and EpiScores profiles in men with cumulative cardiometabolic burden (hypertension, ischemic heart disease, obesity, and dyslipidemia) and comparing these cases with age-matched healthy controls; (2) conducting an EWAS to identify DMPs; (3) performing GSEA to identify biological pathways associated with cardiometabolic disturbances in men. This study is based on an original dataset comprising 60 cases and 60 healthy controls (whole blood, EPICv2 platform). To evaluate generalizability, we replicated the analyses in an independent external cohort of men with generalized atherosclerosis (138 cases and 122 controls) obtained from the Gene Expression Omnibus (GSE220622) [33].

Methods

Data collection and processing

The original dataset comprised 120 men, aged 24–70 years, with clinical information extracted from their medical records. Participants were stratified into two groups. The case group included 60 individuals presenting with at least one documented cardiometabolic condition: hypertensive diseases (I10 Essential (primary) hypertension, I11 Hypertensive heart disease), ischemic heart disease (I25 Chronic ischaemic heart disease), obesity (E66 Obesity), and lipoprotein metabolism disorders (E78 Disorders of lipoprotein metabolism and other lipidaemias). The control group included 60 individuals with no documented history of these conditions. Exclusion criteria included acute-stage chronic diseases, active malignancies, and acute viral respiratory infections present at the time of blood collection.

All participants were informed of the procedure details, as well as potential inconveniences and risks. Each participant signed an informed consent and filled out a consent for personal data processing, taking into account the principle of confidentiality (accessibility only to the research group and presentation of data in a common array). The study was approved by the local ethics committee of Nizhny Novgorod State University (protocol No. 11 dated 26 October 2022). All procedures were conducted in accordance with the Declaration of Helsinki (1964) and its subsequent amendments. The work was performed using the core facilities of the Lopukhin FRCC PCM “Genomics, proteomics, metabolomics” (http://rcpcm.org/?p=2806).

Phenol Chloroform DNA extraction was performed. DNA was quantified using the DNA Quantitation Kit Qubit dsDNA BR Assay (Thermo Fisher Scientific), and 250 ng was bisulfite-treated using the EpiMark Bisulfite Conversion Kit (NEB), with case and control samples randomly distributed across arrays. The Illumina Infinium MethylationEPIC v2.0 BeadChip [50] was used according to the manufacturer’s instructions. Methylation levels were expressed as beta values, ranging from 0 (unmethylated) to 1 (completely methylated) for each probe. Data preprocessing, normalization, and batch effect correction were performed using the standard pipeline in the ChAMP R package (version 2.40.0) [51]. Probes with a detection p-value above 0.01 in at least 10% of samples were excluded. Functional normalization of raw methylation data was performed using the minfi R package function (version 1.56.0) [52]. The ComBat method, implemented in the ChAMP R package, was applied to correct for Slide and Array batch effects [53].

In addition to DNA methylation profiling, the following clinical laboratory parameters were obtained from all participants: glucose, total cholesterol, albumin, creatinine, alkaline phosphatase, C-reactive protein (CRP).

The independent replication dataset corresponded to GSE220622 [33] from the Gene Expression Omnibus [54]. The case group comprised 138 individuals with generalized atherosclerosis, while the control group comprised 122 individuals without atherosclerosis. Raw DNA methylation data (Illumina Infinium MethylationEPIC v1,.idat format) were obtained from the GEO repository and subjected to the same preprocessing pipeline as the original dataset.

Epigenetic data analysis

Epigenetic data analysis comprised three main steps: (1) calculation of epigenetic age acceleration, epigenetic metrics, blood cell estimates and EpiScores; (2) identification of differentially methylated positions (DMPs); and (3) gene set enrichment analysis (GSEA).

First, epigenetic age and corresponding epigenetic age acceleration were calculated for 36 epigenetic clock models (full list provided below): Hannum DNAmAge [26], Horvath DNAmAge [27], Lin [55], DNAmPhenoAge [28], SkinAndBlood [56], GrimAge [29], ZhangEN, ZhangBLUP [57], Han [58], ABEC, cABEC, eABEC [59], AltumAge [60], PCHannum, PCHorvath, PCPhenoAge, HRSInCHPhenoAge, PCSkinAndBlood, PCGrimAge [61], GrimAge2 [62], PipekElasticNet, PipekFilteredH, PipekRetrainedH [63], DNAmFitAge [64], ENCen40 [65], YingAdaptAge, YingCausAge, YingDamAge [66], StocH, StocP, StocZ [67], RetroelementAgeV1, RetroelementAgeV2 [68], IntrinClock [69], Bernabeu [70], EpInflammAge [71]. Epigenetic age estimation models are machine learning algorithms trained to predict chronological age using DNA methylation data. Most models examined here are linear and rely on the ElasticNet approach, whereas EpInflammAge and AltumAge are based on deep neural network architectures. Definitions for the majority of models were obtained from the pyaging library (version 0.1.30) [72], with the exception of Bernabeu and EpInflammAge, which were implemented using software provided in the original publications. Age acceleration was defined as the residual deviation from the linear regression of epigenetic age on chronological age, fitted using the control group. We additionally calculated epigenetic metrics – estimates of health status not expressed in years – including DunedinPACE [30], ZhangMortality [73], DNAmTL [74], PCDNAmTL [61], epiTOC1 [75], stemTOC [76]. Definitions for these metrics were also obtained from the pyaging library.

