Effects of early- and mid-life stress on DNA methylation of genes associated with subclinical cardiovascular disease and cognitive impairment: a systematic review

Cardiovascular disease (CVD) is the leading cause of death in the United States, with more than an estimated 1 in 3 Americans diagnosed with at least one type of CVD [1]. Epidemiologic studies suggest a strong association between CVD and type 2 diabetes mellitus (T2DM), the latter of which affects over 25 million Americans [2]. The Diabetes Heart Study (DHS) contains a large cohort of these dual diagnosis individuals, and follow-up studies done with this population have found markers of subclinical CVD as predictors of cognitive function in diabetic patients, warranting early intervention to prevent CVD and related cognitive deficits [3]. Independent of T2DM, CVD and cognitive impairment (CI) are frequently co-occurring conditions, with empirical research supporting the association between subclinical CVD markers and dysfunction in cognitive domains such as memory, attention, and decision making [3–8]. Markers such as atrial fibrillation and coronary artery calcium may also serve as risk factors for future cognitive decline and dementia [5, 9, 10]. In a similar fashion, the degree of CI can predict the risk and severity of CVD events and mortality [6, 7]. Efforts aimed at identifying commonalities between subclinical CVD and cognition may serve as one approach to prevent clinical diagnosis of CVD and cognitive decline, which is especially prevalent in individuals with T2DM.

Several traditional and novel risk factors have been reported for CVD. Traditional ones include family history, smoking, lipid levels, diabetes mellitus, physical activity, blood pressure, and stress [11]. There may be similar overlap between CVD risk factors and risk for cognitive decline [12]. Novel risk factors for CVD include increased levels of inflammatory markers such as C-reactive protein and interleukin-6 [13]. Sex, race, and ethnicity also serve as potential moderators, with greater CVD prevalence among males than females, and greater CVD-related death rates among non-Hispanic blacks than non-Hispanic whites. However, there is still much variability, warranting research into the mechanisms that can explain residual disparity among individuals.

Stress (i.e. job strain, parental loss of child, illness diagnosis) is a risk factor for both CVD and CI, with increased risk of coronary heart disease mortality in those who experienced three or more stressful life events [14]. Exposure to stress can modify DNA methylation, which may alter gene expression and therefore contribute to disease phenotypes [15]. Early-life stress, such as childhood abuse and stress-related disorders, have lasting effects on methylation that may persist into adulthood [16–19]. Childhood adversities and stress at other points throughout the lifespan may induce health problems by dysregulating physiological systems like the immune, nervous, metabolic, and neuroendocrine systems. The broad and often subjective nature of stress makes it difficult to accurately measure, which is further complicated when considering temporal differences in stress exposure: acute and chronic stressors have different effects on health. Whether it be from a single experience or an accumulation of experiences, greater stress duration can contribute to more toxic and long-term changes in physiological responses in contrast to acute events [20, 21]. Since DNA methylation may mediate the effects of stress on susceptibility to CVD and CI, it is paramount that studies comprehensively gauge stress. There is a lack of research identifying any shared genetic risk factors for clinical CVD and CI, with CI being independent of a neurological disorder [22], but recent research has focused on the identification of CpG sites specifically methylated in genes that are associated with subclinical CVD and CI [15, 23, 24]; however, there is great inconsistency with methods used to identify and quantify methylation. To our knowledge, no review elucidates whether stress can act to modify the methylation status of genes implicated in both subclinical CVD and CI, and therefore contribute to future manifestation of disease. Toward this end, we conducted a systematic review of the literature investigating these relationships (Fig. 1).

To the best of our knowledge this is the first review to examine the effects of stress on DNA methylation associated with subclinical CVD and CI. Existing reviews have analyzed differentially methylated genes in either outcome of interest, but not with the stress component discussed in this review [15]. For reviews that have evaluated some stress exposure, either only one outcome of interest was examined, or genes were indirectly associated with one of the outcomes [19, 52, 53]. Therefore, we conducted a review to highlight what information already exists and what gaps still remain in the literature (Fig. 1). DNA methylation can be a powerful biomarker for early detection of onset and progression of CVD and CI.

The biggest difficulty with this review was in determining what would classify as a stressful exposure. The experience of stress can be subjective, with the same incident being stressful for one person and not at all stressful for another. Psychological and physiological stress can be considered as distinct concepts, so we kept our inclusion of stress as broad as possible. Independent of stress, life events like illness would indisputably have some direct biological effects on methylation and progression of CVD or CI, but this doesn’t mean that the effects of stress would be negligible. Inclusion of animal studies with hypoxia as a stressor may be comparable to obstructive sleep apnea in humans, and methylation changes specific to stress may be parsed out from those that underly the biological effects of the illness or other factors.

