A systematic review of childhood maltreatment and DNA methylation: candidate gene and epigenome-wide approaches

Childhood maltreatment is a major risk factor for chronic and severe mental and physical health problems across the lifespan. Increasing evidence supports the hypothesis that maltreatment is associated with epigenetic changes that may subsequently serve as mechanisms of disease. The current review uses a systematic approach to identify and summarize the literature related to childhood maltreatment and alterations in DNA methylation in humans. A total of 100 empirical articles were identified in our systematic review of research published prior to or during March 2020, including studies that focused on candidate genes and studies that leveraged epigenome-wide data in both children and adults. Themes arising from the literature, including consistent and inconsistent patterns of results, are presented. Several directions for future research, including important methodological considerations for future study design, are discussed. Taken together, the literature on childhood maltreatment and DNA methylation underscores the complexity of transactions between the environment and biology across development.


Introduction
Childhood maltreatment is a highly prevalent public health problem that often has devastating effects on physical and mental health. There are now numerous studies linking childhood maltreatment and other early adversities to nearly all forms of mental illness, as well as chronic medical conditions that cut across multiple organ systems [1][2][3][4][5][6][7] . Unfortunately, the negative sequelae of childhood maltreatment often begin early in life, and persist across adulthood 1,6 , posing risk for premature mortality. Indeed, adults with numerous adverse experiences in childhood die nearly 20 years earlier than those with no early adversity 1 . Understanding the mechanisms of biological influence is therefore critical to developing innovative targets for intervention to enhance health outcomes among highly vulnerable children with this major adverse exposure.
Over the past decade a rapidly growing body of literature has underscored the significant role of epigenetics in the sequelae of childhood maltreatment. Epigenetic processes allow the body to respond to environmental influences by altering gene expression through chemical modifications that regulate chromatin structure and/or DNA accessibility without inducing changes to the DNA sequence 8 . DNA methylation is among the most commonly studied epigenetic processes and involves the addition of a methyl group at sites in the DNA where a cytosine nucleotide occurs next to a guanine nucleotide (CpG dinucleotides). Initial work focused on understanding associations of early life stress and DNA methylation examined these processes in rodents and found that low levels of maternal care (licking and grooming and arched-back nursing) was associated with greater methylation of the glucocorticoid receptor (GR) gene in offspring 9,10 . Furthermore, this work suggested that methylation was a mechanism of the effect of maternal care on the offspring HPA stress response. Emerging from this groundbreaking work with animal models, some of the earliest studies in humans documenting altered DNA methylation in association with childhood maltreatment focused on the glucocorticoid receptor (GR) gene, NR3C1, which modifies responsiveness of the HPA axis to stress exposure. Our group was the first to demonstrate altered leukocyte DNA methylation of NR3C1 in adults with childhood maltreatment 11 , associations of NR3C1 methylation with behavior problems and symptoms in children with early adversity 12 , and maltreatment as a predictor of change in NR3C1 methylation over time 13 . In 2016, we completed a literature review focused on altered methylation of glucocorticoid signaling genes in association with childhood maltreatment 14 . Several other recent reviews have also focused on methylation of glucocorticoid signaling genes [15][16][17] , as well as methylation of genes associated with the serotonin system 18,19 in relation to adverse exposures. We expand upon these prior reviews to consider the recent explosion of research in this area that has examined additional genes as well as epigenome-wide effects. Cecil and colleagues recently conducted a systematic review of research focused on childhood maltreatment, specifically experiences of abuse and neglect, and DNA methylation 20 . The current review provides an important addition to the literature by utilizing a broader conceptualization of childhood maltreatment that includes any experience that involved potential for harm to the child. Our systematic approach to identify and summarize the literature related to childhood maltreatment and other adversities in association with DNA methylation captures research through March, 2020 and includes research focused on candidate genes and studies that leveraged epigenome-wide data in both children and adults.

Methods
In accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines 21 , we conducted a systematic review of human studies investigating the relationship between childhood maltreatment and other adversities and DNA methylation. For the purposes of this review, childhood was defined as events occurring between birth and age 18. Maltreatment was defined as any experience that involved potential for harm to the child, including emotional, sexual, and physical abuse, and emotional and physical neglect. Other adversities included exposure to intimate partner violence or other violence, early parental death or separation, institutional deprivation, and indentured labor. Empirical articles investigating DNA methylation in association with other forms of childhood stress such as low socioeconomic status or famine, without consideration of other adversities, were not included. Studies focusing on experimentally induced stressors were also not included. In addition, because our focus was on childhood maltreatment after birth; we also did not include studies focusing on prenatal stress or substance use and DNA methylation, which have been addressed in other reviews 22,23 .
Studies were identified by searching PubMed/Medline for empirical articles published in March 2020 or earlier using the following search terms: (((methylation OR epigenetic) AND (child AND (maltreat OR adversity OR trauma OR early life stress OR abuse OR neglect OR ACE))), filtered by Humans. Additional searches in PubMed/Medline for studies published during this timeframe were also conducted using the following search terms: (1) emotion regulation AND methylation; (2) abuse AND methylation NOT substance NOT alcohol; (3) trauma AND methylation; (4) early life stress AND methylation; (5) maltreatment AND methylation; (6) adversity AND methylation. Additional articles published during this timeframe identified through other sources (Google Scholar alerts) were also included.
