Meta-analysis of DNA methylation aging signatures in 17 human tissues
Main
Aging is a universal yet profoundly individualized process. While chronological age progresses at a steady pace, its biological impact differs substantially among and within individuals, leading to varying risks of disease, functional decline and mortality[1](https://www.nature.com/articles/s43587-026-01164-5#ref-CR1 "Lopez-Otin, C., Blasco, M. A., Partridge, L., Serrano, M. & Kroemer, G. Hallmarks of aging: an expanding universe. Cell 186, 243–278 (2023)."),[2](https://www.nature.com/articles/s43587-026-01164-5#ref-CR2 "Lopez-Otin, C., Blasco, M. A., Partridge, L., Serrano, M. & Kroemer, G. The hallmarks of aging. Cell 153, 1194–1217 (2013)."). These differences stem from cumulative molecular changes, with epigenetic modifications, particularly DNA methylation (DNAm) (the addition of methyl groups to cytosine bases), among the most reliable indicators of biological aging trajectories[1](https://www.nature.com/articles/s43587-026-01164-5#ref-CR1 "Lopez-Otin, C., Blasco, M. A., Partridge, L., Serrano, M. & Kroemer, G. Hallmarks of aging: an expanding universe. Cell 186, 243–278 (2023)."),[2](https://www.nature.com/articles/s43587-026-01164-5#ref-CR2 "Lopez-Otin, C., Blasco, M. A., Partridge, L., Serrano, M. & Kroemer, G. The hallmarks of aging. Cell 153, 1194–1217 (2013)."),[3](https://www.nature.com/articles/s43587-026-01164-5#ref-CR3 "Horvath, S. & Raj, K. DNA methylation-based biomarkers and the epigenetic clock theory of ageing. Nat. Rev. Genet. 19, 371–384 (2018).").
Age-related DNAm changes fall into two main categories. Differentially methylated positions (DMPs) are sites exhibiting consistent directional changes across individuals with age, either gaining or losing methylation, and have been widely studied as potential biomarkers of biological aging[3](https://www.nature.com/articles/s43587-026-01164-5#ref-CR3 "Horvath, S. & Raj, K. DNA methylation-based biomarkers and the epigenetic clock theory of ageing. Nat. Rev. Genet. 19, 371–384 (2018)."),[4](https://www.nature.com/articles/s43587-026-01164-5#ref-CR4 "Seale, K., Horvath, S., Teschendorff, A., Eynon, N. & Voisin, S. Making sense of the ageing methylome. Nat. Rev. Genet. 23, 585–605 (2022)."),[5](https://www.nature.com/articles/s43587-026-01164-5#ref-CR5 "Seale, K., Teschendorff, A., Reiner, A. P., Voisin, S. & Eynon, N. A comprehensive map of the aging blood methylome in humans. Genome Biol. 25, 240 (2024)."),[6](https://www.nature.com/articles/s43587-026-01164-5#ref-CR6 "Voisin, S. et al. Exercise is associated with younger methylome and transcriptome profiles in human skeletal muscle. Aging Cell 23, e13859 (2024)."),[7](https://www.nature.com/articles/s43587-026-01164-5#ref-CR7 "Campisi, J. et al. From discoveries in ageing research to therapeutics for healthy ageing. Nature 571, 183–192 (2019)."),[8](https://www.nature.com/articles/s43587-026-01164-5#ref-CR8 "Hannum, G. et al. Genome-wide methylation profiles reveal quantitative views of human aging rates. Mol. Cell 49, 359–367 (2013)."). Variably methylated positions (VMPs) instead show increased inter-individual variability with aging methylation levels among individuals, and this variability tends to increase with age, reflecting stochastic influences such as environmental exposures and genetic factors rather than uniform directional change[4](https://www.nature.com/articles/s43587-026-01164-5#ref-CR4 "Seale, K., Horvath, S., Teschendorff, A., Eynon, N. & Voisin, S. Making sense of the ageing methylome. Nat. Rev. Genet. 23, 585–605 (2022)."),[5](https://www.nature.com/articles/s43587-026-01164-5#ref-CR5 "Seale, K., Teschendorff, A., Reiner, A. P., Voisin, S. & Eynon, N. A comprehensive map of the aging blood methylome in humans. Genome Biol. 25, 240 (2024)."),[8](https://www.nature.com/articles/s43587-026-01164-5#ref-CR8 "Hannum, G. et al. Genome-wide methylation profiles reveal quantitative views of human aging rates. Mol. Cell 49, 359–367 (2013)."),[9](https://www.nature.com/articles/s43587-026-01164-5#ref-CR9 "Teschendorff, A. E. et al. Age-dependent DNA methylation of genes that are suppressed in stem cells is a hallmark of cancer. Genome Res. 20, 440–446 (2010)."). A single CpG site can exhibit both properties simultaneously, showing a consistent directional shift across the population (DMP) while also becoming increasingly variable between individuals (VMP).[5](https://www.nature.com/articles/s43587-026-01164-5#ref-CR5 "Seale, K., Teschendorff, A., Reiner, A. P., Voisin, S. & Eynon, N. A comprehensive map of the aging blood methylome in humans. Genome Biol. 25, 240 (2024)."),[7](https://www.nature.com/articles/s43587-026-01164-5#ref-CR7 "Campisi, J. et al. From discoveries in ageing research to therapeutics for healthy ageing. Nature 571, 183–192 (2019).").
A third lens, Shannon entropy, captures the overall disorder of DNAm across the genome at the level of individual samples. Borrowed from information theory, entropy is highest when methylation at a given site is close to 50%, meaning the signal is maximally uncertain, and lowest when sites are fully methylated or unmethylated. As aging progresses, methylation patterns become less precise and more disordered, and entropy increases accordingly. Importantly, entropy integrates signals from both DMPs and VMPs, making it a complementary, system-wide measure of epigenetic aging that neither metric captures alone[4](https://www.nature.com/articles/s43587-026-01164-5#ref-CR4 "Seale, K., Horvath, S., Teschendorff, A., Eynon, N. & Voisin, S. Making sense of the ageing methylome. Nat. Rev. Genet. 23, 585–605 (2022)."),[9](https://www.nature.com/articles/s43587-026-01164-5#ref-CR9 "Teschendorff, A. E. et al. Age-dependent DNA methylation of genes that are suppressed in stem cells is a hallmark of cancer. Genome Res. 20, 440–446 (2010)."),[10](https://www.nature.com/articles/s43587-026-01164-5#ref-CR10 "Vershinina, O., Bacalini, M. G., Zaikin, A., Franceschi, C. & Ivanchenko, M. Disentangling age-dependent DNA methylation: deterministic, stochastic, and nonlinear. Sci. Rep. 11, 9201 (2021).").
