Age-varying DNA methylation patterns associated with blood pressure in mid-to-late adulthood Authors: Gang Xu, Xiaowei Zhuang, Amei Amei, Zuoheng Wang, Edwin C Oh Published: 2026 ABSTRACT AbstractBackgroundEpigenome-wide association studies (EWAS) have identified associations between DNA methylation and blood pressure, yet most rely on single-time-point data and cannot capture how methylation and blood pressure relationships change with age.MethodsWe conducted a longitudinal EWAS of 1945 blood samples from 976 participants in the multi-ethnic study of atherosclerosis using a spline-based varying-coefficient model to detect age-dependent associations between DNA methylation in blood and blood pressure traits. Findings were evaluated for replication in 1187 samples from the Framingham heart study. Models were adjusted for sex, ancestry, and leukocyte composition to account for cellular heterogeneity.ResultsSix CpG sites showed significant age-dependent associations with systolic or pulse pressure after correction for multiple testing. These included loci within STIP1 and CSRP1 that replicated in the Framingham cohort. Several CpG sites demonstrated a reversal of effect direction with advancing age, where higher methylation was associated with higher systolic pressure in younger adults but lower pressure in mid-to-late adulthood. Pathway enrichment analyses identified focal adhesion, actin cytoskeleton remodeling, and Wnt/β-catenin signaling, which are processes relevant to vascular aging. Drug target mapping identified 23 FDA-approved agents interacting with genes at these loci.ConclusionsBlood-derived DNA methylation shows dynamic age-related associations with blood pressure, most pronounced in mid-to-late adulthood, that likely reflect systemic or vascular aging processes rather than direct cellular mediation. Longitudinal analytical frameworks can reveal temporal patterns in epigenetic variation that are not detectable in single time point studies and may inform the discovery of biomarkers for age related cardiovascular risk.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13148-026-02092-2.Keywords: Epigenome‑wide association study (EWAS), Age‑varying DNA methylation effect, Age‑dependent epigenetic regulation, Blood pressure, Longitudinal cohort Introduction Hypertension is a chronic condition characterized by persistently high blood pressure and serves as one of the primary preventable risk factors for cardiovascular disease and dementia [1–3]. Although the genetic and environmental factors that contribute to hypertension have been extensively studied, growing evidence indicates that DNA methylation, an epigenetic process that modifies gene expression without altering DNA sequence, may reflect how genetic susceptibility and environmental exposures jointly influence vascular health [4–6]. Current insights largely stem from cross-sectional genetic and epigenetic studies, revealing significant methylation changes associated with vascular aging and cardiovascular risk [5, 6]. However, these studies only capture static snapshots of methylation at single time points, potentially missing key temporal dynamics. Understanding how DNA methylation is associated with hypertension progression across mid-to-late adulthood could improve mechanistic insight and support the identification of biomarkers and therapeutic targets that vary with age. Conventional epigenome-wide association studies (EWAS) typically employ analytical approaches, such as generalized estimating equations (GEE) and generalized linear mixed models (GLMM), that inherently assume methylation effects remain fixed throughout life. This assumption may overlook biologically critical temporal variations and dynamic gene-environment interactions, limiting the detection of age-specific epigenetic modifications crucial for understanding hypertension progression [7, 8]. For example, age-stratified analyses of global leukocyte DNA methylation have shown that the inverse association with diastolic blood pressure is much stronger in adults > 45 years than in younger adults [9], and cord-blood methylation at HIF3A has been linked to higher systolic blood pressure in early childhood [10], indicating that the impact of methylation on blood pressure can differ across the life course. Recent advances in statistical modeling allow the evaluation of dynamic genomic and epigenetic effects over time. One such method, the Varying coefficient Mixed Model Association Test (VMMAT), employs spline-based modeling to effectively capture smooth temporal variations in dynamic genomic effects across age intervals [11–13]. Other longitudinal methods, including functional principal component analysis (FPCA), Bayesian hierarchical modeling, and penalized spline regression, have also shown promise for identifying complex age dependent molecular patterns [14–16]. Incorporating such dynamic longitudinal techniques into EWAS analyses could substantially improve our understanding of epigenetic regulation across mid-to-late adulthood, helping to identify critical windows of susceptibility and inform targeted therapeutic strategies for hypertension. DNA methylation influences hypertension progression by modulating both established physiological mechanisms and emerging epigenetic pathways. Hypertension involves well-characterized biological pathways, such as the renin–angiotensin–aldosterone system (RAAS), renal sodium handling, and vascular endothelial function, all crucial in regulating blood pressure