Our qualitative synthesis included empirical articles that: (1) examined the relationship between childhood maltreatment/adversity and DNA methylation; (2) provided statistical indicators to examine the impact of maltreatment during childhood on DNA methylation; and (3) described maltreatment/adversity that occurred prior to age 18. Empirical articles were excluded if they: (1) did not provide information on childhood maltreatment/ adversity and DNA methylation; (2) childhood maltreatment/adversity was not clearly distinguished from adult adversities in study measurement/hypothesis testing; (3) only quality (or quantity) of parental care or support was included as a predictor; (4) were not written in English; (5) were only conducted with animals; and (6) did not use a quantitative approach to summarize research findings (i.e., were qualitative, reviews, comments, or other editorials). We did not omit articles based on small sample size; this information is included in the tables so that it can be considered when evaluating results. Figure 1 documents the methods of our systematic review to generate the final empirical articles included in our qualitative synthesis. Of note, consistent with PRISMA guidelines, records screened includes screening of both titles and abstracts.

Results
A total of 100 empirical articles were identified in our systematic review focused on the relationship between childhood maltreatment and DNA methylation. This includes 69 empirical articles focused on candidate genes and 31 empirical articles that leveraged epigenome-wide data. Twenty-eight empirical articles measured DNA methylation in childhood and 72 measured DNA methylation in adulthood. Findings are summarized in Tables 1-4, and examples are described below. Major results of each empirical article are included in the tables, though results of specific CpG sites are not detailed given significant design and analytic variability across studies. Given the focus of the current review, only the major study findings related to the association of childhood maltreatment and other adversities with DNA methylation are included in the tables. Empirical articles that focused on another condition (e.g., parental substance use, psychiatric conditions, as described in Sample description in the tables) but included a result pertaining to maltreatment are presented. Importantly, several empirical articles drew upon overlapping samples. We included information regarding the name of the study in the tables if the study name was provided in the empirical article. However, we were unable to denote all articles that may have utilized overlapping samples as the relevant information was not consistently provided in the literature.

Childhood maltreatment and methylation of candidate genes Children
As displayed in Table 1, 16 empirical articles focused on childhood maltreatment and methylation of candidate genes in children. Most of these studies involved saliva DNA, but one used buccal cell DNA and several examined DNA from blood. The most commonly studied candidate genes were those that regulate glucocorticoid signaling, including NR3C1 which encodes the glucocorticoid receptor (GR) (six studies represented in eight empirical articles) and FKBP5 which modulates sensitivity of the GR 24 (two studies represented in three empirical articles).
Most studies of children support the hypothesis that childhood adversity is associated with higher levels of methylation of NR3C1. In our own study of preschoolers, we found that childhood maltreatment status and the number of stressful life events was associated with greater NR3C1 methylation 25 , and NR3C1 methylation mediated effects of maltreatment and stressful life events on child internalizing behavior problems 12 . Likewise, in 534 school-aged children with low socioeconomic status, children who experienced maltreatment beginning in  Children who were exposed to CM had higher methylation than children not exposed to CM; methylation was negatively correlated with gray matter volume in the left orbitofrontal cortex; children exposed to CM showed lower gray matter volume compared to the non- infancy or early childhood had greater NR3C1 methylation than children who had no maltreatment history 26 .
Interestingly, there were no differences between children who experienced maltreatment that started in the preschool period or later, and those children with no maltreatment history. However, more chronic maltreatment and more types of maltreatment, were associated with greater NR3C1 methylation. Using data from the Bucharest Early Intervention Project, time in institutional care was associated with lower levels of methylation of FKBP5, which modulates sensitivity of the GR, when children were 12 years old 27 . Lower levels of FKBP5 methylation in association with childhood maltreatment were also observed in our sample of preschoolers 28 .
Another study of Tanzanian children found evidence that maltreatment was associated with hypermethylation of CRH, the gene that encodes the hypothalamic corticotropin releasing hormone (CRH) and its downstream target, POMC, which encodes the pituitary precursor of ACTH 29 . Taken together, these findings indicate that childhood maltreatment may be associated with altered methylation of genes that regulate the child stress response, suggesting a possible mechanism for the effect of maltreatment and other adversities on poor health outcomes. Epigenetic alterations with childhood maltreatment have been documented in additional candidate genes. Methylation of serotonin signaling genes, including SLC6A4, which encodes the serotonin receptor, and HTR2A, the gene that encodes the serotonin receptor subtype, 5-HT 2A , have been investigated in association with childhood maltreatment and other interpersonal adversities 27,[30][31][32] . Childhood adversity was associated with greater SLC6A4 methylation in two studies 31,32 , but lower levels of methylation in association with institutional care 27 . Using a subsample (n = 785) of the Avon Longitudinal Study of Parents and Children (ALSPAC), Barker et al. 33 generated an inflammation-related epigenetic polygenic risk score (i-ePGS). Adversity between birth and 7 years of age was related to higher i-ePGS, which, in turn, was an indirect pathway by which adversity was associated with internalizing behavior problems in later childhood. The receptor gene for the pituitary hormone oxytocin (OXTR), best known for its role in social behavior and attachment, has also been studied in relation to early adversity in children. Fujisawa et al. 34 demonstrated greater methylation of OXTR in Japanese children and adolescents with a maltreatment history (n = 44) compared with controls (n = 41), and found that OXTR methylation was negatively associated with gray matter volume in the left orbitofrontal cortex.

