Unraveling the complexity of skin’s biological aging utilizing epigenetic clocks

Abstract

Background

DNA methylation (DNAm) plays a pivotal role in regulating gene expression and tissue function in the skin, which exhibits a high degree of responsiveness to environmental and lifestyle factors. These factors are believed to contribute to epigenetic drift, a hallmark of aging marked by increased methylation variability and changes in regulatory regions. While epigenetic clocks have advanced our understanding of skin aging, the effects of many modifiable factors on the skin methylome remain largely unknown.

Methods

We analyzed DNAm data from 851 participants in a population-based cohort and comprehensive phenotyping of 326 lifestyle, physiological, and pharmacological factors. The DNAm age was estimated using a published skin-specific epigenetic clock, and associations with individual factors were tested using regression models. Epigenome-wide association studies have identified differentially methylated positions linked to significant factors, with further analyses examining their genomic context. The broader relevance of these findings was assessed using other established non-skin-specific epigenetic clocks and phenotypic skin aging measures.

Results

Our analysis identified 20 factors associated with decelerated and 17 with accelerated DNAm age in human skin, reflecting both positive and negative associations with epigenetic aging. We observed that factors associated with DNAm age acceleration tended to coincide with reduced methylome variance, a feature of epigenetic drift, while factors associated with DNAm age deceleration were mapped to methylation differences in transcription elongation regions, supporting transcriptional integrity. Intervention analyses showed that compounds, such as dihydromyricetin and aspirin, are associated with methylation patterns consistent with decelerated epigenetic aging. Several associations were validated in an independent cohort and were consistent across both skin-specific and general epigenetic clocks, suggesting broader relevance of these findings.

Conclusions

These findings suggest that both environmental exposures and certain interventions are associated with variations in the epigenetic trajectory of skin aging. The identified modifiable factors associated with DNAm age and skin phenotypes generate testable hypotheses regarding potential determinants of skin longevity. Longitudinal studies and interventional study designs will be needed to evaluate causality.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13148-026-02101-4.

Background

Epigenetics refers to regulation of gene expression without altering DNA sequence, mediated by DNA methylation (DNAm), histone modifications, and non-coding RNA activity, collectively influence the structure of chromatin [1, 2]. DNAm is one of the most extensively studied epigenetic mechanism. This mechanism involves the covalent addition of a methyl group to the fifth carbon of cytosine, predominantly at CpG dinucleotides, catalyzed by DNA methyltransferases (DNMTs) [3]. The maintenance of DNAm is facilitated by DNMT1, which is established de novo by DNMT3A/3B, and can be reversed through TET-mediated oxidation [4].

DNAm exhibits a dynamic responsiveness to a wide range of environmental stimuli [5]. The impact of methylation changes depends on their genomic context, with DNAm in promoters most often leading to gene silencing [6, 7], whereas methylation outside promoters can have more variable and less predictable effects on gene networks and cellular function [8, 9]. For instance, hypermethylation in promotors and other regulatory regions has been linked to the transcriptional repression of genes [10, 11]. In contrast, alterations in repressed regions may not directly modulate the expression of neighboring genes but could serve as indicators of broader epigenomic instability [12, 13]. With aging, DNAm undergoes epigenetic drift characterized by a stochastic loss of methylation precision and a shift away from the typical bimodal methylation pattern [14, 15]. This process is shaped by risk and protective factors [16, 17], and is associated with an increased incidence of impaired regenerative capacity and decline in skin homeostasis [18, 19]. Collectively, these phenomena contribute to the visible and functional decline that is observed in the aging process [20]. Therefore, a comprehensive understanding of the drivers and consequences of epigenetic drift is fundamental to our understanding of adaptation, resilience, and age-related decline.

As age-related DNAm deviations accumulate [21], they induce predictable alterations at specific CpG sites, which can be quantitively integrated into epigenetic clocks that estimate DNAm age [22]. The Horvath pan-tissue epigenetic clock marked a major advance established this concept by using 353 CpGs [23]. However, Horvath’s analysis and subsequent studies revealed that this pan-tissue epigenetic clock performs notably worse in skin and skin-derived fibroblasts, with higher prediction errors and lower correlation to chronological age compared to other tissues [24, 25]. This limitation is attributed to the unique, tissue-specific methylation dynamics of skin, which are not fully captured by pan-tissue models. To address these limitations, tissue-specific epigenetic clocks have been developed, enabling more precise assessments of DNAm age and health [26]. For instance, the Bormann clock, which was trained specifically on epidermal methylomes, demonstrates superior accuracy in skin, with lower mean absolute errors and higher correlation coefficients compared to the pan-tissue epigenetic clock [25, 27]. Similarly, the Horvath “skin & blood” clock, which uses 391 CpGs (with minimal overlap with the original pan-tissue set), outperforms the pan-tissue epigenetic clock in skin, fibroblasts, and blood, achieving a mean absolute error of 2.6 years and a correlation coefficient of 0.91 in fibroblasts [24]. These findings are consistent with those from multiple studies, with tissue-specific epigenetic clocks demonstrating enhanced performance in their target tissues. While pan-tissue epigenetic clocks generally exhibit optimal performance in blood, they exhibit reduced accuracy in skin, lung, and other non-blood tissues. The enhanced precision of epigenetic clocks has important practical implications. These tools are increasingly applied in observational studies to determine correlates of accelerated or decelerated biological aging and to monitor potential modulation of DNAm age [28–31]. Recent trials have leveraged these clocks to evaluate the impact of pharmacological agents, lifestyle modifications, and bioactive compounds on biological aging rates. For instance, interventions such as caloric restriction [32], exercise [30, 31], and specific drugs [33, 34] have been demonstrated to decelerate or even reverse DNAm age acceleration, as measured by alterations in DNAm patterns in blood.

While epigenetic clocks provide targeted insights into biological aging by aggregating information from selected or weighted CpG sites, Epigenome-Wide Association Studies (EWASes) offer a complementary, unbiased strategy for exploring the broader landscape of DNAm patterns. EWAS is a genome-wide analysis that interrogates hundreds of thousands of CpG sites across the methylome to identify differentially methylated positions (DMPs) and regions (DMRs) associated with phenotypes, exposures, or diseases and uncovering the effects of diverse variables [35]. These methylation changes, identified by EWAS, are not static. They evolve in response to both endogenous processes and external stimuli, thus serving as biomarkers of health and disease states [36, 37]. For instance, significant associations between DNAm in the ELOVL2 gene and aging have been reported, indicating that specific loci may serve as biomarkers for DNAm age [38]. Furthermore, integrating EWAS findings with epigenetic clock analyses allows for a comprehensive view of how these methylation profiles correlate with aging processes across various populations. As the cumulative effects of environmental factors and lifestyle choices shape DNAm patterns, EWAS can pinpoint specific mechanisms by which these factors influence epigenetic aging and disease susceptibility [39].

