DNA methylation signatures of Life’s Essential 8 and their implications for dementia

Abstract

Background

As dementia cases continue to rise, effective prevention strategies are urgently needed. However, objective biomarkers that directly reflect lifestyle factors remain limited. Life’s Essential 8 (LE8) is a composite of modifiable cardiovascular health metrics, and lower LE8 has been consistently associated with increased risk of dementia. In this study, we aimed to identify DNA methylation biomarkers associated with LE8 scores and investigate their relevance for dementia risk.

Methods

We performed an epigenome-wide association study of 273 stroke-free, self-identified Hispanic adults aged 40 and older from the Northern Manhattan Study (NOMAS), a community-based urban cohort study. DNA methylation (DNAm) was assessed using Illumina MethylationEPIC arrays. Robust linear models identified CpGs associated with LE8 score, a composite score on eight health metrics including diet quality, physical activity, nicotine exposure, sleep health, body mass index, blood lipids, blood glucose, and blood pressure.

Differentially methylated regions were identified by combining P-values in sliding windows while accounting for spatial correlations across the genome. We also performed functional annotation, pathway analyses, and integrative analyses with gene expression, genetic variants, brain-blood correlations, and comparisons with previous dementia studies to identify the most biologically meaningful DNAm sites.

Results

After adjusting for age, sex, APOE ε4, immune cell composition, and ancestry, we found 11 CpGs with suggestive evidence of association with LE8 (P-value < 1 × 10^–5) and 37 differentially methylated regions that passed multiple-testing correction. These LE8-associated loci mapped to genes and pathways that support vascular integrity and regulate inflammation, key biological processes relevant to both cardiovascular disease and dementia. Integrative analyses highlighted several CpGs in the HOXA5 gene promoter with converging evidence supporting their potential as dementia biomarkers, including strong blood–brain DNAm correlations, association with gene expression and genetic variants, and prior associations with Alzheimer’s disease neuropathology.

Conclusions

Our comparison with published results showed that a number of LE8-associated DNA methylation sites are associated with dementia, highlighting the possible connection between cardiovascular health and dementia risk and pointing to potential actionable targets for dementia prevention. Moreover, DNAm biomarkers have clinical potential as objective measures to identify individuals at elevated risk, stratify participants based on biologically informed risk profiles, and monitor epigenetic responses to lifestyle interventions in dementia prevention trials. Future studies in larger and more diverse cohorts are needed to validate and refine these methylation biomarkers for clinical applications.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13195-025-01903-7.

Introduction

As the U.S. population ages, the number of individuals with dementia continues to rise, posing a major public health concern. Because the underlying neurodegenerative processes are difficult to halt, the escalating healthcare burden associated with dementia underscores a critical need for more effective prevention strategies. Recent studies have increasingly recognized the close connection between cardiovascular health and brain health [1–5]. In particular, multiple studies demonstrated that modifiable lifestyle factors contributing to cardiovascular health, such as those captured by Life’s Essential 8 (LE8), are significantly associated with cognitive decline and future onset of dementia [6, 7]. The LE8 score includes key metrics of blood pressure, non-high-density lipoprotein (HDL) blood cholesterol, blood glucose, smoking, physical activity, diet, body mass index (BMI), and sleep duration [8]. Each component is scored on a scale of 0 to 100, with higher values indicating better cardiovascular health.

We focused on LE8 because it is a contemporary, behavior-centered construct of cardiovascular health that captures modifiable lifestyle factors. In contrast to event-risk calculators (e.g., Framingham Risk Score) that estimate 10-year cardiovascular events and are heavily age-driven, LE8 directly quantifies behaviors most relevant to prevention. Moreover, LE8 components are widely harmonized, facilitating reproducibility and cross-study comparability.

Here we studied DNA-methylation (DNAm) biomarkers associated with LE8 scores and their relevance to dementia. DNAm is an epigenetic mechanism influenced by both genetics and environment. DNAm can be assayed from peripheral blood at scale and is stable over months to years to reflect cumulative exposures yet remains responsive to lifestyle (diet, smoking, physical activity), making it an excellent source of biomarkers [9]. Because LE8 are significantly associated with incident dementia, DNAm-based biomarkers may also help objectively identify individuals at elevated dementia risk. Although numerous biomarkers, including neuroimaging, CSF/plasma proteins, metabolomics, and polygenic risk scores have advanced dementia research, each has limitations as biomarkers for lifestyle-focused prevention. Genotypes are static and cannot capture behavioral change; proteins and metabolites are susceptible to degradation during sample collection and storage; self-reported health behaviors are prone to recall bias and missingness; and neuroimaging often reflects downstream pathology and is costly for population-level screening. Importantly, while plasma AD biomarkers such as pTau primarily reflect downstream neurodegenerative processes, DNAm can capture upstream, cumulative, and potentially modifiable effects of lifestyle and cardiovascular risk, offering a complementary biomarker for early risk detection, participant stratification, and intervention response monitoring.