Despite advancements in the field, most epigenetic aging research has primarily focused on blood samples, with few specific tissue investigations available, resulting in a limited body of work in the field[5](https://www.nature.com/articles/s43587-026-01164-5#ref-CR5 "Seale, K., Teschendorff, A., Reiner, A. P., Voisin, S. & Eynon, N. A comprehensive map of the aging blood methylome in humans. Genome Biol. 25, 240 (2024)."),[11](https://www.nature.com/articles/s43587-026-01164-5#ref-CR11 "Levine, M. E. et al. An epigenetic biomarker of aging for lifespan and healthspan. Aging 10, 573–591 (2018)."),[12](https://www.nature.com/articles/s43587-026-01164-5#ref-CR12 "Lu, A. T. et al. DNA methylation GrimAge version 2. Aging 14, 9484–9549 (2022)."),[13](https://www.nature.com/articles/s43587-026-01164-5#ref-CR13 "Lu, A. T. et al. DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging 11, 303–327 (2019)."),[14](https://www.nature.com/articles/s43587-026-01164-5#ref-CR14 "Zhu, T., Zheng, S. C., Paul, D. S., Horvath, S. & Teschendorff, A. E. Cell and tissue type independent age-associated DNA methylation changes are not rare but common. Aging 10, 3541–3557 (2018)."). As a result, fundamental questions remain unresolved: to what extent is the aging methylome shared across tissues, and where does it diverge? Are conserved methylation signatures reflective of systemic aging, or are they tissue-restricted phenomena? Understanding these dimensions is critical, as each tissue exhibits unique functional roles, cell-type compositions and regenerative capacities that shape distinct epigenetic aging trajectories. Identifying conserved patterns would also support the use of blood as a minimally invasive proxy to infer biological age in less accessible organs such as brain, heart or muscle.
To address these gaps, we assembled a comprehensive atlas of DNAm across 17 human tissues throughout the adult lifespan. Building on our detailed blood studies with over 30,000 participants[5](https://www.nature.com/articles/s43587-026-01164-5#ref-CR5 "Seale, K., Teschendorff, A., Reiner, A. P., Voisin, S. & Eynon, N. A comprehensive map of the aging blood methylome in humans. Genome Biol. 25, 240 (2024)."), we merged numerous open-source and collaborator datasets and performed integrated analyses across different organs. We characterized the landscape of age-related DMPs, VMPs and Shannon entropy dynamics across tissues and applied co-methylation network analyses and in silico perturbation simulations to identify key genes and pathways associated with epigenetic aging. Summary data and results are openly available at https://bioinformatics.erc.monash.edu/apps/human-aging-atlas/.
Results
A pan-tissue landscape of age-associated DNA methylation
We assembled a cross-sectional atlas of DNAm across 17 human tissues, totaling over 15,000 samples from 131 datasets (Supplementary Table 1). Global mean methylation ranged from 38% (cervix) to 63% (retina), though regional distributions were more uniform (Supplementary Table 2). Age-associated changes were evaluated using three complementary metrics: DMPs, VMPs and methylation Shannon entropy, each applied within individual datasets and meta-analyzed within tissues (Supplementary Data Fig. 1). DMPs were identified using multivariate linear regression, controlling for biological and technical covariates (for example, sex, body mass index (BMI) and batch). By contrast, VMPs were identified using a two-step Breusch–Pagan heteroscedasticity framework, detecting age-associated changes in methylation variance. Shannon entropy was computed from covariate-adjusted _β_ values genome-wide and across CpG subsets and modeled as a function of age to capture sample-level epigenetic disorder. Together, these three layers provide complementary views of methylome remodeling, mean-level shifts, variance-level shifts and system-wide disorder across both locus-specific and tissue-wide scales.
Multivariate linear analysis of age reflects distinct tissue-specific aging signatures
Multivariate linear regression models, adjusting for relevant covariates, identified CpG sites significantly associated with chronological age across tissues. DMP counts (that is, age-associated CpGs) varied markedly (Fig. 1); brain, liver and lung exhibited the largest numbers (brain >180,000 DMPs), while pancreas, retina and prostate showed minimal or no significant age associations at a false discovery rate (FDR) <0.005. Some tissues, such as skin and the colon, exhibited strong signatures despite moderate sample sizes. All tissues except skeletal muscle and lung showed a predominance of age-associated hypermethylation.
**Fig. 1: DMPs with age across 17 human tissues.**
Each volcano plot depicts a unique tissue and its methylation changes with age. Each dot corresponds to a distinct CpG; colored dots indicate a significant association with age at an FDR <0.005, while black dots indicate CpGs that do not exhibit significant age-associated changes.
The distribution of methylation fractions at DMPs between younger (<30 years) and older (>60 years) individuals revealed that CpGs with low (<25%) or intermediate (25–75%) methylation in youth predominantly exhibited age-associated hypermethylation, with some highly methylated regions (>75%) gaining further methylation with age (Fig. 2). Hypermethylated CpGs were more frequently located within CpG islands and shores, while hypomethylated CpGs were predominantly enriched in CpG-poor regions such as shores and open sea across tissues (Extended Data Fig. 1). Cell-type adjustment had a minimal impact on DMP detection, with _t_-statistics highly correlated between adjusted and unadjusted models (_r_ = 0.81–0.99 across tissues), confirming that identified methylation signals are robust to cell-type composition (Extended Data Fig. 2 and Supplementary Table 3). Brain tissue was biologically distinct, showing a balanced distribution of hyper- and hypomethylated DMPs across all methylation fractions (28–40%), with neither direction predominating. The prostate, rectum and stomach lacked sufficient samples from individuals under 30 years to be included in this analysis.