Among the various tissues where these approaches are being applied, skin stands out as a particularly valuable model tissue for epigenetic research. As the body’s primary barrier to environmental insults, it is constantly exposed to ultraviolet radiation, pollutants, and microbes. Because of this direct exposure, along with the skin’s accessibility and the visibility of its phenotypic changes, skin provides a valuable context for studying how environmental and lifestyle factors may influence epigenetic regulation and aging processes. These epigenetic dynamics are modulated in a complex manner by a variety of factors, including diet, physical activity, and exposure to environmental stressors [40, 41]. These variables, in conjunction with lifestyle factors, influence the methylome, contributing to the phenomenon of epigenetic drift [42–44]. In this context, emerging work indicates that targeted strategies can modulate the cutaneous methylome, with evidence of partially restoring youthful patterns and delaying age-related phenotypes [45]. Although the study of DNAm continues to develop, several aspects of skin epigenetics remain incompletely characterized. To date, research has predominantly concentrated on the effects of aging or sun/UV light exposure on DNAm patterns, which are critical factors in skin health and disease [46, 47]. Additionally, there has been a focus on skin DNAm in the context of specific skin-related diseases, including atopic dermatitis (AD), leprosy, psoriasis vulgaris, acne vulgaris, systemic lupus erythematosus (SLE), malar melasma, cutaneous squamous cell carcinoma (cSCC), and systemic sclerosis [48–55]. Although intrinsic aging and UV exposure have been studied extensively, the roles of lifestyle factors, pharmacological interventions, and physiological changes in shaping skin’s DNAm age remain underexplored. Their established effects in other tissues suggest they may also influence epigenetic regulation in skin [45, 56, 57]. Addressing this aspect could serve as a pivotal step towards decoding the epigenetic complexities that underly skin health and disease, potentially leading to personalized epigenetic therapies.

To directly address this gap, we identified associations between epidermal DNAm age and 326 lifestyle, physiological, and pharmacological factors in a large, deeply phenotyped population-based cohort. We hypothesized that these variables could induce distinct, region-specific epigenetic signatures that shape the pace of skin aging and the integrity of key regulatory elements (age-accelerating factors: positively associated with skin DNAm age acceleration; age-decelerating factors: negatively associated with skin DNAm age acceleration). We adopted an integrated approach, incorporating skin-specific epigenetic clock predictions [25], EWAS to identify factor-associated DMPs, and analyses of methylation changes across diverse genomic regions. To further validate and extend our findings, we examined associations with phenotypic skin aging in an independent cohort using the Skinly device, an AI-powered platform that integrates optical imaging and moisture sensing to provide objective measurements of skin age [58]. Furthermore, we assessed the broader relevance of these factors by examining their associations with established non-skin-specific epigenetic clocks, chosen for their complementary insights into epigenetic aging. The panel included clocks developed by Hannum [59], Horvath (skin-blood) [23], Shireby [60], McEwen [61], and Levine [62], each characterized by distinct tissue sources and tailored to predict either chronological age or healthspan-related phenotypic age. This comprehensive framework aims to advance our understanding of the molecular determinants of skin aging and to inform the development of targeted interventions for skin longevity.

Methods

Dataset employed in the study

This study utilized genome-wide DNAm data generated from epidermal samples collected via suction blister method from 851 participants (mean age 56.82 years, SD 12.45, range 28–88 years, 54.88% male, 45.12% female) enrolled in the population-based SHIP-TREND-1 cohort from Western Pomerania, Germany [63]. The age distribution was non-uniform, with a higher proportion of participants in the 45–70 years age group. Within the cohort, the prevalence of diabetes mellitus was 8.22%, and 65.80% of participants had at least one chronic illness. Current smoking was reported by 17.63% of participants, with a mean of 15.32 smoking pack-years (SPY) among current smokers. The mean daily alcohol intake was 10.95 g. Obesity (BMI ≥ 30) was present in 31.02% of participants, and 27.61% were classified as physically inactive. Serum 25-hydroxyvitamin D levels were measured as part of the laboratory assessment, with the majority of participants classified as being in the optimal, suboptimal, or deficient ranges (Supplementary Fig. S1 [min: 6.5, median: 23.6, max: 63.7]). For each participant, a comprehensive set of 326 lifestyle, physiological, and pharmacological factors was systematically assessed, encompassing anthropometric, cardiovascular, laboratory, lifestyle, medical history, psychological, dietary, sleep, and other health-related variables, all collected through standardized clinical examinations, structured interviews, and laboratory assessments [63] DNA was extracted from the epidermal samples, and genome-wide DNA methylation profiling was performed using the Illumina Infinium MethylationEPIC v1.0 BeadChip array, with all sample collection, processing, and analysis procedures as previously described [64].

Data pre-processing and quality control

The raw .idat files from the EPIC arrays were pre-processed using the R package minfi [65, 66]. Detection p-values were calculated for each methylation locus (probe) in each sample via a one-sided normal test comparing observed probe intensity to a background distribution estimated from negative control probes [65, 66], and probes with high detection p-values (p > 0.01) were excluded. Additional filtering removed probes known to be cross-reactive, as identified by EPIC-specific annotations provided by maxprobes R package [67]. SNP-overlapping probes were subsequently filtered using the ‘dropLociWithSnps’ function from minfi [65, 66], which removes loci containing known genetic variants at CpG sites, single base extension positions, or within probe bodies based on dbSNP annotation data [68]. The methylation data were then normalized together using the ‘preprocessQuantile’ function [65, 66]. Ultimately, the beta values were utilized for downstream analyses.

Determining age with epigenetic clocks

To estimate DNAm age, several epigenetic clocks were applied to the DNAm data. For the skin-specific Bormann clock [25], beta values obtained from preprocessing were first converted to M values using the R package lumi [69]. These M values were then used as input for the Bormann clock to generate age predictions.

Additionally, we incorporated a panel of well-established, non-skin-specific epigenetic clocks, selected for their broad applicability to different tissue types and their capacity to capture diverse aspects of biological aging. This panel comprised several chronological age predictors, including the Hannum (whole blood) [59], Horvath (skin-blood) [23] , Shireby (brain cortex) [60] , McEwen (buccal epithelial cells) [61] clocks. Complementing these, we included the Levine clock [62] which was tailored to predict mortality risk and integrate clinical phenotypes (PhenoAge) based on blood DNAm patterns. The R package dnaMethyAge [70] was used to estimate DNAm ages with beta values using these clocks.

Assessing the impact of various factors on epigenetic age prediction

Linear regression models were used to assess the relationship between each of the 326 systematically assessed factors and DNAm age predicted by the Bormann clock [25]. These factors encompassed a comprehensive range of lifestyle, physiological, and pharmacological variables. For each factor, a separate model was fitted with DNAm age as the dependent variable and the investigated factor as the main independent variable, adjusting for chronological age, gender, BMI, and SPY to control for potential confounders. The general formula of the model is given by:

1

where factor_i represents the investigated factor in each separate model.

Statistical significance for the associations was determined based on the raw p-value of the coefficient for factor_i in Eq. 1, with p < 0.05 considered significant (t-test). No multiple testing correction was applied at this stage. All reported associations are based on raw p-values, and the analysis is explicitly hypothesis-driven. As such, these results should be interpreted as exploratory and do not imply causality or directionality of effect.

