Epigenetic mechanisms and their effects on gene activation and silencing are becoming increasingly relevant to phenotype expression and the development of different diseases (Bird, 2007; Gluckman et al., 2009). The epigenome, along with the genome, instructs the unique gene expression program of each cell type in order to define its functional identity during development or disease (Rivera and Ren, 2013). While epigenetic changes are functionally relevant to the genome, they do not involve changes in a DNA sequence. One of such epigenetic modifications is the methylation of cytosine molecule, usually at CG dinucleotides (CpGs), called DNA methylation. Thus, the addition of methyl groups to the DNA molecules, motivate changes in the DNA segment activity without changing its sequence. When the CpGs of promoter regions are methylated, gene expression are often silenced (Hashimshony et al., 2003). Specifically, DNA methylation may affect the transcription of genes in two ways: (a) physically impede the binding of transcriptional proteins to the gene; (b) may be bound by proteins known as methyl-CpG-binding domain proteins resulting in compact, inactive chromatin (heterochromatin) (Choy et al., 2010). There is a growing understanding that environmental manipulation (e.g., diet and exercise) can influence cell behavior and disease states through epigenetic alterations of gene expression.

Exercise is a well-known physiological stimulus resulting in health and functional improvements. Regular exercise has numerous health benefits and can help reduce the risk of common ailments such as cardiovascular disease, type II diabetes, several forms of cancer (Matheson et al., 2013), and importantly, neurodegeneration and cognitive deterioration. Until recently, there has been paucity of knowledge on how aerobic exercise can induce epigenetic modifications in humans. A global study of DNA methylation in human skeletal muscle from relatives of Type 2 Diabetes patients demonstrated that a 6-month exercise resulted in epigenetic changes (Nitert et al., 2012). However, whether aerobic exercise can influence DNA methylations in elderly AA MCI participants has not been examined. It is also unknown whether such changes will include genes having essential roles in neurodegeneration and cognitive decline.

Therefore, we examined the impact of a 6-month aerobic exercise-training on human DNA methylation in mild cognitively impaired (MCI) elderly African Americans (AA)s; identify DNA methylated genes affected by aerobic exercise, and investigated the biological pathways affected by aerobic exercise training-related changes in CpG intensities. We then used quantitative real-time polymerase chain reaction (qRT-PCR) to confirm the effects of these methylation changes on the mRNA levels of the associated genes.

The Howard University Institutional Review Board (IRB) approved the protocols used for this investigation. As required for studies involving human subjects, all participants completed a signed informed consent form before enrollment in the study. The details of the Gene, Exercise, and Memory Study (GEMS-I) protocol have previously been published (Iyalomhe et al., 2015).

Genome-wide methylation levels at CpG sites from participants at baseline and at 6 months after aerobic and stretch training were profiled using the Infinium HumanMethylation450 BeadChip assay (Illumina, San Diego, CA, United States). The raw intensity data of all samples were imported into the software Partek (St. Louis, MO, United States). SWAN method (Maksimovic et al., 2012) was selected to normalize array intensities. SWAN-normalized β values, which correspond to the percentage of methylation at a CpG site, were then calculated. Premised on the exclusion of probes with a detection p-value > 0.01 in one or more of the 22 samples for the downstream analysis, we discarded 6,942 probes while 478,570 probes remained.

Analyses examining baseline genome-wide CpG methylation and the influence of a 6-month aerobic exercise-training on methylations were conducted in STATA 14 software (StataCorp, 2015). First, all variables measured at baseline and post-6 months of exercise were examined descriptively, and their distributions were further inspected using Boxplots and Scatterplots (McGill et al., 1978; Royston and Cox, 2005). Student’s t-test was used to compute mean estimates of continuous variables between groups of baseline measures; nonparametric Mann–Whitney rank test to derive p-values for group differences in non-normal continuous variables; and Chi-square and Exact tests to compare group differences in categorical variables. For continuous measures, and within each group (aerobic and stretch), a paired t-test was used to assess significant differences in means between baseline and six months.

Effects of exercise-induced changes in VO₂max (i.e., VO₂max at baseline and after a 6-month training) on genome-wide CpG methylation were examined within-individual level using linear mixed-effects analysis (West et al., 2014), and robust variance-covariance approach adjusted for standard errors of the coefficients. This is robust to differences in variances between groups and to violation of normality assumption. Because sampling distributions of test statistics are known to be t and F-distributed in simple cases in small sample studies, we also used Satterthwaite and Kenward-Roger denominator-degrees-of-freedom (DDF) adjustments for small-sample inference. Given the sample size, we report standard errors derived from Kenward-Roger DDF more closely similar to the bootstrap estimation of standard errors of randomly selected CpG methylations (Satterthwaite, 1946; Kenward and Roger, 1997).

