Accelerated DNA methylation age and the use of antihypertensive medication among older adults

The discrepancy of DNA methylation age (DNAmAge) with chronological age (termed as age acceleration, AA) has been identified to be associated with many aging-related health outcomes including hypertension. Since taking antihypertensive medication (AHM) could prevent aging-related diseases caused by hypertension, we hypothesized that using AHM could also reduce the AA. We examined this hypothesis among 546 males aged 55–85 years by exploring the associations of AHM use with AA and its change rate (ΔAA) in two visits with a median follow-up of 3.86 years. Horvath DNAmAge was derived from DNA methylation profiles measured by Illumina HumanMethylation450 BeadChip and information on AHM use was collected by physician interview. A general decreasing pattern of AA was observed between the two visits. After the fully adjusting for potential covariates including hypertension, any AHM use showed a cross-sectional significant association with higher AA at each visit, as well as a longitudinal association with increased ΔAA between visits. Particularly, relative to participants who never took any AHM, individuals with continuous AHM use had a higher ΔAA of 0.6 year/chronological year. This finding underlines that DNAmAge and AA may not be able to capture the preventive effects of AHMs that reduce cardiovascular risks and mortality.


INTRODUCTION
DNA methylation, a major form of epigenetic modification, is known to play an important role in aging and the development of age-related health outcomes [1,2]. Recently, a DNA methylation-based biological age predictor, "DNA methylation age (DNAmAge)", has been established and found to be The use of antihypertension medication (AHM) reduces the risk of adverse age-related health outcomes caused by hypertension. Specifically, observational studies, clinical trials, and systematic reviews mostly suggested that effective antihypertensive therapy greatly reduces the risk of CVD in patients with hypertension [18,19], and may also be associated with a decreased risk of cognitive decline and incident dementia [20]. As DNA methylation is a durable and reversible modification, we hypothesized that the use of AHMs might also be able to influence the biological aging reflected by the epigenetic AA. Therefore, we assessed the associations of AHM use with AA and further determined whether the change of AHM use could modify the change rate of AA (ΔAA) during a median follow-up of 3.86 years. This investigation was carried out in the Normative Aging Study (NAS), which is an all-male longitudinal study of a cohort of older veterans living in the Greater Boston area.

Participant characteristics
Characteristics of the 546 participants at each visit are shown in Table 1. Overall, the average age at first and second visits was 72 and 75 years, respectively. More than 60% of the participants were former smokers and less than 5% were current smokers, and the majority of participants were overweight or obese, consumed no or low amounts of alcohol, reported low physical activity and had less than 16 years of education. During a median follow-up of 3.86 years, the prevalence of hypertension increased from about 68% to 75%, stroke from nearly 6.0% to 8.6%, coronary heart disease (CHD) from 26% to 34%, diabetes from 12% to 16% and cancer from 50% to 59%. In particular, 313/374 and 383/411 participants with hypertension took the medication for elevated blood pressure at each visit, respectively. ACE inhibitors and beta blockers were the two most widely used medications. According to our definition of hypertension, the participants using AHM were considered to be have hypertension. Among the 546 participants, after the first visit, 13 stopped using any AHM and 83 started to use.

