Aging and epigenetics: Longitudinal changes in gene-specific DNA methylation

Abstract

DNA methylation has been associated with age-related disease. Intra-individual changes in gene-specific DNA methylation over time in a community-based cohort has not been well described. We estimated the change in DNA methylation due to aging for nine genes in an elderly, community-dwelling cohort of men. Seven hundred and eighty four men from the Veterans Administration Normative Aging Study who were living in metropolitan Boston from 1999–2009 donated a blood sample for DNA methylation analysis at clinical examinations repeated at approximately 3-5 year intervals. We used mixed effects regression models. Aging was significantly associated with decreased methylation of GCR, iNOS and TLR2 and with increased methylation of IFNγ, F3, CRAT and OGG. Obstructive pulmonary disease at baseline modified the effect of aging on methylation of IFNγ (interaction p = 0.04). For participants who had obstructive pulmonary disease at their baseline visit, the rate of change of methylation of IFNγ was -0.05% 5-methyl-cytosine (5-mC) per year (95% CI: -0.22, 0.13), but was 0.14% 5-mC per year (95% CI: 0.05, 0.24) for those without this condition. Models with random slopes indicated significant heterogeneity in the effect of aging on methylation of GCR, iNOS and OGG. These findings suggest that DNA methylation may reflect differential biological aging.

Keywords: :

• aging
• DNA Mmthylation
• epigenesis
• genetic

Epigenetics has been defined as the study of any potentially stable and, possibly, heritable change in gene expression or cellular phenotype that occurs without changes in the base-pairing of DNA.¹ DNA methylation is the best studied of the epigenetic processes. Hypomethylation of regulatory sequences tends to correlate with increased gene expression, while increased methylation usually results in transcriptional suppression.²

Epigenetic modifications in disease processes have been well studied. The largest body of literature is in the area of cancer research but such changes in the epigenome are increasingly being linked to other age-related diseases, such as Alzheimer disease^3,4 and atherosclerosis.^5-7 Epigenetic changes may also occur during normal aging. The relationship between aging and DNA methylation was originally proposed in a study of humpbacked salmon; since then, the association of decreased global DNA methylation and age has been supported in studies of mouse, rat, and human tissues.⁸ In a study of twins varying in age from 3 to 74 y, global and locus-specific epigenetic differences between the pairs were found to increase with age;⁹ however, the study did not follow the same individuals over time. Hypomethylation of repetitive elements (i.e., DNA sequences mostly located in intergenic regions) has also been associated with age in epidemiologic studies.^10,11

We are aware of only one study that has reported changes in gene-specific methylation over time within the same individual.¹² This study examined both global and gene-specific methylation. While the population-average global DNA methylation did not change over time, there were intra-individual changes, with about the same number of individuals showing a decrease as an increase.¹² Gene-specific methylation in this study was analyzed in a subset of 41 individuals in the cohort and results were only presented for a family of five.

Given the lack of studies reporting on changes in DNA methylation in community-based cohorts in the literature, we set out to describe the intra-individual change in gene-specific methylation in a cohort of over 700 men. Within the context of an ongoing study, we used a candidate gene approach to examine DNA methylation. We selected the following genes for analysis: glucocorticoid receptor (GCR), inducible nitric oxide synthase (iNOS), intercellular adhesion molecule (ICAM), interleukin‑6 (IL-6), toll-like receptor‑2 (TLR2), interferon-gamma (IFN-gamma), coagulation factor-3 (F3), carnitine O‑acetyltransferase (CRAT), and 8‑oxoguanine DNA glycosylase 1 (OGG). These genes were chosen because they are expressed with variable degree in leukocytes,¹³ the DNA source used in our study, and because of their relation to inflammation, endothelial function, and oxidative metabolism, processes known to be associated with aging and age-related diseases. We hypothesized that change in methylation over time would be dependent on the age of the participant at the baseline visit and wanted to distinguish between these two effects—cross-sectional age difference between subjects and change in age in each subject over time. We further hypothesized that the direction of change would vary by gene and by individual due to chronic conditions associated with age or other factors. We were able to test these hypotheses through secondary analyses of interaction and random effects models. Here, we describe the changes in methylation of these nine genes in a longitudinal cohort of elderly, community-dwelling men.