Next, we calculated 115 EpiScores – dimensionless estimates of circulating plasma protein levels (e.g., CRP, MMP-9, adiponectin) and lifestyle or anthropometric risk factors – derived from DNA methylation data [42]. Their dimensionless nature imposes limitations on clinical interpretation and direct comparability with empirically measured values. EpiScores were calculated using software provided in the original publication. Like most epigenetic clock models, these estimates rely on ElasticNet linear models.

We also used the pyaging package to estimate the relative abundance of major blood cell types from DNA methylation data. Cell proportions were derived using a standard deconvolution algorithm based on reference methylation panels [77, 78]. For each sample, estimates of the proportions of six cell types were obtained: B cells, CD4 + T cells, CD8 + T cells, monocytes, neutrophils, and NK cells.

Data on phenotype, epigenetic ages, metrics, blood cell estimates, EpiScores are available in the Supplementary Table S1 for the original dataset and Supplementary Table S10 for the replication dataset.

Two statistical tests were applied to assess the significance of between-group differences in the estimated parameters. The Mann–Whitney U test, a nonparametric test for two independent groups, was used to evaluate whether the samples originated from the same population under the two-tailed null hypothesis that the groups are equal [79]. This test did not account for covariates. ANCOVA combines analysis of variance (ANOVA) with linear regression to compare group means of a dependent variable while accounting for the influence of one or more continuous covariates. For the original dataset, we constructed two models: the first included chronological age as a covariate, and the second included chronological age along with all blood biomarkers. For the replication dataset, only an age-adjusted model was fitted, owing to the absence of clinical data. Statistical significance across all tests was assessed using the Benjamini–Hochberg procedure to control the false discovery rate (FDR) at a threshold of 0.05 [80]. To quantify the magnitude of between-group differences, effect sizes were calculated. For the Mann–Whitney U test, we used the Common Language Effect Size (CLES), defined as , where U is the U-statistic, and are the sample sizes of the respective groups. It represents the probability that a randomly selected observation from one group exceeds a randomly selected observation from the other group. For ANCOVA models, we calculated partial eta-squared () using the t-statistic for the group effect: , where (N is the total sample size, k is the number of groups, c is the number of covariates).

We performed the search for differentially methylated positions using the R limma package (version 3.66.0) [81]. To analyze the association between blood DNA methylation and the studied diseases, we constructed linear models using the lmFit function to perform pairwise comparisons of beta values between the case and control groups. Moderated t-statistics, p-values, and log-odds of differential methylation were calculated using the eBayes function. Probes with Benjamini-Hochberg-corrected p-values < 0.05 were considered statistically significant.

To reduce dimensionality and project DMPs into two-dimensional space, we applied t-SNE (t-distributed Stochastic Neighbor Embedding) [82], PCA (Principal Component Analysis) [83], IsoMap (Isometric Mapping) [84] and MDS (Multidimensional Scaling) [85]. t-SNE is a nonlinear probabilistic method that transforms Euclidean distances between points into conditional similarity probabilities. PCA is a linear method that identifies new orthogonal axes (principal components) capturing the maximum variance in the data. IsoMap is a nonlinear method based on preserving geodesic distances along the data manifold. MDS aims to preserve pairwise distances between objects when projecting them into a low-dimensional space.

GSEA was performed using the R package methylGSA (version 1.28.0), via the methylglm function [86]. The analysis was conducted using the GO and Reactome gene sets, restricted to those comprising 5 to 1000 genes. Terms with adjusted p-values < 0.05 were considered statistically significant.

Results

Study design

The dataset collected for this study contains only men aged 24–70 years, (Fig. 1A). The case group included 60 individuals with documented cardiometabolic burden, presenting with at least one of the following conditions: primary hypertension, hypertensive heart disease, ischemic heart disease, obesity, dyslipidemia. The control group included 60 individuals with no documented history of these conditions. DNA methylation data were generated using the Illumina Infinium MethylationEPIC v2.0 platform. The GSE220622 dataset [33] from the Gene Expression Omnibus [54] served as the independent replication dataset. We extracted a male-only subset from this dataset (Fig. 1A), owing to the limited number of female samples (only 8 in the case group) and to maximize concordance with our original cohort. Participants ranged from 40 to 55 years of age. In this replication cohort, the case group included 138 individuals with generalized atherosclerosis, while the control group included 122 individuals without atherosclerosis. The DNA methylation data were generated using the Illumina Infinium MethylationEPIC v1 platform.

Fig. 1.