Adults
As displayed in Table 2, 53 empirical articles focused on childhood maltreatment and methylation of candidate genes in adults. Most studies involved analysis of DNA from blood, with a handful of studies examining DNA from saliva or buccal cells, and a few studies of DNA from brain tissue. The most commonly studied candidate genes included NR3C1 (20 empirical articles), FKBP5 (eight empirical articles), SLC6A4 (six empirical articles), and OXTR (six empirical articles).
Taken together, the majority of studies focused on NR3C1 methylation observed significant associations of childhood maltreatment and methylation of this gene, but there is considerable variability in the findings. Several studies found increased NR3C1 methylation with childhood maltreatment, but others found no effect, or reduced methylation with maltreatment. Melas and colleagues 35 found that childhood adversity was associated with saliva DNA NR3C1 hypermethylation, with childhood parental loss linked to higher methylation near a transcription factor binding site. Two reports analyzed DNA from the Quebec Suicide Brain Bank in relation to childhood maltreatment retrospectively assessed via psychological autopsy, including a group with suicide and childhood maltreatment, a group with suicide but no maltreatment, and a control group. The suicide with maltreatment group had greater hippocampal NR3C1 methylation and reduced NR3C1 expression across exon 1 F 36 , reduced NR3C1 gene expression across exons 1 B , 1 C , and 1 H , and reduced NR3C1 methylation at exon 1 H 37 , in comparison with the other two groups. In contrast, both suicide groups differed from controls in regions of exon 1 B , with hypermethylation at two CpG sites and hypomethylation at one CpG site 37 . In a diverse sample of adults, childhood maltreatment was associated with greater methylation of leukocyte NR3C1, and maltreatment was associated with lower NR3C1 gene expression 38 . Likewise, in two samples with psychiatric disorders, childhood maltreatment was associated with greater NR3C1 methylation 39,40 . In contrast, in a recent study, we found that early adversity, and the number of adverse exposures, was associated with lower NR3C1 methylation in healthy adults 41 . Other studies found no association of maltreatment and NR3C1 methylation [42][43][44][45][46] , but one of these found that NR3C1 methylation moderated effects of maltreatment on cortisol response to the Trier Social Stress Test.
The literature focused on FKPB5 methylation in adults also demonstrates inconsistent effects. Klengel et al. 47 found that childhood maltreatment was associated with lower methylation of FKBP5 in adults with the rs1360780 T risk allele, and this effect was observed in both a subsample of participants from the Grady Trauma Project and a replication sample. This effect was more recently replicated by Tozzi and colleagues 48 who also reported that childhood maltreatment was associated with lower FKBP5 methylation among those with the T risk allele of       the rs1360780 SNP. In contrast, Harms et al. 49 found significant positive associations between stress in childhood and methylation of FKPB5 in young adulthood using a prospective longitudinal design, and a number of studies did not find significant associations among childhood maltreatment and FKBP5 methylation 50-53 , or moderation by FKBP5 risk allele 50,51 .
Turning to the serotonin system, several studies have examined whether childhood maltreatment is associated with methylation of SLC6A4, the gene for the serotonin transporter. Most studies have found a positive association of childhood maltreatment with methylation of regions of this gene [54][55][56][57][58] although some studies show equivocal or trend-level effects 59,60 or no effect 61 of maltreatment on SLC6A4 methylation.
Based on the hypothesis that oxytocin may play a protective role in the biological response to stress and trauma 62 , five studies investigated methylation of the oxytocin-receptor-gene (OXTR) in adulthood; all five reported no overall association of childhood maltreatment with OXTR methylation. A number of these studies did report indirect or moderation effects. In a sample of 309 African American men, childhood adversity had a significant indirect effect on OXTR methylation through socioeconomic instability 38 . No significant direct effect of childhood adversity on OXTR methylation was observed. Likewise, while Smearman and colleagues 63 did not observe simple associations of child abuse history and OXTR methylation when accounting for multiple comparisons in their sample of 393 African American adults, OXTR methylation moderated the association of childhood abuse and psychiatric symptoms.
Several other candidate genes have been examined in individual studies. For example, childhood trauma was associated with lower methylation of the proinflammatory IL-6 promoter, and lower methylation in turn was associated with greater salivary IL-6 in response to the Trier Social Stress Test 64 . Although this study drew upon a relatively small sample, the findings are consistent with other work demonstrating effects of childhood trauma and maltreatment on inflammatory processes.