For those associations that reached statistical significance, effect sizes were calculated using the R package effectsize [71] and interpreted to determine both the magnitude and the direction of the relationship. Continuous predictors are presented as effect sizes per unit increase, and their units can be found in Supplementary Table S1. Reference categories for binary predictors are provided in the same table. Additionally, significant factors were further examined by applying linear models in the same way replacing the outcome variable (DNAm age) by the prediction of each of the non-skin-specific epigenetic clocks.

To assess whether associations reflected independent effects among correlated variables, we fitted multivariable models including related factors simultaneously. Collinearity among predictors was evaluated using variance inflation factors (VIFs). A VIF threshold of 5 was used to indicate moderate collinearity, and values above 10 were considered high.

Ggenome-wide determination of DNAm association with factors of interest

For each factor that was significantly associated with DNAm age as predicted by the Bormann clock [25], a separate EWAS was performed. In this context, each EWAS consisted of fitting a linear model to every CpG site across the genome, using the R package limma [72]. In these models, the investigated factor was included as the variable of interest in the design matrix, with adjustments made for chronological age, sex, BMI, and SPY to account for potential confounders. The ‘lmFit’ function was used to fit a linear model to each CpG site across the genome. The general formula of the model for each CpG site is given by:

2

where GpG_j represents the β value at the jth CpG site, and factor_i is the investigated factor.

Empirical Bayes statistics were then used to accurately identify DMPs with function ‘eBayes’. P-values were adjusted via the Benjamini–Hochberg (BH) procedure and an adjusted p-value of less than 0.05 was deemed significant.

Chromatin state analysis

The top five age-accelerating and age-decelerating factors based on the highest number of DMPs were used to analyze DMP abundance in chromatin regions. To determine the location of DMPs in specific regulatory regions, the ChromHMM annotations for keratinocytes (NHEK cell line) were obtained from the UCSC database [73, 74] and used for the grouping of DMPs with defined chromatin states according to the genomic coordinates. Finally, the abundance of DMPs on chromatin regions was determined.

Integration of differentially methylated probes with gene annotations, transcription factor targeting and pathway enrichment analysis

Identified DMPs were annotated to genes based on the Infinium EPIC manifest file provided by Illumina [75]. Only genes with at least 4 DMPs associated to the same factor were used for further analysis. The corresponding length and number of transcript variants for each associated gene was extracted from the gtf-file (human genome version GRCh37) provided by Ensembl release 87 [76].

The observed ratios were calculated for different types of genes: housekeeping [HRT Atlas v1.0, [77]], skin-specific (547 genes that display elevated expression compared to other tissues, downloaded from The Human Protein Atlas [78, 79], epigenetic factor [EpiFactors Database v2.0, [80, 81]], and transcription factor (TF) genes [TFLink, [82]]. Then the calculated observed ratios were divided by the expected ratios derived from all genes on the EPIC methylation array resulting in the observed to expected (o/e) value.

The identified transcription factors were grouped into their respective families using the annotation provided by TFClass [83] and the corresponding observed to expected (o/e) value of each transcription factor family was calculated. Additionally, the pre-defined table provided by TFLink, in which transcription factors were linked to their target genes, was used to determine the extent to which the transcription factors targeted the DMP associated genes. Only transcription factors targeting at least 1,500 genes present on the EPIC methylation array were used for calculating the proportions of genes targeted by each transcription factor and the corresponding observed to expected (o/e) ratios.

Finally, gene set overrepresentation analyses were performed with the list of genes associated to age-decelerating and age-accelerating factors separately using the R package clusterProfiler [84–87] with default settings when applied to pathways from publicly available databases KEGG [88–90] and Gene Ontology (GO) terms [91, 92] . When applied to a defined set of Hallmarks of Aging pathways [93], the cutoff parameters for p-value and q-value were set to 1 and the maxGSize to 5,000 to obtain results for visualization of all pathways assigned to the Hallmarks of Aging. In downstream analyses the overrepresentation in a Hallmark of Aging pathway was considered significant with an adjusted p-value < 0.05.

Analysis of DNAm alterations linked to chronological aging and aging dynamics

The intra-methylome variance was determined by calculating the variance for each CpG site across the samples corresponding to each of the two age groups, with “young” (individuals younger than 40 years) and “old” (individuals older than 70 years). To compare the overall distribution of these per-CpG variances between the two groups, we used the Wilcoxon rank sum test (p < 0.05), which assessed whether the median variance differs between young and old individuals.

To investigate if the methylation changes of the DMPs associated with factors aligned with the methylation change upon chronological aging, the log2 fold change of DMPs associated with the top seven age-decelerating and top seven age-accelerating factors that exhibit the highest fraction of DMPs overlapping with CpGs associated with age-dependent methylation changes were extracted from the EWASes. The methylation change upon chronological aging was determined between the “young” (< 40 years) and “old” (> 70 years) age group as describe previously [45].

Intervention strategies

To investigate the effects of metabolic and epigenetic interventions on factor-associated CpGs, we analyzed two DNA methylation datasets generated using the Illumina Infinium MethylationEPIC BeadChip v1.0 array. In the in vitro dataset, primary human keratinocytes were treated for 14 days with either bezafibrate (n = 4 treated, n = 4 controls) or nicotinamide adenine dinucleotide (NAD, n = 5 treated, n = 5 controls) [94]. In the in vivo dataset, 30 female volunteers aged 31–65 years participated in a vehicle-controlled study. Dihydromyricetin (DHM) was topically applied to the lower back for 2 weeks, followed by repetitive low-dose UV exposure as an aging stimulus. Epidermal samples were subsequently collected for methylation profiling [45]. Differentially methylated probes induced by the DHM-treatment were determined as described previously [45] and the DMPs induced by the Bezafibrate and NAD treatment, respectively, were determined by applying the same analysis processes as describe previously [45].

Determining age with photo age clocks and assessing the impact of identified factors on phenotypic age prediction

Skinly is a handheld IoT device which combines a moisture sensor with a multi-light-source camera to capture detailed skin images, which were analyzed using deep learning algorithms. The device is paired with a smartphone app, enabling users to perform daily comprehensive skin analyses and measurements at home, providing results comparable to traditional laboratory methods [58].

Using the Skinly ecosystem, a study was conducted in Germany with 2,854 female and 222 male participants, aged between 18 and 87 years, all of whom reported their smoking status. Participant data included demographic details such as age, gender, weight, height and ethnicity, as well as lifestyle information such as dietary preferences and sleep habits. Responses to various surveys were gathered through the Skinly smartphone application. Skin specific measurements including redness, hydration and wrinkledness were measured using the Skinly device. Photo clocks [95] were used to predict participants’ body age based on images captured by the Skinly device and participants’ selfie age based on selfies obtained from participants smartphones.