Datasets overview and study design. A Datasets analyzed in this study: the original dataset (left) and the independent replication dataset GSE220622 (right). For each dataset, the panels display overlapping histograms of age distributions for cases and controls, along with the disease compositions of the respective case groups. In the original dataset, cases and controls are depicted in light red and light green, respectively, with histogram overlaps shown in orange. In the replication dataset (GSE220622), cases and controls are shown in dark red and dark green, with overlaps in brown. B The main steps of the analysis included calculation of age acceleration for 36 epigenetic age models, 6 epigenetic metrics, 115 EpiScores, 6 blood cell estimates; and performance of EWAS and GSEA

Both datasets were analyzed using an identical protocol (Fig. 1B). We first calculated epigenetic ages from 36 clock models, which were then used to derive age acceleration estimates for comparison between cases and controls. Additionally, six epigenetic metrics – estimates of aging not expressed in years – were calculated using the pyaging library [72]. We also calculated 115 EpiScores – dimensionless estimates of circulating plasma proteins and risk factors [42]. Proportions of six major blood cell types (B-cells, CD4 + T-cells, CD8 + T-cells, Monocytes, Neutrophils, NK cells) were also estimated. EWAS was performed for both datasets to identify DMPs using the limma package [81]. GSEA was performed on the genes annotated to the identified DMPs using the methylGSA package [86]. Finally, we compared findings across the two datasets to identify consistent patterns in the epigenetic profiles of men with cardiometabolic disturbances. To our knowledge, this is the first study to simultaneously evaluate 36 clock models, six epigenetic estimators, and 115 EpiScores within a single analytical pipeline in a male cohort with cardiometabolic burden. This comprehensive approach enabled us to systematically compare the sensitivity of different clock generations and identify markers associated with cardiometabolic risk in men.

Epigenetic clocks and metrics

Comparison of laboratory parameters between groups in the original dataset revealed that only total cholesterol (CLES = 0.282; η²_p = 0.132) and glucose (CLES = 0.353; η²_p = 0.010) demonstrated moderate‑to‑large effect sizes, confirming the expected metabolic burden in the case group. By contrast, markers of systemic inflammation (C‑reactive protein), renal function (creatinine), and hepatic function (albumin, alkaline phosphatase) exhibited negligible effect sizes, indicating no meaningful between‑group differences (Fig. 2A). Detailed statistical results are provided in Supplementary Table S2. Next, we applied 36 epigenetic clock models (see Methods for the full list) to compare epigenetic age acceleration between cases and controls (Fig. 2B). Using the Mann–Whitney U test, the case group showed statistically significant age acceleration (adjusted p-value < 0.05) across all three GrimAge versions (GrimAge, GrimAge2, and PCGrimAge) as well as PCPhenoAge, a modified version of the original DNAmPhenoAge clock. The largest effect sizes were observed for GrimAge (CLES = 0.297), PCGrimAge (CLES = 0.307), GrimAge2 (CLES = 0.321), PCPhenoAge (CLES = 0.342), indicating that in approximately 70% of random pairwise comparisons, epigenetic age acceleration estimates were higher in cases than in controls. Our findings align with recent reports of GrimAge acceleration in patients with cardiovascular disease [34, 35, 37], atherosclerosis [33], stroke and heart failure [31], and in individuals with lower cardiovascular health scores [36, 38, 39]. Metabolic disorders, including obesity and dyslipidemia, have also been previously linked to increased GrimAge [87–89]. The Mann–Whitney test did not account for age as a covariate. However, age is a key risk factor in cardiovascular risk scores such as SCORE2 [90], ASCVD [91], QRISK3 [92], and others. It is therefore advisable that its influence be taken into account. To adjust for its potential confounding effect, we performed ANCOVA with chronological age as a covariate. In this age‑adjusted model, significant age acceleration persisted across all GrimAge versions (GrimAge, GrimAge2, and PCGrimAge), two DNAmPhenoAge versions (PCPhenoAge and HRSinCHPhenoAge) and DNAmFitAge. Effect sizes remained consistent, with PCGrimAge (η²_p = 0.097), GrimAge (η²_p = 0.078), and PCPhenoAge (η²_p = 0.074) showing the largest partial eta‑squared values, corresponding to medium‑to‑large effects. Given the availability of clinical blood biomarkers in the original dataset, we repeated the ANCOVA with both age and all biomarkers as covariates. Results were largely consistent with the age‑only model, with the same clocks showing significant differences, except for AltumAge, which reached significance only in the extended model. Effect sizes for key models remained robust: PCGrimAge (η²_p = 0.079), GrimAge (η²_p = 0.071), and PCPhenoAge (η²_p = 0.093). Notably, PCPhenoAge exhibited a slight increase in effect size when biomarkers were added, suggesting sensitivity to metabolic status. Collectively, these three statistical approaches demonstrate the consistent performance of GrimAge and DNAmPhenoAge models in capturing cumulative cardiometabolic risk. Notably, none of the first‑generation clocks trained on chronological age (e.g., Hannum, Horvath, and others) showed sensitivity to the uncomplicated diseases in the original dataset, with ANCOVA effect sizes close to zero (η²_p < 0.03; see Supplementary Table S3 for full details).

Fig. 2.

Comparison of epigenetic profiles in cases and controls. A (left) Adjusted p-values for blood biomarkers in the original dataset. (right) Distributions of blood biomarker values (violin plots) in cases (red) and controls (green). B (left) Adjusted p-values for 36 epigenetic clock models. (right) Distributions of epigenetic ages (scatter plots) and epigenetic age acceleration (violin plots) in cases (red) and controls (green). The black dotted line represents the identity line (chronological age = epigenetic age), and the green solid line represents the regression line fitted to the control group. C (left) Adjusted p-values for 6 epigenetic metrics. (right) Distributions of epigenetic metrics (violin plot) in cases (red) and controls (green). D (left) Adjusted p-values for 6 blood cell estimates. (right) Distributions of blood cell estimates (violin plot) in cases (red) and controls (green). E Adjusted p-values for 115 EpiScores. For all subplots, magenta denotes the Mann–Whitney test, cyan denotes age‑adjusted ANCOVA, and olive denotes ANCOVA adjusted for age and blood biomarkers. The red dotted line indicates the significance threshold (p = 0.05)