**Fig. 2: Methylation state and directional change of age-associated DMPs across 13 human tissues.**
**a**, The direction of methylation change in older individuals (>60 years). For each tissue, bars show the proportion of DMPs exhibiting increased methylation (red) or decreased methylation (blue) with age. **b**, The methylation-state distribution in younger individuals (<30 years) for the same DMPs. Bars show the proportion of DMPs classified as low methylation (<25%; green), intermediate methylation (25–75%; orange) or high methylation (>75%; purple). Tissues shown are adipose, brain, breast, buccal, cervix, colon, heart, kidney, liver, lung, pancreas, retina and skin. Prostate, rectum and stomach had insufficient samples from individuals under 30 years and are not included. For example, in adipose tissue, 13,868 age-associated DMPs were identified; in younger individuals for **b**, 65% showed low methylation, 25.3% showed intermediate methylation and 9.5% showed high methylation, while in older individuals for **a**, the majority of these DMPs showed increased methylation with age.
Sensitivity analysis confirmed that DMP yield correlated significantly with age range (_R_ = 0.66, _P_ = 0.0072) but not mean or median age (_R_ = 0.13 and 0.22, respectively), indicating that broader lifespan sampling enhances detection power without introducing systematic bias (Extended Data Fig. 3b). Power analysis[15](https://www.nature.com/articles/s43587-026-01164-5#ref-CR15 "Mansell, G. et al. Guidance for DNA methylation studies: statistical insights from the Illumina EPIC array. BMC Genomics 20, 366 (2019).") revealed that limited DMP discovery in tissues such as heart, pancreas, prostate and stomach reflects insufficient sample sizes rather than biological stability, as detectability for median effect sizes was extremely low at FDR ≤0.005 in these tissues (Supplementary Table 4). By contrast, brain tissue (_N_ = 7,451) demonstrated high power for upper-quartile effects, resulting in substantially more significant associations. Lung and liver tissues showed comparatively higher detectability (61% and 49%, respectively) despite moderate sample sizes, reflecting comparatively larger effect sizes and favorable variance structure. Skeletal muscle (_N_ = 1,568), despite an adequate sample size, exhibited smaller effect sizes overall, leading to more modest detectability. Together, tissue-specific discovery rates are driven by both sample size and the underlying distribution of effect sizes.
VMPs are scarce and highly tissue-specific
Compared with DMPs, VMPs were much less frequent across tissues. Only brain, buccal, cervix, lung and skin had substantial numbers of age-VMPs, whereas most tissues showed few or none at FDR <0.005 (Fig. 3). Cross-tissue VMP overlap was limited, with fewer than 100 VMPs shared between any tissue pair (Extended Data Fig. 4). To determine whether the observed overlap of age-associated VMPs across tissues exceeded what would be expected by chance, we performed a permutation-based analysis. For each tissue comparison, we randomly selected sets of CpGs matched in number and genomic distribution to the observed VMPs and repeated this process 1,000 times to generate a null distribution of overlaps. The actual observed overlaps were then compared with this distribution to compute empirical _P_ values. This analysis revealed that half of the overlaps between tissue pairs occurred by chance, except for the pairs involving brain and buccal, buccal and cervix, buccal and lung, brain and skin, and buccal and skin (Supplementary Table 5a). VMP detection showed no significant correlation with age range, mean age, median age or sample size (_R_ ≈ 0.05, _P_> 0.8), supporting biological tissue specificity rather than technical artifacts (Extended Data Fig. 3a).
**Fig. 3: Age-related changes in VMPs across 17 human tissues.**
Each scatter plot illustrates age-associated VMPs for a specific tissue. Only brain, buccal, cervix, liver, lung and skin tissues had significant age-associated VMPs at FDR <0.005; the remaining 11 tissues showed no significant VMPs and are displayed for reference. Each dot represents a unique individual, with the _x_ axis showing age and the _y_ axis showing methylation level (_β_ value). Individual data points are shown throughout; no summary statistics or error bars are presented.
Among nonchance overlaps, most shared VMPs showed concordant directionality and similar trends, such as increases or decreases in variability (Extended Data Fig. 4). An exception was brain–buccal VMPs, where only 15% of the 13 overlapping VMPs were directionally consistent. Pathway enrichment analysis showed no enriched pathways for most individual tissues, except for buccal and lung tissues. Buccal VMPs were enriched across 210 pathways (FDR <0.005), particularly in system development (Supplementary Table 6), while in lung tissues, VMPs were enriched in only three pathways: external encapsulating structures, the extracellular matrix and collagen-containing extracellular matrices. VMPs were particularly sensitive to cell-type correction in lung and liver, while colon displayed increased VMPs following adjustment (Supplementary Table 3 and Supplementary Fig. 1). Collectively, these results demonstrate that age-associated hypermethylation is a consistent feature across tissues, with the contribution of cellular composition varying by tissue.
Shannon entropy and epigenetic disorder reveal tissue-specific aging patterns
Shannon entropy captures the overall disorder and unpredictability of DNAm patterns across the genome, offering a single quantitative measure that integrates all age-related differential changes including those from low-variability DMPs. While VMPs shift entropy toward higher or lower states, many of these CpGs are already intermediately methylated[16](https://www.nature.com/articles/s43587-026-01164-5#ref-CR16 "Sharif, J. & Koseki, H. Hemimethylation: DNA’s lasting odd couple. Science 359, 1102–1103 (2018).") in youth, meaning entropy dynamics are not strictly age-dependent at these sites. This analysis therefore provides a complementary layer of insight beyond DMP- or VMP-based approaches, revealing tissue-specific trajectories of epigenetic disorganization during aging (Fig. 4).