To assess the relationship between factors and the phenotypic age, linear regression models were performed in a similar way as described for the DNAm age, again adjusting for age, gender, BMI, and SPY to control for potential confounding effects. The outcome variable in these analyses were Body Skin Age and Selfie-Based Facial Age derived from the photo clock. Ordinary Least Squares (OLS) regression was conducted using the statsmodels package in Python [96].

Results

Diverse associations of lifestyle, physiological, and pharmacological factors with epigenetic age changes in the skin

To investigate factors influencing DNAm age in human skin, we performed an analysis of epidermal samples from 851 participants in the SHIP-TREND-1 cohort [63] (Fig. 1A). Epidermal skin samples were collected, and information on 326 lifestyle, physiological, and pharmacological factors was obtained through clinical exams, lab tests, and questionnaires. This comprehensive dataset allowed us to explore how various factors impact DNAm aging in the skin.

Fig. 1.

External and internal factors identified as associated with the DNAm age of human skin. A Schematic representation of the design of SHIP-TREND-1 study comprising 851 volunteers (females and males). Data types included questionnaire responses, collection of epidermal skin samples (suction blisters) for downstream EPIC Methylation array analysis, blood samples for various tests, and biophysical skin measurements. B Pie chart showing the fraction of tested factors classified as age-accelerating (red), age-decelerating (blue), and those not significantly associated with skin DNAm age (gray), as determined via linear regression models and raw p-value < 0.05 (see Methods for details). C Bar plots depicting the effect size of age-decelerating (left panel) and age-accelerating (right panel) factors. Factors were categorized into “lifestyle” (purple), “pharmacological” (green) and “physiological” (red) factors

Linear regression models, adjusted for gender and chronological age as well as known key confounders SPY [97, 98] and BMI [99, 100], were used to test the association between each factor and DNAm age predicted by the Bormann skin-specific epigenetic clock [25]. The magnitude of the observed associations was subsequently quantified using effect size calculations. This analysis identified 20 age-decelerating and 17 age-accelerating factors (raw p < 0.05; Fig. 1B). Notably, DNAm age deceleration was observed in individuals with higher skin elasticity, greater melanin content, higher skin moisture, regular swimming activity, higher levels of educational attainment, regular use of low-dose acetylsalicylic acid (aspirin), and a higher tanning response. In this context, a higher tanning predisposition score reflects a tendency to tan rather than redden when exposed to sunlight without protection. Individuals who almost always tan and rarely experience redness showed slower epigenetic aging in the skin.

Conversely, DNAm age acceleration was associated with an increased propensity for skin sensitivity, a documented history of eczema, redness upon unprotected sun exposure, regular use of solarium, a higher number of childhood sunburns, a longer duration of diabetes mellitus, and higher blood levels of 25-hydroxyvitamin D (Fig. 1C). These findings suggest that both lifestyle and physiological factors, as well as preventive pharmacological interventions such as regular low-dose aspirin use, are associated with variation in the rate of epigenetic aging in human skin.

When related factors were included together in the models, we observed moderate to high multicollinearity within certain variable clusters, specifically among medication use (e.g., aspirin and anticoagulants), tanning-related characteristics (tanning predisposition and erythema after sun exposure), and education-related factors (Supplementary Table S2). However, due to the observational nature of our study, we cannot infer causality of these associations.

Chromosomal disparities among DMPs reflect complex aging dynamics in skin biology

To elucidate the associations between DNAm patterns and the set of lifestyles, physiological and pharmacological factors that are significantly associated with DNAm age changes, we performed a series of EWASes. Although the analyzed factors were significantly associated to DNAm age, the EWAS analysis did not identify a significantly associated CpG for each factor. Specifically, for 25% of age-decelerating factors and 24% of age-accelerating factors, no associated CpG was detected (Fig. 2A). Furthermore, the number of DMPs strongly varied between the factors, independent of their effect size (Supplementary Fig. S2), whereas age-decelerating factors tend to affect more CpGs compared to age-accelerating factors (Fig. 2B), indicating that age-decelerating and age-accelerating factors might have a different impact on the DNAm pattern.

Fig. 2.

Chromosomal disparities among DMPs reflect complex aging dynamics in skin biology. A Pie chart displaying the fraction of factors without significantly associated DMPs (adjusted p < 0.05) among the age-decelerating (blue, right) and age-accelerating (red, left) factors. B Bar plot showing the number of significantly associated DMPs (adjusted p < 0.05) for the top five age-decelerating (blue) and age-accelerating (red) factors, respectively. C Heatmap illustrating the chromatin states annotation [73, 74] of DMPs (adjusted p < 0.05) for each of the top five age-decelerating (top) and age-accelerating (bottom) factors. The number of DMPs were scaled and displayed as the row z-score

To investigate the chromosomal distribution of DMPs for each associated factor, we set a minimum threshold of 500 DMPs to ensure robust enrichment analysis. For age-accelerating factors, only five met our threshold. Although nine age-decelerating factors exceeded this threshold, we also selected five with the highest number of DMPs from this group to maintain a balanced comparison between age-accelerating and age-decelerating factors. Enrichment analysis of DMPs across chromosomes was conducted for the selected factors.

To explore methylation changes associated with factors on specific regulatory regions, we annotated the DMPs to functionally distinct segments of the keratinocyte genome using ChromHMM segmentation (see Methods for details). Many of the factor-associated CpGs were in heterochromatin regions, which was independent of the direction of the association of the factor with DNAm age (Fig. 2C). Notably, comparing the location of CpGs associated with age-decelerating factors to CpGs of age-accelerating factors, age-decelerating factors were more often found in transcription elongation and weakly transcribed regions, while those of age-accelerating factors, except of “work score (Baecke)” (a measure of occupational physical activity) [101], are more frequent in repressed regions (Fig. 2C). A Mann Whitney U-test confirmed this impression (Supplementary Fig. S3). These results suggest different mechanisms via which age-decelerating and age-accelerating factors might act. To ensure the robustness of these findings, we repeated the chromatin state enrichment analysis including all nine age-decelerating factors with more than 500 DMPs, and observed that the key qualitative patterns remained unchanged (see Supplementary Fig. S4).

Age-associated genes highlight the role of epigenetic and transcriptional regulation

To further investigate the regulatory implications of differential methylation within the context of age deceleration and acceleration, we annotated the factor-associated CpGs to genes based on the EPIC manifest file (see Methods for details). A higher number of genes were annotated to age-decelerating factors compared to age-accelerating factors, which is likely due to the higher number of DMPs associated to age-decelerating factors (Fig. 3A). Further characterization of the annotated genes clearly showed that factor associated genes were significantly longer (Wilcoxon p < 0.001) with significantly more transcript variants (Wilcoxon p < 0.001) than non-factor associated genes (Fig. 3B), which was independent of the number of CpGs within the gene body covered by the EPIC chip (Supplementary Fig. S5). This indicates that the identified factors may target specific types of genes.

Fig. 3.