homeostasis [17–19]. Consistent methylation changes have been identified in key genes within these canonical pathways, including angiotensin-converting enzyme (ACE), aldosterone synthase (CYP11B2), and endothelial nitric oxide synthase (eNOS), further implicating epigenetic mechanisms in hypertension pathogenesis [20–22]. Additionally, recent data have highlighted novel methylation-driven regulatory processes influencing blood pressure through pathways previously underappreciated, such as genes involved in cytoskeletal remodeling, stress response signaling, inflammatory regulation, and epigenetic modulators themselves [6, 23, 24]. These emerging findings demonstrate how environmental exposures—including diet, psychosocial stress, physical activity, and lifestyle behaviors—can trigger methylation changes across both classical and newly identified pathways, thereby shaping individual susceptibility to hypertension throughout the life course. Integrating these diverse physiological and epigenetic perspectives could reveal critical mechanisms and potential intervention points, improving clinical strategies for hypertension prevention and management. In this study, we applied the VMMAT approach to identify age-dependent associations between DNA methylation and blood pressure, addressing limitations of traditional cross-sectional EWAS methods. To our knowledge, this represents the first longitudinal EWAS investigating DNA methylation dynamics in hypertension. Using longitudinal data from the Multi-Ethnic Study of Atherosclerosis (MESA) [25]—a diverse cohort with comprehensive repeated measurements of blood pressure and DNA methylation—we identified significant age-dependent epigenetic associations with systolic blood pressure (SBP) and pulse pressure (PP). These associations were subsequently validated in an independent cohort, the Framingham Heart Study (FHS) [26, 27]. Pathway analyses contextualized the identified methylation sites within biological networks, highlighting key cardiovascular-related pathways, such as focal adhesion, actin cytoskeletal remodeling, and catenin signaling. Integration with publicly available drug and gene databases further highlighted the potential therapeutic relevance of several identified loci. Together, these results demonstrate that longitudinal modeling can uncover dynamic age related associations between DNA methylation and blood pressure that are not captured by cross-sectional approaches and may reflect molecular features of vascular aging. Results Study population The discovery DNA methylation dataset comprised 1945 blood samples collected longitudinally from 976 participants across two examination periods within the multi-ethnic study of atherosclerosis (MESA) [25, 28]. The replication methylation dataset consisted of 1187 blood samples from 1187 independent participants in the Framingham heart study (FHS) [26, 27]. Detailed demographic and clinical characteristics comparing participants from the MESA and FHS cohorts are summarized in Table 1. Table 1Demographic and clinical characteristics of study participants from the multi-ethnic study of atherosclerosis (MESA) and Framingham heart study (FHS)CharacteristicMESAFHSn = 975n = 1187Age at baseline (years), mean (SD)60.34 (9.66)72.11 (8.08)Male, n (%)461 (47.28)562 (45.66)SBP at baseline (years), mean (SD)122.89 (20.41)126.78 (16.55)DBP at baseline (years), mean (SD)71.52 (10.16)70.68 (9.58)PP at baseline (years), mean (SD)51.36 (15.95)56.10 (15.40)MESA, multi-ethnic study of atherosclerosis; FHS, Framingham heart study; SD, standard deviation; SBP, systolic blood pressure; DBP, diastolic blood pressure; PP, pulse pressure Demographic and clinical characteristics of study participants from the multi-ethnic study of atherosclerosis (MESA) and Framingham heart study (FHS) MESA, multi-ethnic study of atherosclerosis; FHS, Framingham heart study; SD, standard deviation; SBP, systolic blood pressure; DBP, diastolic blood pressure; PP, pulse pressure Detection of age-varying methylation effects on blood pressure Dynamic DNA methylation effects on blood pressure traits were assessed using the varying coefficient mixed model association test (VMMAT) [13], which explicitly models age-dependent methylation effects. For comparison, we also performed a conventional mixed-effects epigenome-wide association analysis (EWAS) assuming age-invariant methylation effects. Genome control inflation factors [29] (λ) were computed to assess potential systemic bias in our association test statistics. Ideally, λ should equal 1; values below 1.1 are typically considered acceptable in genome-wide analyses [30], with modest inflation reflecting polygenicity rather than confounding. In our study, SBP analyses yielded λ = 1.10 (VMMAT) and 1.00 (GLMM); PP analyses had λ = 1.13 (VMMAT) and 1.06 (GLMM); and DBP analyses showed λ = 0.99 (VMMAT) and 0.85 (GLMM). Although the VMMAT λ for PP slightly exceeded the conventional threshold, it is still within a range unlikely to compromise result validity, as thresholds up to ~ 1.2 are often tolerated in polygenic traits [31] (Fig. 1). Analytical pipeline of the longitudinal methylation study. DNA methylation sites associated with blood pressure were identified using the varying coefficient mixed model association test (VMMAT) in the Multi-Ethnic Study of Atherosclerosis (MESA). Significant CpG sites underwent subsequent analyses, including: (1) functional enrichment analyses to identify relevant biological pathways; (2) drug repurposing