Results

The mean age of participants at baseline was 72 y (Table 1). Subjects had between one and five measurements of DNA methylation for each gene. Table 2 describes the characteristics of each gene at each visit. The correlations between mean DNA methylation of each gene measure at the first and second visit were also calculated (Table 2). Correlations ranged from high for IL6 (rho = 0.705, p < 0.0001) to moderate for ICAM (rho = 0.557, p < 0.0001) to low for GCR (rho = 0.116, p = 0.0118).

Table 1. Baseline Characteristics of Participants in the Normative Aging Study, Boston, MA, 1999 - 2009 [mean (SD) or n (%)]

	Baseline Visit
	(n = 784)
Age (years)	72.3 (6.8)
Body mass index (kg/m^2)	28.2 (4.1)
Systolic blood pressure (mmHg)	132.6 (16.9)
Diastolic blood pressure (mmHg)	77.8 (9.3)
Fasting blood glucose (mg/dL)	108.6 (28.0)
Treatment with statins	280 (35.7)
Smoking status
Never	222 (28.3)
Former	530 (67.6)
Current	32 (4.1)
Cumulative smoking, pack-years^a	30.7 (27.3)
Season of clinical exam
Spring	192 (24.5)
Summer	218 (27.8)
Fall	241 (30.7)
Winter	133 (17.0)
Day of week of clinical exam
Monday	69 (8.8)
Tuesday	179 (22.8)
Wednesday	342 (43.6)
Thursday	194 (24.7)

^a Among current or former smokers

Table 2. Descriptive Statistics of Gene-specific Methylation by Visit for Participants in the Normative Aging Study, Boston, MA, 1999 - 2009

			Correlation (visit 1 and 2)		Methylation, % of 5-methyl cytosine
Gene	Visit Number	n	rho	P	Minimum	Maximum	Mean	Standard Deviation
GCR	1	735	0.116	0.0118	14.7	72.8	46.8	6.0
	2	471			22.3	87.2	47.2	6.4
	3	230			28.1	59.9	45.6	5.5
	4	29			28.8	54.1	45.1	5.4
	5	1			45.4	45.4	45.4	NA
iNOS	1	661	0.372	< 0.0001	24.5	87.2	69.4	7.3
	2	356			44.0	86.5	67.7	7.2
	3	148			37.8	79.4	65.7	7.8
	4	11			55.8	67.8	62.6	3.9
ICAM	1	707	0.557	< 0.0001	1.7	16.1	4.4	1.8
	2	428			1.6	12.0	4.1	1.8
	3	180			2.2	8.5	4.4	1.1
	4	24			2.5	5.9	3.5	0.7
	5	1			6.0	6.0	6.0	NA
IL6	1	780	0.705	< 0.0001	10.3	86.6	43.7	10.7
	2	547			12.3	81.3	42.8	9.9
	3	303			10.9	72.8	43.0	10.9
	4	45			18.3	66.2	43.6	11.1
	5	1			44.0	44.4	44.3	NA
TLR2	1	722	0.259	< 0.0001	0.0	8.9	3.1	1.3
	2	426			0.0	7.5	3.0	1.3
	3	166			0.0	7.2	2.7	1.2
	4	26			0.9	5.1	2.0	1.1
IFNγ	1	784	0.467	< 0.0001	30.9	95.7	84.4	5.5
	2	541			51.6	94.4	85.1	5.0
	3	295			59.0	95.0	85.3	4.6
	4	42			77.1	94.6	86.6	3.6
	5	1			80.1	80.1	80.1	NA
F3	1	759	0.210	< 0.0001	0.0	14.8	2.3	1.3
	2	478			0.0	11.1	2.6	1.3
	3	191			0.0	6.9	2.6	1.2
	4	27			0.6	3.7	1.7	0.7
CRAT	1	770	0.487	< 0.0001	1.0	7.4	3.2	1.0
	2	519			1.2	7.0	3.1	1.2
	3	268			1.5	6.2	3.3	0.9
	4	37			1.6	5.9	3.1	0.9
OGG	1	633	0.171	0.0024	0.0	11.8	2.2	1.4
	2	314			0.0	7.7	2.2	1.3
	3	100			0.0	7.8	2.6	1.6
	4	6			1.4	4.8	3.2	1.1