Childhood maltreatment and epigenome-wide association studies
Arising in part out of replication inconsistencies as well as interest in identifying novel variation, the past decade has included an increase in the number of Genome Wide Association Studies (GWAS), which are agnostic by design and interrogate genetic variation across millions of common genetic variants the entire genome. However, a limitation of GWAS is the extremely large samples needed to detect effects after adjusting for multiple testing. Similar to GWAS approaches, epigenome-wide association studies interrogate markers across the entire genome.
Consideration of the biology of DNA methylation as well as optimization of power has led to the development of new frameworks for analyzing multiple markers, including examination of CpG sites that are in close proximity to one another (differentially methylated regions, DMRs), examination of DNA methylation in biological pathways known to be involved in the condition, and application of analytic tools that use a ranking approach rather than relying on p values. Researchers have also developed quality control standards and approaches for addressing both concerns with Type I error as well as population stratification that may arise in diverse samples 65 .
Given that biological processes function as part of a larger interrelated system, research has increasingly focused on how either systems or patterns of alterations in methylation may be important in efforts to understand epigenetic modifications. Studies have begun to explore DNA methylation as a measure of molecular aging, with evidence that this epigenetic "clock" is associated with age-related disorders and mortality 66-70 as well as with age-related development such as menopause 71,72 . Several analytical methods have been developed to measure epigenetic age as indicated by epigenome-wide methylation profiles [73][74][75][76][77] . This emerging area of research has the potential to provide exciting evidence for understanding the impact of early adversity on molecular age.

Children
As displayed in Table 3, 12 empirical articles focused on epigenome-wide effects of maltreatment and other adversities in children. These studies suggest that maltreatment is associated with variation in methylation across the genome, and several differentially methylated regions and genes have been identified. In a racially and ethnically diverse sample of 548 children, Cicchetti et al. 78 found differential whole-genome methylation in children with a maltreatment history relative to children with no maltreatment. Maltreated children had higher levels of methylation at sites where methylation was generally low, and lower methylation at sites where methylation was generally high, compared to non-maltreated children. Using data from a subsample (n = 774) of the ALSPAC, Dunn et al. 79 examined associations among childhood adversity and epigenome-wide methylation in children at 7 years of age. Thirty-eight differentially methylated CpG sites were identified in association with early adversity, and the developmental timing of adversity was the most salient predictor of methylation. Neither the simple presen-ce/absence of adversity, recency of adversity, nor accumulation of adverse experiences was associated with altered methylation.
Studies of early adversity and epigenetic age are just emerging in samples of children. Very recently in a sample of 247 children who were 8-16 years of age, threat-related  early adversity (including abuse and violence exposure), but not deprivation (including neglect), was associated with accelerated DNA methylation age 80 . Furthermore, threat-related early adversity exerted a significant indirect effect on depressive symptoms through accelerated DNA methylation age, suggesting that this measure of early adversity is clinically relevant to psychiatric outcomes. This is consistent with Jovanovic et al. 81 that demonstrated that violence exposure was associated with greater DNA methylation age acceleration in African American children who were 6-13 years of age.

Adults
As displayed in Table 4, 19 empirical articles focused on epigenome-wide effects of childhood maltreatment in adults. Collectively, these studies suggest that childhood maltreatment exerts epigenetic effects across the genome, yet these studies differed in their overall focus and methodology. In the ALSPAC and the MRC National Survey of Health and Development, Houtepen et al. 82 identified nine differentially methylated regions (DMRs) in the genome that replicated across the two cohorts and were associated with childhood adversity. No individual CpG sites in their epigenome-wide analysis replicated across the cohorts. O'Donnell et al. 83 used a different methodological approach to describe variation in DNA methylation, and found that childhood maltreatment was associated with variation in methylation at 27 years of age utilizing principal components scores to describe variation in methylation. Lutz et al. 84 examined genome-wide DNA methylation and gene expression in postmortem brain samples of adults with a history of depression who died by suicide. This study found that child abuse history was associated with differential methylation specifically in oligodendrocytes in the cingulate cortex, as well as expression of myelin-related genes.
Investigations using epigenome-wide data to explore effects of early adversity on accelerated epigenetic aging have shown mixed outcomes. Lawn et al. 85 used the ALSPAC and the MRC National Survey of Health and Development to examine associations of childhood psychosocial adversity, including abuse and neglect, and DNA methylation age acceleration. Childhood sexual abuse was associated with methylation age acceleration in ALSPAC, and this effect remained significant when controlling for socioeconomic position. Data regarding sexual abuse was not available in the MRC survey, however Tamman et al. 86 also found that childhood sexual abuse was associated with increased DNA methylation age. Neither individual adversity types nor a cumulative measure of adversity were associated with methylation age acceleration in ALSPAC and the MRC National Survey of Health and Development 85 . Han et al. 87 also reported a positive association between a history of childhood trauma and epigenetic aging in adults with MDD. DNA methylation age was also examined in the sample of 27 year olds described above, with analyses revealing no differences in epigenetic age as a function of childhood adversity 83 . Although there have been more studies focused on childhood adversity and accelerated epigenetic age in adults, few studies overall have been completed.