analyses for potential therapeutic relevance; and (3) evaluation of age-dependent methylation effects on systolic and pulse pressures. Replication analyses were performed using the independent Framingham heart study (FHS) dataset Applying a stringent Bonferroni-corrected genome-wide significance threshold of 1.47 × 10−7 (0.05/339,122 CpGs), we identified two significant CpG sites with age-varying effects on blood pressure: cg08023851 within the STIP1 gene (p = 1.37 × 10−7) significantly associated with SBP, and cg02165526 within the CSRP1 gene (p = 1.39 × 10−7) significantly associated with PP. To explore additional methylation signals at a less conservative threshold, we applied the Benjamini–Hochberg false discovery rate (FDR) correction [32], identifying four additional CpG sites significantly associated with SBP (Table 2 and Fig. 2A). In total, six CpG sites passed FDR control (five for SBP and one for PP), and all six are reported in Table 2. Despite evaluating associations with DBP, no CpGs reached genome-wide significance. Consequently, further analyses focused solely on SBP and PP, for which significant age-varying effects were evident. In contrast, the benchmark EWAS using a conventional non–age-varying mixed-effects model did not identify any CpG sites reaching genome-wide significance for SBP, DBP, or PP. Table 2Genome-wide significant CpG sites associated with blood pressure traits identified in the multi-ethnic study of atherosclerosis (MESA) and replicated in the Framingham heart study (FHS)PhenotypeChrGene regionCpGPositionDiscovery VMMATDiscovery VMMAT_FDRDiscovery GLMMReplication VMMATSBP7LFNGcg081962672,559,4715.24E−070.041.50E−031.59E−0111STIP1cg0802385163,953,1931.37E−070.043.50E−062.17E−0611CTTNcg2724312170,281,5845.31E−070.043.83E−042.15E−0222RTN4Rcg0347542020,255,6676.28E−070.042,41E−041.54E−01XKDM6Acg0555058844,732,7335.00E−070.042.12E−016.76E−01PP1CSRP1cg02165526201,478,3581.39E−070.057.46E−061.41E−02Discovery p-values are bold if they are smaller than CpGs. Replication p-values are bold if they are smaller than discovery CpGs for SBP and discovery CpGs for PP Genome-wide significant CpG sites associated with blood pressure traits identified in the multi-ethnic study of atherosclerosis (MESA) and replicated in the Framingham heart study (FHS) Discovery p-values are bold if they are smaller than CpGs. Replication p-values are bold if they are smaller than discovery CpGs for SBP and discovery CpGs for PP Genome-wide association analysis of methylation effects on blood pressure. Manhattan plots illustrate significant CpG sites associated with A systolic blood pressure (SBP) and B pulse pressure (PP). Genome-wide significant thresholds (Bonferroni-corrected, p < 1.47 × 10−7) and false discovery rate (FDR)-corrected significant sites are indicated In addition, we compared our benchmark, non–age-varying mixed-effects EWAS results with CpG sites previously reported in a 5-year longitudinal twin study of blood pressure [33]. Of the 7 CpGs reported as associated with SBP, 3 were excluded during probe-level quality control in our data, while among the remaining 4, three showed the same direction of effect and one showed the opposite direction. Of the 3 CpGs reported for DBP, two were retained in our dataset and both exhibited concordant effect directions. None of these CpGs reached nominal significance in our benchmark analysis (all raw p-values > 0.05), which may reflect differences in study design, sample size, and population composition. Replication analyses in the independent FHS cohort validated our findings. Specifically, CpG cg08023851 (STIP1) demonstrated significant replication for SBP (p = 2.17 × 10−6), surpassing the Bonferroni-corrected threshold (0.05/5 CpGs tested). Similarly, CpG cg02165526 (CSRP1) significantly replicated for PP (p = 1.42 × 10−2). These robust replications underscore the reliability and generalizability of our primary findings. Several significant CpGs mapped to genes previously implicated in blood pressure regulation. Specifically, KDM6A was associated with elevated blood pressure in mouse models [34], STIP1 variants were recently linked to DBP in a large-scale genome-wide association study (GWAS) [35], RTN4R polymorphisms correlated with elevated systolic and diastolic pressures in the DASH trial [36], and LFNG was identified as a locus for SBP in recent GWAS studies (Table 2) [37]. To gain insight into the regulatory context of the six age-varying CpGs, we have now systematically annotated the six genome-wide significant dmCpGs using the Illumina EPIC manifest, and eFORGE-TF [38–40]. First, using the Illumina annotation, we report for each CpG its genomic position, mapped gene(s), and UCSC RefGene and regulatory feature categories. Two of the six CpGs (cg08023851 in STIP1, cg08196267 in LFNG) fall in promoter-associated regions (TSS1500/TSS200, 5′UTR and/or 1st exon), while cg03475420 (RTN4R) lies in the 5′UTR/1st exon of RTN4R with an “Unclassified” regulatory feature and cg27243121 (CTTN) is located in the 3′UTR, and cg05550588 (KDM6A) spans the 5′UTR, first exon, and gene body of KDM6A. These annotations are now summarized in a new CpG-level Supplementary Table (Supplementary Table 10). Then, we used eFORGE-TF to evaluate overlap with DNase I hypersensitive footprints and predicted transcription-factor (TF) motifs. Four CpGs (cg08023851 in STIP1, cg08196267 in LFNG, cg03475420 in RTN4R, and cg05550588 in KDM6A) overlap high-confidence TF footprints in DNase-hypersensitive