**Fig. 4: Shannon entropy of DNAm with age across 17 human tissues.**
Each plot represents a specific tissue using the largest available datasets. Tissues are grouped into three categories on the basis of the presence of age-associated DMPs and VMPs. Tissues without VMPs (adipose, breast, cervix, colon, heart, kidney, skeletal muscle, pancreas, rectum and retina) show two conditions: ‘Entropy_all’ (entropy calculated across all CpGs genome-wide; shown in yellow) and ‘Entropy_none’ (entropy calculated across CpGs not associated with age; shown in gray). Tissues with both DMPs and VMPs (brain, buccal, liver, lung and skin) show three conditions: ‘Entropy_DMPsonly’ (entropy at CpGs that are DMPs only; shown in blue), ‘Entropy_VMPsonly’ (entropy at CpGs that are VMPs only; shown in orange) and ‘Entropy_DMPVMP’ (entropy at CpGs that are both DMPs and VMPs; shown in pink). Prostate and stomach tissues showed no age-related DMPs or VMPs and are therefore not represented. Lines show fit regression of entropy against age; shaded bands represent 95% confidence intervals.
Entropy increased preferentially at DMPs in several tissues, such as adipose, breast and kidney, suggesting the progressive loss of epigenetic fidelity at aging-associated sites, consistent with models of epigenetic drift in metabolically active and proliferative tissues[4](https://www.nature.com/articles/s43587-026-01164-5#ref-CR4 "Seale, K., Horvath, S., Teschendorff, A., Eynon, N. & Voisin, S. Making sense of the ageing methylome. Nat. Rev. Genet. 23, 585–605 (2022)."),[5](https://www.nature.com/articles/s43587-026-01164-5#ref-CR5 "Seale, K., Teschendorff, A., Reiner, A. P., Voisin, S. & Eynon, N. A comprehensive map of the aging blood methylome in humans. Genome Biol. 25, 240 (2024)."),[17](https://www.nature.com/articles/s43587-026-01164-5#ref-CR17 "Jeong, H., Mendizabal, I. & Yi, S. V. Human brain aging is associated with dysregulation of cell type epigenetic identity. Geroscience 47, 3759–3770 (2025)."). By contrast, brain, buccal epithelium, liver, lung and skin tissues exhibited entropy increases at positions that were both differentially and variably methylated (that is, DMP–VMP overlaps), indicating a more stochastic mode of epigenetic aging in these tissues, where both directionality and variability of methylation are disrupted.
WGCNA highlights tissue-specific hubs and conserved aging modules
To characterize co-regulatory methylation frameworks across tissues, we applied weighted gene co-expression network analysis (WGCNA) to skeletal muscle, adipose tissue, blood and brain methylation data, constructing tissue-specific co-methylation networks. Each tissue generated specific modules, with eigengene expression trajectories showing distinct aging signatures (Extended Data Fig. 5a,c,e,g). Subsequently, we constructed a signed network and conducted hierarchical clustering on the topological overlap matrix to identify modules. The number and size of modules varied among tissues, reflecting biological complexity in a data-driven approach (Extended Data Fig. 5b,d,f,h). Module eigengenes, the first principal component of each module, were calculated and correlated with age to identify those with strong associations with aging. Significant positive and negative age-associated modules were included for further functional interpretation (Extended Data Fig. 6a,c,e,g).
Gene Ontology (GO) biological processes categorized the enriched pathways according to module direction, allowing us to distinguish between age-related increases in methylation (positive modules) and decreases in methylation (negative modules). In adipose tissue, age-associated positive modules exhibited heightened methylation at loci related to forebrain development and cell adhesion, while hypomethylated modules were enriched for synaptic signaling and NF-κB regulation, indicating a potential loss of regulatory control in immune and neuronal pathways (Extended Data Fig. 6b). In blood, hypermethylated modules were related to embryonic development, cell-adhesion, apoptotic signaling and immune pathways, while hypomethylation modules reflected homeostasis regulation, including cell shape and regulation of body fluid levels (Extended Data Fig. 6d). The brain displayed increased methylation in modules enriched for GTPase activity, synaptic activity and organization, cell communication and adhesion, among others, while decreased methylation impacted multiple morphogenesis-related processes and developmental pathways (Extended Data Fig. 6f). In skeletal muscle, gains in methylation were observed in GTPase signaling pathways and some nonmuscle developmental pathways. At the same time, losses were noted in pathways regulating muscle development and differentiation, as well as metabolic-related pathways (Extended Data Fig. 6h). Beyond tissue-specific patterns, we also identified converging modules that revealed shared epigenetic aging programs across tissues. Across tissues, homophilic cell adhesion via plasma membrane adhesion molecules was enriched in adipose, blood and brain tissue (Fig. 5a,b,d,e), implicating a common loss of cell–cell interaction integrity during aging. Small GTPase-mediated signal transduction was enriched in both brain and skeletal muscle, indicating conserved age-associated changes in intracellular communication and cytoskeletal regulation.
**Fig. 5: WGCNA co-methylation network analysis across four tissues.**
**a**,**b**,**d**,**e**, The leading enriched pathways for positive (hypermethylated with age; **a**,**d**) and negative (hypomethylated with age; **b**,**e**) co-methylation modules in adipose tissue (**a**,**b**) and brain (**d**,**e**), respectively. Each panel corresponds to a distinct tissue, with colors indicating module affiliation. Those identified as module-influential are highlighted in green and represented as triangles; associated pathways are highlighted in gray. **c**,**f**, Dot plots of the frequency with which individual genes were ranked among the top module-influential genes across age-associated WGCNA modules for adipose tissue (**c**) and brain tissue (**f**), stratified by direction of module–age associations. The _y_ axis indicates gene names and the _x_ axis reflects the number of pathways influenced by each gene within its module. kME was used to define module-influential genes as those with the highest intramodular connectivity and strongest association with age. Recurrently identified module-influential genes represent network-central loci consistently implicated in aging-associated methylation programs across tissues.
Module membership (kME), defined as the correlation between a gene’s methylation profile and the module eigengene, was used to identify module-influential genes, those exhibiting both high intramodular connectivity and strong association with age. Across all tissues (that is, adipose and brain), members of the protocadherin gamma (PCDHG) gene family were consistently identified as primary module-influential genes (Fig. 5c,f). While PCDHG genes are traditionally associated with neural development and synaptic patterning, their consistent epigenetic regulation in nonneural tissues, including blood and skeletal muscle, suggests a broader role in maintaining structural and signaling stability across organ systems during aging.