Age-associated genes highlight the role of epigenetic and transcriptional regulation. A Bar plot displaying the number of annotated genes among the top five age-accelerating (red, right) and age-decelerating (blue, left) factors represented as percentage of EPIC genes (> 3 probes) linked to each factor, with genes assigned only if more than three DMPs (adjusted p < 0.05) were associated. B Box plots of gene length (left) as well as number of transcript variants (right) of genes associated to no (gray), age-decelerating (blue), and age-accelerating (red) factors. C Bar plot showing the observed-to-expected (O/E) ratios of specific gene types (housekeeping, skin-specific, epigenetic factors and, transcription factors, see Methods for details) among the age-deceleration-associated genes (blue) and age-acceleration-associated genes (red). D Scatter plot comparing the observed-to-expected ratios of transcription factor families (see Methods for details) among the age-deceleration-associated genes (x-axis) and age-acceleration-associated genes (y-axis). E Scatter plot showing the comparison of the observed-to-expected ratios of transcription factors targeting age-decelerating (x-axis) and age-accelerating (y-axis) factor-associated genes (see Methods for details). F Scatter plot comparing the fold enrichment of the Hallmarks of Aging pathways [93] obtained by overrepresentation analysis for age-accelerating (y-axis) and age-decelerating (x-axis) factor-associated genes. The dotted lines in panel D, E and F highlight the enrichment in age-decelerating (blue) and age-accelerating (red) factor-associated genes, respectively

To identify the potential type of targeted genes we categorized the factor-associated genes into housekeeping, skin-specific, epigenetic factor and TF genes using publicly available lists of genes and analyzed their enrichment via calculating their observed to expected ratios (see Methods for details). Although we analyzed the association of different lifestyle, physiological and pharmacological factors on human skin aging using the skin specific Bormann clock [25] and epidermal samples, skin specific genes were not enriched in factor-associated genes (Fig. 3C). Notably, epigenetic factor and especially TF genes showed higher enrichment scores (Fig. 3C). Analysis of the respective TF families revealed distinct groups for age-accelerating and age-decelerating factors (Fig. 3D). The top TF families among the genes associated with age-accelerating factors were TBX-related factors, DEAF and LRRFIP-factors. In contrast, among the genes associated with age-decelerating factors the top TF families were CpG-binding proteins, Foxhead box factors and Myb/SANT domain factors.

We then expanded our TF analysis by assigning factor-associated genes to downstream targets of TFs to determine TFs which could potentially regulate more factor-associated genes than expected (see Methods for details). This analysis revealed that, in contrast to TF families, genes associated with age-decelerating and age-accelerating factors share some of the upstream regulating TFs (Fig. 3E). Transcription factors like METTL3, HMGB1, CRY1 and ZNF398 were among the top ten enriched TFs targeting factor-associated genes. These results indicate that the factor-associated genes might be regulated by similar upstream mechanisms.

As the physiological decline upon aging impairs a common set of cellular processes known as hallmarks of aging [102], we investigated the enrichment of factor associated to genes within those nine hallmarks of aging by applying a combined list of genes from GO and Reactome pathways as described previously [93]. The hallmarks “stem cell exhaustion”, “altered intracellular communication” and “deregulated nutrient sensing” where among the top enriched hallmarks of aging for age-deceleration as well as age-acceleration factor-associated genes (Fig. 3F), whereas all three hallmarks were found to be significant when analyzing the age-decelerating and only the first two were significant for the age-accelerating factor-associated genes (Supplementary Fig. S6). When utilizing gene sets from the established database KEGG [88–90], pathways like “focal adhesion”, “Wnt signaling” and “PI3K-Akt signaling” were among the top significantly enriched pathways (Supplementary Fig. S7A), which supports the results observed for the gene sets assigned to the hallmarks of aging (Fig. 3F). Of note, significantly enriched GO [91, 92] molecular functions included “DNA-binding transcription factor binding” and “histone kinase activity” (Supplementary Fig. S7B). This is consistent with the observed enrichment of genes related to epigenetic factor and transcription factors (Fig. 3C).

Analysis of methylation shifts indicates complex interplay between aging dynamics

In further analyses, we also investigated the relationship between methylation dynamics upon aging and DMPs associated with lifestyle, physiological and pharmacological factors. Analysis of the beta values showed a clear bimodal pattern for age-deceleration-associated CpGs as well as non-factor CpGs, whereas the pattern is more pronounced for non-factor CpGs (Fig. 4A). Notably, the bimodal pattern was lost for the age-acceleration-associated CpGs resulting in a more uniform distribution with an average beta-value around 0.5 (Fig. 4A). These results indicate that the age-acceleration-associated CpGs might contribute to the age-related reduction of the intra-methylome variance, which is one feature of age-related epigenetic drift in human skin [25]. We next addressed the question whether factor-associated genes are equally affected by erosion of the methylome with age. We therefore calculated the intra-methylome variance for the age-deceleration and age-acceleration-associated CpGs separating our cohort into young and old age-groups (see Methods for details). In general, the intra-methylome variance was significantly higher (Wilcoxon rank sum test, p < 0.05) for the age-deceleration-associated CpGs than for the age-acceleration-associated CpGs and only slightly decreased upon aging (Fig. 4B). In contrast, the intra-methylome variance of the CpGs associated with age-accelerating factors was significantly reduced for the old compared to the young age-group (Wilcoxon rank sum test, p < 0.05, Fig. 4B) further suggesting a potential contribution of the age-acceleration-associated CpGs to the epigenetic drift. The results were consistent across both sexes (Supplementary Fig. S8).

Fig. 4.

Analysis of methylation shifts indicates complex interplay between aging dynamics. A Violin plot illustrating the distribution of methylation beta values for CpGs categorized into three groups: age-accelerating (red), age-decelerating (blue), and non-factor associated (gray). B Box plot depicting the intra-methylome variance of age-deceleration-associated CpGs (blue, left panel) and age-acceleration-associated CpGs (red, right panel) between young individuals (chronological age < 40 years; lighter color shade) and older individuals (chronological age > 70 years, darker color shade), with significance depicted as (Wilcoxon test, p < 0.05). Significance is indicated by asterisks as follows: * for p ≤ 0.05 and *** for p ≤ 0.001. C Dot plot of the percentage of overlapping factor-associated CpGs that show consistent trends in methylation change between young (< 40 years) and old (> 70 years) groups, with dot sizes representing the number of corresponding DMPs. For age-accelerating factors, the correct directional methylation change is depicted towards an aged methylome (right, red) and for age-decelerating factors towards a young methylome (left, blue). D Bar plot showing the fraction of non (gray), age-decelerating (blue), and age-accelerating (red) factor CpGs among the DMPs upon treatment of keratinocytes with bezafibrate (in vitro), NAD (in vitro) and DHM (in vivo). Compounds for which either a lifespan elongation in C. elegans [106, 107] or a rejuvenation of human keratinocytes have been reported [45, 94]

To further characterize the methylation dynamics of factor-associated CpGs in the aging context, we compared the direction of the methylation change as identified by the EWASes for the top seven associated factors displaying the highest fraction of age-associated CpGs with the direction upon aging (see Methods for details). Of note, for all age-accelerating factors, more than 90% of their associated DMPs shift towards an old methylome (Fig. 4C). These results indicate that age-accelerating factors might indeed lead to a prematurely aged methylome. Even though we identified DMPs associated with age-decelerating factors, their methylation changes only slightly overlapped (< 60%) with the direction towards a young methylome (Fig. 4C). Only the factor “level of education” perfectly (100%) aligned with a methylation shift into the direction of a young methylome (Fig. 4C). These findings suggest that age-decelerating factors might not directly lead to a rejuvenated methylome indicating that they probably act via different mechanisms as the age-accelerating factors.

