Epigenetic clocks and longitudinal plasma biomarkers of Alzheimer's disease

Abstract

INTRODUCTION

Chronological age is the strongest risk factor for Alzheimer's disease and related dementias (ADRD). However, the association of accelerated biological aging relative to chronological age with ADRD pathology is unclear.

METHODS

In a cohort of 2366 (873 with longitudinal data) cognitively unimpaired older women, we examined associations of seven baseline measures of epigenetic age acceleration (EAA) and pace of aging with 15‐year changes in plasma ADRD biomarkers.

RESULTS

At baseline, higher AgeAccelHorvath and AgeAccelPheno were associated with lower amyloid beta (Aβ) 42 to Aβ40 (Aβ42:Aβ40) ratio, and higher AgeAccelGrim2, PCPhenoAge, and PCGrimAge were associated with elevated neurofilament light (NfL). Longitudinally, higher baseline DunedinPACE – capturing the pace of biological aging – was associated with faster increases in tau phosphorylated at threonine 181 (p‐tau181), p‐tau217, NfL, and glial fibrillary acidic protein (GFAP) over 15 years.

DISCUSSION

Accelerated biological aging, particularly DunedinPACE, was associated with increasing levels of plasma ADRD biomarkers over time.

Highlights

• We studied 2366 older women from the Women's Health Initiative Memory Study.

• AgeAccelHorvath and AgeAccelPheno were linked to lower plasma Aβ42:Aβ40 at baseline.

• AgeAccelGrim2, PCPhenoAge, and PCGrimAge were linked to higher plasma NfL at baseline.

• DunedinPACE was associated with faster increases in p‐tau181, p‐tau217, NfL, and GFAP.

1. BACKGROUND

Although chronological age is the strongest risk factor for Alzheimer's disease and related dementias (ADRD), it remains unclear whether accelerated biological aging relative to chronological age is associated with ADRD pathology. ¹ Epigenetic clocks are DNA methylation (DNAm)‐based biomarkers measuring an individual's biological age. ² Several distinct epigenetic clocks have been developed and validated, including first‐generation clocks trained to predict chronological age, such as those developed by Horvath and Hannum; second‐generation clocks developed to predict clinical phenotypes (PhenoAge) and mortality (GrimAge/GrimAge2); and a third‐generation clock developed to predict the pace of biological aging across different organ systems (DunedinPACE). ³ , ⁴ , ⁵ , ⁶ , ⁷ , ⁸ , ⁹ Epigenetic age acceleration (EAA), indicating faster biological aging relative to chronological age, has been associated with higher risk of age‐related diseases and all‐cause mortality. ² , ³ , ⁴ , ⁵ , ⁶ , ⁷ , ⁹ , ¹⁰ Longitudinal studies in multiple cohorts, including the Women's Health Initiative Memory Study (WHIMS), have shown that EAA is associated with cognitive decline. ¹¹ , ¹² , ¹³ , ¹⁴ , ¹⁵ , ¹⁶ EAA has also been associated with increased risk of mild cognitive impairment (MCI) and dementia in some studies, including in WHIMS. ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹

Over the past decade, significant progress has been made in identifying blood‐based biomarkers of ADRD that are non‐invasive and more cost‐effective compared to cerebrospinal fluid and positron emission tomography (PET) in vivo measures of ADRD pathology. ²² , ²³ , ²⁴ , ²⁵ , ²⁶ , ²⁷ , ²⁸ These biomarkers include plasma biomarkers of AD‐specific pathology, including amyloid beta 42 (Aβ42), Aβ40, and tau phosphorylated at threonine 181 (p‐tau181) and at threonine 217 (p‐tau217). Non‐specific markers of neuronal injury (neurofilament light chain protein [NfL]) and of neuroinflammation (glial fibrillary acidic protein [GFAP]) have also been associated with increased dementia risk. ²⁴ , ²⁹ , ³⁰ , ³¹ Only one prior study examined associations of epigenetic clocks of biological aging with plasma biomarkers of ADRD. ³² However, that study was limited by its cross‐sectional design; examined only Hispanic/Latino adults, thereby limiting generalizability to other racial and ethnic groups; did not examine changes in plasma biomarkers over time; and did not evaluate plasma p‐tau217. A better understanding of whether accelerated biological aging is associated with elevated levels of plasma ADRD biomarkers over time could help identify high‐risk populations and inform precision medicine approaches targeting specific aging pathways to prevent ADRD.

In this study, we examined associations of baseline EAA measured using several well‐studied epigenetic clocks with plasma ADRD biomarkers, including the ratio of Aβ42:Aβ40, p‐tau181, p‐tau217, NfL, and GFAP. Importantly, we had plasma biomarker data available from two visits an average of 15 years apart, allowing us to examine the extent to which epigenetic clocks predict longitudinal changes in plasma ADRD biomarkers. We examined several different clocks that captured different aspects of biological aging. We hypothesized that EAA would be associated with greater indications of ADRD pathology at baseline and greater changes in markers of pathology over time (i.e., lower Aβ42:Aβ40 and higher p‐tau181, p‐tau217, NfL, and GFAP). We hypothesized that the second‐ and third‐generation clocks would show greater association with accelerated biological aging than the first‐generation clocks and that the strongest associations would be found for DunedinPACE, as we previously found that this epigenetic clock showed the strongest association with cognitive decline in WHIMS. ¹¹

2. METHODS

2.1. Study design and population

WHIMS was an ancillary study of the Women's Health Initiative Hormone Therapy trials. WHIMS was designed to investigate the effects of hormone therapy on cognitive outcomes among 7479 women aged 65 to 79 years at enrollment who were cognitively unimpaired at randomization in the period 1995–1998. Details on the WHIMS design and protocols have been published. ³³ , ³⁴ , ³⁵ Women were randomized either to conjugated equine estrogens (CEE) plus medroxyprogesterone acetate (MPA) versus placebo among women with an intact uterus, or CEE alone versus placebo among women with prior hysterectomy. The trials were stopped in 2002 and 2004, respectively, but follow‐up for outcomes continued. Annual follow‐up for in‐person cognitive assessments continued through 2007. In 2008, WHIMS transitioned to annual telephone‐administered cognitive assessments in the WHIMS Epidemiology of Cognitive Health Outcomes (WHIMS‐ECHO) study, which followed participants for cognitive outcomes through 2021. ³⁶

RESEARCH IN CONTEXT

1. Systematic review: The authors conducted a comprehensive literature review using PubMed and Google Scholar to identify studies examining relationships between epigenetic clocks and plasma biomarkers of ADRD. Only one study looked at these relationships cross‐sectionally, with other studies focusing on associations of epigenetic clocks with cognitive function or diagnosis of ADRD.

