One-carbon metabolism-related compounds are associated with epigenetic aging biomarkers: results from the cross-sectional National Health and Nutrition Examination Survey 1999–2002

Abstract

Background

One-carbon metabolism (OCM), a biochemical pathway dependent on micronutrients including B vitamins, plays an essential role in aging-related physiological processes. DNA methylation–based aging biomarkers may be influenced by OCM.

Objectives

This study investigated associations of OCM-related biomarkers with epigenetic aging biomarkers in the cross-sectional National Health and Nutrition Examination Survey (1999–2002).

Methods

Blood DNA methylation was measured in adults aged ≥50 y. The following epigenetic aging biomarkers were included: Horvath1, Horvath2, Hannum, PhenoAge, GrimAge2, Dunedin Pace-of-Aging (DunedinPoAm), and DNA methylation telomere length (DNAmTL). We tested for associations of serum folate, red blood cell folate, vitamin B12, homocysteine (Hcy), and methylmalonic acid (MMA) concentrations with epigenetic age deviation (EAD) among 2346 participants with epigenetic and nutritional status biomarkers using adjusted survey-weighted general linear regression models.

Results

A doubling of serum folate concentration was associated with −0.82 y (95% confidence interval: −1.40, −0.23) lower GrimAge2 EAD, −0.13 SDs (−0.22, −0.03) lower DunedinPoAm, and 0.02 kb (0.00, 0.04) greater DNAmTL EAD. Conversely, a doubling in Hcy concentration was associated with 1.05 y (0.06, 2.04) greater PhenoAge EAD, 1.93 y (1.16, 2.71) greater GrimAge2 EAD, and 0.26 SDs (0.10, 0.41) greater DunedinPoAm. Associations of serum folate with EAD were attenuated after adjusting for smoking status, alcohol intake, and estimated glomerular filtration rate. Furthermore, smoking modified the associations of Hcy with GrimAge2 EAD. Chronic kidney disease modified associations of B12 and MMA with Horvath1 and GrimAge2 EAD, respectively.

Conclusions

In a nationally representative sample of United States adults, higher concentration of folate, a carbon donor, was associated with lower EAD, and higher concentration of Hcys, an indicator of OCM deficiencies, was associated with greater EAD; however, some associations were influenced by smoking and renal function. Future research should focus on high-risk populations. Long-term randomized controlled trials are also needed to establish causality and investigate the clinical relevance of changes in EAD.

Introduction

One-carbon metabolism (OCM) is a biochemical pathway essential for supporting many physiological processes. OCM produces the universal methyl donor S-adenosylmethionine (SAM) and includes the trans-sulfuration pathway, which generates the antioxidant glutathione and provides 1-carbon units for purine and thymidylate biosynthesis. OCM plays an essential role in aging-related processes by supplying methyl groups, producing building blocks for DNA synthesis and repair, and maintaining redox balance in cells [1]. Deficiencies or imbalances in OCM increase risk of aging-related diseases [[1], [2], [3]] and may influence aging-related biomarkers.

OCM is dependent on diet-derived micronutrients including folate (vitamin B9) and cobalamin (vitamin B12) [4] (Figure 1; adapted from Bozack et al. [5]). Folate facilitates the recruitment of methyl groups to OCM, which are subsequently used to remethylate homocysteine (Hcy) to methionine, the precursor of SAM, by methionine synthase using B12 as a cofactor. Upon donation of a methyl group, SAM is converted to S-adenosylhomocysteine, which can be hydrolyzed to Hcy and remethylated to methionine given adequate folate and B12. B12 is also involved in the metabolism of methylmalonic acid (MMA), a byproduct of propionate metabolism [6]. High Hcy concentration is an indicator of folate or B12 deficiency, and high MMA concentration is an indicator of B12 deficiency [7,8].

Figure 1.

One-carbon metabolism. Compounds included in this study are shown in orange. DHF, dihydrofolate; MMA, methylmalonic acid; SAH, S-adenosylhomocysteine; SAM, S-adenosylmethionine; THF, tetrahydrofolate. Adapted from Bozack et al. [5].

The prevalence of folate and B12 deficiencies increases with age due to changes in diet, intestinal absorption, and antagonistic therapeutic drugs [9,10]. Low folate concentrations have been associated with age-related conditions, including dementia [11], cognitive function [12], and cancers [13,14]. Folate and B12 deficiencies also increase Hcy concentration. Elevated Hcy can induce oxidative stress, inflammation, and impair endothelial function [15], increasing risk of cardiovascular disease (CVD) incidence and mortality [16], cognitive impairment and decline [17], and all-cause mortality [16].

The study of aging and the health span has advanced with the development of epigenetic aging biomarkers, also known as epigenetic clocks. These biomarkers leverage DNA methylation levels at specific cytosine–guanine dinucleotide sites, which change with age. Epigenetic clocks can be broadly classified as predictors of chronological age (i.e., first-generation clocks) [18,19], aging-related phenotypes and mortality risk (i.e., second-generation clocks) [[20], [21], [22]], and rate of aging [23]. Epigenetic age deviation (EAD), or the difference between epigenetic and chronological age, is strongly associated with age-related morbidities and all-cause mortality [24,25]. Second-generation clocks have proven to be particularly strong predictors of cancer, CVD, comorbidity, and time to death [20,22]—aging-related outcomes that may have etiologies related to OCM deficiencies.