A number of previous studies, including cohorts with substantial representation of non-Hispanic White participants, have associated composite cardiovascular health metrics with blood DNAm and epigenetic aging. In particular, LE8 and its predecessor Life’s Simple 7 (LS7) were shown to be significantly associated with epigenetic age acceleration in multiple cohort analyses, underscoring the molecular connection between cardiovascular health and DNA methylation [10–13]. Moreover, there is also extensive literature on epigenetic markers of individual LE8 components. For example, cigarette smoking shows the most robust and reproducible differential methylation signals at loci such as AHRR and F2RL3 [14]. Blood lipid levels associate with methylation at CPT1A, ABCG1, and SREBF1 in large EWAS [15], while glycemia and HbA1c consistently link to methylation at TXNIP [16]. Furthermore, emerging studies have also associated objectively measured physical activity to differential methylation [17, 18]. Together, these findings indicate that both composite cardiovascular health and its constituent behavioral factors are reflected in the blood methylome.

In this study, we aimed to identify DNAm biomarkers associated with LE8 scores and evaluate their relevance to dementia risk in this study. We hypothesized that DNAm signatures reflecting better cardiovascular health, as measured by LE8 scores, will correlate with decreased dementia risk due to shared underlying biological pathways. To this end, we analyzed DNAm and LE8 data from 273 self-identified Hispanic participants in the Northern Manhattan Study (NOMAS), a multi-ethnic cohort investigating stroke and age-related health outcomes in northern Manhattan residents [19]. In addition to identifying LE8-associated differentially methylated CpGs and regions (DMRs), we also compared our findings with results from previously published dementia studies and multiple additional external resources, including eQTM (methylation and gene expression associations), mQTL (methylation and genetic variant associations), GWAS summary statistics, brain-blood correlations, biological pathway database, to better understand the functional roles of LE8-associated DNA methylation changes. Our findings highlight shared molecular pathways between cardiovascular disease and dementia, underscoring the potential of these DNA methylation biomarkers as actionable targets for dementia prevention and objective measures in lifestyle-based intervention trials.

Methods

Study subjects

The Northern Manhattan Study (NOMAS) is a community-based cohort study that investigates the incidence and risk factors for stroke, dementia, and other vascular outcomes. Stroke free adults aged ≥ 40 years who had resided in the Northern Manhattan community for at least three months were enrolled between 1993 and 2001. Detailed descriptions of the NOMAS design and protocols have been published previously [19]. The study was approved by the Institutional Review Boards of Columbia University and the University of Miami, and all participants provided written informed consent.

DNA methylation data

For the present analysis, we included DNA methylation data generated from blood samples of 273 self-identified Hispanic participants. These blood samples were processed in two separate DNA methylation experiments conducted in 2019 and 2021, resulting in two non-overlapping sample sets: the Year2019 dataset (n = 134) and the Year2021 dataset (n = 139). For each experiment, genomic DNA (500 ng) was extracted from buffy coats and bisulfite-converted using the EZ-96 DNA Methylation™ Kit (Zymo Research). CpG methylation was measured using the Illumina Infinium MethylationEPIC v1.0 Beadchip arrays at the Center for Genomic Technology (CGT), John P. Hussman Institute for Human Genomics (HIHG).

We implemented stringent quality control (QC) procedures to ensure reliable DNAm estimates. Supplementary Table 1 shows detailed probe and sample counts at each step. Low-quality samples were excluded if they showed poor bisulfite conversion (< 85%) or sex mismatch, and problematic probes were removed if they were cross-reactive [20], located near SNPs, mapped to sex-chromosomes, or lacked annotation. Missing values and failed probes (detection P-values > 0.01) were imputed using the methyLImp2 R package [21], which is designed specifically for methylation arrays. To reduce probe-design bias, we normalized beta values using β-mixture quantile normalization (BMIQ) as implemented in the watermelon R package [22]. To reduce technical confounding, we corrected batch effects using the Harman R package [23]. Next, we performed principal component analysis (PCA) on the sample-by-CpG matrix (using methylation M-values). For each sample we computed Z-scores on the first two principal components (PC1 and PC2). We flagged a sample as an outlier if it exceeded ± 3 standard deviations on either PC1 or PC2. These steps minimize technical artifacts so that downstream associations reflect biology rather than measurement noise. Finally, as blood cell composition is a strong driver of DNAm, we estimated immune cell proportions using the EpiDISH R package. Granulocyte proportions were computed as the sum of neutrophils and eosinophils proportions, as both cell types are classified as granular leukocytes. Similarly, ancestry-informative components were estimated using the EPISTRUCTURE software [24], so we could adjust for them in subsequent association analysis models.

Life’s essential 8

Lifestyle factors and clinical measurements were collected during baseline visits. The LE8 scoring system comprises eight components: diet, physical activity, nicotine exposure, sleep, body mass index (BMI), blood lipids, blood glucose, and blood pressure. LE8 scores were computed following the methodology described in Lloyd-Jones et al. [8]. The specific thresholds used for NOMAS participants are detailed in Supplementary Table 2.

Specifically, diet was assessed using a modified food frequency questionnaire. A diet score was calculated based on caloric intake and intake of beneficial and detrimental food groups, reflecting adherence to the Dietary Approaches to Stop Hypertension (DASH) diet [25]. Physical activity was measured with a modified National Health Interview Survey questionnaire that captured the frequency and duration of 14 activities over the prior two weeks, converted to minutes per week of moderate or vigorous activity [26]. Nicotine exposure was self-reported as current (within the past year), former, or never smoker of cigarettes, cigars, or pipes. Sleep health was based on self-reported average nightly sleep hours. BMI was calculated from measured weight and height (kg/m²). Blood lipids were obtained after an 8-h fast. Total cholesterol was measured enzymatically (Hitachi 705 autoanalyzer), and HDL cholesterol after precipitation of apolipoprotein B-containing lipoproteins [27]. Non-HDL cholesterol was calculated as total minus HDL cholesterol. Blood glucose was measured after an 8-h fast [28]. Because HbA1c was not available, fasting plasma glucose was converted to estimated HbA1c using validated equations [29]. Blood pressure was measured in a seated position after 5 min of rest using a random-zero sphygmomanometer, with two readings 10 min apart averaged [30]. Self-reported use of lipid-, glucose-, and blood pressure-lowering medication was incorporated in LE8 scoring.