Cross-tissue overlap reveals core and divergent epigenetic aging programs
To determine whether epigenetic aging signatures are consistent across tissues or predominantly tissue-specific, we analyzed the overlap of DMPs across all tissues in our atlas. We also included the blood DMPs identified in our previous blood atlas publication[5](https://www.nature.com/articles/s43587-026-01164-5#ref-CR5 "Seale, K., Teschendorff, A., Reiner, A. P., Voisin, S. & Eynon, N. A comprehensive map of the aging blood methylome in humans. Genome Biol. 25, 240 (2024)."). The intersection heat map (Fig. 6a, bottom diagonal half) depicts a complex scenario, showing that DMP overlap between tissues varies from 0% to 25% of the total DMPs, as determined by the tissue with the fewest DMPs. This suggests that most pairwise comparisons show a restricted number of common DMPs, emphasizing distinct aging pathways unique to each tissue, or possibly reflecting a reduced ability to detect DMPs, especially in tissues with limited samples. When evaluating the directionality of the intersected DMPs across tissues (Fig. 6a, top diagonal half), a majority exhibit changes in the same direction, with direction concordance exceeding 59% for all tissue pairs except the brain and retina. The brain and retina demonstrated the highest level of discordant directionality, with only 28% of the 25 overlapping DMPs moving in the same direction. In our search for DMPs with the greatest overlaps among tissues, we identified cg16867657 as significant in 15 tissues (excluding the rectum). This specific CpG site is located on chromosome 6 at position 11044643–11044645, harboring the well-known aging-related gene _ELOVL2_. A total of 37 DMPs were identified across 13 or more tissues (Supplementary Table 9).
**Fig. 6: Cross-tissue overlap of age-associated DMPs reveals both conserved and tissue-specific aging signatures.**
**a**, A combined similarity heat map of pairwise DMP overlap across 17 human tissues. The lower diagonal shows pairwise Jaccard indices of DMP overlap; darker blue indicates greater overlap. The upper diagonal shows effect-size correlation across tissues to assess directionality concordance of overlapping DMPs; darker red indicates greater concordance. **b**, Pathway enrichment analysis of CpGs identified as DMPs in nine or more tissues. Dot color indicates FDR strength and dot size reflects the number of genes associated with each pathway. **c**, A heat map of the 50 most enriched pathways from the mitch multi-tissue pathway enrichment analysis. Red indicates hypermethylation and blue indicates hypomethylation of genes within each pathway. Enriched pathways are listed on the _y_ axis and tissues are shown on the _x_ axis.
To ensure that the observed overlaps were not merely due to random chance, we performed permutation tests (as described in the Methods), by randomly rearranging the sample age labels within each tissue and recalculating DMP overlaps over 1,000 iterations. As shown in Supplementary Table 5b, the results indicated that the empirical overlaps between several pairs of tissues significantly surpassed the 95th percentile of the null distribution, providing compelling evidence against coincidental concordance. The exceptions were buccal versus rectum, cervix versus retina, liver versus retina, rectum versus retina, retina versus blood, and retina versus skin. Notably, only one dataset was found in the analysis of the retina and rectum, which significantly limits the ability to identify age-related changes in these tissues, probably resulting in observed overlaps that do not meet the null distribution.
Finally, we conducted a sensitivity analysis to confirm that our DMPs’ identification was not influenced by technical artifacts such as age ranges, mean age or median age. The updated scatter plots examining DMPs versus age (Extended data Fig. 3b) illustrate our tests for correlations between the number of DMPs and the age range, median age and mean age across different tissues. Our analysis revealed that only the age range demonstrated a significant correlation with DMP count (_R_ = 0.66, _P_ = 0.0072), while mean and median age showed no significant correlation (_R_ = 0.13 and 0.22, respectively). This suggests that a broader sampling across the lifespan enhances detection power without introducing systematic bias.
Cross-tissue network level analysis through module enrichment identifies key pathways and genes as module-influential genes in aging
We subsequently investigated the functional significance of genes that consistently showed age-related changes in methylation across various tissues. We established a rigorous consensus set of genes differentially methylated in nine or more tissues (>50% overlap across tissues) and conducted pathway enrichment analysis on this core group. Notably, a prominent signature surfaced: almost all enriched pathways were associated with transcriptional regulation, chromatin organization and gene regulation (Fig. 6b).
While compelling, this method has constraints owing to statistical power; tissues with fewer samples or minor methylation changes may not be adequately represented (Supplementary Table 1). In addition, it restricts interpretability, as it does not enable comparison of directionality across tissues in each pathway. To tackle these issues, we employed mitch[18](https://www.nature.com/articles/s43587-026-01164-5#ref-CR18 "Kaspi, A. & Ziemann, M. mitch: multi-contrast pathway enrichment for multi-omics and single-cell profiling data. BMC Genomics 21, 447 (2020)."), which combines the magnitude and direction of gene-level shifts across different tissues using _t_-statistics rather than depending only on significance thresholds. This approach allowed underpowered tissues to meaningfully contribute to the overall analysis landscape. The integration based on mitch essentially validated the transcriptional regulation signature while broadening it to include additional pathways relevant to aging, such as mitochondrial function, cell-cycle pathways and DNA damage response (Supplementary Table 7). A thorough examination of the top 50 pathways has unveiled three distinct clusters that illustrate a consistent shift in agreement across nearly all tissues (Fig. 6c). For instance, pathways associated with transcription factor activity, as observed in the analysis above, DNA binding and transcriptional regulation were identified as being enriched in hyper-DMPs across all tissues. In addition, pathways related to cell–cell signaling, synaptic signaling and neurogenesis were observed to be enriched in hyper-DMPs across all tissues, with the exception of the retina and lung. Conversely, pathways involved in intracellular processes, such as those pertaining to mitochondria and macromolecule metabolism, were found to be enriched in hypo-DMPs in all tissues, except for breast and colon.