Clinical trials to elongate the human lifespan and thus promoting longevity have been conducted in humans by either introducing a positive lifestyle e.g., dietary habits or by oral intake of a certain compound like metformin [33, 103–105]. However, it was never investigated if these compounds may modulate the same age-related DNAm patterns as lifestyle choices. We therefore analyzed the DMPs upon compound treatment for which either an extension of the lifespan in C. elegans [106, 107] or a rejuvenating effect in human keratinocytes have been reported [45, 94] including bezafibrate, nicotinamide adenine dinucleotide and DHM. For all three compounds more than the half of their DMPs overlapped with factor-associated CpGs, with the highest fraction found among the DMPs upon DHM treatment (Fig. 4D), which is higher than the overall fraction of factor-associated CpGs present on the EPIC chip (Supplementary Fig. S9). The statistical significance and enrichment of these overlaps for age-deceleration-associated CpGs, which are most relevant to potential “rejuvenation-like” shifts, was confirmed by assessing over-representation using Fisher’s exact test (Supplementary Table S3). These findings indicate that intervention strategies to rejuvenate the skin can to a certain extent reverse the negative and support the positive influence of lifestyle, physiological and pharmacological factors. Interestingly, the fraction of age-deceleration-associated CpGs was higher than the fraction of age-acceleration-associated CpGs, which could be observed for all three compounds (Fig. 4D). This result supports the notion that age-decelerating factors might act in a similar way as rejuvenating intervention strategies.

Diverse associations of factors with skin phenotypic age changes

To investigate if the identified factors associated with DNAm skin age may also impact the skin age phenotype, we made use of the independent Skinly study [58] comprising of 3,076 volunteers (Fig. 5A). The Skinly device and its smartphone application were used to collect the biophysical parameters and lifestyle information as well as images to predict participants’ selfie and body age based on established photo age clocks. We then used these data to perform the association analysis in the same way using the photo age clocks as outcome variables (see Methods for details). We identified five factors which were significantly associated with the skin age phenotype based on at least one of the two applied photo age clocks (raw p < 0.05, Fig. 5B). Specifically, higher body moisture, higher level of education, and longer sleep duration were found to be associated with age deceleration, whereas aesthetic procedures and higher potato consumption were associated with age acceleration. These results are in line with the observed associations between the factors and the DNAm skin age, suggesting that the identified factors might not only impact the DNAm age of the skin but also its phenotypic aging.

Fig. 5.

Image-based clock associations highlight impact of skin aging factors on skin aging phenotype. A Schematic representation of the Skinly study involving 3,076 volunteers. Data types comprised, collection of images for downstream biophysical measurements, tracking, and questionnaire responses. B Dot plot illustrating significant (raw p < 0.05) associations between various factors and two image-based age clocks. Dot size represents strength, and color indicates direction of the association with blue for age-decelerating and red for age-accelerating

Cross-clock associations reveal broader implications of skin aging factors on biological age

To explore whether factors associated with DNAm age in skin also linked to DNAm aging in other tissues, we analyzed age predictions from five established non-skin-specific epigenetic clocks: Hannum (whole blood) [59], Horvath (skin-blood) [23], Shireby (brain cortex) [60], McEwen (buccal epithelial cells) [61], and Levine (PhenoAge) [62], all of which were developed and validated in non-epidermal tissues. All the evaluated epigenetic clocks showed statistically significant positive correlations between their predicted DNAm ages and the chronological ages of the epidermal samples from the SHIP-TREND-1 study. A detailed summary of these correlation strengths and other performance metrics for all clocks, including the Bormann clock [25], is shown in Supplementary Fig. S10. This exploratory cross-clock approach revealed that several factors initially identified in skin such as tanning predisposition, years of diabetes mellitus, skin elasticity, and skin sensitivity, also showed consistent associations with DNAm age across multiple epigenetic clocks (Fig. 6A). Interestingly, regular use of low-dose aspirin was consistently associated with lower DNAm age estimates across all evaluated epigenetic clocks (p < 0.05; Fig. 6B). Collectively, these findings suggest that some factors associated with DNAm age in skin may also be relevant to epigenetic aging in other tissues, although the lack of tissue-specific validation for these epigenetic clocks warrants cautious interpretation. This analysis should be considered exploratory, and the biological interpretation of their predictions in epidermal samples remains uncertain.

Fig. 6.

Cross-clock associations reveal broader implications of skin aging factors on DNAm age. A Dot plot illustrating associations between various factors and the skin-specific epigenetic clock [25], with the x-axis representing t-values derived from linear models (raw p < 0.05). Color indicates the direction of the association (red for age-accelerating and blue for age-decelerating factors), while dot size corresponds to the number of significant validations by other epigenetic clocks [23, 59–62]. B Bar plot illustrating the effect of acetylsalicylic acid on DNAm age deceleration, displaying t-values derived from linear models with the predictions of six different epigenetic clocks (see methods for details)

Discussion

This study presents a systematic, hypothesis-generating assessment of the molecular and phenotypic factors associated with skin aging. Analyses were performed in a large, population-based cohort with deep phenotyping and epidermal DNAm profiling [63]. We tested associations between epidermal DNAm age [25] and 326 lifestyle, physiological, and pharmacological factors. The findings of this investigation suggest a complex landscape in which various factors are associated with differences in the skin methylome and DNAm-based estimates of epigenetic aging. To our knowledge, this is the first systematic screen of potentially modifiable factors associated with DNAm age in human skin. The findings are in line with the results from an independent cohort, where several of the same factors were also linked to objectively measured skin aging phenotypes using the Skinly device [58]. Although not all associations could be confirmed due to differences in measuring skin aging (DNAm vs. image-based), the overlap lends support to the reproducibility of a subset of associations.

