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

The reporting in this systematic review is consistent with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (Additional file 1: Table S1) [25]. The Cochrane Database of Systemic Reviews was checked to rule out any possibility of another review conducted on a similar topic. PubMed was the only database used to retrieve studies. Due to the lack of standardization among studies in this review (explained further in Discussion), a meta-analysis was not feasible. Our searches were not limited by year of publication since the study of epigenetics as a contributor to complex diseases is relatively new. All included studies were published within the last six years, with the exception of one published in 2008 [26]. Only articles in English and those investigating humans or other mammals were included. Animal studies were included so as to compare methylation differences observed between similar stress exposures (with humans). Subclinical CVD and cognitive impairment were not limited to any particular conditions due to the already narrowed focus of the research question and is reflected by the broad use of terms used in the search strategy. Clinical CI and CVD were excluded so that only preclinical risk factors were captured. Subclinical CVD terms are common clinical markers for CVD and were abstracted from review papers and papers cited within those papers [3, 5, 27‐29]. CI terms were those that related to cognition, without regard for any specific condition. All terms were refined using MeSH categories in the Pubmed database. A complete list of the combination of methylation terms AND either cognitive impairment OR subclinical cardiovascular disease terms is shown in Additional file 2.

With the intent to include a broad spectrum of stress exposures, terms related to stress were not specified in the search and instead studies were screened based on certain selection criteria (Additional file 3: Tables S2-S3). The primary requirement was that direct exposure to stress be experienced postnatally, from early- to mid-life. To remain within the scope of this review, parental and prenatal stress exposures were excluded. If there is a parental stressor that also affects the child/pup postnatally (e.g. maternal separation), then this is included. We defined our maximum age for mid-life to be 50 years for humans and 8 months for animals, at which point the animals would have reached social maturity [30]. We included illness as a stressor if onset occurred prior to the ages defined as mid-life. Living with chronic illness comes with emotional and physical daily stressors, contributing to varying amounts of subjective stress among individuals [14]. With the already limited scope of studies expected to be within the literature that pertained to our research question, we wished to keep the stress requirement broad. Although there may be biological effects of the illness itself that could confound any results observed and thus make it difficult to parse out what contributes to changes in DNA methylation, the key interest is consistent DNA methylation patterns among studies. Environmental chemical exposure was excluded as a stressor due to the unlikely possibility that an affected individual would have any subjective stress directly related to the awareness of any chemical exposure. Studies with the CI outcome as a symptom of a specific disease (e.g. Alzheimer’s) were excluded so that results could be applicable to the general population. Our initial search retrieved 902 articles. The first and second authors screened the titles and abstracts of all articles based on the following criteria: (1) evidence of quantitatively measured DNA methylation; (2) evidence of early- or mid-life stress; and (3) evidence of either subclinical CVD or CI. A total of 50 studies remained for further full-text screening, 14 of which were included for narrative analysis. The authors screened the bibliographic references of full-text studies, giving a final total of 16 studies. A flow diagram of this process and details on excluded studies are available in Additional file 4: Figure S1 and Additional file 3: Tables S2-S3.

General characteristics of human studies were extracted based on eight variables: 1) study design; 2) location of study; 3) sample size; 4) sex; 5) age; 6) risk factors; 7) outcome(s); and 8) quality of study (Additional files 5-6: Tables S4-S5). General characteristics of animal studies were extracted based on five variables: 1) experimental group; 2) sample size; 3) sex; 4) age at outcome assessment; and 5) outcome(s) (Additional file 7: Tables S6-S7). Information on stress exposure, outcome, and DNA methylation from each study was extracted based on six variables: 1) exposure; 2) outcome; 3) gene(s) with greatest association; 4) regions analyzed for methylation 5) methylation method; and 6) summary of associations (Tables 1 and 2). Information on human- and animal-based studies was analyzed separately.

For human studies, bias was assessed using the validated Newcastle-Ottawa Scale, which evaluates the quality of non-randomized studies [31, 32]. A modification of the case-control scoring form, adopted from a previous systematic review [33], was used for the evaluation of three cross-sectional studies. Cross-sectional studies that received a score of six stars were considered to have low risk of bias; five or four stars were of medium risk of bias; and three stars or less were of high risk of bias. For four studies that had both case-control and cross-sectional designs, the scoring forms were combined into one form. This adjustment took the original case-control form and added the category “ascertainment of outcome.” Non-randomized studies with a score of ten were of low risk of bias; eight or nine were of medium risk of bias; and seven or less were of high risk of bias. One study [26] was excluded from bias assessment because the design was a randomized controlled trial but only the baseline characteristics of participants were used for analysis. Although prospective studies would have been the best study design to assess the temporal influence of stress on DNA methylation and subsequent outcomes of interest, none qualified to be in this review.

For animal studies, bias was evaluated using the SYRCLE’s risk of bias tool, with the use of signaling questions to facilitate the assessment [34]. Analysis was based on ten entries that relate to selection, performance, detection, attrition, reporting, and other biases. Within-table markings and bias scores for each risk item category were adopted from a previous systematic review [34, 35].