A total of 10 samples from the 2019 batch and 32 samples from the 2021 batch were excluded due to missing LE8 scores. Comparison of participants with available LE8 data (i.e., subjects included in our analysis) to the overall sample (including those with and without missing LE8), the demographic distributions were very similar across groups (Supplementary Table 3).

Statistical analyses to identify CpGs significantly associated with LE8 score

Our primary objective was statistical inference, that is, to identify CpG sites whose methylation levels are associated with cardiovascular health (LE8). To this end, for each CpG, we fitted a robust linear model with DNAm M-values as the outcome, the LE8 score (scaled as a z-score) as the main independent variable, and adjusted for covariates including age, sex, APOE ε4 allele count, proportions of major immune cell types (B, NK, CD4, Gran, Mono), and the first three principal components of ancestry. We used M-values (a logit transform of beta) because they better satisfy linear-model assumptions (homoscedasticity) [31]. Robust linear regression was chosen to reduce the influence of outliers.

Inflation assessment and correction

We estimated genomic inflation factors (lambda values) using both the conventional approach [32] and the bacon method [33], which was specifically designed for EWAS. Briefly, the bacon method fits a Bayesian three-component normal mixture model to the observed test statistics (e.g., t-statistics from the regression of methylation values). One component represents the null distribution, while the other two capture positive and negative associations. From the estimated null component, bacon derives its mean and standard deviation, which quantify bias and inflation in the test statistics, respectively.

Using the conventional approach, the estimated λ values were 1.01 for Year2019 dataset and 0.93 for Year2021 dataset. The inflation factors estimated by the bacon approach (λ.bacon) were 1.00 and 0.96 for Year2019 and Year2021 datasets, respectively. The estimated bias from the bacon method were 0.08 for Year2019 dataset and 0.03 for Year2021 dataset.

After genomic correction using the bacon method [33], the estimated bias were 1.10 × 10^–4 and 1.92 × 10^–5, the estimated inflation factors were λ = 1.02 and 1.00, and λ.bacon = 1.00 and 1.00 for the Year2019 and Year2021 datasets, respectively. The bacon method was then used to compute bacon-corrected effect sizes, standard errors, and P-values for each dataset.

Meta-analysis

To meta-analyze individual CpG results across the Year2019 and Year2021 datasets, we applied the inverse-variance weighted fixed-effects model [34] using the meta R package. To correct for multiple comparisons, we computed the false discovery rate (FDR). CpGs that showed consistent effect directions across both datasets and achieved a nominal P-value less than 1 × 10^–5 were considered to have suggestive significance.

Differentially-methylated regions analysis

For region-based analysis, we used the comb-p method [35], which scans genome-wide CpG locations and P-values to identify regions enriched with clusters of low P-values. We used P-values from the meta-analysis as input for comb-p and parameter settings with –seed 0.05 and –dist 750 (a P-value of 0.05 is required to start a region and extend the region if another P-value was within 750 base pairs). These parameters were shown to have optimal statistical properties in our previous comprehensive assessment of the comb-p software [36]. As comb-p uses the Sidak method to account for multiple comparisons, we selected DMRs with Sidak P-values less than 0.05. To further reduce false positives, we imposed two additional criteria in our final selection of DMRs: (1) the DMR also has a nominal P-value < 1 × 10^–5; (2) all the CpGs within the DMR have a consistent direction of change.

Functional annotation and pathway analysis

Significant individual CpG methylation signals and differentially methylated regions (DMRs) were annotated using gene annotations from Illumina and the Genomic Regions Enrichment of Annotations Tool (GREAT) software [37], which associates genomic regions with target genes.

To identify biological pathways enriched for LE8-associated DNAm, we used the methylRRA function from the methylGSA R package [38], which analyzes single CpG P-values as input. Briefly, methylGSA computes a gene-wise value by aggregating P-values from multiple CpGs mapped to each gene, adjusts for the different numbers of CpGs per gene using Bonferroni correction, and then performs Gene Set Enrichment Analysis [39] in pre-ranked mode to identify pathways enriched with significant CpGs. We analyzed pathways from the KEGG and REACTOME databases, restricting analyses to pathways containing between 5 and 200 genes. To avoid gene sets where the enrichment signal is driven by only one or two genes, we additionally required that significant gene sets include at least three genes in the “core enrichment” subset. Pathways with an FDR less than 0.05 were considered statistically significant.