In the adjusted analysis between age at visit and DNA methylation for the 9 genes (Model 1), increased age was associated with increased DNA methylation for TLR2 at the p < 0.05 level (Table 3). Increased age at visit was also associated with decreased DNA methylation of iNOS at p < 0.05. Models with a term for the cross-sectional effect of age and a term for the longitudinal effect of age (Model 2) were examined next. After adjustment, there was a statistically significant association at the p < 0.05 level between DNA methylation and the longitudinal effect of age for many of the genes examined. For GCR, iNOS, and TLR2, age was associated with decreased methylation. For IFNγ, F3, CRAT, and OGG, age was associated with increased methylation (Table 4). The cross-sectional effect of age was statistically significantly associated with decreased DNA methylation of iNOS and increased DNA methylation of TLR2 in adjusted models. After restricting the sensitivity analysis to participants who did not have a diagnosis of IHD at baseline, the associations persisted, except for TLR2, which lost statistical significance but remained in the same direction for both the cross-sectional and longitudinal effect of age.

Table 3. Association Between Gene-specific Methylation and Age at Visit in the Normative Aging Study, Boston, MA, 1999 - 2009

	Unadjusted		Adjusted ^b
Gene	Change in % 5mC ^a	95% CI	Change in % 5mC ^a	95% CI
GCR	-0.042	-0.090, 0.006	-0.044	-0.099, 0.011
iNOS	-0.164	-0.230, -0.097*	-0.154	-0.229, -0.079*
ICAM	-0.016	-0.031, -0.001*	-0.010	-0.028, 0.007
IL6	-0.033	-0.113, 0.047	0.000	-0.096, 0.095
TLR2	0.005	-0.006, 0.016	0.015	0.002, 0.027*
IFNγ	0.017	-0.024, 0.058	-0.019	-0.062, 0.024
F3	0.004	-0.006, 0.014	-0.001	-0.013, 0.010
CRAT	0.010	0.001, 0.018	0.008	-0.001, 0.018
OGG	0.006	-0.007, 0.019	0.004	-0.011, 0.019

% 5mC: % 5-methyl-cytosine * p < 0.05 ^a Change in % 5mC associated with a 1-y increase in age at visit ^b Adjusting for BMI, SBP, DBP, fasting blood glucose, % lymphocytes, % neutrophils, use of statin, season, day of week, smoking status, and pack-years

Table 4. Association Between Gene-specific Methylation and Cross-sectional Age and Change in Age Over Time in the Normative Aging Study, Boston, MA, 1999 - 2009