Discussion
This systematic review examined associations of childhood maltreatment and DNA methylation in children and adults. One hundred empirical articles focused on humans were identified. These studies included both candidate gene and epigenome-wide approaches. Strengths of the literature included: (1) rigorous approaches to measure childhood maltreatment in studies focused on DNA methylation in children, including record review methods; (2) several racially and ethnically diverse samples in the child studies; (3) diverse and innovative approaches to measuring epigenome-wide effects of maltreatment, including exploration of how early adversity may lead to epigenetic age acceleration; and (4) several replication studies focused on childhood maltreatment and methylation of glucocorticoid signaling genes in children and adults. Collectively, these studies provide evidence that childhood maltreatment and other adversities are associated with DNA methylation. Genes and pathways observed to have altered methylation in relation to childhood maltreatment, and the expected epigenetic pathways from maltreatment to health and mental health outcomes, are displayed in Fig. 2.
Studies of childhood adversity and DNA methylation often focused on methylation of candidate genes that regulate glucocorticoid signaling including NR3C1 and FKBP5. Most studies in children demonstrated increased NR3C1 methylation with maltreatment. In adults, several studies documented greater NR3C1 methylation in those exposed to childhood maltreatment, whereas others demonstrated no associations of maltreatment and NR3C1 methylation. Furthermore, in our own recent work we found lower NR3C1 methylation in association with childhood adversity 41 , and other studies have identified CpG sites in NR3C1 that are hypomethylated in association with maltreatment or suicide 37 and posttraumatic stress disorder (PTSD) 88,89 . For FKBP5, Klengel and colleagues first published findings of an interaction of childhood maltreatment and FKBP5 genotype such that maltreatment was associated with lower FKBP5 methylation in adults with the T risk allele 47 . Lower FKBP5 methylation with maltreatment-although not interacting with genotype-has been seen in the two studies with published data in children. The maltreatment by genotype interaction predicting lower FKBP5 methylation in adults was recently replicated 48 , but other studies have not Whole blood Child abuse and neglect retrieved from substantiated reports from child CM, and participation in NFP, were associated with variation in methylation observed an effect of maltreatment, or an interaction of maltreatment and genotype, and one study of young adults found that stressful life events in childhood were associated with greater FKBP5 methylation, which in turn mediated effects of early stress on prefrontal brain activity 49 . Many studies in adults utilized samples with chronic disease and mental health diagnoses, and did not exclude participants with consistent medication use, which may have influenced the findings. Overall levels of methylation of the NR3C1 gene promoter are generally very low, and this may also limit the ability to reliably detect group differences. However, there is evidence that NR3C1 methylation at exons 1 F , 1 B , and 1 C , is inversely associated with GR gene expression 36,37 . Since GR mediates negative feedback to the HPA axis, higher methylation-and lower GR gene expression-is expected in conditions with elevated cortisol and there is some evidence for this 90,91 . But variable findings for methylation of NR3C1 and FKBP5 are perhaps not surprising given that maltreatment, trauma, and stress-related psychiatric disorders have been linked to both exaggerated and diminished basal and provoked cortisol concentrations, which may in part depend on measured or unmeasured individual and trauma characteristics 92,93 . Moreover, many genes and gene products regulate HPA axis function, and complex interactions among participant characteristics (including age and genetic ancestry), genotype (with allele frequencies that may differ based on genetic ancestry), and stress exposure may best characterize these effects.
Interaction effects may also characterize methylation of oxytocin system genes, given findings of indirect or moderation effects in the absence of simple effects 63,94 . Although the potential risk for statistical artifacts provide important cautions to investigations exploring interaction effects, interactions and conditional effects are known to occur in biological systems and may be appropriate to examine in large samples and/or using stratification or model invariance strategies. Methylation of serotonin signaling genes was also examined in multiple studies.
Most studies of adults and some work in children found elevated methylation of SLC6A4 with maltreatment, but there are also reports of lower SLC6A4 methylation at some CpG sites with adversity, or no difference between adversity groups. Differences in epigenetic findings in SLC6A4 as well as other candidate genes may also be related to inconsistencies in the ways that the number of differentially methylated regions are assessed. For example, SLC6A4 has 81 CpG sites with studies adopting a range of approaches to assessing this large region, including selection of target sites based on the literature, multiple testing of sites, and strategies for binning methylation at multiple sites 95 . Given evidence for ancestry-related differences in methylation, differences across studies related to ancestry may also potentially contribute to apparent inconsistencies 96 . Studies using agnostic approaches to exploring multiple markers across the epigenome also suggested that maltreatment is associated with variation in methylation, though these epigenome-wide studies have generally failed to identify commonly studied candidate genes, leading to questions about candidate approaches. Several reasons for these inconsistencies have been advanced, including large effect sizes needed to detect an effect after multiple testing corrections, biological considerations such as the potential combinatory impact of genotype and DNA methylation (see ref. 83 for an example of integration of genotype and epigenetic influences), interactions among systems of genes, and differences in aspects of the phenotyping and heterogeneity of different studies. Due to traditional approaches to corrections for multiple testing, studies that analyze methylation of multiple markers across the genome tend to require large samples which are often heterogeneous and may have less intensive phenotyping and measurement of maltreatment and other relevant exposures. Promising markers identified with epigenome-wide approaches show some consistency in terms of their roles in neural cell development (BDNF, KITLG, and POU3F1), signaling and apoptosis (LINGO3 and 2NPFF2), neural influences on movement (ALS2), neuroinflammation (ITGB1), and some tentative evidence for immune markers (CXCL1).