regions, with significant motif matches (e.g., Bcl6b_secondary or Zfp161_secondary; FIMO p-values ~ 10−7 to 10−9). In contrast, cg27243121 (CTTN) and cg02165526 (CSRP1) showed no significant TF footprints in eFORGE-TF. These TF-footprint annotations (presence/absence, motif name, FIMO p-value, and footprint genomic coordinates) have been added as new columns to Supplementary Table 10. Age-varying methylation effects We next evaluated the dynamic changes in methylation effects across different age groups for six significant CpGs identified in Table 2. The five CpGs associated with SBP demonstrated a progressive decrease in effect size and further changed in direction with advancing age, indicating that higher methylation at these loci correlated positively with systolic pressure in younger adults, transitioning to a negative correlation in older age groups (Fig. 3A, Supplementary Table 8). The CpG associated with PP, cg02165526, on the other hand, exhibited a progressively stronger positive association between methylation and increased pulse pressure across age strata (Fig. 3B, Supplementary Table 9). Collectively, these findings underscore the complex and dynamic age-dependent nature of DNA methylation's role in blood-pressure regulation. Age-dependent methylation effects on blood pressure. Linear regression coefficients illustrating the strength and direction of methylation effects on A systolic blood pressure (SBP) and B pulse pressure (PP) across different age groups. Each CpG is represented separately to highlight dynamic shifts in methylation impact across mid-to-late adulthood Pathway analysis To contextualize the identified CpG associations biologically, we examined pathway enrichment among genes annotated to CpGs associated with SBP and PP (see Methods for details). Utilizing Gene Ontology (GO) terms—including biological processes, molecular functions, and cellular components—we identified three significantly enriched pathways related to blood pressure (FDR-p < 0.05 in Fisher’s z test) [41] (Table 3, Supplementary Tables 6 and 7). The first significant pathway was focal adhesion (SBP: p = 2.59 × 10−5, FDR-p = 3.04 × 10−3; PP: p = 2.95 × 10−6, FDR-p = 7.37 × 10−4). Focal adhesion kinase (FAK) activation is known to influence vascular smooth muscle contraction and arterial stiffness, essential contributors to hypertension pathogenesis [42–44]. The second enriched pathway was actin cytoskeleton remodeling (PP: p = 7.59 × 10−6, FDR-p = 1.38 × 10−3), a pathway repeatedly implicated in both pulmonary and systemic hypertension due to its critical role in vascular tone regulation [45, 46]. Lastly, we observed enrichment of the catenin complex pathway (SBP: p = 7.29 × 10−4, FDR-p = 4.28 × 10−2), specifically involving the Wnt/β-catenin signaling cascade, which modulates blood pressure in animal models and is actively investigated for its therapeutic potential in hypertension management [47, 48]. Table 3Blood pressure related pathways based on the two gene lists linked to CpG sites associated with systolic blood pressure (SBP) or pulse pressure (PP)PhenotypePathwaysP-valuesFDR-pSBPFocal adhesion2.59E−053.04E−03Endosome2.87E−053.18E−03Glutamatergic synapse1.48E−041.41E−02Catenin complex7.29E−044.28E−02PPFocal adhesion2.95E−067.37E−04Actin binding3.21E−063.28E−03Myofibril7.21E−061.44E−03Actin cytoskeleton7.59E−061.38E−03Contractile muscle fiber1.31E−051.87E−03Sarcomere8.49E−056.78E−03Cation channel complex4.45E−042.61E−02Calcium channel complex6.06E−043.18E−02 Blood pressure related pathways based on the two gene lists linked to CpG sites associated with systolic blood pressure (SBP) or pulse pressure (PP) Drug repurposing analysis To identify therapeutic opportunities, we next evaluated whether the high-confidence, replicated loci might be directly targetable by existing drugs. Using two genes (STIP1 and CSRP1) mapped from CpGs that (i) passed FDR correction in the age-varying EWAS for SBP or PP and (ii) replicated in FHS (see Methods), no FDA-approved drugs were identified that directly target these genes. To generate hypotheses beyond this stringent set, we additionally conducted an exploratory screen based on the broader overlap of SBP- and PP-associated genes (273 genes; Supplementary Table 5). In this exploratory analysis, 23 FDA-approved drugs were found to interact with at least one gene in the broader set (Table 4). Among these, salmon calcitonin, though lacking clinical antihypertensive applications, increases blood pressure via the renin-angiotensin system in animal models [49]. Additionally, protamine sulfate, clinically used to counteract heparin anticoagulation, has demonstrated rapid blood pressure lowering through nitric oxide-mediated sympathetic inhibition and vasodilation [50, 51]. Furthermore, antithrombin alfa, used clinically in thrombotic conditions, can prevent endotoxin-induced hypotension in animal models through suppression of inducible nitric oxide synthase [52], highlighting potential relevance for clinical blood-pressure modulation. Table 4Drug repurposing analysis identifying FDA-approved drugs targeting genes linked to CpG sites associated with both systolic blood pressure (SBP) and pulse pressure (PP)Target nameApproved repurposable drugIndicationCALCRSalmon calcitoninOsteoporosisF10Antihemophilic factor, human recombinantHemophilia ACoagulation factor VIIa Recombinant HumanBleeding in individuals with Hemophilia A or B with inhibitorsCoagulation Factor IX (Recombinant)Hemophilia BTuroctocog alfaHemophilia