To further investigate the signatures identified by mitch, we generated a gene–gene correlation matrix across tissues utilizing Spearman correlation. This method evaluated the consistency of each gene’s aging profile across various tissues. For instance, if two genes increase in methylation levels with age in most tissues, they are considered to have a strong positive correlation. Conversely, if the methylation of one gene increases in some tissues while in another decreases, the correlation is regarded as negative. Consequently, we were able to model ‘_t_-patterns’, indicating whether a gene is uniformly upregulated or downregulated with age across tissues, or exhibits fluctuations that are specific to certain tissues. We utilized hierarchical clustering and dynamic tree cutting on the gene–gene correlation matrix to extract biological structures from these patterns, grouping genes into modules on the basis of co-aging behavior. Unlike WGCNA modules, these modules arise from cross-tissue coherence in aging trajectories as indicated by _t_-statistics, rather than solely from co-methylation patterns in a single tissue.
Each resulting module captures a group of genes that age uniformly across different tissues. Some modules exhibited significantly high positive _t_-scores in nearly all tissues, indicating a widespread hypermethylation linked to aging (Supplementary Table 6). By contrast, other modules were distinctly tissue-specific, showing aging-related hypermethylation in one tissue while demonstrating hypomethylation in another, suggesting varying aging mechanisms. This analysis uncovered several important findings: for instance, module 2 showed elevated average _t_-statistics in blood, indicating significant hypermethylation associated with aging, while also revealing negative _t_-statistics in the brain, suggesting hypomethylation. Only one module (module 5) exhibited a universal pattern, indicating hypermethylation across all tissues (Extended Data Fig. 7 and Supplementary Table 6). Importantly, the module-influential gene in this module belongs to the _PCDHG_ gene family (_PCDHGA1_), reinforcing the idea that this gene family may play a key role in aging processes throughout all human tissues. In summary, our findings highlight that specific gene programs (module clusters) are regulated in a highly tissue-specific manner during aging, while others may indicate shared core aging mechanisms. This may also present tissue-specific or even contradictory regulations (for example, beneficial in one tissue but harmful in another).
In silico robustness analysis reveals functionally fragile genes driving module architecture
To advance from correlation to testing functional effects, we established an in silico validation framework. In this framework, we computationally adjusted methylation levels for each gene, simulating conditions, such as hypermethylation, and reassessed the structural integrity of aging-associated modules. This simulation-driven strategy identified a limited set of genes (defined here strictly in terms of network topology and simulated mathematical disruption, not biologically validated causal effects) that modified the arrangement of various modules across different tissues (Supplementary Table 7). The _PCDHG_ gene family, _MEST_, _HDAC4_ and genes from the _HOX_ family were predominant in the set of genes identified. When combined with additional key genes from each module, they instigated significant changes in modular connectivity, with some modules exhibiting alterations exceeding 50% when key genes were modified. Considering the enrichment pathways of each module, it is reasonable to assert that alterations in DNAm result in changes in function pertinent to these pathways.
To better understand the biological effects of these modulations, we integrated module-level enrichment findings and classified gene modifications as potentially beneficial or detrimental. For instance, if a gene affects a module linked to aging pathways and the effect diminishes the module’s influence, it is deemed beneficial; conversely, if the modulation enhances the module’s effect, it is considered harmful. Only ten modules presented a beneficial effect, while all others appear to have a harmful effect if modified. These findings suggest that most aging-related modules are vulnerable to modifications, with only a small subset showing potential for beneficial modulation.
Discussion
This meta-analysis of over 15,000 methylomes across 17 human tissues, demonstrates that aging is associated with both tissue-specific and conserved cross-tissue effects on the methylome, characterized across three complementary layers: DMPs, VMPs and Shannon entropy. The heterogeneity of DMPs across tissues reflects the organ-specific nature of the aging process[19](https://www.nature.com/articles/s43587-026-01164-5#ref-CR19 "Horvath, S. DNA methylation age of human tissues and cell types. Genome Biol. 14, R115 (2013)."),[20](https://www.nature.com/articles/s43587-026-01164-5#ref-CR20 "Slieker, R. C. et al. Age-related accrual of methylomic variability is linked to fundamental ageing mechanisms. Genome Biol. 17, 191 (2016)."),[21](https://www.nature.com/articles/s43587-026-01164-5#ref-CR21 "Teschendorff, A. E., West, J. & Beck, S. Age-associated epigenetic drift: implications, and a case of epigenetic thrift? Hum. Mol. Genet. 22, R7–R15 (2013).").