Integrative analyses leveraging external resources on gene expression, genetic variants, and brain-to-blood correlations

We next examined the biological relevance of the LE8-related DNAm signals through several complementary analyses leveraging external datasets. Supplementary Table 4 shows details for each dataset, including the purpose of analysis, reference, tissue, sample size, ancestry, age, and sex distribution. First, we tested whether these CpGs correlated with nearby gene expressions in blood (expression quantitative trait methylation, eQTMs). Second, we assessed whether methylation levels at the same CpGs co-varies between blood and brain within individuals, which would strengthen their potential as biomarkers for brain disorders. Third, we evaluated whether these CpGs were under genetic control (mQTLs) and whether those genetic variants are also associated with dementia risk. When the same variant influenced both DNAm and dementia, we used colocalization analysis to test if a single causal variant could account for both associations. Finally, we cross-referenced our CpGs against prior reports associating DNAm with AD biomarkers, neuropathology, or diagnosis. Together, these integrative analyses helped us to prioritize the most biologically meaningful signals.

Specifically, to evaluate the effect of DNAm on the expression of nearby genes in blood samples, we overlapped our LE8-associated DNAm, including both significant individual CpGs and those located within DMRs, with eQTm analysis results in Supplementary Tables 2 and 3 of Yao et al. [40].

To assess the correlation of DNAm levels between blood and brain samples at LE8-associated CpGs, we used the London dataset, which includes 69 matched samples from blood and brain prefrontal cortex tissues [41]. Brain-blood correlations were assessed using two approaches: (1) unadjusted correlations based on methylation beta values and (2) adjusted correlations using residuals from regression models accounting for covariates.

In unadjusted analysis, we calculated Spearman rank correlations between DNA methylation beta values measured in brain and blood samples. For adjusted analysis, we accounted for potential confounders by removing the effects of covariates. Specifically, we removed effects from estimated neuron proportions in brain samples (or estimated immune cell-type proportions in blood samples), array type, age at death (for brain samples) or age at blood draw (for blood samples), and sex, by fitting linear models separately for brain and blood samples and extracting residuals. Spearman correlations were then calculated using these residual values to assess the relationship between brain and blood methylation levels independent of confounders.

For correlation and overlap with genetic loci, we searched blood mQTLs using the GoDMC database (http://mqtldb.godmc.org.uk/downloads) [42]. Using the same criteria from the original GoDMC study [42], we considered a cis P-value smaller than 10^–8 and a trans P-value smaller than 10^–14 as significant. The genome-wide summary statistics for genetic variants associated with dementia described in Bellenguez et al. [43] were obtained from the European Bioinformatics Institute GWAS Catalog (https://www.ebi.ac.uk/gwas/) under accession no. GCST90027158. Colocalization analysis was performed using the coloc R package.

In these integrative analyses, we cross-referenced LE8-associated CpGs with published summary statistics. These comparisons indicate concordance with prior reports but do not constitute replication.

Investigating the role of LE8-associated DNAm in dementia using independent datasets

To compare our results with previous findings, we searched for LE8-associated CpGs (both significant individual CpGs and those located in DMRs) using the CpG Query tool in the MIAMI-AD database [44] (https://miami-ad.org/). For the input on phenotype, we selected “AD Biomarker”, “AD Neuropathology”, “Dementia Clinical Diagnosis”, and “Mild Cognitive Impairment”.

Results

Study datasets

In the 2019 dataset, LE8 scores ranged from 260 to 690, with a mean of 480.15 (± 7.33). In the 2021 dataset, LE8 scores ranged from 190 to 685, with a mean of 450.97 (± 8.21). The mean ages were 62.6 (± 7.2) years for the Year2019 dataset and 61.5 (± 7.2) years for the Year2021 dataset, with females comprising 50.7% (n = 68) and 77.7% (n = 108) of each dataset, respectively (Table 1). Approximately 25.4% (n = 34) of subjects in the Year2019 dataset and 24.5% (n = 34) in the Year2021 dataset carried at least one APOE ε4 allele.

Table 1.

Characteristics of the study participants

Blood DNAm differences at individual CpGs and DMRs are significantly associated with the LE8 score

After adjusting for covariates (age, sex, APOE ε4 allele count, immune cell type compositions, and genetic ancestry) and correcting genomic inflation (Methods), our robust linear model identified 11 CpGs associated with LE8 at the suggestive significance threshold of P < 1 × 10^–5. No CpG passed the 5% false discovery rate (FDR) significance threshold. Among the 11 CpGs with suggestive significance, about half (6 CpGs) are located in the gene bodies of ZNF621, PF4, ABCA1, MAP3K13, IPO13, GALNS, while 2 CpGs are located in 5’UTR of the BANP and ANAPC11 genes (Supplementary Table 5, Fig. 1).

Fig. 1.

Manhattan plot of significant DNA methylation differences associated with Life’s Essential 8 (LE8) score in the meta-analysis of Year2019 and Year2021 blood sample datasets. The X-axis indicates chromosome number. The Y-axis shows –log₁₀(P-value) of meta-analysis. Red dots mark CpGs with P-value < 10^–5 and blue dots mark CpGs that fall within the gene promoter regions associated with the top 20 most significant DMRs. Gene symbols indicate the nearest annotated gene for each highlighted CpG

Using P-values for individual CpGs as input, comb-p [35] software identified 37 differentially methylated regions (DMRs), which achieved both a nominal P-value < 1 × 10^–5 and a Sidak multiple comparison-adjusted P-value < 0.05. We also required all the CpGs within each DMR have a consistent direction of change in estimated effect sizes in the individual CpG analysis (Table 2, Supplementary Table 6). The number of CpGs in these DMRs ranged from 3 to 27. Among these DMRs, just over half (56.8%, 21 DMRs) showed hypermethylation associated with increased value of the LE8. About 40% (15 DMRs) are located in CpG islands or shores.