A recent area of epigenome-wide research involves DNA methylation as an indicator of molecular age. A number of analytic strategies have been used to examine how early life trauma (and/or symptoms associated with trauma such as depression or post-traumatic stress disorder) may accelerate aging. A number of adult and child studies find age acceleration among individuals exposed to maltreatment (meaning that DNA methylation age is older than the chronological age). Childhood maltreatment and other adverse exposures alter epigenome-wide profiles, thereby likely contributing to chronic disease and physical aging [97][98][99][100][101] . Interestingly, the particular markers observed to be relevant in methylation age acceleration research may point toward biological systems implicated in the link between maltreatment and disease. For example, the method described by Horvath includes CpG sites enriched for glucocorticoid response elements 99 . New methods for capturing DNA methylation age continue to be developed, and very recently Belsky and colleagues 102 reported a new algorithm that was derived from longitudinal data of 18 biomarkers of organ-system integrity to capture the rate of aging up to the time of measurement. This new algorithm, but not other previously established measures of DNA methylation age, was associated with childhood maltreatment in sample of 1658 young adults who were longitudinally followed since childhood. As new methods to capture age acceleration continue to develop, systematic reviews and meta-analytic approaches will continue to be important to synthesize associations of child maltreatment and the range of measures of methylation age.
Studies of maltreatment and methylation in children were often characterized by careful measurement of adverse exposures using structured record review techniques and in-depth interview methods, and they often utilized prospective and longitudinal designs. In children, DNA methylation was most frequently measured in saliva or buccal epithelial cells, with fewer studies measuring methylation in blood. Many of the samples were racially and ethnically diverse, and these studies covered the full developmental spectrum, from early childhood through adolescence. In contrast, nearly all the adult studies drew upon retrospective reports of childhood maltreatment. Utilization of prospective longitudinal designs and record review techniques would address a significant gap in the literature. Furthermore, in contrast to studies of maltreatment and methylation of candidate genes in children, most adult studies measured methylation in leukocyte DNA.
Moving beyond simple effects of maltreatment on DNA methylation, several studies highlighted the importance of the developmental timing of exposure to adversity 26,27,78,79 . More research is needed to understand how early adversity at each developmental epoch may be associated with differences in epigenetic marks. Moreover, longitudinal research is needed to explore whether epigenetic changes secondary to early adversity may "reset" during later developmental periods or perhaps may be impacted by later experiences, which either attenuate or compound early adversity. Retrospective research with adults who report childhood adversity has numerous benefits in terms of cost-and time-effectiveness; however, reliance on retrospective reports of childhood adversity present important methodological challenges related to recalling the timing of childhood events, ability to recall very early life events, and systematic recall biases, introducing measurement limitations that are difficult to overcome 103 . Many studies use the Childhood Trauma Questionnaire 104 or other self-report measures. Some self-report measures, such as the Traumatic Life Events Questionnaire 105 and the Maltreatment and Abuse Chronology of Exposure 106 , include trauma characteristics such as the age or frequency of occurrence, which may be important determinants of epigenetic and phenotypic outcomes. Future research may also benefit from interview approaches that can help with retrospective recall such as participant-tailored anchoring.
Although many adult studies reviewed here relied on recall, much can be learned from the few studies that capitalized on longitudinal cohorts. For example, Harms et al. 49  Likewise, very few studies drew upon repeated assessments of DNA methylation over time. In our own research, we demonstrated that childhood maltreatment and other adversities are associated with change in methylation of glucocorticoid signaling genes over time in early childhood 13,107 . Future research should further examine maltreatment as a predictor of change in methylation across development.
Importantly, the majority of studies focused on exposure to childhood maltreatment without consideration of the prenatal environment. Prenatal exposures exert epigenetic effects on several biological systems, particularly the development of the child stress response. For example, intimate partner violence in pregnancy is associated with increased NR3C1 in late childhood and adolescence 108 . Smoking and depression in pregnancy have also been associated with altered methylation of placental stress-related genes, such as NR3C1 and HSD11B2, which encodes the enzyme that inactivates cortisol [109][110][111][112] . Associations of childhood maltreatment and DNA methylation may be partially accounted for by prenatal environmental factors. Conversely, maltreatment may exert a unique and independent effect on DNA methylation above and beyond prenatal exposures. Barker et al. 33 found that child adversity between birth and 7 years of age was associated with an inflammation-related epigenetic polygenic risk score (i-ePGS) at age 7, but there was no association of prenatal adversity and the i-ePGS. Future research should disentangle prenatal exposures and adversities experienced in childhood to better understand effects of maltreatment on DNA methylation in both childhood and adulthood.