AProtamine sulfateAnticoagulant effect of heparinAntithrombin alfaPeri-operative and peri-partum Thromboembolic events in hereditary antithrombin deficientCoagulation factor VII humanBleeding episodes in individuals with hemophilia A or BAnti-inhibitor coagulant complexHemophilia A and BCoagulation factor IX humanHemophilia BAntihemophilic factor humanHemophilia AAlbutrepenonacog alfaHemophilia BEmicizumabBleeding episodes in patients with hemophilia ANonacog beta pegolHemophilia BLonoctocog alfaHemophilia AMoroctocog alfaBleeding episodes in people with hemophilia ATuroctocog alfa pegolHemophilia AEfanesoctocog alfaHemophilia AERBB2TrastuzumabBreast cancer and other cancersTrastuzumab emtansineBreast cancerPertuzumabBreast cancerMargetuximabBreast cancerDMDGolodirsenDuchenne muscular dystrophy Drug repurposing analysis identifying FDA-approved drugs targeting genes linked to CpG sites associated with both systolic blood pressure (SBP) and pulse pressure (PP) Discussion Our longitudinal epigenome-wide association study (EWAS) identified several novel DNA methylation sites associated with blood pressure (BP), specifically highlighting significant CpG loci near the STIP1, CSRP1, LFNG, KDM6A, RTN4R, and CTTN genes. These loci exhibited genome-wide significant dynamic, age-dependent associations, where higher methylation was correlated with lower systolic blood pressure (SBP) in younger adults, but this relationship reversed in older individuals. Additionally, we observed a progressively stronger positive association between methylation and pulse pressure (PP) with advancing age, indicative of increased arterial stiffness. Pathway analyses further implicated focal adhesion, actin cytoskeletal remodeling, and Wnt/β-catenin signaling pathways, reinforcing their central roles in vascular aging and hypertension pathogenesis. Importantly, these associations were robustly replicated in an independent cohort from the Framingham Heart Study (FHS), underscoring their reliability and broader relevance. A key innovation of our study was the development and application of a longitudinal design coupled with the Varying coefficient Mixed Model Association Test (VMMAT), providing insights inaccessible through conventional cross-sectional approaches. Previous large-scale meta-EWAS, including the CHARGE consortium [6] and a subsequent EWAS involving ~ 4820 individuals of European and African ancestries [53], identified several CpGs associated with blood pressure, predominantly within metabolic and inflammatory pathways. These studies demonstrated substantial overlap in CpG findings across populations, and validation in twin cohorts further reinforced shared genetic-epigenetic associations [53–55]. While our primary aim was to uncover dynamic, age-dependent methylation patterns, we noted partial overlap with previously reported loci, such as PTPRN2, NR3C1, GABBR1, and FURIN; however, these loci emerged as less significant findings in our analyses. For example, PTPRN2 was identified in trans-ancestry DMR (Differentially Methylated Region) analyses of SBP [5], and NR3C1 methylation has been linked to blood pressure regulation in human cohort studies [56]. These earlier studies assumed constant methylation–phenotype relationships and did not account for temporal dynamics. In contrast, our VMMAT approach revealed CpG loci whose associations with blood pressure shift across mid-to-late adulthood, suggesting protective effects early in life may become harmful later. This introduces a new dimension in epigenetic epidemiology, emphasizing that risk markers can exert age-dependent beneficial or harmful effects [6]. While some overlap in genes reinforces the general relevance of epigenetic mechanisms, our most significant findings uniquely emphasize the importance of age-dependent changes, thereby filling an essential knowledge gap and demonstrating that traditional cross-sectional methods may miss critical epigenetic indicators associated with hypertension. The mechanistic implications of these age-varying methylation–BP links point to pathways of vascular aging and remodeling. Genes annotated to our top CpGs converge in pathways of focal adhesion, actin cytoskeleton dynamics, and catenin signaling, all fundamental to arterial structure and function [42, 57, 58]. For example, CTTN encodes cortactin, an actin-binding protein that stabilizes endothelial adherens junctions by linking VE-cadherin to the actin cytoskeleton through α/β-catenins [59, 60]. Hypermethylation in the CTTN locus (likely reducing its expression) in younger individuals might enhance junctional stability and lower BP, whereas in older adults the same methylation could reflect or induce impaired repair of junctions under stress, contributing to higher BP. Similarly, CSRP1 encodes a LIM-domain protein (CRP1) expressed in vascular smooth muscle; while not essential for baseline development, CRP1 modulates the smooth muscle response to arterial injury and stress [61, 62]. The enrichment of Wnt/β-catenin signaling in our pathway analysis further supports a role in vascular aging; Wnt–catenin is a key driver of cell proliferation and senescence in the vasculature, and its dysregulation is linked to arterial calcification and stiffness in aging arteries [63, 64]. Importantly, KDM6A, a histone demethylase identified here, has recently been implicated in blood pressure regulation and vascular integrity: loss of KDM6A in mice leads to increased arterial collagen deposition and stiffness [6], raising