Table 2.

Top 20 most significant differentially methylated regions (DMRs) associated with LE8 score. For each DMR, annotations include location of the DMR (DMR) and nearby genes based on GREAT. Comb-p results include the number of probes (nProbes), nominal P-value (pValue), multiple comparison corrected P-value based on Sidak method (Sidak-P), and direction of each CpG within the DMR (direction). All P-values are two-sided. In GREAT annotation, the numbers in parentheses indicate distance from the TSS. Highlighted in red text are gene promoter regions associated with the DMRs

Pathway analysis revealed LE8-associated DNA-methylation are enriched in biological pathways involved in vascular integrity and inflammation

To further understand the biological processes underlying the LE8, we next performed pathway analysis using the methylGSA software [38]. At a 5% false discovery rate (FDR), we identified 1 KEGG pathway and 4 Reactome pathways significantly enriched with LE8-associated DNAm (Table 3). The most significant pathway is KEGG pathway Adherens junction (AJ) (P-value = 2.2 × 10^–5, FDR = 4.83 × 10^–3), which regulates essential vascular functions, such as the control of permeability and transmigration of circulating leukocytes, and the maintenance of existing vessels and formation of new ones. Impaired endothelial AJ, which regulates lipid entry and immune-cell infiltration, is a critical early event in atherosclerosis [45]. Moreover, AJs partner with tight junctions to maintain the blood–brain barrier (BBB). Degradation of the adherence junctions is an early feature of AD and other neurodegenerative disorders that involve BBB breakdown [46]. Other significant pathways include CD209 (DC-SIGN) signaling and the Turbulent-flow induced activation of PIEZO1 and integrin signaling in endothelial cells pathways, which drive endothelial and systemic inflammation, consequently promotes atherosclerotic plaque formation and cognitive decline [47, 48]. Finally, impaired insulin processing disrupts glucose homeostasis and is associated with a higher risk of both cardiovascular events and dementia [49, 50].

Table 3.

A total 5 pathways were significantly enriched with LE8-associated DNAm differences at 5% FDR. This analysis was performed using methylGSA software, which used the continuous CpG-to-LE8 meta-analysis P-values as input

Correlation of significant DNAm with expression of nearby genes in blood samples

To evaluate the functional relevance of our significant DMRs and CpGs, we compared them with established DNAm to gene-expression associations (i.e., eQTMs) computed from matched blood DNAm and gene-expression data in over 4000 participants from the Framingham Heart Study [40]. Among the 11 significant individual CpGs and 228 CpGs within the 37 DMRs, 40 CpGs showed significant cis (within 500 kb) and 17 showed significant trans associations with gene expression (Supplementary Tables 7–8). Notably, target genes associated with cis-eQTMs include PF4, MRPL21, CD300A, ACACB, RUFY1, MIR196B, HOXA2, C11orf21, and PRDM2. Among the 11 significant individual CpG, only cg01447579, located in the promoter of PF4 gene, showed significant correlation with its target gene’s expression.

Correlation of DNAm levels in blood and brain samples at LE8-associated CpGs

We next evaluated cross-tissue concordance of DNAm levels at LE8-associated CpGs using the London dataset, which includes matched pre-mortem blood and post-mortem prefrontal cortex DNAm profiles from 69 subjects [41]. At 5% FDR, among the 11 significant individual CpGs and 228 CpGs mapped within the identified DMRs, 49 CpGs (20.5%) showed significant brain-blood correlations in both unadjusted and adjusted analyses that accounted for covariate variables (Supplementary Table 9). Notably, among the 11 significant individual CpGs, again, only cg01447579, located in the promoter of PF4, showed significant brain-blood correlation (Fig. 2). Other CpGs with significant correlations were located in DMRs and mapped to promoter regions of TSPAN32, ACY3, ABAT, CD300A, SH3RF3, EIF4E, HOXA4, HOXA5 genes and other genomic regions. Except for 4 CpGs, located in ACY3 and CD300A genes, all others showed positive brain-blood methylation correlations.

Fig. 2.

Brain-blood correlations for DNA methylation levels at cg01447579 located in gene body of the PF4 gene. These figures were obtained using the Blood Brain DNA Methylation Comparison Tool (https://epigenetics.essex.ac.uk/bloodbrain/index.php?probenameg=cg01447579). Abbreviations PFC: Prefrontal Cortex, EC: Entorhinal Cortex, STG: Superior Temporal Gyrus, CER: Cerebellum

To further prioritize these 49 CpGs, we overlapped them with the eQTM associations described above. Approximately one-third (18 CpGs, 36.7%) also showed significant cis DNAm-to-mRNA correlations in blood. These CpGs were mapped to the promoter regions of genes including PF4, CD300A, HOXA4, and HOXA5 (Supplementary Table 9).

Methylation quantitative trait loci and intersection with genetic risk loci in dementia

We next searched methylation quantitative trait loci (mQTLs) for the LE8-associated DNAm using the GoDMC database [42]. Among the 11 significant individual CpGs (Supplementary Table 5) and the 228 CpGs located in significant DMRs (Supplementary Table 6), 144 CpGs had 56,152 mQTLs in cis and 35 CpGs had 2435 mQTLs in trans in the blood (Supplementary Table 10).