2. Interpretation: Our longitudinal results provide novel evidence that accelerated pace of aging, as measured by DunedinPACE, was associated with faster increases in plasma p‐tau181, p‐tau217, NfL, and GFAP over 15 years. This suggests that DunedinPACE may signal early ADRD pathology before symptoms emerge.

3. Future directions: Validation of our findings in other cohorts of men and women is needed. Future work should examine whether DunedinPACE provides prognostic information in predicting ADRD beyond plasma biomarkers and whether it could be a useful tool for monitoring ADRD risk in prevention trials and intervention studies.

Among 7479 WHIMS participants, we excluded 240 with only one WHIMS cognitive assessment, 519 who did not consent to DNA sharing, and 304 with no baseline DNA or buffy coat available, leaving 6416 participants whose DNA was available at the baseline visit (1995–1998; Figure S1). DNA was extracted from baseline blood samples in 2022 and quantified for DNA methylation measurement between 2023 and 2024 at the University of Minnesota Genomics Center. As described in prior literature, archived DNA (e.g., stored for 20 years) is suitable for analysis in DNA methylation studies. ³⁷ DNAm was measured using the Illumina Infinium MethylationEPIC version 2.0 using standard protocols. After quality control (Supporting Information), baseline epigenetic data were available for 6069 participants. For the present analysis, we further selected women with data on plasma biomarkers of ADRD available at baseline for cross‐sectional analyses (N = 2366) and a subset with measurements at both baseline and a second time point for longitudinal analyses (N = 873). This study was approved by the Institutional Review Board of the University of California San Diego. All participants provided written informed consent.

2.2. Plasma biomarkers

Fasting blood was drawn at baseline and at a second time point an average of 15 years later. The second blood sample was collected between 2012 and 2013 as part of the WHI Long Life Study, with a mean (standard deviation [SD]) interval of 15.10 (0.92) years between visits (range, 13.54 to 17.36 years). Samples were processed, frozen at −70°C, and then shipped to a repository in Rockville, MD, maintained by Fisher Bioservices. All plasma biomarker assays were performed at the Advanced Research and Diagnostics Laboratory (ARDL) at the University of Minnesota in 2024. Samples were shipped to the laboratory on dry ice. The detailed WHI protocol for blood collection and processing is available on the WHI website (Supporting Information). Plasma EDTA samples were used for the collection of plasma biomarkers. All blood samples were shipped to the lab in one batch to minimize batch effects. Levels of plasma biomarkers were measured using the Quanterix HD‐X platform. The Simoa Human Neurology 4‐Plex E platform was used to measure levels of Aβ40, Aβ42, NfL, and GFAP. P‐tau181 was measured using the Simoa p‐tau181 version 2 assay. P‐tau217 was measured using the ALZpath Simoa pTau‐217 version 2 assay. Samples were assayed in singlets with the inclusion of 192 duplicates; laboratory personnel were blinded to the inclusion of these duplicate samples, and they were also blinded to cognitive impairment status. Baseline and follow‐up samples were measured in the same laboratory at the same time using a single lot of reagents for each biomarker. The average intra‐assay coefficients of variation for Aβ40, Aβ42, NfL, GFAP, p‐tau181, and p‐tau217 derived from these duplicates were 9.2%, 9.3%, 7.1%, 9.8%, 10.2%, and 11.4%, respectively.

The following inter‐assay laboratory coefficients of variation (CVs) were derived from an ARDL pooled sample and the two kit controls, which were run on every plate along with the samples: (1) Aβ40: 4.8%, 5.7%, and 13.3% at mean concentrations of 19.0, 97.6, and 42.0 pg/mL, respectively; (2) Aβ42: 3.6%, 4.9%, and 12.8% at mean concentrations of 8.3, 38.0, and 2.4 pg/mL, respectively; (3) NfL: 6.4%, 9.0%, and 9.4% at mean concentrations of 23.5, 499.2, and 8.2 pg/mL, respectively; (4) GFAP: 10.6%, 9.8%, and 15.7% at mean concentrations of 188.4, 3744.0, and 72.3 pg/mL, respectively; (5) p‐tau181: 7.2%, 6.4%, and 10.7% at mean concentrations of 42.4, 1130.9, and 14.9 pg/mL, respectively; and (6) p‐tau217: 11.4%, 11.2%, and 12.9% at mean concentrations of 0.75, 0.39, and 0.15 pg/mL, respectively.

The lower limits of detection (LOD) for each biomarker were as follows: 1.54 pg/mL for Aβ40, 0.54 pg/mL for Aβ42, 1.76 pg/mL for GFAP, 0.36 pg/mL for NfL, 0.620 pg/mL for p‐tau181, and 0.012 pg/mL for p‐tau217. No woman in our analytic sample had values less than the LOD for any biomarker.

2.3. Epigenetic clocks

We examined the most widely studied using the first‐, second‐, and third‐generation epigenetic clocks: AgeAccelHorvath, AgeAccelHannum, AgeAccelPheno, AgeAccelGrim2, principal component (PC)‐based versions of AgeAccelPheno and AgeAccelGrim (PCPhenoAge and PCGrimAge, respectively), and DunedinPACE. Details on the calculation of the epigenetic clocks are provided in the Supporting Information.

AgeAccelHorvath, derived from the multi‐tissue Horvath epigenetic clock, and AgeAccelHannum, derived from the blood‐based Hannum epigenetic clock, are first‐generation clocks developed on chronological age. ³ , ⁴ AgeAccelPheno is a second‐generation clock capturing phenotypic age, consisting of age and nine clinical biomarkers (e.g., levels of albumin, creatinine, glucose, C‐reactive protein, and immune markers); it has been shown to outperform the first‐generation clocks in predicting healthspan indices. ⁵ AgeAccelGrim2, also a second‐generation clock, is a composite biomarker of DNAm‐based surrogates of plasma proteins (e.g., cystatin C, CRP, GDF‐15), a DNAm‐based estimator of smoking pack years, age, and sex. ⁷ PCPhenoAge and PCGrimAge are PC‐based versions of AgeAccelPheno and AgeAccelGrim. ⁸ DunedinPACE, Dunedin (P)ace of (A)ging (C)alculated from the (E)pigenome, a third‐generation clock, was developed using longitudinal data from 19 biomarkers assessing cardiovascular, metabolic, renal, hepatic, immune, dental, and pulmonary systems to measure the pace of biological aging across organ systems among middle‐aged adults followed for 20 years and has been shown to predict morbidity, disability, and mortality. ⁹ For the epigenetic clocks, higher values indicate accelerated biological aging relative to chronological age. For DunedinPACE, values >1 indicate faster pace of aging (e.g., a value of 1.10 indicates a pace of aging 10% faster than the average), while values <1 indicate slower pace of aging (e.g., a value of 0.90 indicates a pace of aging 10% slower than the average).