Testing the relationship between markers of OCM and EAD can inform the application of epigenetic aging biomarkers in research and clinical settings. In this study, we aimed to investigate associations of OCM-related biomarkers with EAD among participants in the NHANES, a nationally representative sample of adults in the United States. Our study focused on concentrations of folate, B12, Hcy, and MMA and 7 well-established epigenetic aging biomarkers. We hypothesized that promoters of OCM (i.e., folate and B12) would be associated with lower EAD, whereas indicators of OCM deficiencies (i.e., Hcy and MMA) would be associated with increased EAD.

Methods

Study population

We leveraged existing data from the 1999–2000 and 2001–2002 cycles of NHANES. NHANES is an ongoing program conducted by the National Center for Health Statistics (NCHS), which includes interviews, physical examinations, and laboratory measurements and is designed to evaluate and monitor the health and nutrition among noninstitutionalized adults and children in the United States [26]. In the 1999–2000 and 2001–2002 cycles of NHANES, DNA methylation was measured in blood samples of a subset of adults aged 50 y or older. To protect participant confidentiality, participants ≥ 85 y were top coded as 85 y old. Consequently, all participants assigned an age of 85 y were excluded from our analyses as we were unable to determine their true chronological age. Our study included 2346 adults with available DNA methylation data and nutritional biomarkers after excluding sex mismatches and participants 85 y and older (Supplemental Figure 1). All participants provided written informed consent, and the study protocols were approved by the NCHS Research Ethics Review Board [27].

DNA methylation

A detailed description of DNA methylation laboratory methods and data processing can be found on the NHANES website [28]. In the 1999–2000 and 2001–2002 cycles of NHANES, DNA methylation was measured in blood samples of 2532 adults. Samples were selected for DNA methylation measurement among participants aged 50 y or older with available biospecimens and included a random sample of approximately half of eligible non-Hispanic White participants and all eligible participants from other ethnic and racial groups [28]. DNA was extracted from whole blood and stored at −80°C until DNA methylation measurement. DNA underwent bisulfite conversion, and DNA methylation was measured using the Illumina Infinium MethylationEPIC BeadChip v1.0 (Illumina) following the manufacturer’s recommendations. Raw DNA methylation data were imported into R for preprocessing and quality control (QC). The preprocessing and QC pipeline included color correction, background subtraction, and removal of outlier samples based on median intensity values.

Epigenetic age and biomarkers

Epigenetic age estimates and aging biomarkers were provided by NHANES [28]. Before biomarker calculation, DNA methylation data were normalized using the beta mixture quantile (BMIQ) method, which adjusts for probe type bias [29]. For the calculation of the Horvath1, Horvath2, PhenoAge, and GrimAge2 clocks, a modified BMIQ method was applied using a “gold standard” based on the largest training data set in the Horvath1 clock [18]. The epigenetic age estimates or scores were calculated for each participant using the respective published coefficients and intercepts. We focused on widely studied and established epigenetic aging biomarkers including 3 first-generation clocks: Horvath1 [18], Horvath2 [19], and Hannum [30]; 2 second-generation clocks: PhenoAge [20] and GrimAge2 [21]; Dunedin Pace-of-Aging (DunedinPoAm) [23]; and DNA methylation telomere length (DNAmTL) [31]. GrimAge2 is an updated version of the GrimAge biomarker of mortality risk, which includes training on additional DNA methylation–based estimates of plasma proteins (high-sensitivity C-reactive protein [CRP] and hemoglobin A1C). Due to the high correlation between GrimAge and GrimAge2 in NHANES (r = 0.99), we chose to only analyze GrimAge2.

Nutritional biochemistries

Methods for nutritional biochemistries are described on the NHANES website with the respective data releases [32,33]. Briefly, serum folate and B12 concentrations were measured using the Quantaphase II Folate/vitamin B12 radioassay kit (Bio-Rad Laboratories) [32]. The limits of detection (LODs) were 0.2 ng/mL and 20 pg/mL for folate and B12, respectively [34,35]. Red blood cell (RBC) folate was measured by diluting samples in ascorbic acid in water and incubating or freezing to hemolyze RBCs. Samples were diluted again in protein diluent, and RBC folate was measured using methods analogous to serum folate. The coefficients of variation (CVs) were 3%–7%, 2%–4%, and 3%–6% for serum folate, B12, and RBC folate, respectively. Plasma Hcy was measured using the Abbott Homocysteine assay, a fluorescence polarization immunoassay, and the Abbott IMX analyzer (Abbott Diagnostics). Samples with concentrations <2 or >15 μmol/L were rerun to confirm results, and the CV was 3%–6% [36,37]. MMA was extracted from plasma or serum with a strong anion exchange resin, and concentrations were measured by gas chromatography and a mass selective detector (Hewlett–Packard). Samples with concentrations >20 μmol/L were rerun to confirm results, and the CV was 4%–10% [38,39].