Next, we evaluated if the mQTLs overlapped with genetic risk loci implicated in dementia, by comparing them with the genetic variants nominated in a recent Alzheimer’s Disease and Related Dementias (ADRD) meta-analysis [43]. We found that one mQTL (rs2526377), located about 4 kb upstream of TSPOAP1 gene and within an intron of the antisense lncRNA TSPOAP1-AS1, overlapped with a genome-wide-significant AD locus (P-value < 5 × 10^–8). Three additional mQTLs, all located near the same TSPOAP1 gene, overlapped with genetic variants reaching a suggestive genome-wide significance threshold at P < 10^–5 (Supplementary Table 11).

Given the observed overlap between the mQTLs and ADRD genetic risk loci, we next sought to determine whether the association signals at these loci (variant to CpG methylation levels and variant to clinical ADRD status) were due to a single shared causal variant or distinct causal variants close to each other. To this end, we performed a co-localization analysis using the method described in Giambartolomei et al. (2014) [51]. The results of this co-localization analysis strongly suggested [52] (i.e. PP3 + PP4 > 0.90, PP4 > 0.8, and PP4/PP3 > 5) that one genomic region located in the TSPOAP1 gene and including the causal variant rs2526377, influences both phenotypes (i.e., ADRD status and CpG methylation levels) (Supplementary Table 12).

Comparison LE8-associated DNAm with prior studies of CSF AD biomarkers, AD brain neuropathology, or clinical AD status

To evaluate the relevance of our findings in dementia, we compared the 11 significant individual CpGs and 228 CpGs located within DMRs from our LE8 DNAm analysis with results from previous research in the MIAMI-AD database (https://miami-ad.org/). These are cross-study comparisons using published results (not de novo analyses), they indicate consistency with prior reports rather than replication. At the Bonferroni-corrected significance threshold for 239 CpGs (P-value < 2.09 × 10^–4), 31 CpGs (13.0%) were also significantly associated with AD neuropathology in prior studies. These CpGs were located in the TSPAN32, ACY3, BTBD17, HOXA4, HOXA5 genes and intergenic regions (Supplementary Table 13).

Using a more relaxed nominal threshold (P-value < 0.05), 147 CpGs (61.5%) overlapped with previously reported AD-associated CpGs, with changes in the expected directions (Supplementary Table 13). In prior AD studies, these CpGs were significantly associated with CSF AD biomarkers (36 CpGs), AD brain neuropathology (113 CpGs), and clinical dementia status (31 CpGs). The overlapping CpGs were located in promoter regions of ANO4, ABAT, CDH5, BTBD17, CD300A, HOXA5, LYPD8, TSPAN32, ACY3, SH3RF3 genes and other genomic regions.

Moreover, 15 CpGs, located in the promoter regions of CD300A, HOXA4, and HOXA5 genes were supported by converging evidence, including cis eQTMs, significant brain-blood correlations in DNAm levels, and prior associations with AD (Supplementary Table 13). Among them, two CpGs (cg14058329, cg02005600) were located in the HOXA5 promoter and were influenced by AD-associated variants rs2526377 that showed trans-colocalization signals described above.

Discussion

In this community-based study of 273 Hispanic participants from the NOMAS cohort, we conducted an epigenome-wide association analysis to identify DNAm differences associated with Life’s Essential 8 (LE8), a composite metric of cardiovascular health. Our pathway analysis revealed that LE8-associated DNAm changes were significantly enriched in biological pathways related to endothelial junctional complexes (adherens/tight junctions) and blood–brain barrier (BBB) stability, innate and adaptive immune signaling, and metabolic regulation (Table 3). These results are consistent with a model in which better cardiovascular health preserves vascular barrier integrity. In addition, enrichment in inflammation-related pathways (DC-SIGN-mediated antigen presentation and PIEZO1-integrin signaling) suggests that healthier LE8 profiles may limit endothelial activation and reduce chemokine-driven recruitment of myeloid cells implicated in both atherosclerosis and neuroinflammation. Finally, enrichment of the insulin processing pathway underscores the central role of glucose homeostasis in both vascular disease and dementia biology. Taken together, these findings support the hypothesis that improving LE8 scores may simultaneously reduce cardiovascular and neurodegenerative risk by stabilizing adherens junctions and the BBB, attenuating DC-SIGN/PIEZO1-associated inflammatory signaling, and enhancing insulin processing and lipid clearance.

A review of recent literature revealed that several top LE8-associated CpGs and DMRs have been associated with lifestyle factors, vascular health and cognitive function in prior studies. For example, the most significant CpG is located on ZNF621, which encodes a zinc finger transcription factor. In the cardiovascular domain, a rare intronic variant in ZNF621 (rs34412695) showed one of the strongest associations with hypertension [53]. In dementia, ZNF621 emerged as one of the most consistently differentially expressed genes between AD cases and controls across multiple brain regions, including the hippocampus and entorhinal cortex [54]. The second most significant CpG is located within the gene body of PF4, which encodes Platelet Factor 4, a chemokine released by activated platelets that promotes atherosclerosis through plaque formation. Paradoxically however, recent studies showed that injection of PF4 to older mice reduced brain inflammation and restores synaptic plasticity [55, 56]. In humans, decreased level of serum PF4 levels are significantly correlated with cognitive decline and cerebrospinal fluid biomarkers in AD [57]. These findings suggest that PF4 functions as a context-dependent immunomodulator with cell-type specific effects, a key reason it is being explored as a potential therapeutic target in neurodegeneration, despite its pro-inflammatory role in cardiovascular disease. Previous studies have also shown that physical exercise increases platelet-derived PF4 levels [58], which is consistent with our observed association between increased LE8 scores and hypermethylation at the PF4 gene body.