2.4. Covariates

Baseline questionnaires assessed age, race, ethnicity, education, smoking status, diabetes (defined as ever having been diagnosed and treated with pills or insulin shots), cardiovascular disease (CVD), including myocardial infarction and stroke, and total energy expenditure from recreational physical activity (in metabolic equivalent [MET] hours/week). Hypertension was defined as self‐report of physician‐diagnosed hypertension, use of hypertensive medications, or measured systolic blood pressure ≥ 130 mmHg or diastolic blood pressure ≥ 80 mmHg. Height and weight were measured with a stadiometer and balance beam scale, respectively, to calculate body mass index (BMI; kg/m²). Apolipoprotein E epsilon 4 (APOE ε4) carrier status, defined as the presence of at least one ε4 allele versus non‐carrier, was determined in women with available genome‐wide genotyping data based on two single nucleotide variants, rs429358 and rs7412. Imputation was performed using the 1000 Genomics Project reference panel and the MaCH algorithm implemented in Minimac. ³⁸ Both single nucleotide polymorphisms (SNPs) had high imputation quality (R ² > 0.97 for rs429358 and R ² > 0.97 for rs7412). ¹¹ The hormone therapy treatment arm in the original WHIMS trial (estrogen alone, estrogen placebo, estrogen plus progestin, or estrogen plus progestin placebo) was also included as a covariate. Fasting blood samples from previous WHI ancillary studies were sent to the University of Minnesota Medical Center Advanced Research and Diagnostics Lab for measurement of serum lipids, including low‐density (LDL; mg/dL) and high‐density (HDL; mg/dL) lipoprotein cholesterol. HDL was measured using the HDL‐C plus third‐generation direct method on the Roche Modular P Chemistry Analyzer. LDL was calculated using the Friedewald formula. ³⁹ Creatinine was assayed using the Creatinine Plus Reagent (Roche Diagnostics) on the Modular P Chemistry Analyzer (Roche Diagnostics). We calculated estimated glomerular filtration rate (eGFR) using the 2021 Chronic Kidney Disease Epidemiology Collaboration equation based on serum creatinine measurements. ⁴⁰ White blood cell (WBC) counts were estimated for CD4+ T cells, CD8+ T cells, natural killer cells, B cells, monocytes, neutrophils, and granulocytes using the Identifying Optimal Libraries (IDOL) algorithm. ⁴¹

2.5. Statistical analysis

2.5.1. Descriptive statistics

We calculated the ratio of plasma Aβ42 to Aβ40 because the ratio is a better measure of amyloid pathology than either biomarker alone. ⁴² Since plasma biomarkers were non‐normally distributed, Aβ42:Aβ40, NfL, GFAP, p‐tau181, and p‐tau‐217 were standardized using log₂‐based Z‐transformation in all analyses, consistent with prior literature. ⁴³ To minimize the influence of outliers, we applied winsorization at the 99th percentile across all plasma ADRD biomarkers.

Means and SDs, or counts and proportions, were reported for continuous and categorical baseline covariates, respectively, in the overall sample and across quartiles of DunedinPACE. For descriptive purposes, we displayed cohort characteristics by quartiles of DunedinPACE, the clock we previously found to be most associated with cognitive decline in WHIMS. ¹¹ Differences in baseline continuous and categorical variables across DunedinPACE quartiles were examined using Kruskal–Wallis rank‐sum tests and chi‐squared tests, respectively. Fisher's exact tests with simulated p values (based on 2000 replicates) were used for categorical variables with low expected counts. Means and SDs of plasma ADRD biomarkers were reported for participants (N = 873) with longitudinal data.

2.5.2. Associations of epigenetic clocks with plasma ADRD biomarkers at baseline

In the analyses, we examined continuous values of epigenetic clocks in relation to plasma ADRD biomarkers. Associations of each epigenetic clock with plasma ADRD biomarkers at baseline were assessed by linear regression models to estimate β coefficients and their respective 95% confidence intervals (CIs) for 1‐SD increases in EAA and the pace of aging. Progressively adjusted models were used to examine the influence of potential confounders. Model 1 was adjusted for age only. Model 2 was adjusted for age, race, ethnicity, education, and hormone therapy arm. Model 3 was additionally adjusted for physical activity and smoking status. Model 4 was additionally adjusted for BMI. Model 5, the fully adjusted model, was additionally adjusted for diabetes, CVD, total cholesterol, HDL cholesterol, eGFR, and estimated blood cell composition (CD8 T, CD4 T, natural killer, B cell, monocyte, neutrophil). These potential confounders were selected from prior literature and include factors that may influence levels of plasma ADRD biomarkers (eGFR, BMI, co‐morbidities, and cholesterol levels). ¹¹ , ⁴³ Models also incorporated inverse propensity score weights to account for sample selection for plasma biomarker measurements by utilizing the following variables in the full WHIMS dataset (N = 7479): participation in various WHI substudies, MCI or dementia incidence, age, region, race, ethnicity, smoking status, hormone therapy trial arm assignment, CVD, diabetes, any non‐melanoma cancer, depressive symptoms, hysterectomy, history of female hormone use, BMI, and systolic blood pressure. Covariates with missing data were imputed using multivariate imputation by chained equations with 20 imputations and 20 iterations using the “mice” package in R. ⁴⁴

2.5.3. Associations of baseline epigenetic clocks with longitudinal changes in plasma ADRD biomarkers

To examine associations of baseline epigenetic clocks with rates of change per decade in plasma ADRD biomarkers over follow‐up, we used linear mixed‐effects models with random intercepts and slopes, with each plasma ADRD biomarker modeled as the dependent variable in separate models. ⁴⁵ , ⁴⁶ An unstructured covariance matrix was specified. Epigenetic clocks and plasma ADRD biomarkers were similarly standardized by Z‐transformation and log₂ Z‐transformation as above, respectively. Model fit comparison based on the Akaike information criterion indicated that linear time trend models were the most appropriate. Minimally and fully adjusted models included the same baseline covariates as described above, in addition to time from baseline to the second visit measuring plasma biomarkers. The linear mixed‐effects model equation was as follows:

$$\begin{array}{ccc}\hfill Y_{ij}& =& {\beta }_0+{\beta }_1\times {\mathrm{epigenetic}\mathrm{age}\mathrm{acceleration}}_i+{\beta }_2\times \mathrm{tim}{\mathrm{e}}_{ij}\hfill \\ & & +{\beta }_3\times \left(\mathrm{epigenetic}\mathrm{age}\mathrm{acceleratio}{\mathrm{n}}_i\times \mathrm{tim}{\mathrm{e}}_{ij}\right)\hfill \\ & & +\sum _{k=1}^K{\beta }_k\times \mathrm{covariat}{\mathrm{e}}_{ki}+b_{0i}+b_{1i}\times \mathrm{tim}{\mathrm{e}}_{ij}+{\epsilon }_{ij}\hfill \end{array}$$

where $Y_{ij}$ is the outcome (standardized log₂ plasma biomarker) for participant i at time j, ${\beta }_0$ is the fixed intercept, ${\beta }_1$ is the fixed effect for EAA, ${\beta }_2$ is the fixed effect for time, ${\beta }_3$ is the fixed effect for the interaction between EAA and time, ${\beta }_k$ is the fixed effect of the kth covariate, $b_{0i}$ is the random intercept for participant i, $b_{1i}$ is the random slope for time for participant i, and ${\epsilon }_{ij}$ is the residual error for participant i at time j.