Covariates

Demographic, socioeconomic, anthropometric measures, and self-reports of health-related behaviors were collected as part of the NHANES questionnaire and physical examination. As defined by NHANES, chronological age at the time of the screening interview was calculated from the reported date of birth. Self-reported race and ethnicity were classified by NHANES as Non-Hispanic White, Mexican-American, other Hispanic, Non-Hispanic Black, or other race including Multiracial [40,41]. Participants were asked if any of the following groups represented their national origin or ancestry: “Mexican-American/Mexican,” “other Hispanic or Latino,” “both Mexican and other Hispanic,” or “not Hispanic.” Participants were coded as Mexican-American if they responded “Mexican-American/Mexican” or “both Mexican and other Hispanic,” and other Hispanic if they responded “other Hispanic or Latino.” Participants who reported “not Hispanic” were further asked if they considered themselves “American Indian or Alaskan Native,” “Asian,” “Black or African-American,” “Native Hawaiian or Pacific Islander,” “White,” or “other,” and were coded as such. Smoking status was based on the Smoking and Tobacco Use Questionnaire Section and classified as never (not having smoked ≥100 cigarettes in life), former (having smoked ≥100 cigarettes in life but not currently smoking), or current (having smoked ≥100 cigarettes in life and currently smoking every day or some days). Alcohol intake in average number of drinks per day in the last 12 mo was calculated from the Alcohol Use Questionnaire Section. Occupational status was obtained from the Occupation Questionnaire Section and classified as white-collar and professional work, white-collar and semiroutine work, blue-collar and high-skill work, blue-collar and semiroutine work, or no work as described [42]. Education level was classified as less than high school, high school diploma (including GED), or more than high school. The poverty to family income ratio (PIR) was based on the Department of Health and Human Services’ poverty guidelines and calculated as the family income divided by the poverty guidelines for family size and year. BMI in kg/m² was calculated from height and weight measurements collected by trained technicians during examinations.

Data on selected protein concentrations measured in blood were also available in NHANES. Serum β-2 microglobulin (B2M) concentrations were measured with a B2M immunoassay, and serum cystatin C concentrations were measured with a Cystatin C immunoassay on an automated multichannel analyzer (Siemens Healthcare Diagnostics) [43]. The lower and upper LODs for B2M and cystatin C were 0.72 and 23.0, and 0.23 and 8.00 mg/L, respectively. The CVs were 3.4%–3.8% and 3.5%–4.3%, respectively. CRP concentrations were measured using latex-enhanced nephelometry [44,45]. The lowest reportable concentration was 0.02 mg/dL, and the CVs were 5%–9% [46,47]. Glycohemoglobin (hemoglobin A1C) concentrations in whole blood were measured using an HPLC-based automated glycohemoglobin analyzer (Primus) and had CVs < 3% [48,49]. Serum creatinine was measured using a modified version of the Jaffe reaction [50,51]. The analytical range was 0.1–25 mg/dL, and the CVs were 1%–4% [52]. Serum creatinine was standardized for the 1999–2000 cycle based on the recommendations provided by NHANES; standardization was not needed for the 2001–2002 cycle [50,51]. Estimated glomerular filtration rate (eGFR) in mL/min/1.73 m² was calculated based on the 2021 Chronic Kidney Disease (CKD) Epidemiological Collaboration creatine equation [53]. CKD was classified as eGFR < 60 mL/min/1.73 m².

Statistical analyses

Among 2532 participants with epigenetic age estimates, we excluded those with a recorded age of ≥85 y (n = 130) or with a mismatch between recorded sex and a DNA methylation–based sex estimate (n = 56). A total of 2346 adults with DNA methylation data and nutritional biomarkers were available for analysis (Supplemental Figure 1). We calculated EAD, also known as epigenetic age acceleration [54], as the residuals of regressing chronological age in years on epigenetic age. DunedinPoAm was analyzed centered and scaled. Blood cell type proportions (CD8+ T cells, CD4+ T cells, neutrophils, monocytes, B cells, and natural killer cells) were estimated using regression calibration [55] based on IDOL probe selection [56,57].

Descriptive statistics before survey weighting were calculated as means and SDs for continuous variables and frequencies and proportions for categorical variables. We evaluated performance of the epigenetic aging biomarkers by calculating correlations and median absolute errors (MAEs) between chronological age and estimated epigenetic age. Survey weights were provided with the NHANES epigenetic biomarkers data set [28]. The survey design was specified using the Survey R package for statistical analyses [58,59]. To preserve the study design, the survey design was specified prior to dropping participants ≥85 y old and sex mismatches. Correlations between OCM-related biomarkers were calculated using the svycor function using the bootstrap procedure in the jtools R package [60].

We evaluated associations of OCM-related biomarkers with epigenetic aging biomarkers (EAD residuals or DunedinPoAm) using survey design–weighted generalized linear regression models. Analyses were conducted using the svyglm function in the Survey R package [58,59]. The concentrations of serum folate, RBC folate, B12, Hcy, and MMA were log₂ transformed to reduce the influence of outliers and meet model assumptions. Base models were adjusted for potential confounders and precision variables identified a priori: chronological age, chronological age², sex, BMI (continuous), race and ethnicity, smoking status, education level, occupation, and PIR (continuous).