The third most significant LE8-associated CpG is located in the gene body of ABCA1, which encodes a cholesterol transporter protein crucial for maintaining cholesterol homeostasis. Importantly, ABCA1 facilitates clearance of Aβ from the brain, thus reducing neuroinflammation and the risk of AD [43, 59, 60]. Relative to LE8 related lifestyle factors, hyperglycemia downregulates ABCA1 gene expression in vascular and renal cells [61], smoking suppresses macrophage ABCA1 expression [62]; obesity and insulin resistance correlate with lower ABCA1 expression in visceral adipose tissue [63], and circadian disruption downregulates its expression in clock-mutant mice [64]. In contrast, regular physical exercise is associated with higher leukocyte/muscle ABCA1 expression [65] and diet rich in omega 3 fatty acids upregulates ABCA1 expression [66].

Similarly, among the top 20 most significant DMRs associated with LE8, the ACY3 gene encodes aminoacylase-3, an enzyme involved in catecholamine metabolism and blood-pressure regulation [67]. In a recent AD study, a DMR in ACY3 was the only DMR identified by two independent analytical methods and correlated with both rate of cognitive decline and conversion to AD in the ADNI cohort [68]. Notably, a recent pharmacogenetics study provided replicated evidence that a genetic variant on the ACY3 gene influences blood pressure response to beta-blockers [67].

Another noteworthy gene associated with our top DMRs is HOXA5, which encodes a homeobox transcription factor that regulates cell differentiation and morphogenesis. In vascular tissue, HOXA5 protects against cardiovascular disease by maintaining endothelial integrity and promoting a contractile, anti-inflammatory smooth-muscle phenotype that resists plaque formation [69]. In the brain, hypermethylation at a 40-kb hypermethylated block spanning the HOXA cluster on chromosome 7, which includes HOXA5, has been consistently associated with AD neuropathology across multiple independent datasets [70]. In peripheral blood, we previously found that hypermethylation at several CpGs within the HOXA5 DMR is associated with increased CSF pTau181 levels, tau pathology in the brain, as well as methylation levels in the brain [71]. Among LE8-related lifestyle factors, obesity and higher BMI are associated with increased HOXA5 methylation in both blood and adipose tissue [72], while hyperglycemia induces hypermethylation and downregulation of HOXA5 expression in endothelial cells [73]. On the other hand, exercise was shown to upregulate HOXA5 in a rat model [74], suggesting that modifiable health behaviors may epigenetically regulate this gene’s activity relevant to vascular and neurodegenerative disease.

By integrating multiple external resources, we found two HOXA5 promoter CpGs (cg14058329, cg02005600), that are supported by converging evidence of significant correlation with gene expression, brain-blood DNAm, and have been previously associated with AD pathology. Moreover, these two CpGs were influenced by AD-associated variant rs2526377, which is located at about 4 kb upstream of the TSPOAP1 gene and within an intron of the antisense lncRNA TSPOAP1-AS1 on chromosome 17. This SNP reached genome-wide significance in a recent ADRD GWAS (OR = 0.95, P-value = 1.6 × 10^–12) ⁴³. In our mQTL scan, the G allele of rs2526377 (the AD-protective allele) was associated with reduced methylation at both CpGs in the HOXA5 promoter. Furthermore, colocalization analysis suggested the shared causal variant rs2526377 drives both the AD risk and methylation levels at HOXA5. Taken together, these results demonstrated that rs2526377 may influence dementia risk through trans regulation of HOXA5, rather than by directly altering the activity of TSPOAP1 or TSPOAP1-AS. This underscores the importance of considering alternative mechanisms when identifying target genes of non-coding variants, and highlights that both genetic and lifestyle factors may converge on HOXA5 to modulate AD risk.

Another recent study also examined DNAm associated with LE8 in the context of cardiovascular disease (CVD) [75]. However, there are notable differences that distinguish Carbonneau et al. (2024) from the present study. While Carbonneau et al. (2024) identified LE8-associated CpGs using Framingham Heart Study samples measured by the older 450k arrays, we analyzed the NOMAS samples measured by the newer 850k EPIC arrays. Additionally, Carbonneau et al. (2024) demonstrated DNAm signature of LE8 were associated with cardiovascular outcomes, including stroke, incident CVD, CVD mortality, and all-cause mortality. In contrast, our study specifically focused on evaluating the role of LE8-associated CpGs in dementia. Finally, while the samples from the Framingham Heart study included mostly European individuals, we analyzed samples from self-reported Hispanic subjects in NOMAS. Hispanics represent one of the fastest-growing ethnic groups in the US and have approximately 1.5 times the risk of dementia compared to non-Hispanic whites [76]. Recent studies also indicated a higher prevalence of cardiovascular risk factors such as hypertension, diabetes, and obesity among Hispanic Americans [77, 78], which are closely associated with cognitive decline [79].