For studies aiming to identify the moderating effects of stress on DNA methylation, it is advisable that a subjective survey be used to quantify study participants’ perception of stress. Studies can better gauge life stress through use of interviews like the Stress and Adversity Inventory (STRAIN), due to its breadth of coverage. This assessment measures life course stress by covering various life domains and quantitatively calculates the number of stressors within each domain, from early-life throughout adulthood, and the self-reported severity of those stressors [54]. One study from this review [46] concluded that duration of HIV infection contributed to differential DNA methylation levels, with viral DNA/RNA levels not contributing to this effect since there were no observed differences between control and experimental groups. This could imply that stress duration associated with the illness influenced these results. However, the sample size for the neurocognitive control group was 8, and 18 for the HAND group. Larger sample sizes might have increased the power and thus revealed a significant difference in viral levels; therefore, experienced stress due to illness duration might not actually be the driving factor. With depression and stress explaining some variability in the disease progression of HIV [55], it is possible that independent stressful life events might have moderated some of the effects observed with DNA methylation. This might also have been the case in one study [36] in this review which analyzed brain samples of schizophrenic patients. Larger sample sizes, along with comprehensive stress survey results, might have captured differences in DNA methylation, due to the relation of stress with psychiatric illness [14]. It is highly recommended that future studies use adequate sample sizes, so to omit this as a limiting factor in the absence of any observed effects of stress.

With the brain being more sensitive to plasticity at certain points throughout development, it is important to consider the temporal nature of stress [56]. A stressed individual during a particular developmental period may be more vulnerable to enduring epigenetic alterations that could predispose them to disease onset. Recent research with participants from the Multi-Ethnic Study of Atherosclerosis (MESA), after accounting for socioeconomic status (SES), found that living in a disadvantaged neighborhood during early-life was associated with increased DNA methylation of stress- and inflammation-related genes assessed later in life [57]. It is possible that additional changes might have been observed had more stressful experiences been assessed, with cumulative exposure previously reported to affect methylation [52]. Previous research has shown that the occurrence of three or more childhood adversities was associated with adult-onset of physical conditions such as heart disease and T2DM [58]. One study in our review utilized stressful life events questionnaires with a life-history calendar in conjunction with self-reported obstetric complications, and scores were differentiated into lower or greater stress [47]. Greater stress was associated with lower methylation in COMT and greater CI than with lower stress. However, there was no prospective study design which measured methylation before onset of any CI or subclinical CVD, so it cannot be adequately concluded whether DNA methylation is a predisposing factor for the outcomes reported or a consequence of them.

Incongruent results among studies could also be attributed to the effects of other risk factors on methylation. For CVD these include blood pressure, smoking, nutrition, pollution, cholesterol, and diabetes [59–64]. In this review, blood pressure was only accounted for in two studies [37, 38], smoking in three [26, 37, 39], cholesterol in two [38, 39], and diabetes in two [37, 39]. Information on risk factors could not be determined for two studies [36, 47]. Anxiety and depression are possible modifying factors when assessing cognitive impairment, due to the potential effects these affective disorders have on confounding cognitive testing performance [65]. One study controlled for depressive symptoms, but cognitive performance was not an outcome of interest [37].

Using traditional risk factors to predict CVD proves more difficult in explaining variability in outcomes for women than for men. Empirical evidence for coronary heart disease risk suggests that men have a greater lifetime risk than women for development of disease [66], but higher rates of all-cause mortality exist for women, at any level of coronary calcification, even after adjustment for traditional risk factors [67]. When considering both sex and risk factors in a population of diabetic subjects, men had a 2.1 risk and women a 4.9 risk for cardiovascular death, compared to non-diabetic subjects [68]. CVD typically occurs later in life for women, and there may be some sex-specific epigenetic marks, as reported previously [61]. All but two human studies [36, 37] in this review included both sexes in their study design, with one [45] finding X-linked differentially methylated loci. Only one study [50] in this report included animals of each sex (Additional file 7: Tables S6-S7). Future studies should consider recruitment of both sexes.