Next, we calculated epigenetic metrics – estimates of aging and mortality risk not expressed in years – to compare cases and controls (Fig. 2C). Using the Mann–Whitney test, the case group exhibited shorter telomere length estimates (DNAmTL and PCDNAmTL), higher mortality risk (ZhangMortality), accelerated aging (DunedinPACE), and elevated cancer risk estimates (epiTOC1 and stemTOC). The largest effect sizes were observed for PCDNAmTL (CLES = 0.734) and DNAmTL (CLES = 0.699), confirming shorter telomere length estimates in the case group. ZhangMortality (CLES = 0.307), DunedinPACE (CLES = 0.342), epiTOC1 (CLES = 0.382), and stemTOC (CLES = 0.395) showed moderate effects. DunedinPACE and DNAmTL have previously been shown to be associated with CVD risk and mortality [93, 94], cardiometabolic biomarkers [95], and dyslipidemia [87]. Increased CVD mortality risk has also been reported for the ZhangMortality model [93]. No statistically significant differences were identified in either ANCOVA model (η²_p < 0.03; see Supplementary Table S4 for full details). This lack of significance likely reflects the fact that these metrics inherently incorporate age or mortality risk into their algorithms, rendering additional adjustment redundant.

The statistically significant differences observed with the second- and third-generation models may stem from several factors. First-generation clocks use chronological age as the target variable, whereas second- and third-generation models are designed to estimate phenotypic age, time to death, and the pace of age-related changes. Moreover, second- and third-generation models are linked to clinical parameters, including blood test indicators (PhenoAge), plasma protein levels (GrimAge), lipid profile and cardiorespiratory function (DunedinPACE), and physical activity biomarkers (DNAmFitAge), which are typically dysregulated in adverse cardiometabolic states [96–101]. This underscores the potential utility of second- and third-generation models as tools for assessing cardiometabolic risk.

Analysis of blood cell composition estimated from methylation data (B cells, CD4 + T cells, CD8 + T cells, monocytes, neutrophils, NK cells), revealed no significant between-group differences using either the Mann–Whitney test or ANCOVA adjusted for age and laboratory parameters (Fig. 2D). Effect sizes were consistently small across all cell types (CLES ≈ 0.50, η²_p < 0.02) in both age-adjusted and age + biomarker-adjusted models (Supplementary Table S5). This suggests that cellular heterogeneity is not a significant confounder in this study.

We next examined EpiScores, dimensionless estimates of circulating plasma proteins and risk factors, for between-group differences (Fig. 2E). Two EpiScores showed significant differences across all three statistical tests (Mann–Whitney and both ANCOVA models): Alcohol (CLES = 0.311, η²_p = 0.121 and 0.118) and WFIKKN2 (CLES = 0.729, η²_p = 0.109 and 0.103). Alcohol is a well-established risk factor, with increased consumption elevating the risk of CVD onset and progression [102, 103]. WFIKKN2 is an inhibitor of myostatin, a protein whose elevated levels reflect muscle loss and impair hepatic glucose metabolism [104, 105]. Three additional EpiScores showed significant differences in two of the three tests (Mann–Whitney and age-adjusted ANCOVA, but not the fully adjusted model): Waist:Hip Ratio (CLES = 0.266, η²_p = 0.128 and 0.083), MMP-12 (CLES = 0.208, η²_p = 0.082 and 0.080), and CRP (CLES = 0.263, η²_p = 0.086 and 0.062). The reduction in η²_p upon biomarker adjustment suggests that part of the association for these three markers may be mediated by metabolic factors. Elevated waist-to-hip ratio is associated with hypertension, insulin resistance, and dyslipidemia [106, 107]. MMP-12 has been associated with elastin degradation and extracellular matrix remodeling in atherosclerosis and hypertension, as well as with vascular stiffness and elevated blood pressure [108, 109]. CRP (C-reactive protein), an inflammatory marker, is associated with adverse cardiac events and is considered a marker of subclinical inflammation in obesity [110, 111]. Detailed statistical results are provided in Supplementary Table S6. Collectively, these EpiScores delineate a profile of men with elevated cardiometabolic burden, characterized by systemic inflammation, visceral adiposity, vascular remodeling, myostatin dysregulation, and behavioral factors such as alcohol consumption. All calculated epigenetics ages, metrics, blood cell estimates and EpiScores for the original dataset are provided in Supplementary Table S1.

EWAS and GSEA

Beyond epigenetic ages and metrics, we next compared the methylation profiles of cases and controls. To this end, we performed an epigenome-wide association study (EWAS) to identify differentially methylated positions (DMPs) between cases and controls. Using the limma package [81], we identified 1,221 DMPs at an adjusted p-value threshold of < 0.05 (Supplementary Table S7). Only ten of these DMPs were hypermethylated in cases relative to controls, while the vast majority were hypomethylated (Fig. 3A). The hypermethylated probes mapped to the ADGRG1, FHL2, GRM2, MIR7-3, MKX, and TMEM200B genes, as well as to non-genic regions. ADGRG1 encodes the GPR56 protein, whose dysregulation has been linked to metabolic disorders and cardiovascular diseases [112]. FHL2 regulates intracellular signaling pathways implicated in the pathogenesis of heart disease and metabolic disorders [113]. Increased MKX expression in subcutaneous adipose tissue has been associated with obesity and metabolic disorders [114]. Projection of all DMPs onto a two-dimensional plane using standard dimensionality reduction methods revealed separation between cases and controls, indicating distinct methylation profiles (Fig. 3B).