Carbonneau et al. [75] analyzed 3688 discovery samples and reported 609 CpGs at FDR < 0.05 that were replicated across multiple cohorts totaling 9807 additional samples. Among the > 10,000 samples included in discovery and replication, only 224 were Hispanic. We compared our 11 suggestive CpGs and 37 Sidak-significant DMRs from NOMAS with their 609 CpGs and found two overlapping sites: cg23594345 (intergenic) and cg10402995, which maps to the promoter of LYPD8 gene. LYPD8 encodes a colon-enriched glycoprotein secreted by epithelial cells, it binds flagella of Gram-negative, flagellated bacteria to inhibit motility and help maintain the mucosal barrier, implicating host–microbe–immune interfaces potentially relevant to vascular and neuroinflammatory pathways. Given differences in array content (450 K vs. EPIC), ancestry composition, and analytic pipelines, exact CpG-level concordance is challenging without harmonized re-analysis. Nevertheless, both studies observed enrichment of LE8-associated DNAm in immune-related pathways, underscoring the prominent role of immune signaling in cardiovascular health. The limited CpG overlap may also reflect ancestry-specific biology and different lifetime exposure in Hispanic populations, leading to partially distinct methylation signatures of LE8.

This study is not without limitations. First, our study had a modest sample size, a common issue in epigenetic research. We also did not perform de novo replication in an independent cohort. Instead, we observed consistency with previously published associations in external datasets, which supports biological plausibility but should not be interpreted as replication. While several DMRs passed multiple-testing correction, single-CpG findings did not survive FDR; accordingly, results should be viewed as hypothesis-generating. Second, we were not able to comprehensively evaluate potential modification of LE8-to-dementia associations by age, sex, and APOE genotype because of limited sample size. Nevertheless, these factors were included as covariates in our models, ensuring the findings are broadly applicable across different ages, sexes, and APOE genotypes. Future studies with larger cohorts are needed to replicate our findings and to explore subgroup-specific DNA methylation associations with LE8. Third, to evaluate the functionality of the LE8-associated CpGs and their relevance to dementia, we leveraged several external datasets. As these resources are largely composed of samples from non-Hispanic White participants, we interpret their results as supporting biological plausibility rather than demonstrating cross-ancestry generalizability. Accordingly, we caution that our findings may have limited applicability to other diverse ancestry groups. Fourth, given the modest sample size, we did not perform mediation analyses, although DNAm could plausibly mediate the relationship between cardiovascular health and cognition. An exploratory genome-wide mediation screen would be considerably underpowered and potentially misleading; on the other hand, restricting mediation tests to LE8-associated CpGs within the same cohort would constitute post-selection (“double-dipping”), resulting in inflated Type I error and biased estimates. Future larger, longitudinal Hispanic cohorts with harmonized lifestyle, DNAm, and cognitive outcomes will be essential to enable well-powered and causally valid mediation analyses.

The strengths of this study include stringent quality control procedures, robust linear modeling that accounted for key confounders such as age, sex, APOE ε4 status, immune cell-type proportions, and genetic ancestry, as well as corrections for genomic inflation. To gain deeper biological insight into the LE8-associated DNAm changes, we performed pathway analysis and integrated our findings with multiple external resources, including blood-based eQTM and mQTL datasets, genetic associations for ADRD, and brain-blood DNAm correlations, to prioritize candidate biomarkers most relevant to dementia.

In summary, our study identified potential DNA methylation biomarkers associated with LE8 and demonstrated their relevance to dementia, underscoring shared biological pathways between cardiovascular and neurodegenerative diseases. Although limited by a modest sample size and the current scope of LE8 in capturing psychological and environmental influences due to the lack of standardized metrics for these domains at present, our findings suggest that methylation biomarkers linked to lifestyle factors are biologically relevant in dementia. Future work that integrates emerging consensus measures of psychological and environmental risk will generate a more comprehensive DNAm signature for lifestyle-related dementia risk. Importantly, an ideal AD biomarker should be actionable; because LE8-based biomarkers reflect modifiable behaviors, they represent promising candidates for secondary prevention trials. Such biomarkers could be used to identify individuals at elevated risk, stratify participants by lifestyle-related biological risk, monitor epigenetic responses to lifestyle interventions, and point to actionable targets for modifying cardiovascular pathways implicated in dementia. Future research involving larger and more diverse cohorts is needed to validate these methylation biomarkers and clarify their roles in dementia risk.

Authors’ contributions

L.W., T.R, H.G., C.A., J.Y., E.R.M. designed computational analyses. D.L., W.Z., L.G., M.A.S., L.W., N.D. analyzed the data. L.W., J.Y, H.G., C.A., T.E., C.G., S.S., B.W.K., J.Y., E.R.M, X.S.C., S.B. contributed to the interpretation of the results. L.W., D.L. wrote the paper, and all authors participated in the review and revision of the manuscript. L.W. conceived the original idea and supervised the project.

Funding

This research was supported by US National Institutes of Health grants R61NS135587 (L.W.), R01NS128145 (L.W.). NOMAS is funded by the National Institutes of Health/National Institutes of Neurologic Disorders and Stroke (NIH/NINDS) (R01NS029993). The authors would also like to thank Evelyn F. McKnight Brain Institute for support.

Data availability

The NOMAS DNA methylation datasets have been deposited in the NCBI Gene Expression Omnibus database (accession: GSE305883). The scripts for the analysis performed in this study can be accessed at https://github.com/TransBioInfoLab/DNAm-and-LE8.

Declarations

Ethics approval and consent to participate

The NOMAS study was approved by the University of Miami Institution Review Board. Written informed consent was obtained from all the participants or their authorized representatives.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.