To aid in interpretation of plasma ADRD biomarker change over time associated with baseline epigenetic clocks, we divided the coefficients of significant epigenetic clock $\times$ time interactions by the main effects of time and multiplied these values by 15 years, yielding the equivalent number of years of average plasma ADRD biomarker change associated with 1‐SD increases in baseline epigenetic clocks.

2.5.4. Sensitivity analyses

Sensitivity analyses were conducted to perform subgroup analyses by race (Black and White), APOE ε4 genotype (carrier vs non‐carrier), and baseline age (<70 vs ≥70 years). Participants from racial groups other than Black or White were excluded from subgroup analyses by race because of insufficient sample size. To formally test whether associations of EAA with plasma ADRD biomarkers differed by race, APOE ε4 carrier status, or age, we included interaction terms in the models. In further sensitivity analyses, we excluded participants with eGFR < 60 mL/min/1.73 m², as chronic kidney disease may be associated with levels of plasma ADRD biomarkers. ⁴⁷ Finally, we conducted sensitivity analyses excluding outliers defined as values exceeding 5‐SD above the mean, rather than applying winsorization, to assess the robustness of our findings to extreme values for plasma ADRD biomarkers. All sensitivity analyses were performed using the fully adjusted model.

Analyses were conducted using R 4.4.3 in RStudio 2024.12.1. Because our analyses were hypothesis‐driven and involved correlated plasma ADRD biomarkers, we did not apply Bonferroni correction, which would be overly conservative. We interpret findings in the context of 95% CIs and present nominal p values in the Supporting Information. Thus, cautious interpretation of findings is encouraged.

3. RESULTS

3.1. Participant characteristics

Differences between WHIMS participants in our analytic sample relative to those not included in the sample are shown in Table S1. In the analytic sample (N = 2366), the mean (SD) baseline age was 69.8 (3.8) years; 74.4% were White, 17.3% were Black, 4.6% were Asian, 0.6% were American Indian or Alaskan Native, 0.3% were Native Hawaiian or other Pacific Islander, 2.8% were more than one race, and 6.4% were Hispanic or Latino (Table 1). Significant differences (p < 0.05) in covariates were observed across quartiles of DunedinPACE. Higher levels of DunedinPACE were associated with higher BMI, a greater proportion of current or past smokers, fewer years of education, and lower total physical activity levels. The proportion of Black and Hispanic women increased with higher levels of DunedinPACE, while the proportion of White women decreased. Higher levels of DunedinPACE were associated with higher prevalence of diabetes and hypertension and lower levels of total and HDL cholesterol. There were no significant differences by chronological age, CVD, eGFR, or APOE ε4 carriage.

TABLE 1.

Baseline sociodemographic, behavior, and health characteristics by quartiles of DunedinPACE.

Characteristic	Overall N = 2366	Q1: [0.718, 0.977) N = 592	Q2: [0.977, 1.046) N = 591	Q3: [1.046, 1.121) N = 591	Q4: [1.121, 1.642] N = 592	p value
Age, mean (SD)	69.8 (3.8)	69.7 (3.8)	69.7 (3.7)	69.9 (3.7)	69.7 (3.8)	0.5
Hormone therapy arm, n (%)						<0.001
Estrogen‐alone placebo	475 (20.1%)	99 (16.7%)	108 (18.3%)	120 (20.3%)	148 (25.0%)
Estrogen‐alone intervention	477 (20.2%)	96 (16.2%)	129 (21.8%)	125 (21.2%)	127 (21.5%)
Estrogen plus progestin placebo	682 (28.8%)	190 (32.1%)	172 (29.1%)	154 (26.1%)	166 (28.0%)
Estrogen plus progestin intervention	732 (30.9%)	207 (35.0%)	182 (30.8%)	192 (32.5%)	151 (25.5%)
Race, n (%)						<0.001
American Indian or Alaskan Native	13 (0.6%)	2 (0.3%)	4 (0.7%)	2 (0.3%)	5 (0.9%)
Asian	106 (4.6%)	16 (2.7%)	20 (3.4%)	42 (7.3%)	28 (4.9%)
Native Hawaiian or other Pacific Islander	8 (0.3%)	0 (0.0%)	3 (0.5%)	2 (0.3%)	3 (0.5%)
Black	402 (17.3%)	55 (9.3%)	75 (12.9%)	98 (16.9%)	174 (30.6%)
White	1727 (74.4%)	506 (85.8%)	468 (80.4%)	412 (71.2%)	341 (59.9%)
More than one race	64 (2.8%)	11 (1.9%)	12 (2.1%)	23 (4.0%)	18 (3.2%)
Missing	46	2	9	12	23
Ethnicity, n (%)						<0.001
Not Hispanic or Latino	2198 (93.6%)	574 (97.5%)	555 (94.2%)	543 (92.8%)	526 (89.8%)
Hispanic or Latino	151 (6.4%)	15 (2.5%)	34 (5.8%)	42 (7.2%)	60 (10.2%)
Missing	17	3	2	6	6
BMI, mean (SD)	28.5 (5.6)	26.8 (4.6)	27.6 (5.4)	29.0 (5.6)	30.7 (5.8)	<0.001
Missing	12	5	2	2	3
Smoking status, n (%)						<0.001
Never smoked	1329 (57.0%)	369 (63.3%)	350 (59.6%)	350 (59.9%)	260 (45.0%)
Past smoker	877 (37.6%)	204 (35.0%)	215 (36.6%)	205 (35.1%)	253 (43.8%)
Current smoker	126 (5.4%)	10 (1.7%)	22 (3.7%)	29 (5.0%)	65 (11.2%)
Missing	34	9	4	7	14
Education, n (%)						<0.001
Less than high school equivalent	197 (8.4%)	29 (4.9%)	50 (8.5%)	47 (8.0%)	71 (12.0%)
High school diploma or GED	520 (22.1%)	132 (22.4%)	130 (22.0%)	132 (22.5%)	126 (21.4%)
Vocational, training school, or some college or associate	869 (36.9%)	202 (34.2%)	210 (35.6%)	231 (39.4%)	226 (38.3%)
College graduate or higher	771 (32.7%)	227 (38.5%)	200 (33.9%)	177 (30.2%)	167 (28.3%)
Missing	9	2	1	4	2
Diabetes, n (%)	154 (6.5%)	10 (1.7%)	24 (4.1%)	42 (7.1%)	78 (13.2%)	<0.001
Missing	5	1	1	1	2
Cardiovascular disease, n (%)	101 (4.3%)	18 (3.0%)	21 (3.6%)	27 (4.6%)	35 (5.9%)	0.073
Physical activity (hours/week), mean (SD)	11.6 (13.6)	14.3 (15.2)	11.6 (12.8)	11.4 (13.4)	9.1 (12.4)	<0.001
Missing	7	2	2	1	2
Total cholesterol (mg/dL), mean (SD)	234.0 (39.8)	238.7 (37.2)	236.4 (40.3)	231.5 (38.4)	228.6 (42.5)	<0.001
Missing	270	35	55	88	92
HDL cholesterol (mg/dL), mean (SD)	53.7 (12.6)	57.4 (12.8)	54.2 (12.3)	52.5 (12.2)	50.3 (11.7)	<0.001
Missing	270	35	55	88	92
Hypertension, n (%)	1657 (70.2%)	357 (60.3%)	406 (69.0%)	431 (73.1%)	463 (78.6%)	<0.001
Missing	7	0	3	1	3
eGFR (mL/min/1.73 m2), mean (SD)	83.8 (13.6)	84.5 (11.8)	84.9 (12.2)	83.7 (14.6)	81.8 (15.4)	0.095
Missing	271	35	56	88	92
APOE ε4 carrier status, n (%)						0.3
No ε4 alleles	1189 (73.6%)	361 (74.3%)	337 (75.9%)	277 (73.1%)	214 (69.9%)
At least one ε4 allele	426 (26.4%)	125 (25.7%)	107 (24.1%)	102 (26.9%)	92 (30.1%)
Missing	751	106	147	212	286