Fig. 3.

EWAS and GSEA results. A Volcano plot of differentially methylated positions between cases and controls. Colors indicate significant DMPs: blue for hypomethylated and orange for hypermethylated in cases relative to controls. The black dotted line denotes the significance threshold (p = 0.05). B Projection of DMPs into two-dimensional space using t-SNE, PCA, IsoMap, and MDS. Points represent cases (red) and controls (green) in the reduced space. C Schematic summary of GSEA results obtained using the GO and Reactome libraries

The 1,221 DMPs mapped to 304 unique genes, for which we performed GSEA using the methylGSA package [86]. The Gene Ontology (GO) library [115] yielded 173 significant terms at an adjusted p-value < 0.05 (Supplementary Table S8). These terms clustered into eight broad groups potentially associated with cardiometabolic risk (Fig. 3C): disorders of lipid and energy metabolism; dysregulation of vascular tone and cardiovascular function; chronic inflammation and immune dysregulation; dysregulation of GPCR signaling and second messenger pathways; renal dysfunction and electrolyte regulation; epigenetic reprogramming and post-transcriptional regulation; mitochondrial dysfunction and oxidative stress; and neurohumoral regulation and sensory perception. Lipid metabolism dysregulation can be associated with a range of adverse outcomes, including cardiovascular disease and metabolic syndrome. Indeed, approximately 50% of CVD-related deaths have been attributed to metabolic disorders such as obesity, atherogenic dyslipidemia, and hypertension [116]. Endothelial dysfunction, which precedes hypertension, may contribute to a deleterious cycle in which hypertension exacerbates endothelial dysfunction, promoting further target organ damage [117]. Chronic low-grade inflammation may mediate the link between obesity and its cardiometabolic consequences. Interventions targeting inflammation are therefore considered a cornerstone of therapeutic strategies [104]. GPCR signaling dysregulation underlies many cardiovascular pathologies, including heart failure, cardiac hypertrophy, hypertension, and atherosclerosis. Given that many contemporary drugs target GPCRs, these receptors represent important therapeutic targets for CVD [118]. Reduced renal function can increase the risk of cardiovascular events and mortality via mechanisms involving mitochondrial dysfunction, inflammation, innate immune activation, and oxidative stress [119]. Epigenetic mechanisms integrate genetic predisposition with environmental and behavioral factors, governing the expression of genes involved in inflammation, metabolism, and endothelial function. DNA methylation profiles, in particular, are associated with accelerated biological aging and increased CVD risk [14]. Mitochondrial dysfunction can promote oxidative stress through excessive reactive oxygen species production and activates pro-inflammatory signaling pathways, thereby contributing to endothelial dysfunction [120]. Chronic sympathetic activation, commonly observed in obesity and insulin resistance, promotes hypertension, vasoconstriction, and cardiac remodeling. Sensory pathways, including peripheral receptors, can influence feeding behavior and metabolic profiles [121].

The Reactome library [122] yielded 9 significant terms (Supplementary Table S9), which similarly grouped into four of the GO categories: chronic inflammation and immune dysregulation; GPCR signaling and second messenger pathways; renal dysfunction and electrolyte regulation; and neurohumoral regulation and sensory perception. The recurrence of these four pathway categories across both libraries may suggest a convergence of dysregulated inflammatory, neurohumoral, and renal regulatory axes, manifesting through GPCR-mediated signaling in the context of cardiometabolic risk.

Replication on independent dataset

To replicate the findings from the original dataset, we examined the open-access dataset GSE220622 from the NCBI GEO repository and repeated all steps of the epigenetic profile analysis. We selected a male-only subset and formed the case group based on a diagnosis of generalized atherosclerosis. Notably, the age range in the replication cohort was significantly narrower than that of the original dataset (Fig. 1A).

For this dataset, we first calculated 36 epigenetic ages and compared age acceleration using the Mann–Whitney and ANCOVA tests (see Fig. 4A). Of note, only ANCOVA with chronological age as a covariate was applied here, as blood biomarker data were unavailable for this independent replication cohort. Both tests revealed significant differences between cases and controls for all GrimAge models (GrimAge, GrimAge2, and PCGrimAge), as well as for the original DNAmPhenoAge model and DNAmFitAge. Effect sizes were largest within the GrimAge family: PCGrimAge (CLES = 0.203, η²_p = 0.212), GrimAge (CLES = 0.224, η²_p = 0.193), and GrimAge2 (CLES = 0.229, η²_p = 0.180) all demonstrated large effects, indicating separation between cases and controls. DNAmFitAge also showed a substantial effect (CLES = 0.272, η²_p = 0.130), while DNAmPhenoAge and its modifications exhibited moderate effects (η²_p ranging from 0.013 to 0.043). In contrast, first‑generation clocks consistently yielded negligible effects (η²_p < 0.01), confirming their limited sensitivity to cardiometabolic burden. The original study that provided this replication dataset also reported statistically significant GrimAge age acceleration in patients with atherosclerosis [33]. Notably, the entire GrimAge epigenetic clock family showed significant age acceleration in both datasets. This suggests that these second-generation models can be sensitive to the cardiometabolic diseases and may serve as biomarkers for risk detection, disease progression monitoring, and therapeutic evaluation. Replication further confirmed the statistically significant performance of the DNAmPhenoAge model, its two modifications (PCPhenoAge and HRSInCHPhenoAge), and DNAmFitAge. The effect size patterns in the replication cohort (Supplementary Table S11) mainly reproduced those observed in the original dataset, further strengthening the evidence for the robustness of these associations across independent male cohorts with related cardiometabolic phenotypes.