While this study offers valuable insights, some limitations should be acknowledged. Firstly, our reliance on self-reported questionnaires may introduce subjective bias or inaccuracies in the data, potentially affecting the reliability of the associations drawn, as the accuracy of recall can vary among participants. Secondly, the relatively small sample size may limit the statistical power of some analyses, particularly for less common exposures or lifestyle factors that warrant more extensive investigation. This limitation underscores the potential that some significant factors may not have been identified or adequately assessed due to this participant constraint. In addition, although attempts were made to validate our findings using the Skinly independent cohort [58], this validation included a limited number of the factors that were available for analysis, and not all associations could be confirmed. One must also consider that the Skinly cohort is predominantly female and consists of self-selected users of a specific device. This may introduce selection bias and limit the representativeness of this sample for broader populations. The background population from which our cohort was drawn was not specifically designed for this type of analysis, and thus important confounders related to unique demographic or environmental factors might have been overlooked, limiting the generalizability of our results. Since the SHIP-TREND-1 cohort consists of a relatively homogeneous Northern German population with a limited range of skin phototypes and specific environmental exposures, our findings may be difficult to generalize to other ethnicities, climates, or phototype distributions, particularly with regard to sun- and pigmentation-related factors. Our analysis also did not include ambient air pollution, which is an increasingly recognized environmental exposure relevant to skin biology. Recent evidence suggests that exposure to particulate matter (PM₁₀, PM_2.5) can modulate systemic biomarkers of oxidative stress and inflammation, such as cortisol and TNF-α, even in non-invasive samples like saliva, and these pathways are likely to influence skin aging as well [108]. Moreover, we noted that certain related variables, such as medication use (including aspirin and anticoagulants), tanning (tanning predisposition and sun exposure erythema) and education-related factors, exhibited moderate or high collinearity, as reflected by elevated VIFs (Supplementary Table S2). This collinearity reduces the interpretability of individual effect estimates for these variables and may compromise the robustness of some associations reported in this study. It is also important to note that no multiple testing correction was applied in the epigenetic clock analysis. All reported associations are based on raw p-values, and the analysis is explicitly hypothesis-driven. Consequently, the findings of this study should be regarded as exploratory in nature, as they do not suggest any causal or directional relationship between the factors under investigation. Future longitudinal or interventional studies, as well as genetic approaches such as Mendelian Randomization, will be necessary to rigorously assess causality. Furthermore, studies in more diverse populations, encompassing a wider range of ethnic backgrounds, skin phototypes, and environmental exposures, will be essential to evaluate the external validity and broader applicability of these findings.

Despite the aforementioned limitations of our study, our approach offers an opportunity to gain new insights into the aging process that goes beyond the current approach of clock deconvolution. Conventional deconvolution approaches focus on identifying aging pathways by examining the CpGs that constitute established epigenetic clocks [109]. However, the specific CpGs included in these clocks can vary widely depending on the training data and algorithm used, leading to substantial variability in clock composition [110, 111]. Typically, only a small fraction of the methylome is incorporated into any given epigenetic clock, with fewer than 1,000 CpGs out of more than 800,000 measured. These CpGs are selected primarily based on their statistical association with chronological age rather than their biological relevance. As a result, many CpGs and pathways that may be equally or even more relevant to aging can be inadvertently excluded, limiting the mechanistic insight provided by such clocks [112, 113]. In contrast, our approach used epigenetic clocks not as endpoints, but as a tool to identify factors that are associated with variation in the pace of biological aging. We then systematically analyzed CpGs associated with these factors, allowing us to explore molecular underpinnings of aging, independently of the specific CpGs chosen by any algorithm. This clock-composition-independent strategy enables a broader investigation of aging mechanisms that may capture additional biologically relevant patterns. The value of such an approach was recently highlighted by Ying, Liu [114], who used Mendelian randomization to identify CpGs with putative causal effects on lifespan and health span. Ying and colleagues’ intriguing finding was the observation that these causal CpGs were not significantly enriched in widely used epigenetic clocks, such as the Horvath Blood & Skin clock or GrimAge, underscoring the need for alternative strategies to uncover the true drivers of aging and longevity [114]. Therefore, Ying and colleagues trained a clock called “adaptAge” based on around 1,000 of the identified CpGs with putative causal effects on lifespan and health span. Interestingly, approximately 35% of these causal CpGs overlap with our identified age-deceleration-associated CpGs, which is 2.75 time higher than expected by chance (Supplementary Fig. S11). While this overlap does not validate individual associations or establish causality, it is consistent with the interpretation that our results capture biologically meaningful variation related to epigenetic aging of human skin.

Throughout this study, we identified a range of modifiable factors that are significantly associated with changes in skin’s DNAm age. Environmental exposures, such as childhood sunburns and regular solarium use, were associated with age acceleration of the skin methylome [115]. These associations suggest that modifiable environmental factors, including both natural and artificial UV exposure, may be related to variation in DNAm aging of human skin. Notably, solarium use delivers much higher doses of UVA radiation than natural sunlight, which has been linked to more rapid and severe photoaging and is associated with accelerated epigenetic aging and DNAm changes in the skin [116, 117]. In addition, higher circulating 25-hydroxyvitamin D levels were among the factors associated with DNAm age acceleration. This observation should be interpreted cautiously because 25-hydroxyvitamin D levels can reflect supplement use, UV exposure and correlated health behaviors. Our data do not allow these sources to be distinguished. In our cohort, 25-hydroxyvitamin D concentrations predominantly fell within the guideline categories of deficient, suboptimal, and optimal status, with no indication of pathological increases [118]. Consistent with a potential link to sun-related behavior, prior work has reported that women with lower photodamage scores are more likely to be vitamin D insufficient [119].

In contrast, we observed that regular low-dose aspirin use was associated with decelerated epigenetic aging across multiple pan-tissue epigenetic clocks, which may reflect broader systemic associations. This observation is consistent with longitudinal evidence suggesting that aspirin us is linked to suppressed cancer-related epigenetic aging, potentially through reducing age-associated DNA hypermethylation and slowing DNAm age progression [120]. However, it is important to acknowledge that individuals using low-dose acetylsalicylic acid are likely to have underlying cardiovascular risk factors and may be on multiple medications. Therefore, the observed association between aspirin use and decelerated DNAm age may reflect broader systemic effects or confounding, rather than a direct causal effect of aspirin itself. Furthermore, our findings suggest that certain bioactive compounds or interventions may be associated with methylation patterns indicative of decelerated epigenetic aging. For instance, we observed that DHM not only exhibits overlap with age-deceleration-associated CpGs but was associated with reduced DNAm age and improved skin aging phenotypes [45], positioning it as a promising agent for promoting skin longevity. Importantly, the intervention analyses should be viewed as mechanistic triangulation rather than definitive proof, as these datasets were generated in different experimental settings and the observed overlap is defined at the level of CpG sets without systematic quantification. Collectively, these results underscore the broader potential of targeted lifestyle interventions and bioactive compounds not only to improve skin health but also to extend skin longevity by mitigating the negative impacts of aging at the molecular level.

Before examining the distinct mechanisms underlying age-associated factors, it is important to recognize that several hallmarks identified in our analysis, such as altered intercellular communication, stem cell exhaustion, and deregulated nutrient sensing, are not unidirectional in their effects [102]. These processes can have both protective and detrimental roles depending on their regulation, intensity, and context [121]. For example, intercellular communication is essential for tissue repair and homeostasis, but chronic inflammatory signaling can drive tissue dysfunction and aging [122]. Nutrient sensing pathways like mTOR and IGF-1 support cellular maintenance when appropriately regulated, yet their dysregulation is linked to age-related decline [123]. Stem cell activity is necessary for regeneration, but exhaustion of these pools or loss of regulatory control can impair tissue maintenance or increase cancer risk [124]. This duality may explain why these hallmarks are enriched among both age-accelerating and age-deceleration-associated CpGs in our study, reflecting the complex and context-dependent nature of aging mechanisms.