		Unadjusted		Adjusted ^b
				Full Cohort		No IHD at Baseline
Gene		Change in % 5mC ^a	95% CI	Change in % 5mC ^a	95% CI	Change in % 5mC ^a	95% CI
GCR	Cross-sectional age	-0.022	-0.073, 0.029	-0.031	-0.087, 0.025	-0.038	-0.104, 0.028
	Longitudinal age	-0.151	-0.256, -0.046*	-0.191	-0.321, -0.061*	-0.212	-0.365, -0.059*
iNOS	Cross-sectional age	-0.063	-0.133, 0.007	-0.097	-0.173, -0.022*	-0.120	-0.212, -0.028*
	Longitudinal age	-0.660	-0.795, -0.526*	-0.695	-0.860, -0.531*	-0.661	-0.860, -0.462*
ICAM	Cross-sectional age	-0.012	-0.029, 0.004	-0.010	-0.028, 0.008	-0.007	-0.028, 0.015
	Longitudinal age	-0.030	-0.059, -0.002*	-0.015	-0.051, 0.021	0.003	-0.041, 0.047
IL6	Cross-sectional age	0.001	-0.101, 0.103	0.025	-0.083, 0.133	-0.002	-0.125, 0.122
	Longitudinal age	-0.081	-0.198, 0.037	-0.064	-0.221, 0.094	-0.097	-0.277, 0.082
TLR2	Cross-sectional age	0.016	0.004, 0.028*	0.020	0.007, 0.032*	0.015	-0.001, 0.030
	Longitudinal age	-0.054	-0.078, -0.031*	-0.033	-0.062, -0.004*	-0.025	-0.060, 0.011
IFNγ	Cross-sectional age	-0.033	-0.080, 0.014	-0.040	-0.084, 0.005	-0.031	-0.081, 0.020
	Longitudinal age	0.154	0.081, 0.227*	0.110	0.021, 0.198*	0.130	0.032, 0.229*
F3	Cross-sectional age	-0.002	-0.013, 0.008	-0.005	-0.017, 0.006	-0.001	-0.015, 0.012
	Longitudinal age	0.040	0.017, 0.062*	0.041	0.014, 0.068*	0.055	0.023, 0.086*
CRAT	Cross-sectional age	0.006	-0.003, 0.015)	0.005	-0.005, 0.015	0.006	-0.006, 0.018
	Longitudinal age	0.023	0.006, 0.039*	0.036	0.015, 0.057*	0.033	0.009, 0.058*
OGG	Cross-sectional age	-0.006	-0.019, 0.007	-0.002	-0.016, 0.013	-0.004	-0.022, 0.015
	Longitudinal age	0.084	0.055, 0.113*	0.081	0.045, 0.116*	0.088	0.045, 0.130*

% 5mC: % 5-methyl-cytosine; IHD: Ischemic Heart Disease * p < 0.05 ^a Change in % 5mC associated with a 1-y increase in cross-sectional age or longitudinal age ^b Adjusting for BMI, SBP, DBP, fasting blood glucose, % lymphocytes, % neutrophils use of statin, season, day of week, smoking status, and pack-years

A statistically significant interaction between obstructive pulmonary disease at baseline and change in age was found for methylation of IFNγ (interaction p = 0.04). For participants who had obstructive pulmonary disease at their baseline visit, the rate of change of methylation of IFNγ was -0.05% 5‑mC per year (95% confidence interval: -0.22, 0.13) but was 0.14% 5‑mC per year (95% confidence interval: 0.05, 0.24) for those without this condition. There was no statistically significant interaction between either diabetes mellitus or obesity at baseline and change in age for methylation for any of the genes investigated.

Finally, models with random intercepts and slopes for the effect of age were fit. For three of the genes evaluated (GCR, iNOS, and OGG), there was significant heterogeneity in the effect of change in age, indicating differing rates of change among subjects. Figure 1 displays the distribution of the subject-specific slope estimates for those genes.

Figure 1. Plots of Subject-specific Slopes for Change in Age in the Normative Aging Study, Boston, MA, 1999 – 2009.

Discussion

We examined changes in methylation patterns of nine genes over time in a cohort of elderly men. After adjusting for covariates, we found that change in age was significantly associated with decreased methylation of GCR, iNOS and TLR2 and with increased methylation of IFNγ, F3, CRAT and OGG. By estimating random slopes, we also found that this change was modified by unmeasured characteristics that varied between individuals in our population.

The association between global measures of DNA methylation and age has been previously studied in both animals and humans. Global loss of DNA methylation during aging has been reported in tissue from rats, mice, and cows.¹⁴ Decreased genomic methylation has also been seen in human lymphocytes¹⁵ and peripheral blood cells.¹⁶ In a study examining methylation changes within the same individuals over time, Bjornsson et al.¹² found both losses and gains in global DNA methylation over time in different individuals, resulting in no change in the population average. Bollati et al. found that aging was associated with a decline in repetitive element methylation in the same cohort as we investigated here.¹⁰

A number of human tissue studies have reported associations between cross-sectional age and gene-specific DNA methylation. Hypermethylation of a number of specific genes has been associated with cross-sectional age including the estrogen receptor (ER) (human colonic mucosa),¹⁷ insulin-like growth factor II (IGF2) (colonic tissue),¹⁸ p14 (colonic tissue),¹⁹ p16 (gastric epithelia),²⁰ E-cadherin (urothelial bladder cells),²¹ c‑fos (liver tissue),²² and collagen α 1 (periodontal ligament tissue).²³ In addition, genes associated with colorectal cancer, including MYOD and N33, have been associated with age-related partial methylation in normal colon mucosa.²⁴ A study of over 200 human tissue samples from 10 anatomic sites also found a significant association between aging and DNA methylation.²⁵