Fig. 4.

Results for the independent replication dataset GSE220622. A Adjusted p-values for 36 epigenetic clock models. B Adjusted p-values for 6 epigenetic metrics. C Adjusted p-values for 6 blood cell estimates. D Adjusted p-values for 115 EpiScores. E Volcano plot of DMPs between cases and controls. Colors indicate significant DMPs: blue for hypomethylated and orange for hypermethylated in cases relative to controls. The black dotted line denotes the significance threshold (p = 0.05). F Schematic summary of GSEA results obtained using the GO and Reactome libraries. For all subplots, magenta denotes the Mann–Whitney test, cyan denotes age‑adjusted ANCOVA, and the red dotted line indicates the significance threshold (p = 0.05)

The results for the epigenetic metrics also align quite well across the original and replication datasets, reproducing significant differences between the case and control groups (Fig. 4B). Effect sizes confirmed associations for PCDNAmTL (CLES = 0.707, η²_p = 0.036), DNAmTL (CLES = 0.688, η²_p = 0.038), DunedinPACE (CLES = 0.263, η²_p = 0.116), and ZhangMortality (CLES = 0.350, η²_p = 0.050), while epiTOC1 and stemTOC showed negligible effects (η²_p < 0.005, Supplementary Table S12). DunedinPACE, ZhangMortality, DNAmTL, and PCDNAmTL reached significance in the ANCOVA only for the independent replication dataset, which may be attributable to several differences between the cohorts. Specifically, the replication dataset exhibits lower heterogeneity, as it includes only one disease (atherosclerosis), whereas the original dataset encompasses a broader range of cardiometabolic conditions. The smaller sample size of the original dataset relative to the replication cohort may also account for the lack of significance in the former. Additionally, the replication dataset was generated using the EPIC platform, whereas the original utilized the EPICv2 platform. Blood cell estimates had no significant differences between the groups, like for the original dataset (Fig. 4C), with consistently small effect sizes (η²_p < 0.02, Supplementary Table S13).

Although the independent replication dataset yielded more statistically significant EpiScores than the original one (Fig. 4D), all five EpiScores identified in the original dataset (Alcohol, Waist:Hip Ratio, WFIKKN2, MMP-12, CRP) remained significant upon replication. Effect sizes confirmed these associations: MMP‑12 showed a large effect (CLES = 0.230, η²_p = 0.132), followed by CRP (CLES = 0.321, η²_p = 0.056). WFIKKN2 also demonstrated a substantial effect (CLES = 0.683, η²_p = 0.085), while Alcohol (CLES = 0.358, η²_p = 0.063) and Waist:Hip Ratio (CLES = 0.339, η²_p = 0.073) showed moderate‑to‑large effects (Supplementary Table S14). This suggests that these circulating protein and risk factor markers can be associated with cardiovascular and metabolic pathologies. Several additional EpiScores with large effect sizes were also identified in the replication cohort, including Smoking (CLES = 0.245, η²_p = 0.191), NRTK3 (CLES = 0.760, η²_p = 0.164), Osteomodulin (CLES = 0.743, η²_p = 0.129), Semaphorin‑3E (CLES = 0.739, η²_p = 0.168), and NCAM‑120 (CLES = 0.737, η²_p = 0.137), suggesting that these markers may be sensitive to atherosclerotic burden. MMP-9 and MMP-1, also identified in the replication dataset, have been reported as markers of atherosclerotic plaque stability and adverse outcomes in atherosclerosis [123, 124]. Adiponectin is another notable EpiScore; its reduction has been associated with an increased risk of atherosclerosis and adverse coronary events [125]. The greater number of significant EpiScores in the replication cohort may partly reflect its more homogeneous disease definition (atherosclerosis alone), which could enhance statistical power relative to the broader cardiometabolic burden in the original dataset. Overall, the replication of the five key EpiScores across both cohorts, along with the emergence of additional markers in the more homogeneous replication dataset, supports their potential utility as biomarkers of cardiometabolic risk in men. All calculated epigenetics ages, metrics, blood cell estimates and EpiScores for the replication dataset are provided in Supplementary Table S10.