We further adjusted for behavior and health factors that may be upstream on the causal pathway or confounders of associations of OCM-related biomarkers with EAD: smoking status, alcohol intake, and eGFR, an indicator of renal function. Smoking [61] and alcohol [62] may act as folate antagonists by affecting the absorption, metabolism, and maintenance of circulating levels. Elevated Hcy can impair renal function, and, in turn, impaired renal function increases the risk of CVD [63]. Poor renal function can also affect folate and B12 transport and excretion and has been associated with increased serum folate and RBC folate concentration in a previous study in NHANES [64]. We also tested for interactions between the concentrations of OCM-related biomarkers and smoking status and CKD by conducting stratified analyses and including an interaction term in unstratified analyses.

Considering conflicting evidence of the relationship between excess folate and B12 and adverse health outcomes, including cancer progression and cognition [[65], [66], [67], [68], [69]], we tested for nonlinear associations using fully adjusted models with folate tertiles (serum folate: <11.2, ≥11.2, <17.6, and ≥17.6 ng/mL; RBC folate: <250, ≥250, <352, and ≥352 ng/mL RBC) and B12 tertiles (<398, ≥398, <583, and ≥583 pg/mL). We also conducted analyses with the dichotomous variable hyperhomocysteinemia (Hcy concentration > 15 μmol/L) [70] compared with not.

To evaluate if individual GrimAge2 components were contributing to the associations with GrimAge2 EAD, we conducted adjusted analyses of Hcy concentrations using GrimAge2 components (plasma protein surrogates and smoking pack-years) as the outcomes. We conducted complementary analyses for proteins with laboratory measures, if available.

We performed sensitivity analyses adjusting for cell type proportions to distinguish between intrinsic aging (i.e., intracellular) and extrinsic aging (i.e., reflecting age-related changes in cell type proportions). Our main analyses were conducted using complete cases. We also conducted sensitivity analyses following the imputation of missing covariate data. Data were imputed with multiple imputations by chained equations using the MICE R package [71] and 5 iterations. Following imputation, we reanalyzed associations of OCM-related micronutrients and compounds with aging biomarkers using fully adjusted weighted generalized linear regression models and pooled estimates from each imputed data set with the pool function.

We used 95% confidence intervals (95% CIs) to evaluate precision of associations and P < 0.05 to test for statistical significance. All analyses were conducted in R version 4.4.1 [72]. We used the STROBE-nut checklist when writing our report [73].

Results

Epigenetic aging biomarker and nutritional biomarker data were available for 2346 participants. Participant demographic characteristics and concentrations of OCM-related biomarkers, prior to survey weighting, are shown in Table 1. Participants had a mean (SD) chronological age of 65.1 (9.3) y, and approximately half (51.2%) of the participants were male. Nearly all participants were folate replete; 5 participants had serum folate concentrations less than or equal to the reference value of 3 ng/mL [74]. Twenty-one (0.9%) participants were classified as B12 deficient (concentration < 150 pg/mL), and 329 (14.0%) participants were classified as B12 insufficient (concentration < 300 pg/mL) [8]. Hyperhomocysteinemia was more common, with 192 (8.2%) participants having Hcy concentrations > 15 μmol/L [70].

TABLE 1.

Participant characteristics (N = 2346).

	n (%) or mean (SD)
Male, n (%)	1202 (51.2%)
Age at screening, mean (SD)	65.1 (9.3)
Race and ethnicity, n (%)
Mexican American	681 (29.0%)
Other Hispanic	151 (6.4%)
Non-Hispanic White	922 (39.3%)
Non-Hispanic Black	511 (21.8%)
Other race and multiracial	81 (3.5%)
BMI (kg/m2), mean (SD)	28.8 (5.8)
Missing, n (%)	84 (3.6%)
Smoking status, n (%)
Current	374 (15.9%)
Former	903 (38.5%)
Never	1,063 (45.3%)
Missing	6 (0.3%)
Alcohol intake, drinks/d, mean (SD)	0.4 (1.0)
Missing, n (%)	111 (5.0%)
Chronic kidney disease, n (%)1	318 (13.6%)
Missing, n (%)	4 (0.2%)
Serum folate (ng/mL), mean (SD)	16.5 (12.2)
Missing, n (%)	1 (<0.01%)
Folate deficiency2	5 (<0.01%)
Red blood cell folate (ng/mL RBC), mean (SD)	328 (159)
Missing, n (%)	5 (0.2%)
Vitamin B12 (pg/mL), mean (SD)	734 (4,840)
Missing, n (%)	1 (0.0%)
Vitamin B12 insufficiency, n (%)3	329 (14.0%)
Vitamin B12 deficiency, n (%)4	21 (0.9%)
Homocysteine (μmol/L), mean (SD)	10.1 (5.87)
Missing, n (%)	3 (0.1%)
Hyperhomocysteinemia5	192 (8.2%)
Methylmalonic acid (μmol/L), mean (SD)	0.2 (0.8)
Missing, n (%)	7 (0.3%)

¹Estimated glomerular filtration rate (eGFR) < 60 mL/min/1.73 m².

²Serum folate ≤ 3 ng/mL.

³B12 < 300 pg/mL.

⁴B12 < 150 pg/mL.

⁵Homocysteine > 15 μmol/L.