Ethnic and racial disparities also contribute to differences in CVD and CI risk. In a comparison of European Americans and African Americans (AA) in the Diabetes Heart Study [2], AA had increased carotid artery intima-medial thickness, as well as risk factors like smoking and low-density lipoprotein cholesterol. However, they had lower levels of coronary artery calcium and carotid calcified plaque. This differential pathogenesis might be partly due to epigenetic variations that include DNA methylation. For example, it has been shown that baseline endothelial cell gene expression in healthy young adults is differentially expressed in 31 genes between AA and European Americans [69]. This could contribute to the disparity observed in vascular-related disease among AA. In one study using participants from MESA [70], CVD risk factor ankle-brachial index (ABI) was analyzed in four different ethnic groups [71]. Controlling for traditional and novel risk factors, AA had a 1.47 greater risk for peripheral arterial disease (PAD) compared with non-Hispanic whites, followed by Hispanics and Chinese. When normal ABI levels were analyzed in regards to gender and ethnicity in a subgroup of MESA participants, women had lower ABI values than men, and blacks had lower levels than non-Hispanic whites [72]. One study [46] in this review had participants of two different ethnicities, but no sensitivity analysis was done to identify potential differences in methylation. These results further highlight the need for future studies to control for sex, race, and ethnicity, in addition to assessing differences across strata, such as that of stress on methylated genes in men and women of different ethnic groups.

Variability in methylation changes of the same gene could be attributed to the method of methylation assessment used and which cells or tissues were analyzed. For example, the study by Peter et al. [45] used the Infinium HumanMethylation450 Bead Chip Array, which detects methylation status of over 485,000 individual CpGs in 99% of known genes, including coding regions and island shores [73]. With this method, increased methylation was identified at two sites (cg06860277, cg07194846) within the promoter region of COMT, in addition to sites in shore and sea regions of various genes [74]. These sites might have been overlooked had the methylation assessment been pyrosequencing, for example, which requires initial knowledge of individual target CpG site sequences. Alelu-Paz et al. [36] reported no methylation change in MB-COMT, but analysis was restricted to brain regions. Significant methylation changes in COMT were found in studies which utilized peripheral blood mononuclear cells and whole blood [45, 47]. Results from whole blood analyses should be interpreted with caution, due to reported differences in methylation between white blood cell fractions [75]. Only one study in this review controlled for methylation across different cell fractions [45]. Moreover, Ursini et al. [47] showed that methylation at the rs4680 site and at sites within the promoter region of COMT was dependent on genotype; individuals with two Val alleles had lower methylation at all sites and greater working memory impairment. These sites are likely different than those reported to have increased methylation in the study by Peter et. al [45]. For future studies aiming to find novel genes and specific CpG sites, the best approach would be to search for differentially methylated regions across the genome, with consideration of well-established stress-related risk genotypes like that for FKBP5, NR3C1, and COMT [76].

Our findings supplement work done with epigenetic regulation of genes associated with stress, subclinical CVD, and cognition. With no single study analyzing common genes between subclinical CVD and cognition, and the high comorbid prevalence of the two, future work should aim to combine these areas (Fig. 1). There was one common gene of interest (NR3C1) which was associated with stress and both outcomes, albeit not in the same animal or study [37, 50]. Nonetheless, there lies the implication that other genes with similar association can be identified with methodologically rigorous studies. For example, although COMT methylation was only analyzed in association with CI, it is likely that studies assessing subclinical CVD may also find a relationship. One study [40] in this review found significantly increased methylation at several CpG sites in the gene Txndr, which lies adjacent to COMT in both humans and rats. Therefore, some of these CpG sites of significance might overlap if both genes were analyzed. Since common genes have been reported in both CVD and pathological disease processes like Alzheimer’s Disease [22], future studies with elderly participants may consider controlling for these processes if they wish to explore CI as an outcome that is independent of these processes.

Despite limitations, this review provides background for future population-based studies designed to identify DNA methylation patterns of overlapping genes between subclinical CVD and CI. These patterns may be critical to identifying links between stress and those vulnerable to future health disparities. We limited our literature search to Pubmed, which may have affected the number of included studies. Additionally, no comparison of statistical approaches was performed, largely because some studies lacked complete methodological information. It was beyond the scope of this review to evaluate the statistical approaches for each study, but we understand the importance of this and had there been less variability across studies, we would have addressed this. The exclusion of prenatal and parental stress may have left out pertinent data, but those data are beyond the scope of this review, which was to analyze stress exposed from early- life (neonatal) to mid-life.

Measuring life course stress with validated scales, accounting for number of stressors, utilizing genome-wide methylation analysis, controlling for sex, ethnicity, and traditional and novel risk factors, are all important considerations for future studies. Early detection of vulnerable individuals with biological biomarkers can pave the way for clinical intervention programs to reduce the burden of CVD and comorbidities highly ubiquitous today.