Of the 16 full-text records included for narrative review, human and animal data were each extracted from 8 disparate studies. Of the eight human studies, four had CI as an outcome and four had subclinical CVD. Both male and female participants were represented in all but two studies, which had only male participants [36, 37]. Aside from one study [38] in which participants were, on average, seven years old when the study was conducted, average age was 43–69 years old. For studies with participants older than 50 years of age, age of stress exposure (including illness onset) was less than 50 [26, 36, 37, 39] (Additional files 5-6: Tables S4-S5). All except one human study [36] were judged to be of medium or high quality (Additional files 8-9: Tables S8-S9).

Of the eight animal studies, three had CI and five had subclinical CVD as an outcome. All studies were done with either mice or rats. All except one study [40], which had insufficient data, had the stressor applied before eight months of age (Additional file 7: Tables S6-S7). Overall risk of bias among studies was fairly high, with the greatest risk for allocation concealment and blinding. Low risk of bias was consistently reported for the following items: baseline characteristics, selective outcome reporting, and other bias. Risk items with unclear bias were random housing and random outcome assessment (Additional files 10-11: Tables S10-S11).

In human studies, two conditions analyzed as stressors were schizophrenia and Human Immunodeficiency Virus (HIV) infection [36, 46]. Stressful experiences included postnatal malnourishment [45], and obstetric complications and/or stressful life events (self-reported) [47]. In animal studies, stressors included chronic variable stress [48], repetitive stress [49], and maternal separation [50]. Presence or absence of control groups, as well as risk factors controlled for, are shown in Additional file 6: Table S5.

In one of the three human studies analyzing COMT, methylation in the promoter region was increased and moderately associated with ADHD index (measure of attention) in adults who were malnourished during the first year of life [45]. In another study there was significantly reduced methylation at the site of the Val158 allele of the rs4680 single nucleotide polymorphism (Val158Met) in association with greater impairment in working memory performance [47]. The Val158 allele creates a CpG site, whereas the alternate allele (Met) does not. A different study analyzing methylation of MB-COMT, which is the long form isoform of COMT, found no association between adult-onset schizophrenia, CI, and methylated CpGs in the dorsolateral prefrontal cortex (DLPFC) [36]. This study was considered to be of high risk of bias, with sample size being a large issue of concern (Additional file 9: Table S9). With only 3 DLPFC control samples to compare to an experimental sample size of 15, it is unlikely that any effect could be observed. In the same study, DISC1, DRD1, GABBR2, HINT1, RELN, and SLC6A4 were analyzed in the hippocampus, cerebellum, or anterior cingulate cortex, but no significant associations were observed. SLC6A4 codes for the serotonin receptor and methylation changes in this gene have been implicated in mice socially isolated as adults [51]. Other genes analyzed by studies were ZBTB9 and SYNGAP1, with significantly increased methylation related to higher intelligence [45], but after adjusting for socioeconomic status, significance in CpG sites in the island and shore regions of ZNF57 was lost. An additional 15 X-linked differentially methylated loci in males and one in females were associated with both attention and intelligence. Details on these X-linked loci, as well as on a myriad of additional genes with significant methylation changes, including ABCF1, IFNG, MIR200B, VARS, and WT1, can be found in the supplemental files of Peter et al. [45]. In another study, HIV-associated neurocognitive disorders (HAND) was correlated with increased methylation of 363 CpGs genome-wide in the occipital cortex [46]. However, this was for overall cognitive function and not for individual cognitive domains (learning, memory, attention, motor) (Table 2).

Among animal studies, four genes were analyzed for methylation changes. In animals who were physically, psychologically, and socially stressed, reduced methylation of CpGs located at tss-554 and tss-506 of the Bace1 promoter region in the left side of the hippocampus was negatively correlated with memory performance [48]. In the prefrontal cortex, reduced methylation was examined at tss-506 and tss-518, but the latter occurred only among aged stressed mice (vs. young stressed mice). Lower methylation was observed at tss-506 and tss-554 in the amygdala, with the latter again only seen among aged stressed mice. In the same study, tss-109 of Bdnf exon 4 region had increased methylation in the hippocampus of aged mice, and no effects were observed for Gsk3b [48]. In a study by Zhu et al. [50], 14 out of 17 CpG sites within the promoter region of exon 1₇ of Nr3c1 had increased methylation in the hippocampus of rats with neonatal maternal separation, which was associated with memory impairment [50]. In comparison, global methylation in the hippocampus was negatively correlated with recognition memory in another study where rats were repetitively stressed; however, no specific genes were further identified [49] (Table 2). These results are consistent with previous research showing that regulation of learning, memory, and basic brain function is associated with methylation of stress response genes in the hippocampus, and stressors like childhood abuse and social environment can modify this methylation [19, 52]. Figure 2 illustrates the overlap of genes among studies with CI as an outcome.