Tissues such as the brain, liver, lung, skeletal muscle and skin exhibited a large number of DMPs associated with age, suggesting a reconfiguration of the methylome or vulnerability associated with advanced age[4](https://www.nature.com/articles/s43587-026-01164-5#ref-CR4 "Seale, K., Horvath, S., Teschendorff, A., Eynon, N. & Voisin, S. Making sense of the ageing methylome. Nat. Rev. Genet. 23, 585–605 (2022)."),[22](https://www.nature.com/articles/s43587-026-01164-5#ref-CR22 "Jain, N. et al. DNA methylation correlates of chronological age in diverse human tissue types. Epigenetics Chromatin 17, 25 (2024)."),[23](https://www.nature.com/articles/s43587-026-01164-5#ref-CR23 "Kang, Y. K. et al. Emergence of CpG-cluster blanket methylation in aged tissues: a novel signature of epigenomic aging. Nucleic Acids Res. https://doi.org/10.1093/nar/gkaf354
(2025)."). By contrast, tissues such as the kidney, prostate, rectum and stomach demonstrated minimal to no detectable changes, probably attributed to the limited sample size or an inherently more stable epigenetic landscape. However, power analyses indicated extremely low detectability for median effect sizes in these tissues at FDR ≤0.005, largely owing to small sample sizes and the resulting stringent empirical significance thresholds. These findings suggest that the limited number of DMPs observed in these tissues more likely reflects insufficient statistical power rather than a biologically stable epigenetic landscape. Larger cohorts will be required to determine whether comparable age-associated methylation changes are present in these rarer tissue types. Despite this tissue variability, a discernible pattern emerged: most tissues demonstrated age-associated hypermethylation, particularly in previously unmethylated regions in younger individuals[22](https://www.nature.com/articles/s43587-026-01164-5#ref-CR22 "Jain, N. et al. DNA methylation correlates of chronological age in diverse human tissue types. Epigenetics Chromatin 17, 25 (2024)."),[23](https://www.nature.com/articles/s43587-026-01164-5#ref-CR23 "Kang, Y. K. et al. Emergence of CpG-cluster blanket methylation in aged tissues: a novel signature of epigenomic aging. Nucleic Acids Res. https://doi.org/10.1093/nar/gkaf354
(2025)."),[24](https://www.nature.com/articles/s43587-026-01164-5#ref-CR24 "Welsh, H. et al. Age-related changes in DNA methylation in a sample of elderly Brazilians. Clin. Epigenetics 17, 17 (2025)."). This observation implies a transition toward epigenetic silencing, probably resulting from the closure of open chromatin regions, affecting enhancer activity or gene expression regulation[24](https://www.nature.com/articles/s43587-026-01164-5#ref-CR24 "Welsh, H. et al. Age-related changes in DNA methylation in a sample of elderly Brazilians. Clin. Epigenetics 17, 17 (2025)."),[25](https://www.nature.com/articles/s43587-026-01164-5#ref-CR25 "Lynch, C. J., Richart, L. & Serrano, M. A pattern emerges in chromatin aging: AP-1 steals the show. Cell Metab. 36, 1639–1641 (2024)."). Exceptions to this trend, were skeletal muscle and lung tissues, which demonstrated heightened hypomethylation, which may signify tissue-specific aging mechanisms[6](https://www.nature.com/articles/s43587-026-01164-5#ref-CR6 "Voisin, S. et al. Exercise is associated with younger methylome and transcriptome profiles in human skeletal muscle. Aging Cell 23, e13859 (2024)."),[26](https://www.nature.com/articles/s43587-026-01164-5#ref-CR26 "Ramirez, J. M. et al. The molecular impact of cigarette smoking resembles aging across tissues. Genome Med. 17, 66 (2025)."),[27](https://www.nature.com/articles/s43587-026-01164-5#ref-CR27 "Zhan, Y. et al. DNA hypomethylation-mediated upregulation of GADD45B facilitates airway inflammation and epithelial cell senescence in COPD. J. Adv. Res. 68, 201–214 (2025)."), encompassing demethylation at regulatory or structural loci.
The predominance of age-associated hypermethylation across multiple tissues should not be interpreted (on its own) as evidence against stochastic epigenetic drift. Because many tissues in the younger cohort showed a predominance of low methylation states, age-associated increases in methylation are, in many cases, compatible with drift toward more intermediate methylation levels (Supplementary Table 8). To address this, we examined methylation-state transitions between young and older groups in parallel with the direction of methylation change. This comparison suggests that, in several tissues, a substantial fraction of age-associated increases probably reflects movement toward intermediate methylation states, consistent with entropic drift, whereas in others, the patterns may also be compatible with more structured regional remodeling. In addition, although some age-associated signal was attenuated after deconvolution, many DMPs persisted, indicating that both cell-compositional changes and cell-intrinsic methylation alterations contribute to tissue aging. Together, these findings support a model in which age-related methylation change reflects a mixture of stochastic drift and tissue-specific epigenetic remodeling, rather than a single uniform process[7](https://www.nature.com/articles/s43587-026-01164-5#ref-CR7 "Campisi, J. et al. From discoveries in ageing research to therapeutics for healthy ageing. Nature 571, 183–192 (2019)."),[20](https://www.nature.com/articles/s43587-026-01164-5#ref-CR20 "Slieker, R. C. et al. Age-related accrual of methylomic variability is linked to fundamental ageing mechanisms. Genome Biol. 17, 191 (2016)."),[21](https://www.nature.com/articles/s43587-026-01164-5#ref-CR21 "Teschendorff, A. E., West, J. & Beck, S. Age-associated epigenetic drift: implications, and a case of epigenetic thrift? Hum. Mol. Genet. 22, R7–R15 (2013)."),[28](https://www.nature.com/articles/s43587-026-01164-5#ref-CR28 "Debette, S. et al. Visceral fat is associated with lower brain volume in healthy middle-aged adults. Ann. Neurol. 68, 136–144 (2010)."). This insight expands upon previous aging epigenome studies that primarily quantified methylation changes as a binary gain or loss, and highlights the importance of baseline methylation state in shaping the trajectory of age-associated epigenetic remodeling. Future studies should investigate whether these directional shifts correspond with chromatin accessibility and histone modifications, and whether they confer functional consequences on regulatory landscapes relevant to tissue aging and disease vulnerability.