To account for the multiplicity of tests from the genome-wide inquiry, we used the frequentist q–values approach. The q-value package inputs a variable of p–values and outputs a variable of q-values, equal in each observation to the minimum FWER or FDR that would result in the inclusion of the corresponding p-value in the discovery set if the specified multiple–test procedure was applied to the complete set of input p–values (Newson, 2010). We computed false discovery rates (FDR) using Sidak approaches (Šidák, 1967).

In this sample of elderly AA MCI study participants, a 6-month standardized, and supervised aerobic exercise-training influenced global DNA methylation changes. While the most significant hypomethylations involved VSP52, SACRB1, ARTN, NR1H2, and PPPLR5D genes (Table 2), these genes were not similarly influenced by training effects in the stretch exercise group. We discuss the biological plausibility of our most important findings, consistency with previous studies, and coherence with the known roles of exercise on our molecular findings, and therefore on the pathophysiology of age-associated cognitive decline.

It is now acknowledged that physical activity can enhance brain plasticity and improve cognition and wellbeing (Weinberg and Gould, 1999). Also, physical exercise can induce structural and functional changes in the brain and protect against neurodegeneration (Mandolesi et al., 2018). In a recent comprehensive review, Mandolesi et al. (2018) acknowledged the positive biological and psychological effects of physical activity on the brain through its enhancement of cognition, while Buchman et al. (2012) showed that physical activity reduced inflammation and improved memory in MCI and potentially Alzheimer’s Disease patients (Buchman et al., 2012). However, the mechanisms underlying this beneficial effect of exercise on memory are poorly understood. The present study is a logical extension of our previously published work on the effects of aerobic exercise on global array expression in AA MCI volunteers, showing that aerobic-exercise can down-regulate the expression of pro-inflammatory genes, concomitantly up-regulate anti-inflammatory genes, and promoted the expression of genes involved in axonal growth and neuronal survival (Iyalomhe et al., 2015). While the aerobic group underwent progressive exercise intensity and duration to achieve individualized 70% VO2 Max and endurance capacity, the stretch (control) group only undergone progressive stretch exercise. Overall, exercise training-induced increases in endurance capacity and reduced body weight in the aerobic exercise group affirmed training effects. Conversely, a slight decline in VO2max and unchanged body weight, in the stretch group, underscore less fitness effects. Collectively, this methylation study provides additional insights into the mechanism underlying the advantageous effect of exercise on memory.

While the biological mechanisms contributing to the health benefits of exercise are well-acknowledged, the exact mechanism underlying its effects on memory remains mostly unknown. Exercise can reshape the epigenome and induce significant changes in DNA methylation (Lindholm et al., 2014). Its ability to activate or silence specific genes may, therefore, explain its effects on health. For example, acute intensity exercise can alter the methylation of exercise-responsive genes, resulting in DNA hypomethylation in skeletal muscle (Barres et al., 2012). In contrast, a 6-month exercise program in previously sedentary middle-aged men increased methylation in adipose tissue (Rönn et al., 2013). For example, Rönn et al. (2013) provided a detailed map of the genome-wide DNA methylation pattern in human adipose tissue and linked exercise to altered adipose tissue DNA methylation. Therefore, these epigenetic changes can alter gene expression and metabolism and thus inform the advantageous effects of exercise on health (Ling and Ronn, 2014). Nonetheless, understanding the consequent biologic cascade resulting in long-lasting effects of regular exercise-training on DNA methylation needs a more nuanced understanding. For example, Soci et al. (2017) showed that exercise-induced changes in DNA methylation could alter several molecular mechanisms such as muscle contraction, increased mitochondrial mass, oxidative and non-oxidative mechanisms as well as cardiac and skeletal hypertrophy (Soci et al., 2017). Interestingly, some of these mechanisms have implications for fitness and cerebral hemodynamics.

The most significant effect of a 6-month exercise in our study is the hypomethylation of vacuolar protein sorting homolog 52 (VPS52) gene, supported by a corresponding increase in its mRNA levels. VPS52 is a crucial part of the Golgi-associated retrograde complex that plays an important role in cellular inter-compartment transport and is highly expressed in the brain¹. In one experiment, the cortex of Amyloid Precursor Protein (APP) knockout mice exhibited the upregulation of VPS52 and a related protein. Conversely, lipid dysregulation and elevated cerebral APP are associated with decreased expression of VPS52 (Castello, 2015). Therefore, given VPS52 complex interaction with lipid regulation in the context of changing APP build-up, studies are needed to understand the role of VPS52 on memory and human aging.

Hypomethylation near Nuclear Receptor Subfamily 1 Group H Member 2 (NR1H2) gene was also associated with exercise-associated increases in its mRNA expression. This observation is likely significant in that dysfunctionality of the NR1H2 gene can promote late-onset AD (Patel and Forman, 2004; Adighibe et al., 2016). Because dysregulated cholesterol levels can induce amyloidogenesis (Infante et al., 2010), NR1H2-mediated cellular efflux of cholesterol may protect against AD development.