The search for DMPs identified 86 CpG sites (Supplementary Table S15). Of these, 5 were hypermethylated and 81 were hypomethylated in cases relative to controls (Fig. 4E). In both datasets, the number of hypermethylated DMPs in cases was significantly lower. Seven DMPs were shared between the datasets (cg01940273, cg17739917, cg08262002, cg07803149, cg12009405, cg25430089, cg22108930), all hypomethylated in the case group and mapping to three genes (PRSS23, CLEC3B, LMNA). In the GSE220622 dataset, the hypomethylated in cases DMPs mapped to the MYO1G, BRCA2, FARS2, ZEB2, and C2CD2 genes. ZEB2 has been associated with coronary artery disease risk [126], and BRCA2 has been linked to vascular endothelial function [127]. All DMPs mapped to 50 genes, for which GSEA identified several significant terms in the GO and Reactome libraries (Fig. 4F, Supplementary Tables S16 and S17). The identified terms corresponded well to the pathway groups established for the original dataset, with the exception of the renal dysfunction and electrolyte regulation group. Thus, most findings from the original dataset were confirmed in this independent atherosclerosis cohort, suggesting similar epigenetic profiles associated with cumulative cardiometabolic risk.

Discussion

This study provides a comprehensive epigenetic characterization of a broad cardiometabolic burden in an exclusively male cohort. The case group exhibited significant acceleration of second‑generation (GrimAge, GrimAge2, PCGrimAge) and third‑generation (DunedinPACE) clocks, increased mortality risk (ZhangMortality), and shorter epigenetic telomere length estimates (DNAmTL) compared to healthy controls. Effect sizes confirmed robust associations: the GrimAge family showed the largest effect among epigenetic clocks, while DunedinPACE and ZhangMortality demonstrated moderate-to-large effects that remained consistent across the original and replication cohorts. EpiScore analysis revealed differences in markers of inflammation (CRP), vascular remodeling (MMP‑12), visceral adiposity (Waist:Hip Ratio), and behavioral factors (Alcohol). Notably, Alcohol and WFIKKN2 remained significant even after adjustment for blood biomarkers, suggesting that these EpiScores capture aspects of cardiometabolic risk that are independent of routine metabolic parameters. Gene set enrichment analysis of differentially methylated positions identified pathways that can be related to chronic inflammation, GPCR signaling, metabolic dysregulation and mitochondrial dysfunction. Key findings were reproduced in an independent male atherosclerosis cohort, with effect size patterns that replicated those observed in the original dataset.

Among 36 epigenetic clock models tested, only second‑generation (GrimAge family, DNAmPhenoAge, DNAmFitAge) and third‑generation (DunedinPACE) clocks consistently distinguished cases from controls. In contrast, none of the 19 first‑generation clocks (e.g., Horvath, Hannum) detected significant age acceleration, with effect sizes consistently near zero. This divergence likely reflects their training targets: first‑generation clocks model chronological age, whereas second‑ and third‑generation models incorporate clinical biomarkers, plasma proteins, or longitudinal decline rates. Consequently, the latter are inherently more sensitive to disease‑induced physiological disruption, even in uncomplicated disease states. This pattern is consistent with a study on cardiovascular health, in which first‑generation clocks (Hannum, Horvath) were not significantly associated with cardiovascular health score, whereas second‑generation clocks (DNAmPhenoAge, GrimAge) showed highly significant associations [38]. The robustness of second‑generation clocks in cardiovascular pathology is further supported by large cohort studies: GrimAge acceleration has been linked to declines in cardiovascular health in the CARDIA (Coronary Artery Risk Development in Young Adults) and Framingham Heart studies [36], to incident cardiovascular events, stroke, and heart failure in MESA (Multi-Ethnic Study of Atherosclerosis) [31], and to an increased risk of atrial fibrillation (19% per 5‑year increment) [40].

Most epigenetic studies of cardiovascular disease use mixed‑sex cohorts and adjust for sex as a covariate [32, 35], implicitly assuming that epigenetic‑disease associations are identical in both sexes. The few studies that have stratified by sex consistently report less favorable epigenetic profiles in men, including higher epigenetic age in blood, saliva, and brain tissue across multiple ethnic groups [128], a mean epigenetic age acceleration of + 1.0 years compared to -1.0 years in women accompanied by poorer cardiovascular health scores [129], and positive age acceleration according to GrimAge, Horvath, and DunedinPACE in male but not female residents of Mediterranean Blue Zones [130]. These observations point to a male‑specific vulnerability that mixed‑sex analyses may obscure. By focusing exclusively on men, our study provides a detailed male‑specific reference that would be impossible to extract from a sex-agjusted design. We provide the first comprehensive description of the landscape of epigenetic age acceleration, DMPs, enriched biological pathways, and EpiScore profiles in men with cardiometabolic risk. We further demonstrate that first‑generation clocks are insensitive in this context – a finding that may help reinterpret weak or inconsistent associations previously reported in mixed cohorts, as those signals could be driven by female participants or by sex‑interaction effects. Finally, we validate second‑/third‑generation clocks and certain EpiScores specifically in a male‑only cohort, confirming their utility in a population known to have higher baseline epigenetic age [129] and poorer cardiovascular health profiles [9, 10]. Although the formal discovery of sex‑specific traits awaits the analysis of female DNA methylation data for cardiometabolic risk, our male‑specific findings provide a foundation for future hypothesis‑driven comparisons.

The shortening of DNAmTL observed in our cases corroborates findings from the large NHANES cohort, in which each 1 kb increase in DNAmTL was associated with a 53% reduction in cardiovascular disease risk [94]. The absence of significant between‑group differences in blood cell composition estimates is consistent with data from other population studies that have not identified strong associations between the proportions of major immune cell types and cardiovascular disease risk [131]. This suggests that the epigenetic differences we observed are not driven by shifts in cell populations, but may reflect more subtle molecular changes associated with cardiometabolic risk.