BP, and KDM6A appears essential for normal renal sodium handling and BP homeostasis [34]. Our finding that higher methylation (presumably lower expression) of KDM6A is associated with rising BP in older adults is consistent with this, hinting that age-related epigenetic silencing of an epigenomic regulator might exacerbate vascular aging. In aggregate, these mechanistic links suggest that the CpGs we identified are not merely statistical associations but map to biological processes—cytoskeletal remodeling, cell–matrix adhesion, developmental signaling—that plausibly connect epigenetic drift to vascular aging and hypertension. Despite its strengths, our study has important limitations. First, DNA methylation was measured in peripheral blood, which may not directly represent methylation changes in vascular tissues (endothelium or smooth muscle) that more directly govern BP. While some methylation signals are shared between blood and vascular tissue, others are tissue-specific; thus, our findings could be potential markers of systemic factors or immune processes. Future studies should examine these CpGs in vascular tissue or cell models to confirm functional relevance. Second, although longitudinal, our statistical model still tests the association between BP and methylation rather than casual interpretation. It remains possible that changes in BP lead to altered methylation (rather than vice versa), or that a third factor (e.g., medication, diet, inflammation) influences both. We attempted to mitigate confounding by adjusting for key covariates, but residual confounding and reverse causation are challenging to eliminate which require additional biological meaningful assumptions. Third, our sample size, while sizeable for a longitudinal EWAS (nearly 1,000 participants, ~ 1945 samples), is modest by genome-wide standards, and we may have missed smaller effects. Moreover, multiple testing in EWAS is stringent, possibly omitting biologically relevant associations below significance thresholds. Nonetheless, replication of major findings in an independent cohort and pathway-level validations substantially mitigate these concerns. Additionally, our finding of no significant associations with diastolic blood pressure (DBP) may reflect distinct underlying epigenetic mechanisms for DBP, limitations in statistical power, or age-specific sensitivities, which should be explored further. Furthermore, our analyses focus on single CpG sites rather than differentially methylated regions (DMRs). Region-based methods may identify additional coordinated methylation changes and represent an important direction for future work. Finally, the ancestry composition of our cohorts limits the generalizability of our findings. MESA includes participants of non-Hispanic White, African American, Hispanic, and Asian ancestry, whereas the FHS offsping cohort is predominantly of European ancestry. Although we adjust for population structure using principal components, our results may not capture methylation–blood pressure relationships in other ancestry groups or the full diversity within African populations, underscoring the need for future studies in more diverse cohorts. Looking forward, our study suggests several promising research directions. Future EWAS should routinely integrate longitudinal modeling to uncover dynamic epigenetic effects in other age-related phenotypes, such as glycemic control or cognitive decline. In hypertension specifically, integrating methylation data with transcriptomic [65] and proteomic [66] profiling could determine whether identified methylation shifts translate into gene expression or protein-level changes within vascular tissues. Such multi-omics approaches would greatly strengthen interpretations. Moreover, more frequent longitudinal sampling from younger adulthood through advanced ages could precisely pinpoint critical windows during which methylation exerts maximal influence on BP trajectories. Additionally, exploring relationships between BP-related methylation dynamics and accelerated biological aging (e.g., epigenetic clocks) may provide insights into how epigenetic markers predict cardiovascular risk over the life course [67–69]. Ultimately, our findings introduce a perspective where age-dependent epigenetic regulation dynamically influences cardiovascular trajectories across mid-to-late adulthood, opening new avenues for targeted preventive and therapeutic strategies for hypertension. Methods Study population of discovery dataset Data for the discovery study were obtained from the public available dataset MESA from dbGaP (dbGaP Study Accession: phs000209). The MESA cohort comprises 6,814 asymptomatic adults aged 45–84 years, of whom 38% are White, 28% African American, 22% Hispanic, and 12% Asian—predominantly of Chinese ancestry [25]. The first blood examination spanned July 2000–July 2002 and was followed by four additional examination visits. Study population of replication dataset Data for the replication study were obtained from the public available dataset FHS from dbGaP (dbGaP Study Accession: phs000007). The FHS began in 1948 with 5,209 Framingham, Massachusetts residents aged 30–62 and has examined them every two years ever since [70]. In 1971 the study added 5,124 adult children and spouses (the Offspring cohort) [71], and between 2002 and 2005 it enrolled 4,095 grandchildren (the Third Generation cohort) [72], each cohort undergoing repeat exams roughly every four years. Ethics approval and consent The MESA