The study by Bjornsson et al. is the only other study that we are aware of that has examined intra-individual gene-specific DNA methylation changes due to aging in a human cohort. Gene-specific methylation was analyzed in a subset of 41 individuals in their cohort, though they only reported the gene-specific methylation results for 50 genes that showed the greatest change over time in 5 members of a family. Though none of the genes we investigated were included in this group of 50, we saw both increases and decreases in methylation over time in our study, depending on the gene being analyzed.

When biological markers change over time in aging individuals, the markers will be associated with age at any given time. Therefore, a cross-sectional analysis will be influenced by selection of study participants because values for older participants will represent only those who survived up until that age.²⁶ A. Munoz and S.J. Gange, Methodological issues for biomarkers and intermediate outcomes in cohort studies, Epidemiol Rev. 1998; 20:29–42.²⁷ In this study, we were able to make use of repeated measurements of DNA methylation on a large sample of individuals. This allowed us to distinguish between the cross-sectional effect of age (i.e., baseline effect) and the longitudinal effect (i.e., effect of years since the first visit).

By examining the longitudinal effect, we found that aging was significantly associated with decreased methylation of iNOS, TLR2, and GCR. Methylation of iNOS has been shown to suppress iNOS expression.²⁸ While we cannot conclude anything about expression of iNOS in the present study, iNOS is associated with inflammatory processes and inflammation is common in aging and age-related disease. Similarly, age-associated defects in the innate immune system have been observed and human studies have found an association between aging and decreased TLR induced cytokine production. In a study of rhesus monkeys, glucocorticoid receptor genes were upregulated with aging.²⁹

We also found that aging was significantly associated with increased methylation of F3, CRAT, and OGG. Again, while we did not measure gene expression in our study participants, each of these genes has been associated with markers of aging and age-related diseases. Markers of activated coagulation, such as factor VIII,³⁰ as well as increased oxidative stress damage, have been shown to be increased with age. Carnitine acetyltransferase (CRAT) is a key enzyme in the metabolic pathway in mitochondria, peroxisomes and endoplasmic reticulum, and carnitine decline is a trait of an insulin resistant state associated with increasing age.³¹ Overall, while our study cannot draw conclusions on how gene expression may relate to aging, the findings point to continued DNA methylation changes throughout the life course and indicate that DNA methylation may reflect differential biological aging. How these changes relate to gene expression requires further investigation.

We found a significant interaction between having obstructive pulmonary disease at baseline and change in age when looking at methylation of IFN‑γ . We also found significant heterogeneity between subjects for the longitudinal effect of age on methylation of three different genes—GCR, IL‑6, and OGG. Interestingly, we saw no effect of aging on the methylation of IL‑6 in our main analysis because methylation of this gene increased over time in some participants, and decreased over time in others. Taken together, these findings demonstrate that changes in methylation over time do not only vary by gene, but also by individual.

This study has several limitations. Methylation analysis was performed on white blood cell (WBC) DNA and it is not clear to what extent the changes we observed in WBC DNA reflect similar modification of DNA methylation in target tissue. However, all the genes we selected are known to be expressed and functional in WBCs, and participate in critical cellular functions in WBCs such as inflammation and oxidative metabolism. Therefore, measurements in WBCs may be appropriate biological markers.

There is still much uncertainty regarding factors that may influence DNA methylation, and, therefore, the causal pathway between aging and DNA methylation is not clearly defined. Our multivariate models included a number of covariates that are possible predictors of DNA methylation. While the effect estimates did not change substantially between the unadjusted and adjusted results, it is also possible that these covariates are intermediates in the mechanistic pathway between age and DNA methylation. If so, our adjusted analysis may be too conservative,

Our cohort consists of elderly, white males. While we found an association between aging and DNA methylation, it is important to note that our youngest subject was 55 y old at baseline. Therefore, it is not clear how aging would be associated with DNA methylation in a younger cohort or one with a wider age range. It is also not clear if changes in DNA methylation vary by gender.