Abbreviations: APOE ε4, apolipoprotein E epsilon 4; BMI, body mass index; eGFR, estimated glomerular filtration rate; HDL, high‐density lipoprotein; SD, standard deviation.

3.2. Cross‐sectional associations of epigenetic clocks with plasma ADRD biomarkers at baseline

At baseline, most plasma biomarkers showed moderate to strong correlations (r = 0.30 to 0.69; Figure S2), particularly among the tau proteins (r = 0.69) and between NfL and GFAP (r = 0.51). In contrast, the Aβ42:Aβ40 ratio was moderately to weakly correlated with tau proteins (r = −0.21 to −0.34) and showed little to no correlation with NfL (r = −0.08) and GFAP (r = −0.12).

Cross‐sectional associations of epigenetic clocks with plasma biomarker levels at baseline are shown in Figure 1 and Table S2. In the fully adjusted model, every 1 SD higher in AgeAccelHorvath (SD 5.38; β = −0.050; 95% CI: −0.093 to −0.007) or AgeAccelPheno (SD 6.80, β = −0.048; 95% CI: −0.092 to −0.003) was associated with lower Aβ42:Aβ40. Associations of AgeAccelGrim2, DunedinPACE, PCPhenoAge, and PCGrimAge with Aβ42:Aβ40 were in the same direction but failed to reach statistical significance. There were no significant associations of any epigenetic clock with p‐tau181 or p‐tau217. Every 1 SD higher in AgeAccelHannum (SD 5.01, β = 0.042; 95% CI: 0.001 to 0.082), AgeAccelGrim2 (SD 4.37, β = 0.071; 95% CI: 0.020 to 0.122), PCPhenoAge (SD 6.47, β = 0.073; 95% CI: 0.024 to 0.122), or PCGrimAge (SD 3.20, β = 0.086; 95% CI: 0.033 to 0.139) was associated with higher NfL levels at baseline; no other clocks showed associations with NfL, and there were no significant associations of any epigenetic clock with GFAP levels at baseline.

FIGURE 1.

Associations of epigenetic clocks with plasma biomarkers of ADRD at baseline. Associations between epigenetic clocks and plasma biomarkers were based on linear regression models. Standardized log₂‐transformed biomarker levels are shown with 95% confidence intervals per 1‐SD increase in epigenetic clocks. Epigenetic clocks include AgeAccelHorvath (SD 5.38), AgeAccelHannum (SD 5.01), AgeAccelPheno (SD 6.80), AgeAccelGrim2 (SD 4.37), DunedinPACE (SD 0.11), PCPhenoAge (SD 6.47), and PCGrimAge (SD 3.20). Biomarkers include Aβ42:Aβ40, p‐tau181, p‐tau217, NfL, and GFAP. Standard deviation of log₂‐transformed biomarker values used in model fitting: Aβ42:Aβ40 (SD 0.31), p‐tau181 (SD 0.55), p‐tau217 (SD 0.81), NfL (SD 0.59), and GFAP (SD 0.63). Models were adjusted for chronological age, hormone therapy treatment arm, education, smoking status, race, ethnicity, physical activity, body mass index, diabetes, cardiovascular disease, hypertension, total cholesterol, HDL cholesterol, estimated glomerular filtration rate, and white blood cell counts (CD8T, CD4T, natural killer cells, B cells, monocytes, and neutrophils). ADRD, Alzheimer's diseases and related dementias; Aβ, amyloid beta; CI, confidence interval; GFAP, glial fibrillary acidic protein; NfL, neurofilament light; p‐tau181, tau phosphorylated at threonine 181; p‐tau217, tau phosphorylated at threonine 217; SD, standard deviation.

3.3. Associations of baseline epigenetic clocks with longitudinal changes in plasma ADRD biomarkers

Overall, 129 of 873 women (14.8%) were diagnosed with MCI or probable dementia between baseline and the second study visit an average of 15 years later. Average levels of the six plasma ADRD biomarkers at baseline and the second study visit are shown in Figure 2, and annual changes in biomarker levels are shown in Figure S3. The Aβ42:Aβ40 ratio did not change over time. In contrast, all other plasma biomarkers increased from baseline to the second time point. The correlation pattern of annualized biomarker changes was similar but weaker (r = −0.03 to 0.64) compared to their correlations at baseline (Figure S4). Aβ42:Aβ40 changes were not correlated with changes in other plasma biomarkers (r = −0.03 to −0.11), suggesting that its longitudinal trajectory was relatively independent or had a different time course in relation to other plasma biomarkers.

FIGURE 2.

Changes in plasma biomarkers over an average of 15 years. Distribution of plasma biomarker levels at baseline and a second time point an average of 15 years later among participants with both measurements (N = 873). Biomarkers include plasma Aβ42:Aβ40, p‐tau181, p‐tau217, GFAP, and NfL. Violin plots display the distribution density. Boxplots indicate median and interquartile ranges. Means and SDs are presented above each group. Aβ, amyloid beta; GFAP, glial fibrillary acidic protein; NfL, neurofilament light; p‐tau181, tau phosphorylated at threonine 181; p‐tau217, tau phosphorylated at threonine 217; SD, standard deviation.