Participant characteristics stratified by smoking status and CKD are shown in Supplemental Tables 1 and 2, respectively. Compared with never smokers, former smokers had lower median B12 concentration and greater median Hcy concentration; current smokers had lower median concentrations of serum folate and RBC folate and higher median concentration of Hcy. Compared with participants without CKD, participants with CKD had not only greater median RBC folate concentration but also greater median Hcy and MMA concentrations.

All first- and second-generation clocks were strongly correlated with chronological age, with Pearson correlation coefficients ranging from r = 0.76 for PhenoAge to r = 0.87 for Horvath2 (P < 0.001) (Supplemental Figure 2). The Horvath2 clock had the lowest MAE compared with chronological age (2.71 y), whereas PhenoAge had the largest MAE (10.33 y). As expected, DunedinPoAm was very weakly correlated with chronological age (r = 0.04; P = 0.049) and DNAmTL was negatively correlated with chronological age (r = −0.58; P < 0.001).

After survey weighting, there were moderate correlations of serum folate with RBC folate and Hcy with MMA (r = 0.51 and 0.50, respectively; P < 0.001) (Supplemental Figure 3). Serum folate was significantly but weakly correlated with B12 (r = 0.04; P = 0.034), Hcy (r = −0.07; P = 0.002), and MMA (r = 0.05; P = 0.014). There were also weak correlations of RBC folate with Hcy (r = −0.09; P < 0.001) and B12 with Hcy (r = −0.05; P < 0.001) and MMA (r = −0.02; P < 0.001).

Associations of folate and B12 with epigenetic aging biomarkers

Figure 2 and Supplemental Tables 3 and 4 show results of our analyses of promoters of OCM. Our analyses used weighted generalized linear regression models adjusted for chronological age, chronological age², sex, race and ethnicity, BMI, education level, occupation, and PIR. We additionally adjusted for self-reported smoking status, alcohol intake, and eGFR, an indicator of renal function. Associations are reported in EAD units per doubling of concentration.

FIGURE 2.

Associations of promoters of one-carbon metabolism (OCM) with epigenetic aging biomarkers. Effect estimates (95% confidence intervals [CIs]) and P values are shown for a doubling in concentration of each compound. Results are from weighted generalized linear regression models adjusted for age, age², sex, race and ethnicity, BMI, education level, occupation, and poverty-to-income ratio. Effect estimates for Horvath1, Horvath2, Hannum, PhenoAge, and GrimAge2 EAD are in years; effect estimates for DunedinPoAm are in standard deviations; and effect estimates for DNAmTL EAD are in kilobases. EAD, epigenetic age deviation; eGFR, estimated glomerular filtration rate; RBC, red blood cell.

Each doubling of serum folate concentration was associated with −0.82 y (95% CI: −1.40, −0.23; P = 0.010) lower GrimAge EAD, −0.13 SDs (95% CI: −0.22, −0.03; P = 0.015) lower DunedinPoAm, and 0.02 kb (95% CI: 0.00, 0.04; P = 0.037) greater DNAmTL EAD. After adjusting for smoking, alcohol, and eGFR, associations were in the same direction but no longer significant (P > 0.05). Fully adjusted analyses of serum folate tertiles were also null (Supplemental Table 4). In analyses stratified by smoking status, serum folate was negatively associated with GrimAge2 EAD (B [95% CI]: −1.42 y [−2.51, −0.33]; P = 0.015; p_int = 0.13) and DunedinPoAm (B [95% CI]: −0.24 SDs [−0.46, −0.02]; P = 0.035; p_int = 0.44) among current smokers but not among never or former smokers (Supplemental Table 5). In analyses stratified by CKD, serum folate was negatively associated with GrimAge2 EAD among participants without CKD (B = −0.61 y; 95% CI: −1.03, −0.18; P = 0.010) but not among participants with CKD (p_int = 0.15) (Supplemental Table 6).

RBC folate concentration was positively, but not significantly, associated with PhenoAge EAD (B = 0.88 y; 95% CI: −0.10, 1.85; P = 0.07) (Figure 2 and Supplemental Table 3). The association was similar after adjustment for smoking, alcohol, and eGFR. Among former smokers, RBC folate was significantly associated with GrimAge2 EAD (B = 1.14 y; 95% CI: 0.37, 1.91; P = 0.008; p_int = 0.13); associations were not significant among never or current smokers (Supplemental Table 5).

Vitamin B12 was not significantly associated with any epigenetic aging biomarker in the study population overall (P > 0.05) (Figure 2 and Supplemental Tables 3 and 4). However, there was significant interaction between B12 and CKD status (p_int = 0.040) (Supplemental Table 6). Among participants with CKD, B12 concentration was associated with greater Horvath1 EAD (B = 1.56 y; 95% CI: 0.21, 2.91; P = 0.029); among participants without CKD, the association was null.