and FHS datasets were approved by their respective institutional review boards (IRBs). All participants provided written informed consent for participation and use of their genetic and clinical data. Our use of these publicly available datasets was reviewed and approved under an exempt status by the IRB at our institution (University of Nevada Las Vegas). DNA methylation profile and quality control DNA methylation was profiled in whole blood with the Illumina’s Methylation EPIC BeadChip (850 K) for 976 MESA participants who provided samples at Exams 1 and 5 as part of the National Heart, Lung, and Blood Institute (NHLBI) Trans-Omics for Precision Medicine (TOPMed) MESA Multi-Omics project (dbGaP Study Accession: phs001416.v4.p1). We performed DNA methylation quality control with the R package minfi [73]. For probe-level quality control, we computed detection p-values for all 1,051,539 array probes and removed any CpG probe with a detection p-value > 0.01 in at least one sample. This filtering step excluded 712,417 probes and resulted in 339,122 high-quality CpG sites. The derived beta () values were logit transformed to M-values for the downstream EWAS analysis. For replication, whole-blood DNA methylation in the FHS was assayed with the Illumina Infinium Methylation EPIC BeadChip (850 K sites) in 1,249 Offspring-cohort participants and 547 Generation 3 participants enrolled in the NHLBI TOPMed FHS Multi-Omics project (dbGaP Study Accession: phs000974.v6.p5). Downstream analyses were restricted to the 1,249 Offspring participants, for whom we extracted M-values from the processed methylation data. Cell type compositions (CD8T, CD4T, NK, Bcell, monocyte and granulocyte) were estimated in both MESA and FHS with the estimateCellCounts function [74–76] in R package minfi [73], with the cord-blood reference panel described previously [76], under default settings. Cell type composition estimators for CD4T, NK, Bcell, monocyte and granulocyte were then included as covariates in all EWAS models. Statistical modeling for age-varying methylation effects We modeled longitudinal outcomes using a generalized linear mixed model (GLMM) with varying coefficients allowing for methylation effects to vary smoothly over age. Specifically, let denote the trait value measured for subject at age . Suppose a trait is measured repeatedly on a sample of subjects. For each observation we have a -dimensional covariate vector which can include both static variables such as sex or dynamic variables such as cell type compositions and a DNA methylation measure expressed on the M value scale . The model is defined as: + + Here is the link function (logit for binary traits, identity for continuous traits). The functions and are smooth nonparametric functions of age representing a age-varying intercept and a age-varying methylation effect at the CpG under test, is the effects of the covariates. Correlation among repeat measurements is accommodated by two random terms: a subject-specific intercept and a vector of time-dependent random effects with following an AR(1) structure parameterized by [77, 78]. Conditional on and , the responses are assumed to be independent. The primary hypothesis of interest is whether there is a age-varying methylation effect between the methylation site and the trait. We test using the VMMAT [13], which combines the p-values of the spline basis coefficient and the variance component into a single Cauchy test statistic and provides a computationally efficient genome-wide screen for CpG sites with age-dependent associations. Then, we evaluated associations with systolic blood pressure (SBP), diastolic blood pressure (DBP), and pulse pressure (PP), employing the joint Cauchy combination test [79] implemented in VMMAT. As a benchmark comparison, we also implemented a generalized linear mixed model (GLMM) assuming a static (age-invariant) methylation effect. Both models incorporated age (VMMAT use cubic splines to model age-varying intercept and methylation effect), sex, the first four genomic principal components (PCs) of ancestry [28], and estimated cellular proportions (CD4 T, NK, B cell, Mono, and Gran, excluding CD8T to mitigate multicollinearity [80]). Subject-specific random intercepts and slopes were included to capture within-individual correlations across repeated measures. The overall analytical workflow is depicted in Fig. 1. All primary epigenome-wide analyses were conducted at the single-CpG level; we did not perform formal differentially methylated region (DMR) analyses in this study. Pathway analysis To investigate the biological functions of CpG sites associated with blood pressure, we conducted functional pathway enrichment using the PANTHER Classification System [81]. CpGs with p-values < 5 × 10−4 from either VMMAT or GLMM analyses were mapped to genes located within annotated regions, resulting in 521 genes for SBP and 586 for PP (Supplementary Tables 1–4). Over-representation of Gene Ontology (GO) terms was evaluated using Fisher’s exact test, and statistical significance was controlled via Benjamini–Hochberg FDR (threshold: FDR-p < 0.05). Drug repurposing analysis To identify potential therapeutics, we first focused on the two genes (STIP1 and CSRP1) mapped from CpGs that (i) passed FDR correction in the age-varying EWAS for SBP or PP and (ii) replicated in FHS. These genes were used as the primary, high-stringency input set for drug repurposing. To generate hypotheses beyond this stringent set, we additionally conducted