Epigenetic mechanisms are increasingly being recognized as playing an important role in chronic disease. For the first time, the intra-individual change in DNA methylation of 9 genes has been described in a large, community-based cohort. The longitudinal nature of the study enabled examination of the effect of aging over time, while adjusting for cross-sectional age. This study also found that the effect of aging could vary by gene and individual, either due to known clinical conditions or other unidentified factors. These results have implications for understanding the links between epigenetics, aging, and age-related diseases.

Methods

Study population

This study included 784 elderly men who, as of March 1999, were active participants in the Normative Aging Study (NAS). The NAS cohort was established by the Veterans’ Administration in 1961, which enrolled men 21 to 80 y of age from the greater Boston, MA area who were free of known chronic medical conditions.³² Since the time of enrollment, participants have had comprehensive clinical examinations at 3- to 5‑y intervals. Further details can be found elsewhere.³³ In examinations that took place between March 1999 and May 2009, 784 active participants agreed to donate at least one whole blood sample that was used for DNA methylation analysis. Because of assay failure and the limited amount of DNA available from each subject, DNA methylation analysis was successful on varying numbers of subjects for each gene (Table 2).

Exposure variable and covariates

NAS participants reported to the study center on the morning of their scheduled examinations. Age, height and weight were assessed at the visit. A physician elicited a complete medical history, including daily medications. Current smoking and past history of smoking were assessed using the American Thoracic Society questionnaire.³⁴

Methylation analysis

DNA was extracted from stored frozen buffy coat of 7 mL whole blood. DNA methylation analyses were performed on bisulfite-treated DNA using highly quantitative analysis based on PCR pyrosequencing; 500 ng DNA (concentration 25 ng/μL) was treated using the EZ DNA Methylation-Gold Kit (Zymo Research, Orange, CA, USA) according to the manufacturer’s protocol. Final elution was performed with 200 μL M-Elution Buffer. The assay for methylation was developed by locating the promoters using Genomatix Software (Genomatix Software Inc., Ann Arbor, MI) as described in Table S1. For each assay a 50 μL PCR was performed in 25 μL GoTaq Green Master mix (Promega), 25 ng bisulfite-treated genomic DNA, and water. PCR cycling conditions and reagent amounts are described in Table S2. PCR products were purified and sequenced by pyrosequencing as previously described³⁵ using 0.3 μM sequencing primer. Primers and sequences analyzed for each assay are shown in Table S3. For each assay, primer sets and conditions were evaluated using a two-step approach. In the first step, correct PCR amplification was verified on standard DNA by running the PCR products on an agarose gel. The standard DNA was comprised of a pool of human DNA from healthy volunteers. In the second step, five different samples of normal DNA from healthy volunteers were run starting from bisulfite treatment through PCR amplification and Pyrosequencing to test the entire analytic pipeline. The mean CV was 0.02 for GCR, 0.02 for iNOS, 0.06 for ICAM, 0.02 for IL-6, 0.04 for TLR2, 0.002 for IFNγ, 0.08 for F3, 0.05 for CRAT, and 0.15 for OGG. Primer sets/conditions that did not produce successful pyrosequencing analyses, as indicated by the built-in Pyrosequencing software (Pyro Q‑CpG Software), were modified and retested. In this second step, all samples were run in duplicates. Primer sets and conditions with low reproducibility were modified and retested. DNA methylation analysis measured multiple individual CpG dinucleotide positions for each marker and was repeated twice on each sample to minimize the assay variability. The mean value of all dinucleotide positions was used in all statistical analyses.

Statistical analysis

Due to the repeated measures of DNA methylation for many participants, the association between age at visit and DNA methylation was evaluated with a mixed effects model (Model 1) for each gene, to account for the correlation among measurements within the same subject. The following was assumed:

1)

Y_it = β₀ + u_i + β₁ age + β₂X₂+ … + β_pX_p (Model

where Y_it was the level of gene-specific methylation (measured as % 5‑methyl-cytosine) in subject i at time t, β₀ was the overall intercept, and u_i was the separate random intercept for subject i. In the above model X₂–X_p are the covariates, which were included only in multivariable models.