Very few studies examined sex differences in the effects of childhood maltreatment on DNA methylation. Studies focused on in utero stress exposure often find sex differences in epigenetic pathways and observed outcomes in both preclinical and human models [113][114][115] . For example, Braithwaite et al. 116 found that maternal depressive symptoms in pregnancy were associated with greater methylation of NR3C1 in male, but not female, infants. Stroud et al. 109 found that the moderating effects of placental HSD11B2 methylation on links between prenatal major depressive disorder and infant cortisol response emerged most strongly for newborn daughters, whereas direct and moderating effects of SLC6A4 gene expression were evident only for sons 109 . In one of the few studies identified in this review that examined sex differences, Cicchetti et al. 78 found that boys and girls showed different directions of the effect of maltreatment on methylation of ALDH2, a gene that encodes a key enzyme in the metabolism of alcohol. Sex differences were also observed in the effect of the developmental timing of adversity. As we have shown in a meta-analysis with ADCYAP1R1, the adenylate cyclase activating receptor gene associated with PTSD, depending on the functional outcomes of the gene(s), sex and developmental differences are reasonable to expect 117 . Future work on maltreatment and DNA methylation should consider the role of child sex to ensure that important moderation effects are not being overlooked.
An additional future direction is to move from association to causal inference. Maltreatment is often confounded with additional measures of adversity, including poverty and other sociodemographic factors, as well as personality factors, and potentially genetic factors. Many studies reviewed utilized statistical control for potential confounding factors; however, true random assignment designs are not ethical in humans. Thus, synthesizing across preclinical studies (where random assignment is possible) and human association studies will be critical. Additionally, innovative designs in human studies, including random assignment to interventions to reduce maltreatment, and control groups that are demographically and psychosocially matched to maltreatment groups, may allow the field to move closer to causal inference. Research considering intervention effects on change in methylation over time may also address this gap in knowledge. Indeed, we found that service utilization was associated with increases in FKBP5 methylation over time in preschoolers with early adversity 107 . More recently, Vinkers et al. 118 observed changes in methylation among soldiers successfully treated for PTSD such that changes in methylation were observed among soldiers who had reductions in PTSD symptoms. Taken together, this work provides initial evidence that psychosocial interventions exert influence on the epigenome. DNA methylation has been measured in several tissue types, including blood, saliva, and buccal cells. Yet, the majority of studies focused on DNA methylation in children utilized saliva and buccal cells, and fewer examined methylation in blood. Although some researchers have questioned the value of peripheral markers in research aimed at understanding psychobehavioral outcomes believed to be related to central brain processes, with some researchers suggesting that psychiatric epigenetic research should be limited to brain tissue, a number of studies point to the value of peripheral indicators. Research has shown reasonable concordance between gene-expression signals in blood and brain [119][120][121] and primate research has identified correspondence in DNA methylation profiles in the brain and blood 122 . Similar gene expression was found in the cerebellum and peripheral blood mononuclear cells (PBMCs) across 4000 unique genes 123 . Correspondence between DNA methylation in PBMCs and postmortem brain tissue was identified with respect to a marker of reward and stress-induced responses 124 . Recent research showed good correspondence of DNA methylation in blood and saliva 125 . Studies of methylation age have also shown consistency across tissue types 75 . Interestingly, there is some evidence 126 that saliva samples may be more strongly associated with brain methylation than blood samples, although correlations of methylation in the brain with saliva, blood, and buccal cell DNA were all observed to be high and the strength of these associations appear to depend on the genomic region of interest 127 . These studies have provided helpful signals regarding the utility of blood, brain, saliva, and buccal cells, and Epigenomic Roadmap datasets now provide some insights into the ways that peripheral methylation profiles may map on to those in brain tissue 128,129 . Nonetheless, cell type heterogeneity remains a significant challenge for epigenetic research that has been addressed using a number of strategies such as cell counts, cell separation and examination of single cells, and analytic strategies that account for cell type [130][131][132][133] .
Another concern involves the technology and analytic strategies used to interrogate genome methylation. Power to detect true differences and false positives are both major concerns, so large samples sizes and replication samples are required, but very large studies may not be able to provide rich data on maltreatment. Researchers have also described the importance of considering reliability of BeadChip technology, the most frequently applied technology for interrogating the genome, as a function of: (a) sample type, with lower reliability and replicability in dried blood spots, (b) tissue type, with some probes demonstrating greater cross-tissue concordance, (c) platform (such as Ilumina Infinium HumanMethylation450 BeadChip with nearly 500,000 sites vs Ilumina Methyla-tionEPIC BeadChip with nearly 850,000 sites), and (d) observed variability at each site, with lower reliability at sites with less variability [134][135][136][137] . Research is needed to carefully document reliability considerations for proper assessment of reproducibility; interested readers are directed to a thorough discussion of reliability, replication, and reproducibility in DNA methylation measurement (see ref. 134 ). In addition to the importance of clear site level documentation of findings in publications, suggestions for addressing reliability concerns in ongoing research include analytic approaches such as dropping CpG sites with low observed variability, replicating findings with procedures such as pyrosequencing, and integrating preanalysis reliability metrics 134,136,137 . Focusing the analyses on regions with more variability across participants reduces the likelihood of false positives, and also reduces the number of comparisons, thereby increasing power. It is also important to note that methylation arrays assess only CpG sites; this yields fewer sites than wholegenome bisulfite sequencing approaches, and some research suggests some important variability occurs outside of these CpG islands 137,138 . Standards in this field continue to be refined, and researchers should follow these developments closely. At a minimum, researchers should implement strategies to address quality control (including thorough evaluations of normalization procedures and strategies for addressing batch effects), cell heterogeneity, and sample ancestry 65,[139][140][141] .