AGING
Our main analyses based on the DNAmAge estimated by Horvath's algorithm. Figure 1 shows that the estimates were highly correlated with chronological age at each visit (Spearman coefficients >0.6). While DNAmAge increased between the first and second visit, the overall AA of the second visit was lower than that of the first visit (Table 1), showing declining trajectories with a crude average ΔAA of about -0.03 year/chronological year. The trajectories of AA and distributions of ΔAA were illustrated in Figure 2. Figure  S1 showed strong correlations between the AAs at both visits among those with or without hypertension at both visits (Spearman coefficients ~0.7), while for those who had hypertension or hypertension was controlled after the first visit, their correlation was slightly attenuated (Spearman coefficients = 0.53).
Associations of hypertension with age acceleration and the change rate of age acceleration We evaluated the relationship between hypertension and AA at each visit. Participants with hypertension had higher AA (years) than those without (first visit: 0.18 vs. 0.09; second visit: -0.01 vs. -0.18; p-value=0.016). The similar pattern was also observed for ΔAA that participants with hypertension had a higher ΔAA (year/chronological year) compared to those without (first visit: 0.03 vs. -0.14; second visit: 0.01 vs. -0.12; p-value=0.007). We further adjusted for other potential covariates at each visit in a mixed linear model to validate the observed patterns, including age, body max index (BMI), smoking status, alcohol consumption, years of education, physical activity, leukocyte distribution and the batch of microarray experiments in DNA methylation measurement. As shown in Table S1, the correlations of hypertension with AA and ΔAA remained positive, albeit not statistically significant. We also tested the associations of diabetes, metabolic syndrome and commonly measured clinical biomarkers with AA and ΔAA (Tables S1 & S2). Despite the statistically significant association between diabetes and ΔAA, neither metabolic syndrome nor any biomarkers were strongly associated with AA or ΔAA.
Cross-sectional associations of antihypertensive medication with age acceleration and the change rate of age acceleration We first investigated the cross-sectional associations of AHM use with AA and ΔAA at each visit. Overall, taking any AHM was significantly associated with higher AA and ΔAA ( Table 2). After adjusting for potential covariates at each visit [age, BMI, smoking status, alcohol consumption, years of education, physical activity, leukocyte distribution, total cholesterol, high-density lipoprotein (HDL), triglycerides, fasting glucose, systolic blood pressure (SBP), hypertension, stroke, CHD, diabetes, cancer and the batch of microarray experiments in DNA methylation measurement], people who took any AHM at each visit were about 2.3 years older than the controls in terms of AA, and their AA increased significantly during follow-up at a rate of more than 0.4 year/chronological year compared to the people who did not take any AHMs. However, the effects of specific classes of medications varied. While taking diuretics was negatively associated with AA and ΔAA, calcium channel blockers and alpha and beta blockers were positively correlated with AA and ΔAA. Angiotensinconverting enzyme inhibitors (ACE inhibitors) and angiotensin receptor antagonists (ARBs) both had negative associations with ΔAA, but their relationships with AA were in different directions. An additional subgroup analysis was carried out among people with hypertension at each visit (Table 3), and the use of any AHM remained robustly associated with increased AA and ΔAA. Table S3a showed sensitivity analyses for the crosssectional associations of any AHM use with the AA and ΔAA of Hannum DNAmAge and DNA methylation   Table 3. Cross-sectional associations of antihypertensive medication use with age acceleration and change of age acceleration at first and second visits among the subpopulation who had hypertension. AGING Phenotypic age (DNAmPhenoAge). Taking any AHMs showed positive correlations with the AA of each biomarker and significantly associated with both forms of AA at the first visit. People who took medications were about 2.2 years/1.7 years older in people compared to the controls, respectively. Even though taking any AHMs was not robustly associated with the ΔAA of neither biomarker, the effects were in the same direction as we identified for the Horvath DNAmAge.

Longitudinal associations of antihypertensive medication with age acceleration and the change rate of age acceleration
We further examined the longitudinal association between AHM use and ΔAA (  Table 5). After fully controlling for potential covariates, stopped taking AHM was correlated with a declining ΔAA compared to the continued AHM use after the first visit albeit not statistically significant.
Same longitudinal tests were also performed for the ΔAA of Hannum DNAmAge and DNAmPhenoAge as sensitivity analyses in Table S3b. The ΔAA of neither biomarker showed significant associations with the change of AHM use, but an increasing pattern was still observed for the people with continuous AHMs use in comparison to the people who never used the AHM for each biomarker.