Associations of Hcy and MMA with epigenetic aging biomarkers

Homocysteine

Associations of markers of OCM deficiencies with epigenetic aging biomarkers are shown in Figure 3 and Supplemental Tables 3 and 4. Hcy concentration was positively associated with PhenoAge EAD (B [95% CI]: 1.05 y [0.06, 2.04]; P = 0.039), GrimAge2 EAD (B [95% CI]: 1.93 y [1.16, 2.71]; P < 0.001), and DunedinPoAm (B [95% CI]: 0.26 SDs [0.10, 0.41]; P = 0.003). After adjusting for smoking, alcohol, and eGFR, associations of Hcy with GrimAge2 EAD remained significant but attenuated (B [95% CI]: 1.23 y [0.48, 1.97]; P = 0.005), and hyperhomocysteinemia compared with normal Hcy range was associated with greater PhenoAge EAD (B [95% CI]: 1.55 y [0.01, 3.00]; P = 0.040) and GrimAge2 EAD (B [95% CI]: 1.78 y [0.47, 3.09]; P = 0.010) (Supplemental Table 4). In stratified analyses, Hcy concentration was positively associated with GrimAge2 EAD among former smokers (B [95% CI]: 1.76 y [0.60, 2.92]; P = 0.007; p_int = 0.028) and current smokers (B [95% CI]: 2.07 y [0.70, 3.45]; P = 0.007; p_int = 0.23) but not among never smokers (P > 0.05) (Figure 4 and Supplemental Table 5).

FIGURE 3.

Associations of markers of one-carbon metabolism (OCM) deficiencies with epigenetic aging biomarkers. Effect estimates (95% confidence intervals [CIs]) and P values are shown for a doubling in concentration of each OCM-related compound. Results are from weighted generalized linear regression models adjusted for age, age², sex, race and ethnicity, BMI, education level, occupation, and poverty-to-income ratio. Effect estimates for Horvath1, Horvath2, Hannum, PhenoAge, and GrimAge2 EAD are in years; effect estimates for DunedinPoAm are in standard deviations; and effect estimates for DNAmTL EAD are in kilobases. EAD, epigenetic age deviation; eGFR, estimated glomerular filtration rate; MMA, methylmalonic acid.

FIGURE 4.

Scatter plots and associations of homocysteine concentration with GrimAge2 epigenetic age deviation (EAD). The transparency of points corresponds to their survey weights. Estimates are from weighted generalized linear regression models and shown for average age, BMI, alcohol intake, estimated glomerular filtration rate, and poverty-to-income ratio, and for the reference level of race and ethnicity, education level, and occupation. Interaction P value: never smokers compared with former smokers = 0.028; never smokers compared with current smokers = 0.23. ∗ P < 0.05; ∗∗ P < 0.01.

MMA concentration was not associated with the epigenetic aging biomarkers in the study population overall (Supplemental Table 3). However, among participants with CKD, MMA was positively associated with GrimAge2 EAD (B [95% CI]: 1.10 y [0.09, 2.11]; P = 0.037; p_int = 0.038) (Supplemental Table 6).

Associations with GrimAge2 components

GrimAge2 is calculated as a weighted linear combination of DNA methylation–based surrogates for 9 plasma proteins and smoking pack-years. To evaluate if individual components were driving the associations of Hcy with GrimAge2 EAD, we evaluated associations with each of the components separately. Hcy concentration was positively associated with 4 GrimAge2 components: B2M (B [95% CI]: 0.10 SD [0.02, 0.18]; P = 0.002), tissue inhibitor metalloproteinase-1 (TIMP-1) (B [95% CI]: 0.06 SD [0.01, 0.11]; P = 0.024), adrenomedullin (ADM) (B [95% CI]: 0.17 SD [0.05, 0.29]; P = 0.009), and log(hemoglobin A1C) (B [95% CI]: 0.17 SD [0.03, 0.30]; P = 0.019). Laboratory measures for serum B2M, cystatin C, CRP, and hemoglobin A1C were also available in these NHANES cycles. All GrimAge2 components were moderately correlated with their measured counterparts (without survey weighting: B2M: r = 0.25; cystatin C: r = 0.22; CRP: r = 0.28; and hemoglobin A1C: r = 0.50; P < 0.001). We tested for associations of Hcy with directly measured serum proteins to determine if results may reflect in changes in clinical biomarkers. Overall, results were consistent with those of the GrimAge2 components (Table 2). However, Hcy was also positively associated with cystatin C.

TABLE 2.

Associations of homocysteine with GrimAge2 components and measured proteins.

	GrimAge2 component		Direct measurement
	B	95% CI	B	95% CI
GDF151	0.06	(−0.08, 0.19)	—
B2M	0.10	(0.02, 0.18)	0.132	(0.07, 0.18)
Cystatin C	0.06	(−0.03, 0.14)	0.083	(0.04, 0.12)
TIMP1	0.06	(0.01, 0.11)	—
ADM	0.17	(0.05, 0.29)	—
PAI1	0.07	(−0.13, 0.27)	—
Leptin	0.00	(−0.07, 0.08)	—
log(CRP)	0.10	(−0.06, 0.26)	−0.124	(−0.33, 0.08)
log(Hemoglobin A1C)	0.17	(0.03, 0.30)	−0.035	(−0.05, −0.01)
Smoking pack-years	0.10	(−0.09, 0.29)	—

Effect estimates (95% confidence intervals [CIs]) and P values are shown for a doubling in homocysteine concentration. GrimAge2 components are expressed on a Z-score scale. Results are from weighted generalized linear regression models adjusted for age, age², sex, race and ethnicity, BMI, smoking status, alcohol intake, eGFR, education level, occupation, and poverty-to-income ratio.