To explore potential molecular patterns, we separately analyzed CpGs according to whether their associated factors were linked to accelerating or decelerating skin’s DNAm age. Interestingly, the associated CpGs displayed distinct patterns. CpGs linked to age-accelerating factors in our study, were observed to be enriched in polycomb repressed regions (Fig. 2C), which have been shown to play a role in maintaining cellular specificity by silencing developmental genes [125]. Additionally, among the top enriched TF-families were T-box TFs (Fig. 3D), whose expression is regulated by polycomb repression, suggesting their potential role in controlling developmental processes [126]. Experimental evidence from mouse models indicates that the expression of this non-epidermal T-Box TF in epidermis can suppress the expression of epidermis-specific genes, underscoring their potential influence on cellular specificity [126]. In our data, the link to cellular specificity may also explain the enrichment of “altered intercellular communication” among the hallmarks of aging (Fig. 3F). However, these mechanistic interpretations are only based on positional enrichment and changes in methylation variance or bimodality, and should be considered correlative rather than causal. Furthermore, CpGs associated with age-accelerating factors displayed a uniform methylation level, suggesting a loss of bimodal pattern, which may reflect a reduction in cellular specificity (Fig. 4A). This phenomenon aligns with previous findings observed in aging and various diseases, where loss of epigenetic information is reflected by methylation levels shifting toward the mean [127, 128]. We also observed that the intra-methylome variance of age-acceleration-associated CpGs was decreased with age (Fig. 4B) indicating an erosion of the methylation pattern and a reduced dynamic range in the skin methylome, which contributes to the epigenetic drift [25]. Consistently, the direction of methylation changes at these sites closely matched the trend observed upon chronological aging (Fig. 4C). This is consistent with the hypothesis that lifestyle and physiological age-accelerating factors may influence CpGs in genomic regions that are important for cellular specificity, potentially compounding the loss of specificity that naturally occurs with age. These findings suggest that epigenetic alterations associated with external age-accelerating factors may reflect similar mechanisms to those underlying intrinsic aging, potentially contributing to the decline in tissue function and regenerative capacity observed in aged skin.

In contrast to the CpGs associated with age-accelerating factors, those linked to age-decelerating factors presented distinct characteristics, being enriched in chromatin regions classified as transcription elongation and weak transcription (Fig. 2C). These chromatin states, which are annotated by weak H3K36me3 histone modifications, are typically located between enhancers and transcribed regions with strong H3K36me3 signals [74, 129, 130]. H3K36me3 is deposited by elongating RNA polymerase II [131] and becomes more pronounced in exons further downstream of the transcription start site [132], suggesting that CpGs associated with age-decelerating factors might be located within or near regions of active transcriptional elongation. Intragenic DNAm, which occurs within gene bodies, is known to influence both transcriptional elongation and alternative splicing, thereby affecting gene expression outcomes [133]. Recent work has shown that the pace of RNA polymerase II (Pol-II) increases with age, and that slowing this process can extend the lifespan in model organisms such as C. elegans and D. melanogaster [134], highlighting the biological significance of transcriptional kinetics in aging. Notably, we found that genes linked to age-deceleration-associated CpGs were enriched for DNMTs, MBDs and TETs (Fig. 3D), which are enzymes responsible for establishing, recognizing and removing DNAm [135]. This finding suggests that the regulation of the DNAm machinery during transcriptional elongation may play a crucial role in modulating the pace of biological aging. It should also be noted that our gene and transcript-level analyses are based on gene annotations and isoform counts, not on direct gene expression measurements in the same samples, hence the functional interpretation remains indirect. Furthermore, the methylation levels of these CpGs displayed a bimodal distribution (Fig. 4A) and the intra-methylome variance at these sites only slightly decreased with age (Fig. 4B), suggesting that these CpGs are relatively stable and may not contribute to the age-related epigenetic drift observed at other sites. This is supported by the observation of only a weak trend in methylation changes at these CpGs towards a younger methylome (Fig. 4C) and aligns with previous findings that CpGs with a greater impact on healthy longevity do not necessarily undergo substantial methylation changes during aging [114]. In summary, lifestyle, physiological and pharmacological factors associated with decelerating epigenetic skin aging appear to target CpGs in genomic regions important for transcriptional elongation, and the methylation level of these sites may influence the pace of transcriptional elongation and, by extension, contribute to longevity. These results suggest that epigenetic alterations linked to external age-decelerating factors may act through mechanisms distinct from those driving the natural aging process, potentially offering new avenues for interventions aimed at promoting healthy aging.

To summarize, our findings reveal that the molecular landscape of skin aging is shaped by a dynamic interplay of modifiable age-accelerating and age-decelerating factors, each leaving distinct epigenetic signatures. The observation that interventions such as low-dose aspirin and DHM can modulate these molecular patterns highlights the potential for targeted strategies to not only decelerate but potentially reverse aspects of skin aging. While the results of our study provide a foundation for understanding the epigenetic regulation of skin health, additional longitudinal and interventional studies will be required to clarify the mechanisms underlying these associations and to assess their potential for clinical application. A particularly informative future research direction would be to integrate epidermal DNAm with RNA-seq (or other transcriptomic readouts) from skin to better link factor-associated methylation changes to downstream transcriptional consequences.

Conclusion

This study indicates that skin aging may be driven by mechanistically distinct epigenetic processes. The first of these is a damaging change from polycomb domain disruption that compromises cellular specificity. The second is an adaptive change from transcriptional elongation modulation that supports transcriptional integrity. As suggested by Ying and colleagues, it is important to differentiate between adaptive and damaging epigenetic changes [114]. In other words, in the field of aging research, one might never identify the fountain of youth via comparing the biological differences between young and old ages. This mechanistic distinction is essential for developing targeted interventions that preserve beneficial adaptations while minimizing detrimental alterations, thereby advancing the precision of anti-aging strategies for skin.

Author contributions

HV took the lead in managing the epidemiological study SHIP. SC took the lead in managing the epidemiological study Skinly. AB and KK were in charge of data curation. The bioinformatical analysis and data visualization was conducted by AB, MQ, KK and CF. The administration of the project was under the supervision of CF, EG, MW, MBT, SC, SJ and SG. The draft of the manuscript was written by AB, while CF, LK, KK, and MQ contributed to the manuscript’s writing, conceptualization, and supervision of the manuscript. All authors participated in revising the manuscript, read the final version, and approved it for submission.

Funding

SHIP is an integral part of the Community Medicine Research Network, which is affiliated with University Medicine Greifswald. This network is financially supported by the German Federal State of Mecklenburg-West Pomerania.

Data Availability

Access to SHIP data for research purposes is available upon submission of a formal data access application, which can be completed at [http://www2.medizin.uni-greifswald.de/cm/fv/ship/daten-beantragen/].

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.