an exploratory screen based on the broader overlap of 273 SBP- and PP-associated genes (CpGs p < 5 × 10−4; Supplementary Table 5). These shared genes were cross-referenced with the DrugBank database (v5.1.12) [82] to identify FDA-approved drugs targeting them. DrugBank provides comprehensive information on drug–target interactions, enabling systematic repurposing opportunities. Age-varying effects of significant CpG sites To examine the temporal dynamics in methylation effects, we stratified the cohort into seven sequential age groups. Within each stratum, we refitted regression models to estimate age-specific methylation coefficients for each significant CpG site (as listed in Table 2). Coefficients were then plotted across age strata with lines connecting adjacent age groups to visualize trajectories of methylation effects across mid-to-late adulthood. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1. Acknowledgements The Multi-Ethnic Study of Atherosclerosis is supported by contracts 75N92020D00001, HHSN268201500003I, N01-HC-95159, 75N92020D00005, N01-HC-95160, 75N92020D00002, N01-HC-95161, 75N92020D00003, N01-HC-95162, 75N92020D00006, N01-HC-95163, 75N92020D00004, N01-HC-95164, 75N92020D00007, N01-HC-95165, N01-HC-95166, N01-HC-95167, N01-HC-95168 and N01-HC-95169 from the National Heart, Lung, and Blood Institute, and by grants UL1-TR-000040, UL1-TR-001079, and UL1-TR-001420 from the National Center for Advancing Translational Sciences (NCATS). The authors thank the other investigators, the staff, and the participants of the MESA study for their valuable contributions. A full list of participating MESA investigators and institutions can be found at http://www.mesa-nhlbi.org. This manuscript was not prepared in collaboration with investigators of the MESA and does not necessarily reflect the opinions or conclusions of the MESA or the NHLBI. The Framingham Heart Study is conducted and supported by the National Heart, Lung, and Blood Institute (NHLBI) in collaboration with Boston University (Contract No. N01-HC-25195, HHSN268201500001I and 75N92019D00031). This manuscript was not prepared in collaboration with investigators of the Framingham Heart Study and does not necessarily reflect the opinions or views of the Framingham Heart Study, Boston University, or NHLBI. Author contributions GX, XZ, AA, and EO contributed to the study design and interpretation of the data; GX and XZ collected, curated, and visualized the data; GX, AA, and ZW designed the statistical analyses; GX, XZ, and EO analyzed the data; GX and EO drafted the manuscript. All authors contributed to the article and approved the submitted version. Funding ECO is supported by NIH (MH109706 and AI194218) and the Centers for Disease Control and Prevention (CDC—NH75OT000057-01-00). ZW is supported by NIH (LM014087). XZ is supported by NIH (AG071566). The project contents are solely the responsibility of the authors and do not necessarily represent the official views of the CDC. The corresponding authors have full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Data availability The DNA-methylation and phenotypic data analyzed in this study were obtained from dbGaP under the following study accessions: MESA cohort data, phs000209; MESA TOPMed DNA methylation, phs001416; FHS cohort data, phs000007; and FHS TOPMed DNA methylation, phs000974. Any request to access relevant data in this study should be sent to the corresponding authors. Code availability A R package that implements VMMAT for DNA methylation is provided at https://github.com/xug3/VMMAT_methylation. Declarations Ethics approval and consent to participate This study involves secondary analyses of existing, de-identified data from the Multi-Ethnic Study of Atherosclerosis (MESA) and the Framingham Heart Study (FHS). All original cohort protocols were approved by the local Institutional Review Boards or Ethics Committees at each participating institution, and all participants provided written informed consent at enrollment. The present secondary analysis was reviewed and approved by the Human Research Protection Program Institutional Review Board at Yale University (IRB protocol 2000036485; expedited review; approval date November 21, 2023) and by the Biomedical Institutional Review Board at the University of Nevada, Las Vegas (UNLV-2025–267, “Secondary multi-omics data analyses to understand genetic effect of Hypertension”; expedited review; approval date July 24, 2025). All procedures were conducted in accordance with the Declaration of Helsinki, the U.S. Federal Policy for the Protection of Human Subjects (45 CFR 46), and relevant institutional guidelines and regulations. Consent for publication All authors have provided consent to publication. Competing interests The authors declare no competing interests. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Gang Xu, Email: gang.xu@unlv.edu. Edwin C. Oh, Email: edwin.oh@unlv.edu References Associated Data Supplementary Materials Supplementary Material 1. Data Availability Statement The DNA-methylation and phenotypic data analyzed in this study were obtained from dbGaP under the following study accessions: MESA cohort data, phs000209; MESA TOPMed DNA methylation, phs001416; FHS cohort data, phs000007; and FHS TOPMed DNA methylation, phs000974. Any request to access relevant data in this study should be sent to the corresponding authors. A R package that implements VMMAT for DNA methylation is provided at https://github.com/xug3/VMMAT_methylation.