Figure 3 and Table S3 show associations of baseline epigenetic clocks with changes in plasma ADRD biomarkers in the fully adjusted models. Every 1 SD higher baseline DunedinPACE (SD = 0.11, corresponding to an 11% faster pace of aging compared with the average) was associated with a faster rate of increase in p‐tau181 (β = 0.062; 95% CI: 0.024 to 0.099), p‐tau217 (β = 0.051; 95% CI:  0.015 to 0.087), NfL (β = 0.064; 95% CI:  0.029 to 0.100), and GFAP (β = 0.056; 95% CI:  0.022 to 0.089), corresponding to the equivalent of approximately 1.23 years of increase for p‐tau181, 1.01 years for p‐tau217, 1.07 years for NfL, and 1.41 years for GFAP over the 15‐year follow‐up. Figure 4 illustrates estimated changes in the plasma biomarkers during follow‐up across different levels of baseline DunedinPACE.

FIGURE 3.

Associations of baseline epigenetic clocks with changes in plasma biomarkers of ADRD over an average of 15 years. Associations between baseline epigenetic clocks and rates of changes per decade in plasma biomarkers based on linear mixed‐effects regression models. Standardized log₂‐transformed rates of change in plasma biomarkers are shown with 95% confidence intervals per 1‐SD increase in epigenetic clocks at baseline. Epigenetic clocks include AgeAccelHorvath (SD 5.36 years), AgeAccelHannum (SD 4.93), AgeAccelPheno (SD 6.72), AgeAccelGrim2 (SD 4.29), DunedinPACE (SD 0.11 years of physiologic decline), PCPhenoAge (SD 6.43), and PCGrimAge (SD 3.15). Biomarkers include Aβ42:Aβ40, GFAP, NfL, p‐tau181, and p‐tau217. Standard deviation of log₂‐transformed biomarker values used in model fitting: Aβ42:Aβ40 (SD 0.32), p‐tau181 (SD 0.62), p‐tau217 (SD 0.92), NfL (SD 0.73), and GFAP (SD 0.68). Models were adjusted for chronological age, time since baseline in decades, hormone therapy treatment arm, education, smoking status, race, ethnicity, physical activity, body mass index, diabetes, cardiovascular disease, hypertension, total cholesterol, high‐density lipoprotein cholesterol, estimated glomerular filtration rate, and white blood cell counts (CD8T, CD4T, natural killer cells, B cells, monocytes, and neutrophils). ADRD, Alzheimer's diseases and related dementias; Aβ, amyloid beta; CI, confidence interval; GFAP, glial fibrillary acidic protein; p‐tau181, tau phosphorylated at threonine 181; p‐tau217, tau phosphorylated at threonine 217; NfL, neurofilament light; SD, standard deviation.

FIGURE 4.

Estimated changes in plasma biomarkers of ADRD over an average of 15 years across 3 levels of DunedinPACE from a fully adjusted linear mixed‐effects model. Linear mixed‐effects models contained random intercepts and slopes adjusted for chronological age, time since baseline in decades, hormone therapy treatment arm, education, smoking status, race, ethnicity, physical activity, body mass index, diabetes, cardiovascular disease, hypertension, total cholesterol, HDL cholesterol, estimated glomerular filtration rate, and white blood cell counts (CD8T, CD4T, natural killer cells, B cells, monocytes, and neutrophils). An interaction term between DunedinPACE and time was included. Results are Z‐transformed DunedinPACE set to the mean (1.05), 1 SD below (0.94) and above the mean (1.16) at baseline. Biomarkers include GFAP, NfL, p‐tau181, and p‐tau217. ADRD, Alzheimer's disease and related dementias; GFAP, glial fibrillary acidic protein; NfL, neurofilament light; p‐tau181, tau phosphorylated at threonine 181; p‐tau217, tau phosphorylated at threonine 217; SD, standard deviation.

Every 1 SD higher baseline AgeAccelGrim2 (SD, 4.29) was associated with a faster rate of increase in p‐tau181 (β = 0.050; 95% CI:  0.012 to 0.088) and GFAP (β = 0.035; 95% CI:  0.001 to 0.069), corresponding to 1.01 and 0.89 years of increase over the 15‐year follow‐up, respectively. Every 1 SD higher baseline PCPhenoAge (SD, 6.43) was associated with a faster increase in both p‐tau181 (β = 0.045; 95% CI: 0.008 to 0.082) and p‐tau217 (β = 0.039; 95% CI: 0.004 to 0.073), equivalent to 0.91 and 0.77 years of increase over follow‐up, respectively. Finally, every 1 SD higher in baseline PCGrimAge (SD, 3.15) was associated with a faster increase in p‐tau181 (β = 0.043; 95% CI: 0.006 to 0.081) and GFAP (β = 0.034; 95% CI: 0.000 to 0.068), both corresponding to 0.87 years of increase over follow‐up.

3.4. Sensitivity analyses

In the cross‐sectional analysis, the associations between AgeAccelGrim2, DunedinPACE, PCPhenoAge, and PCGrimAge with NfL, as well as the associations of AgeAccelGrim2 and PCGrimAge with GFAP, were significantly modified by race (Table S4). Among Black women, higher AgeAccelGrim2 was significantly associated with higher NfL (β = 0.184; 95% CI:  0.060 to 0.309) and GFAP (β = 0.145; 95% CI:  0.010 to 0.280), whereas these associations were not significant among White women. Similarly, PCPhenoAge (β = 0.142; 95% CI:  0.029 to 0.255) and PCGrimAge (β = 0.157; 95% CI:  0.023 to 0.290) were each associated with higher NfL in Black but not White women. The association of DunedinPACE with NfL, as well as of PCGrimAge with GFAP, was stronger but not significant in Black versus White women. There were no significant interactions between any of the epigenetic clocks and baseline age or APOE ε4 in cross‐sectional analyses.

In longitudinal analyses, the association of AgeAccelPheno with change in Aβ42:Aβ40 was modified by race (p interaction = 0.011), with a stronger but not significant association among Black versus White women (Table S5). The association of AgeAccelHannum with change in NfL was modified by baseline age (p interaction = 0.028, Table S6), with a stronger but non‐significant association among women <70 versus ≥70 years. No other significant interactions with race or baseline age were observed, and APOE ε4 genotype also did not modify any associations in longitudinal analysis.

After excluding participants with eGFR < 60 mL/min/1.73 m², there remained 1942 women at baseline and 829 with longitudinal data for sensitivity analysis. At baseline, only the association of PCGrimAge with NfL remained significant (β = 0.063; 95% CI:  0.005 to 0.121; Table S7). In longitudinal analyses, the associations observed in the main analyses remained significant; additionally, AgeAccelGrim2 (β = 0.042; 95% CI:  0.007 to 0.077) and PCGrimAge (β = 0.038; 95% CI:  0.003 to 0.073) were each associated with a faster rate of increase in NfL (Table S8). In the sensitivity analyses excluding outliers defined as 5 SD above the mean, findings were similar to the main analysis (Tables S9 and S10).