Unlike DMPs, VMPs are rare and highly tissue-specific, supporting the idea that an aging-related epigenetic drift occurs randomly and locally[4](https://www.nature.com/articles/s43587-026-01164-5#ref-CR4 "Seale, K., Horvath, S., Teschendorff, A., Eynon, N. & Voisin, S. Making sense of the ageing methylome. Nat. Rev. Genet. 23, 585–605 (2022)."),[5](https://www.nature.com/articles/s43587-026-01164-5#ref-CR5 "Seale, K., Teschendorff, A., Reiner, A. P., Voisin, S. & Eynon, N. A comprehensive map of the aging blood methylome in humans. Genome Biol. 25, 240 (2024)."),[29](https://www.nature.com/articles/s43587-026-01164-5#ref-CR29 "Okada, D., Cheng, J. H., Zheng, C., Kumaki, T. & Yamada, R. Data-driven identification and classification of nonlinear aging patterns reveals the landscape of associations between DNA methylation and aging. Hum. Genomics 17, 8 (2023)."),[30](https://www.nature.com/articles/s43587-026-01164-5#ref-CR30 "Ziller, M. J. et al. Charting a dynamic DNA methylation landscape of the human genome. Nature 500, 477–481 (2013)."). Few tissues show many VMPs, and most lack consistent pathway enrichment, except buccal and lung tissues, implying these changes are probably random degradation rather than biological responses[7](https://www.nature.com/articles/s43587-026-01164-5#ref-CR7 "Campisi, J. et al. From discoveries in ageing research to therapeutics for healthy ageing. Nature 571, 183–192 (2019)."),[31](https://www.nature.com/articles/s43587-026-01164-5#ref-CR31 "Tarkhov, A. E. et al. Nature of epigenetic aging from a single-cell perspective. Nat. Aging 4, 854–870 (2024)."). This questions the view that methylation variability is always harmful or meaningful in aging, emphasizing the need to distinguish noise from genuine signals[29](https://www.nature.com/articles/s43587-026-01164-5#ref-CR29 "Okada, D., Cheng, J. H., Zheng, C., Kumaki, T. & Yamada, R. Data-driven identification and classification of nonlinear aging patterns reveals the landscape of associations between DNA methylation and aging. Hum. Genomics 17, 8 (2023)."). Entropy, linked to molecular chaos, varies by tissue and is highest in metabolically active organs such as adipose, breast and kidney, possibly indicating regulatory loss in energy-demanding tissues[7](https://www.nature.com/articles/s43587-026-01164-5#ref-CR7 "Campisi, J. et al. From discoveries in ageing research to therapeutics for healthy ageing. Nature 571, 183–192 (2019)."),[32](https://www.nature.com/articles/s43587-026-01164-5#ref-CR32 "Vaidya, H. et al. DNA methylation entropy as a measure of stem cell replication and aging. Genome Biol. 24, 27 (2023)."),[33](https://www.nature.com/articles/s43587-026-01164-5#ref-CR33 "Xie, H. et al. DNA methylation modulates aging process in adipocytes. Aging Dis. 13, 433–446 (2022)."). The brain’s entropy profile is unique, balancing DMPs and VMPs, hinting at a different aging process that may reduce both directional and random methylation changes[7](https://www.nature.com/articles/s43587-026-01164-5#ref-CR7 "Campisi, J. et al. From discoveries in ageing research to therapeutics for healthy ageing. Nature 571, 183–192 (2019)."),[17](https://www.nature.com/articles/s43587-026-01164-5#ref-CR17 "Jeong, H., Mendizabal, I. & Yi, S. V. Human brain aging is associated with dysregulation of cell type epigenetic identity. Geroscience 47, 3759–3770 (2025).").
It is well established that a subset of CpGs is under genetic control[34](https://www.nature.com/articles/s43587-026-01164-5#ref-CR34 "Oliva, M. et al. DNA methylation QTL mapping across diverse human tissues provides molecular links between genetic variation and complex traits. Nat. Genet. 55, 112–122 (2023).") methylation quantitative trait locus and may exhibit higher inter-individual variability. However, our VMP analysis tested for age-associated changes in variance rather than baseline variability per se. Because methylation quantitative trait locus effects are generally stable across the lifespan, they are unlikely to account for systematic age-dependent variance shifts unless interacting with age. Thus, while genetic regulation may influence absolute methylation variability, the VMPs identified here reflect loci where variance changes with age.
Our WGCNA analysis identified modules of co-methylated CpGs with distinct aging patterns and functions. Modules with higher methylation in older individuals were linked to developmental and adhesion pathways, while those with decreased methylation related to immune, metabolic and neurogenic signaling. This reflects aging hallmarks: reduced regenerative capacity and altered cell communication. GTPase signaling and cell-adhesion pathways were enriched across tissues, indicating aging affects synaptic and cytoskeletal functions. Small GTPases such as Rho, Ras and Rab are key in cytoskeletal, vesicle, and cell communication activities that decline with age. Dysregulation of GTPases is linked to neurodegeneration, immunosenescence and senescence[7](https://www.nature.com/articles/s43587-026-01164-5#ref-CR7 "Campisi, J. et al. From discoveries in ageing research to therapeutics for healthy ageing. Nature 571, 183–192 (2019)."),[35](https://www.nature.com/articles/s43587-026-01164-5#ref-CR35 "Papke, B. & Der, C. J. Drugging RAS: know the enemy. Science 355, 1158–1163 (2017)."),[36](https://www.nature.com/articles/s43587-026-01164-5#ref-CR36 "Westrip, C. A. E., Zhuang, Q., Hall, C., Eaton, C. D. & Coleman, M. L. Developmentally regulated GTPases: structure, function and roles in disease. Cell. Mol. Life Sci. 78, 7219–7235 (2021)."),[37](https://www.nature.com/articles/s43587-026-01164-5#ref-CR37 "Yin, G. et al. Targeting small GTPases: emerging grasps on previously untamable targets, pioneered by KRAS. Signal Transduct. Target. Ther. 8, 212 (2023)."). Growing interest in drugs targeting GTPases as gerotherapeutics supports our findings, demonstrating their relevance.
The PCDHG gene family emerged repeatedly as a module-influential candidate across the tissue-specific WGCNA network and the pan-tissue analysis. PCDHG genes encode cell-adhesion proteins critical for synaptic organization[38](https://www.nature.com/articles/s43587-026-01164-5#ref-CR38 "Meltzer, S. et al. gamma-Protocadherins control synapse formation and peripheral branching of touch sensory neurons. Neuron 111, 1776–1794 (2023)."), and hypermethylation in this family has been associated with reduced brain white matter[39](https://www.nature.com/articles/s43587-026-01164-5#ref-CR39 "Schmithorst, V. et al. Complex regulation of protocadherin epigenetics on aging-related brain health. Preprint at medRxiv https://doi.org/10.1101/2024.04.21.24306143
(2024)."), a marker of cognitive decline. This connection is intriguing as previous studies show abdominal fat can predict decreases in cerebral white matter[28](https://www.nature.com/articles/s43587-026-01164-5#ref-CR28 "Debette, S. et al. Visceral fat is associated with lower brain volume in healthy middle-aged adults. Ann. Neurol. 68, 136–144 (2010)."),[40](https://www.nature.com/articles/s43587-026-01164-5#ref-CR40 "Nam, K. W. et al. Abdominal fatn