Enrichment of GPCR‑related pathways is particularly interesting in the context of sex differences in cardiovascular disease. The G protein‑coupled estrogen receptor (GPER) mediates many of the protective cardiovascular effects of estrogens in premenopausal women, including vasodilation, reduction of cardiac hypertrophy, and attenuation of vascular inflammation. Activation of GPER has been shown to protect against atherosclerosis and pressure overload‑induced cardiac remodeling, and higher GPER expression in females contributes to their relative resistance to these pathologies [132]. Thus, the dysregulation of GPCR signaling pathways observed in our male cohort may be particularly detrimental given the absence of this female‑specific protective axis. Three genes with DMPs in both the original and replication cohorts – PRSS23, CLEC3B, and LMNA – warrant consideration as potential cross‑phenotype epigenetic markers. PRSS23 encodes a serine protease involved in tissue remodeling [133]; CLEC3B (Tetranectin) protects cardiomyocytes from hypoxic damage via the PI3K/Akt pathway and is downregulated in cardiovascular patients [134, 135]; and LMNA variants are a known cause of familial dilated cardiomyopathy and familial partial lipodystrophy type 2 (FPLD2), linking cardiac and metabolic pathology [136].

Although our results reveal consistent associations between markers of epigenetic aging and cardiometabolic burden in men, the current study design precludes definitive conclusions regarding their clinical utility in diagnosis, prognosis, or treatment response. The identified markers may serve as candidates for generating hypotheses in future prospective studies assessing their potential for early risk stratification or monitoring. If these associations are confirmed in longitudinal cohorts with clinical endpoints (e.g., cardiovascular events, mortality), they may influence clinical decision-making.

Several limitations should be acknowledged. The modest sample size of the original dataset limits statistical power and may have contributed to false negatives. Grouping patients with multiple diseases into a single case group precludes conclusions about disease-specific methylation signatures. The study’s restriction to men, while reducing biological heterogeneity, precludes direct sex comparisons and limits generalizability to women. The observational case‑control design does not establish causality: the observed methylation differences may be consequences rather than causes of disease. Covariate adjustment was limited to chronological age and selected blood biomarkers; other factors (medication, lifestyle, genetic variation) that may influence analysis results were not included. DNA methylation was assessed only in whole blood, and tissue‑specific changes in vascular wall, heart, or adipose tissue may differ. The analysis focused solely on DNA methylation and did not integrate transcriptomic, proteomic, or genetic data. Differences in EWAS algorithms or methylation platforms (EPICv1 in the replication dataset vs. EPICv2 in the original) may affect reproducibility. The lack of longitudinal data on cardiovascular outcomes, treatment response, or disease progression limits interpretation of these results to hypothesis generation and requires prospective validation before clinical applications can be considered. Finally, most epigenetic clocks rely on linear models, which may miss non‑linear relationships. Despite these limitations, the reproducibility of core findings in an independent cohort strengthens their validity and suggests that the identified epigenetic signatures reflect fundamental pathophysiological processes in male cardiometabolic risk.

Abbreviations

ANCOVA

Analysis of covariance

ANOVA

Analysis of variance

ASCVD

Athero sclerotic cardiovascular disease

CKD

Chronic kidney disease

CLES

Common language effect size

CRP

C-reactive protein

CVD

Cardio vascular disease

DMP

Differentially methylated position

DNA

Deoxyribo nucleic acid

DNAm

DNA methylation

EWAS

Epigenome-wide association study

GPCR

G protein-coupled receptor

GSEA

Gene set enrichment analysis

IsoMap

Isometric mapping

MD

Metabolic disorders

MDS

Multidimensional scaling

PCA

Principal component analysis

QRISK3

Qresearch RISK estimator version 3

RNA

RiboNucleic acid

SCORE2

Systematic coronary risk evaluation 2

t-SNE

T-distributed stochastic neighbor embedding

Author contributions

Conceptualization: AK, IY, MI; Methodology: AK, IY, MI; Software: AK, IY; Formal analysis: AK, IY; Resources: FB, NB, EZ; Writing—Original Draft: AK, IY, MI; Writing—Review & Editing: AK, IY, FB, NB, EZ, CF, MI; Visualization: AK, IY; Supervision: EZ, CF, MI. All authors have read and agreed to the published version of the manuscript.

Funding

This work was supported by the Ministry of Economic Development of the Russian Federation (grant No 139–15-2025–004 dated 17 April 2025, agreement identifier 000000C313925P3X0002).

Data availability

Data on phenotype, epigenetic ages, metrics, blood cell estimates, EpiScores are available in the Supplementary Tables. Raw DNA methylation data are available upon reasonable request from the corresponding author.

Declarations

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

Each participant signed an informed consent and filled out a consent for personal data processing, taking into account the principle of confidentiality (accessibility only to the research group and presentation of data in a common array). The study was approved by the local ethical committee of Nizhny Novgorod State University (protocol No. 11 dated 26 October 2022). All research procedures were in accordance with the 1964 Helsinki Declaration and its later amendments.

Consent for publication

Not applicable.