4. DISCUSSION

Among older women, we observed that epigenetic clocks were associated with plasma ADRD biomarkers at baseline and longitudinally. AgeAccelHorvath and AgeAccelPheno were associated with lower Aβ42:Aβ40 ratio. AgeAccelGrim2, PCPhenoAge, and PCGrimAge were associated with higher NfL at baseline; however, only PCGrimAge remained significant after removing those with eGFR < 60 mL/min/1.73 m². A 1‐SD higher DunedinPACE corresponded to the equivalent of 1.23 years of increase for p‐tau181, 1.01 years for p‐tau217, 1.07 years for NfL, and 1.41 years for GFAP over the 15‐year follow‐up. Baseline AgeAccelGrim2 and PCGrimAge were associated with faster increases in p‐tau181 and GFAP, and PCPhenoAge was associated with faster increases in p‐tau181 and p‐tau217.

Several studies have observed associations of epigenetic clocks with lower cognitive performance and greater cognitive decline as well as increased dementia risk. ¹¹ , ¹² , ¹³ , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ³² Only one prior study has evaluated associations of epigenetic clocks with plasma biomarkers of ADRD; however, that study was cross‐sectional and did not examine p‐tau217. ³² In the Study of Latinos‐Investigation of Neurocognitive Aging (SOL‐INCA, N = 2625), PCPhenoAge, AgeAccelGrim2, and DunedinPACE were associated with lower Aβ42:Aβ40 ratio among women but not men, while associations of epigenetic clocks with higher NfL levels were observed in both sexes. ³² We found that AgeAccelHannum, AgeAccelGrim2, PCPhenoAge, and PCGrimAge, but not DunedinPACE, were associated with NfL at baseline; however, after removing those with baseline eGFR < 60 mL/min/1.73 m² indicative of chronic kidney disease, only PCGrimAge remained significant. Discrepancies between WHIMS and SOL‐INCA may be partly explained by differences in study populations, as SOL‐INCA included only Hispanic/Latino men and women who had either MCI or were cognitively healthy, while WHIMS consisted of White, Black, and Hispanic/Latina women who were cognitively unimpaired at baseline. In addition, SOL‐INCA did not adjust analyses for white blood cell composition, which may be an important confounder in epigenetic analyses. ⁴⁸

In the ADNI cohort, first‐ and second‐generation clocks were not associated with cognitive function, but higher DunedinPACE was associated with worse scores on tests of memory and executive function. ²⁰ Other studies have found associations of first‐ and second‐generation clocks with cognitive function; however, they did not examine DunedinPACE. ¹² , ¹⁵ , ¹⁹ More recent studies found weaker or non‐significant associations of DunedinPACE with cognitive decline or risk of dementia relative to first‐ or second‐generation clocks. ¹³ , ¹⁶ Among a subset of WHIMS women (N = 795), we previously observed that higher DunedinPACE was associated with faster annual decline in cognitive function. ¹¹ In the larger WHIMS cohort (N = 6069), both DunedinPACE and AgeAccelGrim2 were associated with higher risk of MCI or dementia, ²¹ and the Framingham Heart Study reported similar findings for DunedinPACE. ¹⁴ , ²⁰

Although DunedinPACE was not associated with baseline plasma ADRD biomarker levels, it was the most predictive of longitudinal change across multiple plasma biomarkers in our study. Risk of ADRD‐related pathology increases with age, and with a mean age of 70 years at baseline, our cohort of cognitively healthy women likely contains a low proportion of individuals with ADRD‐related pathology at baseline. ⁴⁹ This proportion is likely to increase over time. The lack of sensitivity of DunedinPACE to baseline levels of biomarkers while being predictive of an increase in biomarker change over time in our study may indicate that DunedinPACE is sensitive to latent variation in risk of ADRD‐related pathology prior to the increase in plasma biomarker levels. Whether DunedinPACE may serve as an even earlier indicator of ADRD risk than plasma biomarker levels is an important area for future research.

The greater sensitivity of DunedinPACE compared to other clocks to biomarker change over time may be partly explained by the fact that DunedinPACE was trained on longitudinal physiological change in 19 biomarkers measuring organ system integrity (e.g., cardiovascular, metabolic, renal, and immune) at four time points spanning two decades, whereas other clocks were trained on chronological age (first‐generation clocks) or clinical phenotypes (second‐generation clocks) at a single time point. ⁹ It is of interest that DunedinPACE was developed to predict the pace of biological aging among a primarily White sample of younger adults – aged 26 to 45 years old, an age range too young to experience any substantial effect of latent neurodegenerative disorders. Notably, none of the biomarkers in DunedinPACE relates specifically to brain health. Nevertheless, a measure that captures 20‐year change in organ health from young adulthood to middle age significantly predicted increase in ADRD pathology over time in our study, with no differences in associations between White and Black women or by APOE ε4 carriage. These findings support the robustness of DunedinPACE for predicting aging phenotypes and support the foundation of the geroscience hypothesis that biological mechanisms of aging may contribute to multiple co‐morbidities of aging, including neurodegenerative disorders. ⁵⁰

The plasma ADRD biomarkers we examined were associated with incident dementia in prior studies ⁴³ , ⁵¹ and reflect different pathologies in the brain, with different temporal dynamics in relation to dementia onset. Plasma Aβ42:Aβ40 is believed to reflect soluble amyloid pathology, which develops early in the course of the disease, decades prior to dementia, and plateaus early in the preclinical period prior to the emergence of brain amyloid plaques. ⁵² Changes in plasma Aβ42:Aβ40 ratio over time are very subtle, with relative change peaking at −1% per year, and detecting such small shifts requires highly precise assays. ⁵² , ⁵³ In contrast, plasma p‐tau181 and p‐tau217, which are believed to reflect fibrillar amyloid pathology or amyloid plaques measured by amyloid PET, increase as AD progresses. ⁵² , ⁵⁴ Recently, we reported that plasma p‐tau217 was associated with over three‐fold higher risk of incident dementia during an average 15‐year follow‐up in the WHIMS cohort. ⁵⁵ In the present study, plasma Aβ42:Aβ40 remained stable over the average 15‐year follow‐up, while plasma p‐tau181 and p‐tau217 increased over time, which is consistent with this dynamic. Our findings showed that, at baseline, EAA as measured by AgeAccelHorvath and AgeAccelPheno was associated with lower plasma Aβ42:Aβ40 ratio, but not with plasma p‐tau181 or p‐tau217. In longitudinal analyses, AgeAccelGrim2, DunedinPACE, PCPhenoAge, and PCGrimAge were associated with a faster increase in p‐tau181, while DunedinPACE and PCPhenoAge were also associated with a faster increase in p‐tau217. These findings suggest that accelerated biological aging, as measured by DunedinPACE and PCPhenoAge, may be associated with progression of AD pathology over time.