Although exercise-training was associated with methylation changes in Scavenger receptor class B member 1 (SCARB1), Protein Phosphatase 2 Regulatory Subunit B’Delta (PPP2R5D), and Artemin (ARTN), corresponding changes in gene expression remained non-significant in our study. Therefore, further studies are needed to discern their biological roles. For instance, SCARB1 encodes high-density lipoprotein receptor that regulates cholesterol efflux from the peripheral tissue to the liver, is present in astrocytes and vascular smooth muscle cells in AD brain, and has been demonstrated to mediate adhesion of microglia to fibrillar amyloid-β (Aβ) (El Khoury et al., 1996; Thanopoulou et al., 2010; Jawaid and Khalil, 2011; Mulder et al., 2012). PPP2R5D is an interesting finding because its mutations can promote intellectual disability (Houge et al., 2015) and the development of spatially restricted tauopathy by deregulating CDK5 and GSK3-beta in mice lacking PPP2R5D subunits (Louis et al., 2011). The PPP2R5D gene encodes the protein B56-delta (B56δ), a part of (B subunit) of the phosphatase 2A (PP2A) enzyme, which dephosphorylates other specific proteins, to control their activation. Since most of the affected proteins are involved in active signaling, the activation/inhibition of PPP2R5D is often tightly controlled (Wang et al., 2008); thereby making its expression levels less informative. Further, ARTN may be critical to neuronal survival, given that protein signals through the RET receptor and GFR alpha three coreceptors encoded by ARTN support the survival of several peripheral neurons and a population of dopaminergic CNS neurons (Zihlmann et al., 2005).

However, we note that gene (ARTN) did not amplify in blood cells, and an alternative primer yielded a similar result. Also, we verified from GeneCards (Stelzer et al., 2016) that, while the ARTN gene is widely expressed in nervous tissues, it is not expressed in blood cells. Because phosphatases and kinases (e.g., PPP2R5D and GSK3β) are regulated more by post-translational modifications, increases in their activities are often associated with compensatory decreased mRNA levels through other pathways, and vice versa. While our methylation result is consistent with an expectation of increased levels of PPP2R5D, the incongruent decrease in mRNA levels suggests other regulatory factors may be involved. Also, our finding of training-induced changes in SCARB1 methylation is notable, given its expression in astrocytes (Mulder et al., 2012). Nonetheless, our findings are significant to current knowledge on exercise training-induced changes on blood methylation profile in MCI subjects.

The inaccessibility of living human brain tissue in AD studies motivated our consideration for using human peripheral blood to evaluate methylation profile following exercise intervention. Several studies have reported that DNA methylation patterns in human blood parallel epigenetic changes in the brain (Horvath et al., 2012; Walton et al., 2016). For example, Horvath et al. (2012) noted a correlation r = 0.9 across whole-blood and postmortem brain tissue methylation datasets, suggesting that blood DNA methylation profiles may be a promising surrogate for brain tissue (Horvath et al., 2012). Using blood obtained premortem, Davies et al. (2012) showed that individual methylation profile correlated with relevant postmortem brain tissue (cortex and cerebellum) methylation levels (correlation coefficient = 0.66–0.76) (Davies et al., 2012). Further, a recent systematic review of the DNA methylation studies showed that epigenetics changes in the peripheral blood of AD patients correlated with AD pathology (Wei et al., 2020). Acknowledged associations of DNA methylation in peripheral blood cells of AD patients with poor cognitive performances and APOE ε4 polymorphism are consistent with this report (Di Francesco et al., 2015). Thus, our approach to investigate methylation changes in peripheral blood following exercise intervention provides valuable insight into the mechanism underlying the effects of fitness adaptation on methylation, and potentially, on experimentally inaccessible human brain and neurodegeneration.

Our study revealed global changes in DNA methylations following 6 months of standardized and supervised aerobic exercise training in MCI subjects. The relevance of some of these genes to AD pathophysiology provides further evidence to the likely significance of aerobic exercise in moderating AD pathophysiology in MCI populations. An important limitation of this pilot study is the relatively small sample size. Nonetheless, we provide novel insight into the effect of aerobic exercise on the epigenome. Our plan includes testing these findings in larger samples and informing the genes’ relationship with cognitive phenotype and disease progression spectrum.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: https://www.ebi.ac.uk/arrayexpress/ with accession E-MTAB-11166.

The studies involving human participants were reviewed and approved by the Howard University Institutional Review Board (IRB). The patients/participants provided their written informed consent to participate in this study.

JN, EN, and TO conceived the study, cleaned, and analyzed the data. FB performed RNA isolation, cDNA preparation, conducted Q-PCR gene expression experimental setup, and analysis. JN, EN, FB, and TO wrote the manuscript. All authors contributed to study design and strategies for analyses, and participated in the review of the final manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.