DISCUSSION
In the present study, we investigated the cross-sectional and longitudinal associations of AHM use with the AA of Horvath DNAmAge in a longitudinal study of older male participants examined over two visits. Even though a general decrease of AA was observed between the two visits, after fully adjusting for hypertension and other potential covariates, any AHM use showed a cross-sectional significant association with higher AA at each visit, as well as a longitudinal association with increased ΔAA between visits. Particularly, relative to participants who never took any AHM, individuals with continuous AHM use had a higher ΔAA of 0.6 year/chronological year. Additional sensitivity analyses on another two DNA methylation-based biomarkers (Hannum DNAmAge & DNAmPhenoAge) showed the similar patterns with the use of AHM as the Horvath DNAmAge.
Recently, Marioni et al. found that DNAmAge increases at a slower rate than chronological age across the life course in five independent cohorts, especially in the older population [21], which suggested that, overall, AA declines as people get older. The global decreasing pattern of AA across the two visits observed in our study is in line with this finding. This pattern may also be explained by survival bias, due to the higher probability that healthier participants with relatively lower AA may stay longer in longitudinal studies. After adjusting for hypertension and other potential covariates that might affect the DNAmAge, we surprisingly found that taking AHMs was associated with faster AA. This finding is at variance with the declining pattern of DNAmAge across the lifespan and is not consistent with the preventive effects of AHMs against age-related health outcomes caused by hypertension [18,19]. Several explanations may account for this discrepancy.
First, measurement bias or inaccuracies may affect the comparability of longitudinal AA and ΔAA estimates. Nevertheless, we have restricted this technical bias to the greatest extent by adjusting for the batch of DNA methylation measurement as a random effect and normalizing the methylation profiles with Horvath's internal normalization method. Furthermore, selection bias due to differential survival rate may cause an underrepresentation of participants with higher AA [22], and may lead to underestimation or even contradicting results of the effect of AHM use on AA and ΔAA, and distort their associations towards the opposite. In our analyses, we accounted for potential selection bias due to the loss of follow-up using inverse probability weighting (IPW) [23]. Point estimates were similar between models using and not using IPW, indicating that little selection bias was introduced due to the loss to follow-up. Additionally, since the use of AHMs is usually treated as the indicator of hypertension severity, our study might report a proxy outcome that indeed reflected the impact of hypertension on aging (confounding by indication). Due to this concern, despite the weak positive but not significant pattern we observed between hypertension and the AA (ΔAA), we additionally adjusted for SBP and hypertension in the fully-adjusted analysis model and performed another sensitivity analysis by adding the interaction term of www.aging-us.com 3217 AGING Table 4. Associations between antihypertensive medication use and change rate of age acceleration from first to second visit.  AGING hypertension and AHMs (data not shown). Neither of the two adjustments altered the patterns of taking AHMs with AA and ΔAA in any relevant manner. In the meanwhile, people might stop taking AHMs after reducing the blood pressure through changing lifestyles and exercises, two factors that are likely to decrease the ΔAA. After controlling for the factors, we still observed a declining pattern between stopping taking any AHM and ΔAA, albeit not significant due to the limited sample size.
One biologically plausible explanation for this inconsistent observation is a potential causal connection among aging, epigenetic biomarkers of age, hypertension and AHM use. We speculate that in these four-corner relationships, aging independently leads to the change of epigenetic biomarkers of age (e.g., DNAmAge, DNAmPhenoAge) and hypertension via two separate pathways. Aging could cause the change of epigenetic biomarkers of age by altering the methylation levels of age-related CpG sites. In parallel, aging could prompt hypertension via a "vicious cycle", which consists of inflammation, oxidative stress and endothelial dysfunction, and might not be closely related to age-related DNA methylation changes [16,17]. Taking AHMs could control blood pressure and reduce the risks of aging-related diseases, such as CVD, kidney failure and dementia, which are caused by hypertension. Nevertheless, as the causation between aging and hypertension is one-way, reducing blood pressure using AHM might not reversely affect biological aging by bringing the aberrant changes of age-related CpG sites back to normal levels. On the contrary, the AHMs' potential side effects, such as abnormal glucose and lipid metabolism [24] and psychological/cognitive disorders [25,26], might have the potential to accelerate epigenetic biomarkers of age as suggested by previous studies [6,10,27]. However, this hypothesis needs to be evaluated by further research with larger populations and multiple follow-ups along with corresponding causal inferences and functional tests.
It is worth noting that the effects of specific medications on AA and ΔAA were not all unfavorable. In our study, ACE inhibitors, ARBs, and diuretics showed weak negative correlation with AA and ΔAA, while beta blockers showed a strong aging accelerating effect. Beyond the effects from specific medications, we should not ignore the adverse drug reactions (ADRs) from combined and inappropriate medication use [28], given most of the participants with hypertension took multiple medications at the same time. ADRs could become more severe and frequent in the elderly due to age-dependent pharmacodynamics and pharmacokinetic changes promoting drug-drug or drug-disease interactions, and such reactions could directly (or indirectly) facilitate the development of aging-related health outcomes including frailty and all-cause mortality [29,30]. This effect might play another key role in accelerating DNAmAge if the ADRs are not identified and treated with further changes in prescriptions.
Major strengths of the present study include the relatively large sample size and repeated measurements with detailed information on a broad range of covariates in a large cohort study. We also acknowledge several limitations in the interpretation of results. First, shifts of leukocyte distribution might affect the DNA methylation changes in whole blood samples [31]. Hence, we adjusted for leukocyte distribution by the Houseman algorithm to restrict potential confounding from differential blood counts to the greatest possible extent [32]. Nevertheless, residual confounding by this and other factors cannot be excluded entirely for the longitudinal analysis. Furthermore, the selected study participants were Caucasians and all male, which limits the generalizability of our results to other racial/ethnic groups and women. Finally, although our overall sample size was relatively large, some of the nonsignificant results may have been due to the lack of statistical power with the relatively smaller size of subsamples for specific AHMs. And given those subgroup analyses did not appear as robust as the whole population and showed contradicted patterns, there might be a risk of getting false positive findings, further research with a bigger sample size and longer multiple follow-ups are warranted to validate our results by eliminating the possibility of false positives.
In summary, we observed cross-sectional and longitudinal associations of any AHM use with increased AA and ΔAA. Our findings suggest that controlling blood pressure by taking medications might not be able to reduce the accelerated epigenetic aging, on the contrary, was associated with accelerated DNAmAge. This study partly reveals the relationship between AHMs and biological aging, and also underlines that DNAmAge and AA may not be able to capture the preventive effects of AHMs that reduce cardiovascular risks and mortality. Future investigations are required to confirm our findings and to elucidate the causal relationship between AHM use and DNAmAge, as well as the underlying pathophysiological mechanisms.

Study design and population
The NAS study is an ongoing longitudinal study of aging, established by the U.S. Department of Veterans Affairs in 1963. Details of this study have been published previously [33]. Briefly, the NAS is a closed cohort of 2,280 male veterans from the Greater Boston area. They were enrolled after an initial health screening that determined that they were free of known chronic medical conditions. Blood samples were collected from 657 participants, most of whom were visited up to 4 times between 1999 and 2013. Participants have been reevaluated every 3-5 years on a continuous rolling basis using detailed on-site physical examinations and questionnaires. We restricted the current analysis to the data of the first two visits of 546 Caucasian participants (aged 55-85 years) who had been visited twice at least, in order to control for the heterogeneity of race and to analyze the change of AA longitudinally. The NAS study was approved by the Department of Veterans Affairs Boston Healthcare System, and written informed consent was obtained from each subject prior to participation.

Data collection
As previously described [34], at each visit, participants were asked to provide detailed information about their lifestyles, dietary habits, activity levels, and demographic factors. Height and weight were measured and were used to calculate body mass index (BMI, in kg/m 2 ). Blood samples were collected for assessing blood-based biomarkers. Participants' status of major diseases was assessed based on the medical history and physicians' diagnosis. In particular, hypertension was defined as a measured SBP of ≥140 mmHg, a measured diastolic blood pressure (DBP) of ≥90 mm Hg, or participants' use of AHMs [35]. SBP and DBP were measured by a physician, and the AHMs included calcium channel blockers, ACE inhibitors, ARBs, alpha blockers, beta blockers and diuretics.

DNA methylation data
DNA of whole blood samples were collected between 1999 and 2013. As previously described [36,37], we used the QIAamp DNA Blood Kit (Qiagen, CA, USA) to extract DNA from buffy coat, and performed bisulfite conversion with the EZ-96 DNA Methylation Kit (Zymo Research, CA, USA). To minimize batch effects, we randomized chips across plates and randomized samples based on a 2-stage age-stratified algorithm so that age distributed similarly across chips and plates. We measured DNA methylation of CpG probes using Illumina HumanMethylation450 BeadChip for all the samples from both visits. After quality control, the remaining samples were preprocessed using the Illumina-type background correction, dye-bias adjustment and BMIQ normalization [38], which were used to generate methylation status. The methylation status of a specific CpG site was quantified as a β value ranging from 0 (no methylation) to 1 (full methylation).
A total of 353 CpG sites were retrieved from the methylation profiles for the estimation of DNAmAge for each participant based on the algorithm proposed by Horvath [3]. This algorithm was derived from a range of tissues and cell types using 353 probes targeted in the Illumina 27k and 450k methylation arrays. In this study, we performed the estimation by using the online calculator (http://labs.genetics.ucla.edu/horvath/dnamage/), where background-corrected beta values were preprocessed by the calculator's internal normalization method [3]. AA was determined as discrepancies between methylation and chronological age in the form of residuals, which have a mean of 0 and represent positive and negative deviations from chronological age in years. The residuals were calculated by a linear regression procedure in which DNA methylation age was the outcome and chronological age was the independent variable. The ΔAA for each participant between the two visits was determined as:

Statistical analysis
Study population of each visit was described with respect to major socio-demographic characteristics, lifestyle factors and detailed AHM use. The correlation between estimated DNAmAge and chronological age at each visit was evaluated by Spearman correlation coefficient.
We first tested the cross-sectional association of hypertension with AA at each visit and the Δ AA between the two visits. A linear mixed model with AA (ΔAA) as outcome was employed, controlling for covariates that have been reported to be associated with DNA methylation changes or with the use of AHM, including age (years), BMI [kg/m 2 , underweight (<18.5, < 1% of the study population) or normal weight (18.5 [32]). The batch of microarray experiments in DNA methylation measurement was controlled for as the random effect. Corresponding AA AGING was additionally adjusted for in the analyses of ΔAA at each visit.
We then investigated the cross-sectional associations of AHM use with AA and ΔAA at each visit in two mixed linear regression models. Model 1 adjusted for age, leukocyte distribution and the random batch effect, and Model 2 further adjusted for BMI, smoking status, alcohol consumption, physical activity, education, total cholesterol (mg/dL), HDL (mg/dL), triglycerides (mg/dL), fasting glucose (mg/dL), SBP (mm Hg), hypertension, stroke, coronary heart disease, diabetes and cancer (yes/no). AA was additionally adjusted for in the analyses of ΔAA at each visit.
Finally, we examined the longitudinal association between AHM use and the ΔAA. AHM use in the longitudinal analysis was classified as: never used, stopped use after first visit, started use after first visit and continuous use according to the reports of AHM use between the two visits, and was treated as a predictor in the analysis models. Two mixed linear regression models were used, adjusting for the same covariates described above.

Sensitivity analyses
We further performed sensitivity analyses for the main results by using another two DNA methylation based aging biomarkers: DNAmAge estimated by Hanumm et al.'s algorithm [39], and DNAmPhenoAge recently developed by Morgan et al. as an "update" of the Horvath DNAmAge for the lifespan [40].
Hanumm DNAmAge was estimated based on 71 agerelated CpG sites reported in 2013 [39] and determined as the sum of the methylation beta values multiplied by the reported effect sizes of the predictors. AA of Hannum's DNAmAge was also determined as discrepancies between DNA methylation and chronological age in the form of residuals.
Another batch of 513 CpG sites was retrieved for the estimation of DNAmPhenoAge for each participant based on the algorithm proposed by Levine et al. [40]. With the coefficient and intercept values provided by the authors, we estimated the DNAmPhenoAge as: As defined by the authors [40], difference between Phenotypic and chronological age (DNAmPhenoAgechronological age) was defined as the AA of DNAmPhenoAge. The ΔAA for the two aging biomar-kers between the two visits was also determined by the formula (1).
We tested the cross-sectional and longitudinal associations of the AA and ΔAA of the two biomarkers with the use of any AHM using the same analysis models employed in the main analyses for Horvath DNAmAge.
Data cleaning and all aforementioned analyses were performed by SAS version 9.4 (SAS Institute Inc., Cary, NC, USA), and all statistical tests were two-sided with p-values of <0.05.

AUTHOR CONTRIBUTIONS
XG contributed to the data analysis and drafting. Other authors contributed to critical revision of the manuscript. All authors read and approved the final manuscript.