ADM, adrenomedullin; B2M, β-2 microglobulin; CRP, C-reactive protein; eGFR, estimated glomerular filtration rate; GDF15, growth/differentiation factor 15; TIMP1, tissue inhibitor of metalloproteinases.

¹N = 1826 for GrimAge2 components.

²N = 1808; units: log(serum B2M in mg/L).

³N = 1810; units: log(serum cystatin C in mg/L).

⁴N = 1826; units: log(CRP in mg/dL).

⁵N = 1826.

Sensitivity analyses

In sensitivity analyses adjusting for estimated cell type proportions, results were similar to those of our main analyses but attenuated (Supplemental Table 7). The association of Hcy concentration with GrimAge2 EAD remained significant (B [95% CI]: 1.17 y [0.28, 2.05]; P = 0.025). Analyses using imputed covariate data were also largely consistent with our main analyses (Supplemental Table 8). A doubling in Hcy concentration was associated with 1.37 y greater GrimAge2 EAD (95% CI: 0.69, 2.06; P = 0.002) and 0.17 SD greater DunedinPoAm (95% CI: 0.01, 0.33; P = 0.036).

Discussion

We tested associations of OCM-related biomarkers with epigenetic aging in NHANES, a nationally representative sample of adults in the United States. The strongest and most robust associations were with EAD measured by GrimAge2. We provide evidence that higher concentration of Hcy, an indicator of OCM deficiencies, is associated with greater PhenoAge and GrimAge2 EAD and DunedinPoAm; the association with GrimAge2 EAD persisted after adjustment for smoking status, alcohol intake, and renal function. We also found that serum folate concentration was associated with lower GrimAge2 EAD and DunedinPoAm and greater DNAmTL EAD; associations with GrimAge2 EAD remained significant among participants without CKD after adjusting for smoking and alcohol. Furthermore, we found that smoking and CKD significantly modified some associations. It is known that smoking, alcohol intake, and kidney disease are associated with increased EAD [[75], [76], [77], [78]], and changes in OCM-related biomarkers may be one mechanism through which this occurs.

Although our study did not investigate clinical end points, our findings support research linking elevated Hcy to adverse age-related health outcomes. Particularly relevant to the current study, in NHANES 1999–2006, Hcy has been associated with increased all-cause mortality and CVD mortality [79]. GrimAge2 is an epigenetic clock developed to estimate mortality risk using a weighted linear combination of sex, age, and DNA methylation surrogates for smoking pack-years and 9 plasma proteins [21]. In addition to the overall positive association of Hcy with GrimAge2 EAD, we found that Hcy was associated with greater DNA methylation surrogates of B2M (a biomarker of CVD, renal function, and inflammation [80]), TIMP-1 (a regulator of cell growth [81] and inflammation [82]), ADM (a vasodilator and prognostic marker in CVD [83]), and hemoglobin A1C. In stratified analyses, we found evidence that Hcy has a stronger association with GrimAge2 EAD among former and current smokers compared with never smokers. We also found significant negative associations of serum folate with GrimAge2 EAD and DunedinPoAm. After adjusting for smoking, alcohol, and eGFR, associations remained negative but did not reach statistical significance. Stratified analyses suggested that folate may still be protective against increased EAD among smokers and individuals with normal kidney function, although we did not find significant effect modification.

In contrast to our hypothesis that higher folate would be protective against epigenetic aging, we observed a trend toward a positive association between RBC folate and EAD measured by second-generation clocks, and a significant positive association of RBC folate with GrimAge2 EAD among former smokers. We hypothesize that RBC folate concentration may capture pathophysiological changes beyond serum folate measurements. RBC abnormalities, including megaloblastic anemia, macrocytosis, and increased mean cell volume and red cell distribution width, are associated with folate and B12 deficiencies [84]. However, cell distribution width has also been positively associated with inflammatory markers and oxidative stress [85,86], and RBCs may be involved in inflammatory processes due to their cytokine-binding capacity [87]. The second-generation clocks are trained on biomarkers of inflammation, such as white blood cell count, mean cell volume, and RBC distribution width, and therefore, these clocks may be sensitive to inflammation-related variation in RBCs.

Unexpectedly, we also found that B12 was positively associated with Horvath1 EAD among participants with CKD. Compromised kidney function and CKD can contribute to B12 deficiency due to metabolic changes, comorbidities, malnutrition, and increased urinary excretion of B12 [88]. However, associations of increased mortality risk with greater B12 concentrations among patients undergoing hemodialysis [89], patients with cancer [90], and the elderly [91] have also been reported. Particularly among individuals with inflammation-related conditions, greater B12 concentrations may indicate functional B12 deficiency due to decreased uptake of B12 by peripheral tissues as a compensatory mechanism to direct B12 away from infections or inflammatory sources in peripheral tissues [89].