Plasma NfL and GFAP also increased over the average 15‐year follow‐up in our study. Plasma NfL is a non‐specific marker of axonal damage that increases with ADRD pathology. ⁵² Because axonal degeneration occurs in aging and in multiple age‐related pathologies, it is not surprising that NfL showed one of the largest increases over time in our community‐based sample. ⁵⁶ Plasma GFAP, a biomarker of reactive gliosis and neuroinflammation, is also a non‐specific biomarker of ADRD. ⁵⁷ Reactive gliosis is associated with amyloid plaques in the brain, and plasma GFAP is predictive of brain amyloid pathology. ⁵⁸ , ⁵⁹ Plasma GFAP increases across the AD severity spectrum from preclinical stages to dementia. ²⁷ , ³¹ , ⁵² We observed moderate correlations between longitudinal plasma p‐tau181, p‐tau217, NfL, and GFAP, which was expected given that they are all sensitive to increasing fibrillar amyloid pathology. ⁵² , ⁶⁰ , ⁶¹ DunedinPACE showed similar magnitudes of association with each of these four biomarkers in our study. In addition, PCPhenoAge outperformed AgeAccelPheno and was associated with a faster rate of increase in NfL. AgeAccelGrim2 and PCGrimAge were both associated with a faster rate of increase in GFAP in our study.

Our study's strengths include a diverse sample with longitudinal measures of plasma ADRD biomarkers at two visits spanning an average of 15 years and examinations of first‐, second‐, and third‐generation epigenetic clocks. We had information on six plasma ADRD biomarkers, including p‐tau217, which has shown better predictive performance of AD pathology and clinical phenotypes relative to other biomarkers, such as p‐tau181 and p‐tau231. ⁶² , ⁶³ The WHIMS dataset is highly comprehensive and allowed us to adjust for a robust set of potential confounders in our analysis. We also acknowledge several limitations. The sample consisted of women only, which limits the generalizability of our findings to men. Plasma ADRD biomarkers were measured at only two time points; thus, we were unable to examine biomarker trajectories over time. We did not attempt to relate the EAA biomarkers to a composite of the plasma proteins, as currently there is no validated multi‐marker index that predicts dementia, and any such multi‐marker index would require rigorous investigation into its ability to predict dementia across multiple cohorts. Finally, because we analyzed multiple epigenetic clocks and multiple plasma biomarkers, findings should be interpreted with caution.

In conclusion, the accelerated pace of biological aging, as indicated by DunedinPACE, was associated with a faster rate of increase in plasma p‐tau181, p‐tau217, NfL, and GFAP over time among older women. Further, AgeAccelGrim2, PCGrimAge, and PCPhenoAge were associated with increases in p‐tau181, PCPhenoAge with increases in p‐tau217 and NfL, and AgeAccelGrim2 and PCGrimAge with increases in GFAP. These findings suggest that accelerated biological aging may be implicated in the future development of ADRD pathology. Further research is needed to determine whether epigenetic clocks provide independent information in predicting ADRD beyond that provided by plasma biomarkers of ADRD and whether these clocks may be useful for monitoring changes in risk of developing ADRD pathology in trials testing interventions for dementia prevention.

CONFLICT OF INTEREST STATEMENT

The Regents of the University of California are the sole owner of patents and patent applications directed at epigenetic biomarkers for which S.H. and A.T.L. are named inventors. S.H. is a founder, paid consultant, and serves on the board of the non‐profit Epigenetic Clock Development Foundation that licenses these patents. S.H. reports funding from NIH 1U01AG060908‐01 and receiving royalty payments surrounding these patents. S.H. is a principal investigator at Altos Labs. A.T.L. reports royalties or licenses for the DNAm GrimAge2 clock. B.S. discloses the following relationships: the NHLBI grant R01 HL164485; funding from the American Heart Association and NINDS; consulting fees from the Women's Health Initiative; medicolegal malpractice review for various firms; and American Heart Association regional chapter president. M.M.M. reports funding from NIH: RF1 AG69052; RF1 AG077386; R01AG079397, U19 AG078109, and U24 AG082930; DOD: W81XWH2110490; and the Alzheimer's Association. M.M.M. has served on scientific advisory boards and/or has consulted for Acadia, Althira, Beckman Coulter, Biogen, Cognito Therapeutics, Eisai, Lilly, Merck, Neurogen Biomarking, Novo Nordisk, Roche, and Siemens Healthineers; received speaking honorariums from Roche, Novo Nordisk, Biogen, and Medscape; and participated in grant reviews for the Alzheimer's Drug Discovery Foundation. B.Z. reports grant funding from R01AG079149 and residual class settlement funds in the matter of April Krueger vs. Wyeth, Inc., Case No. 03‐cv‐2496 (US District Court, SD of Calif.). A.Z.L. reports funding from NIA R01AG079149 and R01AG074345 and NHLBI Contract No. 75N92021D00001. S.M.R. reports NIA/NIH funding as an NIA intramural employee, support from the McKnight Foundation as a keynote speaker, serving on the International Scientific Advisory Boards of the Canadian Consortium on Neurodegeneration in Aging and Dementia Platforms UK, and serving on the External Advisory Board of Adult Aging Brain Connectome. S.N. reports funding from NIA 5K99AG082863‐02. L.K.M. reports funding from NIA R01AG079149 and R01AG074345. M.A.E. reports funding from AG074345, AG058571, and AA‐POINTER‐19‐611541; service on a steering committee for Nestle; support for attending meetings and/or travel from the American Diabetes Association and European Association for the Study of Diabetes; and participation on a data safety monitoring board or advisory board for Acumen, Annovis Bio, and several NIH‐supported studies. A.H.S. reports funding from NIA R01AG079149 and R01AG074345 and residual class settlement funds in the matter of April Krueger v. Wyeth, Inc., Case No. 03‐cv‐2496 (US District Court, SD of Calif.), as well as consulting fees from the WHI as the chair of a scientific interest group and member of the Publications and Presentations Committee. S.R.R., C.M.N., A.X.M., K.B., D.L., J.E.M., and L.F. declare no conflicts of interest. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

This study was performed in accordance with the ethical standards as laid down in the 1964 Declaration of Helsinki and its later amendments and was approved by the Institutional Review Board at the University of California San Diego. All participants provided written informed consent.

Supporting information

Supporting Information

Supporting Information

Supporting Information

Zhang B, McEvoy LK, Nguyen S, et al. Epigenetic clocks and longitudinal plasma biomarkers of Alzheimer's disease. Alzheimer's Dement. 2025;21:e70983. 10.1002/alz.70983