In multivariable models, the following potential confounders or predictors of gene-specific methylation were chosen a priori based on previous literature^10,36and included in the analysis: body mass index (BMI), systolic blood pressure (SBP), diastolic blood pressure (DBP), fasting blood glucose (FBG), use of a statin medication, smoking status (never/former/current) and pack-years smoked, and season and day of week of blood draw. Percent Lymphocytes and % neutrophils were also included to control for potential age-related variation among leukocyte populations in gene-specific methylation.

The coefficient of age in model 1 reflects both the longitudinal change in each subject over time, as well as the cross-sectional difference between subjects, based on their age. If DNA methylation is associated with survival, the latter is likely to underestimate the true rate of change with time. To further explore these two different aspects, a second model for each gene (Model 2) was fit that included both: (i) age at the first visit (baseline age) and; (ii) years elapsed since the first visit (change in age). The following was assumed:

2)

Y_it = β₀ + u_i + β₁ baseline age + β₂ change in age + β₃X₃+ … + β_pX_p (Model

where β₁ was the coefficient to estimate the cross-sectional effect of age and β₂ was the coefficient to estimate the longitudinal effect of age. The correlation between baseline age and change in age was rho = -0.17. The same covariates were included in Model 2 as were included in Model 1.

To determine if clinical characteristics that are prevalent in an aging cohort modified the association between aging and DNA methylation, models with interaction terms for such conditions at baseline were fit, as a secondary analysis. The following three clinical conditions were examined as potential effect modifiers: obesity (defined as BMI ≥ 30), obstructive pulmonary disease (defined as FEV₁/FVC < 0.7), and diabetes mellitus (defined as physician-diagnosed diabetes mellitus or fasting blood glucose > 126 mg/dL). The regression models included cross-product terms for the interaction between those modifying factors and change in age, along with the main effects. If there was a significant interaction (at p < 0.05), effect estimates from each stratum were then calculated.

To determine if there was variation in the effect of aging between participants not captured by the specific clinical characteristics described, models with random intercepts and random slopes for the change in age were fit. These models allowed both the intercept and slope to vary randomly among subjects thereby allowing for heterogeneity of effect between subjects. A Wald test for the covariance of the random slopes was performed to determine if there was significant heterogeneity in the effect of age between subjects.

Finally, to determine if health status at baseline influenced the results, a sensitivity analysis was performed. Since DNA methylation has been associated with development of ischemic heart disease (IHD) in this cohort,³⁷ the main analysis examining the cross-sectional and longitudinal effects of age (Model 2) was repeated only on subjects who did not have diagnosed IHD at baseline (n = 551). IHD was defined as diagnosis with International Classification of Diseases, 9th Revision (ICD9) codes 410 – 414. All mixed effect models were conducted using the PROC MIXED procedure in SAS version 9.2 (SAS Institute, Cary, NC, USA). p values were two-sided.

Abbreviations:
5mC	=	5-methylcytosine
CRAT	=	carnitine O-acetyltransferase
DNA	=	deoxyribonucleic acid
F3	=	coagulation factor-3 gene
GCR	=	glucocorticoid receptor gene
iNOS	=	inducible nitric oxide synthase gene
ICAM	=	intercellular adhesion molecule gene
IFN-gamma	=	interferon-gamma
IL-6	=	interleukin-6 gene
NAS	=	Normative Aging Study
TLR2	=	toll-like receptor-2 gene
OGG	=	8-oxoguanine DNA glycosylase 1 gene
PCR	=	polymerase chain reaction
VA	=	Veterans Administration
WBC	=	white blood cell

Acknowledgments

This work was supported by the National Institue of Environmental Health Sciences at the National Institutes of Health (T32ES07069, ES0002, ES015172–01, ES014663, and T32ES016645). The VA NAS is supported by the Cooperative Studies Program/Epidemiology Research and Information Center of the US. Department of Veterans Affairs and is a component of the Massachusetts Veterans Epidemiology Research and Information Center, Boston, Massachusetts.