Associations of B vitamins and Hcy with EAD have previously been investigated in the Veterans Affairs Normative Aging Study, a cohort of older community-dwelling men (N = 715) [92]. Using the mixture approach, this study identified B6 as important for predicting PhenoAge EAD with a negative direction of association and folate as important for predicting GrimAge EAD and PhenoAge EAD with positive directions of associations. Our findings were not consistent with these results, possibly due to differences in populations and analytical differences. In particular, the mixture approach may not fully capture the complex relationship between nutritional factors and OCM metabolites, including an antagonistic relationship between 1-carbon donors and Hcy. Associations of OCM-related compounds and epigenetic aging have also been analyzed in a supplementation study among older adults with mild cognitive impairment (N = 217). This study reported a positive correlation between baseline plasma Hcy and the rate of epigenetic aging [93].

Both our analyses and these studies focused on OCM-related biomarkers, which may be influenced by factors beyond dietary intake, including behavior, use of pharmaceuticals that are OCM antagonists, genetic factors related to metabolism of OCM micronutrients, and underlying health conditions. Therefore, randomized controlled trials are important for understanding how B-vitamin intake influences epigenetic aging and downstream health. A supplementation study among older adults found evidence that supplementation with a B-vitamin complex (folic acid, B6, and B12) may decrease epigenetic age among individuals with hyperhomocysteinemia [93]. However, evidence for the clinical benefits of supplementation with folic acid or other B vitamins is mixed. In a 7-y study of women with high-risk of CVD (N = 5442), supplementation with a B-vitamin complex did not reduce the risk of cardiovascular events compared with placebo, despite a reduction in Hcy in the treatment group [94]. Similarly, a 3-y study of adults with cerebral infarction (N = 3680) found no association of B-vitamin complex with risk of stroke [95]. Meta-analyses of folic acid trials concluded that supplementation may reduce the risk of stroke but not mortality [96], and that reductions in CVD were greatest among individuals with low baseline folate and high reductions in Hcy [97]. Inconsistent results have been found for studies of B-vitamin supplementation and cognitive function [98,99]; effects may also be modified by baseline nutritional status [100]. Future research on nutritional intake may measure both clinical end points and epigenetic aging biomarkers that represent intermediate changes in complex biological pathways [101].

Our study was strengthened by its large sample size and population representative of middle-aged and older adults in the United States. However, we were limited by cross-sectional data and the inability to evaluate the effects of changes in OCM status on long-term epigenetic aging and health. A portion of our sample had missing data on covariates, and therefore, our main analyses were restricted to complete cases; however, sensitivity analyses using imputed data were consistent. Covariate data on participant race and ethnicity were also limited to that provided by the NHANES. Adjusting for this variable may not fully control for confounding due to dietary differences between ethnic groups or the influence of ancestry on metabolism and epigenetic aging biomarkers. Additionally, we were limited by the nutritional biomarker data available for NHANES. A more comprehensive study including additional OCM-related micronutrients may provide further insights into the relationship between nutrition and epigenetic aging. Our study was also unable to analyze effect modification by genetic factors that affect OCM, such as variants in methylenetetrahydrofolate reductase (MTHFR), the enzyme that processes folate. Although survey weighting was used to ensure representativeness of our sample, survey weighting inflates the variance of model parameter estimates and decreases power [102]. We did not perform multiple testing correction, and therefore focused on precision of estimates along with prehypothesized associations, making such adjustments nonessential. Results should be interpreted in this context, and with a focus on precision of effect sizes rather than P values. Furthermore, our study was limited to a sample representative of the United States adult population, and therefore, findings may not be generalizable to populations with different dietary habits, nutritional status, lifestyle habits, and age.

In conclusion, as the application of epigenetic aging biomarkers becomes more common in research and clinical settings, it is important to understand how intervenable factors, such as nutrition, influence epigenetic aging. In this study of a nationally representative sample of middle-aged and older adults, we found that Hcy concentration is positively associated with epigenetic aging, particularly among former and current smokers. We also found evidence that serum folate is associated with lower EAD. Our results suggest that changes in OCM-related biomarkers may be one mechanism through which behavioral and health factors are related to EAD. Future studies should address the long-term consequences of OCM-related epigenetic aging and investigate associations in subpopulations, such as smokers and individuals with impaired renal function, which may have a greater benefit from nutritional interventions.

Author contributions

The authors’ responsibilities were as follows – AKB: designed and conducted research, analyzed data, performed statistical analysis, and wrote the article; DK, JCN-E, NG, HS, SD, MG, BLN, DHR contributed to the analyses; HS: calculated the epigenetic ages; DK, JCN-E, NG, HS, SD, MG, BLN, DHR, AC: reviewed and edited the manuscript; AC: supervised the work; AKB: has primary responsibility for final content; and all authors: read and approved the final manuscript.

Data availability

All data sets analyzed in the current study are publicly available from the NHANES website.

Funding

AKB is supported by the National Institutes of Health (NIH) National Institute of Environmental Health Sciences (NIEHS) grant K99ES035109. JNE and AC are supported by the NIEHS grant R01ES031259. This research was also supported by the NIH National Institute on Minority Health and Health Disparities (NIMHD) grants R01MD011721 (MPI: BLN and DHR) and R01MD016595.

Conflict of interest

The authors report no conflicts of interest.

Appendix A. Supplementary data

The following is the Supplementary data to this article: