DNA methylation‐based biomarkers of aging were slowed down in a two‐year diet and physical activity intervention trial: the DAMA study

Abstract Several biomarkers of healthy aging have been proposed in recent years, including the epigenetic clocks, based on DNA methylation (DNAm) measures, which are getting increasingly accurate in predicting the individual biological age. The recently developed “next‐generation clock” DNAmGrimAge outperforms “first‐generation clocks” in predicting longevity and the onset of many age‐related pathological conditions and diseases. Additionally, the total number of stochastic epigenetic mutations (SEMs), also known as the epigenetic mutation load (EML), has been proposed as a complementary DNAm‐based biomarker of healthy aging. A fundamental biological property of epigenetic, and in particular DNAm modifications, is the potential reversibility of the effect, raising questions about the possible slowdown of epigenetic aging by modifying one's lifestyle. Here, we investigated whether improved dietary habits and increased physical activity have favorable effects on aging biomarkers in healthy postmenopausal women. The study sample consists of 219 women from the “Diet, Physical Activity, and Mammography” (DAMA) study: a 24‐month randomized factorial intervention trial with DNAm measured twice, at baseline and the end of the trial. Women who participated in the dietary intervention had a significant slowing of the DNAmGrimAge clock, whereas increasing physical activity led to a significant reduction of SEMs in crucial cancer‐related pathways. Our study provides strong evidence of a causal association between lifestyle modification and slowing down of DNAm aging biomarkers. This randomized trial elucidates the causal relationship between lifestyle and healthy aging‐related epigenetic mechanisms.


| INTRODUC TI ON
Population aging is emerging as one of the most critical health issues, leading to medical, social, economic, and political problems. To quantify healthy aging in epidemiological and clinical studies is not straightforward. Among various biomarkers of healthy aging proposed in recent years, the epigenetic clocks, based on DNA methylation (DNAm) data, are getting increasingly accurate in predicting the individual biological age (Horvath, 2013;Horvath & Raj, 2018). The concept of epigenetic age acceleration (AA) has been introduced as the difference between predicted DNAm age and the chronological age: positive values of AA indicate unhealthy aging and vice versa (Horvath, 2013). Recent literature suggests epigenetic AA as a reliable biomarker of healthy aging as it has been associated with longevity Dugué et al., 2018), several pathological conditions , and non-communicable disease risk factors like obesity (Horvath et al., 2014), poor physical activity (PA) (Quach et al., 2017), and low socioeconomic status (Fiorito et al., ,2017. To date, epigenetic clocks that have gained considerable popularity in the scientific community are Horvath (Horvath, 2013) and (Hannum et al., 2013) "first-generation clocks," and Levine's DNAmPhenoAge (Levine et al., 2018) and Lu's DNAmGrimAge (Lu et al., 2019) "next-generation clocks." It has been shown that the "next-generation clocks," DNAmGrimAge particularly, outperform "first-generation clocks" in predicting longevity and the onset of age-related pathological conditions and diseases (Bergsma & Rogaeva, 2020;Lu et al., 2019). Specifically, the DNAmGrimAge is built as a linear combination of seven DNAm-based surrogate markers of plasma proteins: adrenomedullin (ADM), beta-2-microglobulin (B2 M), cystatin C (Cystatin C), growth differentiation factor 15 (GDF-15), leptin (Leptin), plasminogen activator inhibitor-1 (PAI-1), and tissue inhibitor metalloproteinases 1 (TIMP-1) plus DNAmbased biomarkers for smoking pack-years, using DNAm values of 1,030 unique CpG sites.
Additionally, the total number of stochastic epigenetic mutations (SEMs) per individual has been proposed as an alternative biomarker of healthy aging based on whole-genome DNAm data (Gentilini et al., 2015). The total number of SEMs per individual, also known as epigenetic mutation load (EML) (Yan et al., 2020), is defined as the sum of extreme (outliers) DNAm values per sample. Recently, (Gentilini et al., 2015) provided evidence of the exponential relationship between age and SEMs, which occurs naturally during aging as a consequence of the "epigenetic drift." A higher EML has been associated with age-related pathological conditions like X chromosome activation skewing (Gentilini et al., 2015) and risk factors for non-communicable diseases like cigarette smoking, alcohol intake, exposure to toxicants, and low socioeconomic status (Curtis et al., 2019;Fiorito et al., 2019), and it is associated with increased risk of different types of cancer in prospective studies (Gagliardi et al., 2020;Wang et al., 2019). Interestingly, DNAm epigenetic clocks and EML are weakly correlated, suggesting they describe different aspects of epigenetic aging processes (Yan et al., 2020).
A fundamental property of epigenetic is the potential reversibility of the effect, raising questions about the possible slowdown of epigenetic aging by improving lifestyle. Recent observational studies provided evidence that smoking-related DNAm modifications tend to reverse after smoking cessation in a time-dependent manner (Guida et al., 2015), and epigenetic AA due to early-life social adversities can be partially reversed improving lifestyle and social conditions in adulthood (Fiorito et al., 2017). A pilot clinical trial conducted on nine volunteers suggests that the epigenetic clock could be reversed after one-year treatment with a cocktail of drugs based on recombinant human growth hormone (Fahy et al., 2019).
In this study, we aimed to investigate whether modifying dietary habits and increasing PA have favorable effects on biological aging, measured using both the DNAmGrimAge and the EML, in healthy postmenopausal women. This study sample consists of 219 adult post-menopausal women from the "Diet, Physical Activity, and Mammography" (DAMA) study: a single-center, 24-month randomized intervention trial whose primary aim was to investigate whether mammographic breast density (an established independent risk factor for breast cancer development) could be reduced in healthy postmenopausal women by modifying their dietary habits and physical activity levels (Masala et al., 2019).

| RE SULTS
After DNAm data quality controls and filtering, this study sample include 219 DAMA participants, distributed into four trial study arms (arm 1: dietary intervention, arm 2: PA intervention, arm 3: dietary +PA intervention, and arm 4: control group), with wholegenome DNAm measured from blood collected at baseline and after two years of intervention. For each sample, we computed the total number of SEMs and DNAmGrimAge measures. For statistical comparisons, we used a logarithm transformation of the total number of SEMs (referred to as EML henceforth), and DNAmGrimAge Acceleration (referred to as DNAmGrimAA henceforth) was defined as the residuals of the regression of DNAmGrimAge on chronological age as described by Lu and colleagues (Lu et al., 2019).

| Association analyses at baseline
In Table S1, we reported the characteristics of the study sample by study arm at baseline. There were no statistically significant differences among the four groups considering anthropometric and lifestyle characteristics, nor DNAmGrimAA, whereas the EML differed by groups at baseline (ANOVA test p-value =0.01).
In Table 1, we reported the results of two multivariate linear regression models having either baseline DNAmGrimAA or EML used as the outcome, and baseline anthropometric and lifestyle characteristics entered as the predictors. DNAmGrimAA was associated with obesity (β = 0.80 95% CI 0.11-1.49, p = 0.02 comparing overweight with normal-weight; β = 2.53 95% CI 1.28-3.78, p = 0.0001 comparing obese with normal-weight) and smoking (β = 0.88 95% CI 0.23-1.52, p = 0.01 comparing former with never smokers) adjusting for the other risk factors in Table 1, whereas EML was not associated with any lifestyle variables at baseline. Table 2 reports Pearson correlation coefficients (and corresponding p-values) among the two epigenetic aging biomarkers and dietary variables at baseline. Higher consumption of fruit and vegetables was associated with decreased DNAmGrimAA (p = 0.05 and p = 0.002, respectively), whereas a higher consumption of processed meat was associated with increased EML (p = 0.01).

| Association analyses after the intervention
We run a difference-in-difference model to estimate the differential changes of DNAmGrimAA and EML in the treated group compared with the control group during the two-year intervention. We applied a two-step approach: First, we defined the "delta DNAmGrimAA" and "delta EML" as the difference between the two epigenetic aging biomarkers measured after and before the intervention. In Figure 1

| Additional investigation on the eight DNAmGrimAA components
We controls, Figure 2b) and contribute for more than 30% of the explained variability ( Figure 2c). Also, reduction of DNAmLeptin and DNAmGDF15 provided a substantial contribution (20% and 15% respectively, Figures 2d-e-f).

| Enrichment analyses on epimutated CpG sites
We further investigated the biomolecular pathways involved in the reduction of EML caused by the PA intervention and the stability of SEMs over these two years. The majority of the identified CpG sites carrying a SEM at baseline (i.e., before the intervention) were still epimutated after the two-year trial (the average proportion of "stable" SEMs per individual was 69%, ranging from 54% to 89%).
We performed additional investigation of what we named "physical activity-related reversible SEMs" (PArSEMs), that is, those CpG sites in which we found a SEM at baseline but not after the PA intervention trial.

| Sensitivity analyses
For sensitivity analyses, we repeated the previously described lin- Our results on the analyses performed at baseline (before the intervention trial) confirmed previously observed cross-sectional associations between the epigenetic clocks and risk factors for noncommunicable diseases, like obesity, consumption of processed meat (with unfavorable effects), and consumption of fruit and vegetables (with favorable impacts). Interestingly, the two epigenetic biomarkers of aging likely describe different aspects of the aging-related molecular mechanisms, as they are associated with different risk factors TA B L E 3 Average differences and 95% confidence intervals (CIs) of DNAmGrimAA and EML measured before the randomized trial minus DNAmGrimAA and EML measured after the randomized trial (first two columns); and differential changes in the delta DNAmGrimAA is associated with lower consumption of fruit but not with higher consumption of red meat, whereas an inverse pattern of associations was observed for the EML.

DNAmGrimAge biomarker
The main aim of the present study was to compare the DNAm-

| Increasing physical activity slows down the EML biomarker
Increasing evidence indicates that aging is associated with an accumulation of SEMs, and in turn, the total number of SEMs is associated

| CON CLUS IONS
We provided strong evidence of a causal association of improving dietary habits and increasing physical activity on DNAm-based biomarkers of healthy aging. It is worthy to note that DAMA study is intentionally based on non-extreme interventions, meaning that relatively easily achievable changes in one's lifestyle behaviors lead to a significant slowing down of biological aging biomarkers, which in turn are associated with higher longevity, lower risk of developing age-related diseases, and increased quality of life in the older age.
Further, our results indicate that dietary quality and physical activity influence epigenetic aging through complementary molecular mechanisms, suggesting that their effect is potentially cumulative rather than interchangeable. In conclusion, our results provide further evidence about the importance of policy intervention programs to promote a healthy diet and physical activity, leading to a substantial reduction of the burden for many aging-related pathological conditions and diseases. Additionally, our results provide a step forward in understanding the biological mechanism of aging and identifying health-related biomarkers.

| Strength and limitations
Since this was a secondary analysis, the relatively modest sample size is a possible limitation of this study. The original factorial study design included four arms (arm 1: diet, arm 2: PA, arm 3: diet +PA, and arm 4: controls), but for statistical comparisons, we used the two main intervention groups (arms 1 and 3 for investigating the effect of dietary intervention, and arms 2 and 3 for investigating the effect of PA intervention). However, a post hoc power analysis of the study indicates that our analytical strategy makes this study well-powered

| Study sample
The DAMA study was a single-center, 24-month randomized intervention trial (Trial Registration ID: ISRCTN28492718) with a 2x2 factorial design, whose primary aim was to investigate whether mammographic breast density (MBD) could be reduced in high-MBD (>50%) healthy post-menopausal women by modifying their dietary habits and/or PA levels (Masala et al., 2019). Study participants were selected in 2009-2010 among postmenopausal women aged 50-69 years that attended the local breast cancer screening program in Florence, Italy (Masala et al., 2014). Women were eligible for inclusion if they had a negative screening mammogram with MBD >50% (assessed using the BI-RADS classification (Liberman and Menell, 2002)); those selected for a second-stage diagnostic procedure following the screening mammogram were excluded regardless of the final outcome of the diagnostic process. Other exclusion criteria were as follows: recent (past 12 months) hormone replacement therapy use; current smoking, or having quit smoking by <6 months; being previously diagnosed with cancer (except non-melanoma skin cancer) or suffering from any illness that could hamper an active participation in the study activities.
At baseline, all participants provided information on dietary habits and lifestyle (including household, occupational and leisuretime PA) by filling two questionnaires previously used within the EPIC (European Prospective Investigation into Cancer and Nutrition) study (Palli et al., 2003), and had their anthropometric measures taken using standardized procedures. A fasting venous blood sample was taken, divided into plasma, red cells, and buffy-coat aliquots, and stored together with urine samples in the project biobank. Each woman was then randomly assigned to one of the four study arms of no more than one glass of wine daily at meals for those already used to drink alcohol. In addition, women allocated to the dietary intervention study arm were also requested to attend six group meetings and eight cooking classes over the course of the study.
Women randomized to the PA intervention (arm 2) were asked to increase their moderate-level recreational PA up to 1 hour/day (corresponding to about 3 MET-hours day [MET=metabolic equivalent]), to be combined with a more strenuous activity accounting for 6-10 MET-hours weekly. Women were also requested to attend weekly a one-hour session led by trained PA experts in an appropriate fitness facility and were provided with some equipment for home exercises. Finally, the study protocol also included participation in six group sessions and six collective walks supervised by the study team.
Women assigned to the diet +PA intervention (arm 3) were requested to change both their dietary habits and PA levels by combining the protocols of arms 1 and 2.
Study participants assigned to any intervention arm (1, 2, or 3) were requested to keep five written one-week diaries on diet and/ or PA levels (depending on study arm), which were then reviewed by the study personnel to monitor the achievement of the study objectives and provide further tailored counseling to the participants.
Women randomly assigned to the control group (arm 4) received general advice on healthy diet and PA levels according to the recommendations from the World Cancer Research Fund (WCRF) report 2007 (Wiseman, 2008), were invited to attend a group meeting taking place in the first six months of the study, and were distributed ad hoc printed material.
At the end of the study (24±3 months from enrollment, coinciding with the time of the next mammographic screening), all study participants underwent a final visit, in which the same protocol as at the baseline visit was applied.
Compliance with the proposed interventions was good, with an increased consumption of vegetables and legumes, and a reduced consumption of meat and cakes, observed women assigned to the dietary intervention group, and an increase in all type of recreational physical activity for those allocated to the PA group (Masala, 2019).
In the main analysis, a decrease in MBD was observed among women in the dietary intervention and in the PA group compared to controls, while no significant effect on MBD was found among women in the double intervention group (Masala, 2019).

| Genome-wide DNA methylation analyses
Buffy coats stored in liquid nitrogen were thawed, and genomic Matched pairs (pre-and post-intervention) were arranged randomly on the same array. All the chips were subsequently scanned using the Illumina HiScanSQ system. Control probes included in the microarray were used to assess bisulfite conversion efficiency and to exclude lower-quality samples from further analyses.

| Computation of DNAmGrimAge
We computed the epigenetic age acceleration (AA) measures according to the algorithm described by Lu et al. (Lu et al., 2019).

| Identification of stochastic epigenetic mutations (SEMs)
We identified SEMs based on the procedure described by Gentilini et al. (Gentilini et al., 2015). Specifically, for each CpG, considering the distribution of DNA methylation beta values across all samples, we computed the interquartile range (IQR)-the difference between the 3rd quartile (Q3) and the 1st quartile (Q1)-and we defined a SEM as a methylation value lower than Q1-(3×IQR) or higher than In all the regression models, comparisons with p-value lower than 0.05 were considered statistically significant.

| Enrichment analyses
The genomic locations of SEMs were annotated by merging the

Illumina information on the chromosomal position of each probe with ENCODE/NIH Roadmap Chromatin ImmunoPrecipitation
Sequencing (ChIP-Seq) data for chromatin states and transcription factor binding sites (TFBS) in untreated human embryonic stem cells (hESC) (ENCODE Project Consortium, 2012). We investigated whether SEMs were enriched in functional genomic regions using the procedure implemented in the R package regioneR (Gel et al., 2016). Briefly, the algorithm is specifically designed to test whether a set of genomic loci significantly overlap with a set of genomic regions, using a permutation procedure that controls the type I error rate and avoids spurious associations driven by the intrinsic structure of the DNA (i.e., relationship between CG content, gene promoters, and copy number alterations) (Gel et al., 2016).

ACK N OWLED G M ENTS
The study was supported by the following grants from the Italian

CO N FLI C T O F I NTE R E S T
The authors declare that they have no conflict of interest. Masala involved in supervision.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.