Our systematic review examined associations of maltreatment and other interpersonal adversities during childhood with DNA methylation in children and adults. We identified 100 studies, including studies utilizing both candidate gene approaches and epigenome-wide analyses. Several strengths of the literature were identified, as well as directions for future research. Key challenges facing the field and associated recommendations are described in Fig.  3. We included both studies focused on childhood maltreatment and other adversities, as well as studies that were not focused specifically on early adversity but included measurement of adversity in their examination of another topic (e.g., psychiatric or physical health diagnosis). Inclusion of these studies represents both a strength and limitation due to selection biases inherent to these designs, as well as confounds such as the presence of medical conditions and medications that may influence methylation. We acknowledge several other limitations to the current review. Specifically, (1) the present review focused on DNA methylation and did not include studies addressing other epigenetic modifications (e.g., microRNA, histone modifications); (2) the review did not address the functional impact (e.g., gene expression, proteins) of DNA methylation alterations in various studies; and (3) although systematic, the review was qualitative, and thus, does not provide information regarding effect sizes or heterogeneity of included studies. Meta-analytic reports of the most commonly studied candidate genes and EWAS results would build upon our qualitative analysis; however, the extensive number of individual CpG sites both within each candidate gene and across the EWAS studies poses a significant challenge. Furthermore, a critical limitation of the literature is that the location of individual CpG sites is often not consistently identified in empirical manuscripts, further precluding meta-analytic approaches. Future research should more carefully detail the precise location of CpG sites to facilitate future meta-analysis efforts.
In less than two decades, there has been an explosion of research in childhood maltreatment and epigenetics. Mirroring trends observed in genetic research, studies have increasingly focused on sampling methylation across the genome. As with genetic research, advantages of this approach include opportunities to identify key mechanisms that would otherwise have been overlooked. However, this approach generally requires larger samples to account for multiple testing which in turn has been criticized for less nuanced phenotyping of samples permitted in smaller cohorts. Some of these concerns may be somewhat addressed through collaborations such as efforts undertaken by the Psychiatric Genetics Consortium working groups, particularly when care is taken to balance genomic inflation/ deflation and to address concerns that can arise with diverse samples interrogated on differing platforms 65 . Researchers have also begun to explore ways that methodological approaches may improve understanding of epigenetic factors and adversity outcomes such as repeated measurement, sampling across tissues, twin/family approaches, and psychological/pharmacological treatments. Epigenetic approaches are also perhaps best understood in combination with thorough assessments of genotype, epigenome, and gene expression. Many factors related to the type and timing of adversity, availability and quality of buffers to adversity, and new events and time since early adversity are likely critically

Key Challenges Recommendations
• Epigenome-wide studies have generally failed to identify commonly studied candidate genes, leading to questions about candidate approaches • Continue to invest in consortiums that allow for harmonization and pooling of data to create large datasets necessary for detection of effects across the genome • Inconsistent associations of child maltreatment and methylation of candidate genes (e.g., NR3C1, FKBP5) may be due to complex interactions and conditional effects • Interactions among candidate genes and participant characteristics (e.g., sex, genetic ancestry, age), genotype, and stress exposure should be examined in studies with adequate power • Unique studies may pool data through consortiums to develop the necessary large-scale datasets • Childhood maltreatment is difficult to measure, and other adversities occurring throughout development, including prenatally, may also have an important impact. Nearly all studies of methylation in adults have relied on retrospective reports of childhood maltreatment • Utilization of comprehensive, prospective longitudinal designs and record review techniques, and interviews, in addition to subjective self reports, are needed • Very few studies have drawn upon repeated assessments of DNA methylation over time, or have examined causal effects of intervention • Longitudinal studies should examine differences in methylation at each developmental stage, and whether epigenetic changes remain the same, reverse, or strengthen over time, and in response to interventions • New methods to measure epigenetic age are continuing to be developed, and associations with maltreatment and other adversities are not consistently replicated across the various approaches • Systematic reviews and meta-analysis efforts will be important for determining which approaches are most closely linked to maltreatment, other exposures, and health outcomes to move towards standardization of approaches • Reliability of technological approaches to measure genome wide methylation has been observed to be relatively low • Implement analytic strategies for careful quality control and document this in published work • Focus analyses on regions with more variability across participants to reduce likelihood of false positives and increase power • Continue to invest in refinement of technology • Extensive number of individual CpG sites within each candidate gene and across EWAS studies poses a challenge to metaanalysis efforts • Carefully detail precise location of individual CpG sites to facilitate meta-analysis efforts important influences on negative outcomes of adversity. Ongoing recognition of the complexity of biological and environmental factors as well as phenotypic nuances is critically needed-apparent inconsistences do not necessarily negate findings from either of the divergent studies but rather may point to important considerations for ongoing research and theoretical conceptualizations.