Introduction: Prenatal Maternal Stress, Epigenetics, and Development

Human and animal studies have demonstrated that perturbations in early life, especially in the critical developmental window of intrauterine life, have programming effects on health outcomes in later life. A growing body of literature has also demonstrated that prenatal maternal stress (PNMS) can adversely impact a variety of developmental outcomes in the offspring throughout childhood that persist into adulthood [1, 2]. Epigenetics is considered to be one of the most important mechanisms involved in the long-term effects of in utero stress exposure. In the present review, we summarize the body of human research, exploring the association between fetal experience and epigenetic change, especially DNA methylation. Evidence reported here from research on genes that are associated with intrautrine adversity has shed light on how DNA methylation patterns may mediate biological processes involved in stress regulation. Furthermore, we provide evidence about these associations that have been reported from Project Ice Storm, a human model of the effects of PNMS on subsequent child and adolescent development.

The Fetal Programming Hypothesis

The fetal programming hypothesis, originally proposed by Barker [3], has been studied intensely in animals [4–9] and has also been applied to human populations [10–15]. It stipulates that behaviors and health outcomes may result from permanent alterations in the structure and function of fetal organs. The primary mechanism believed to be involved in fetal programming of later outcomes is exposure to maternal cortisol. Animal research has shown that elevated maternal cortisol levels negatively impact the developing fetal hypothalamic-pituitary-adrenal (HPA) axis resulting in less than optimal outcomes. Stress activates maternal HPA axis functioning and triggers glucocorticoid (GC) secretion that reaches the fetus by transplacental passage. In human study, a linear relationship has been reported between maternal and fetal cortisol levels; relatively small increases in maternal cortisol can double fetal concentrations [16]. Furthermore, stress in the mother not only increases her own circulating cortisol levels but also reduces the activity of the GC barrier enzyme in the placenta, HSD11B2 (11β-hydroxysteroid dehydrogenase 2, which converts noxious maternal cortisol into inactive cortisone before reaching the fetus), leaving her fetus less well protected [17].

PNMS

Maternal exposure to a severe stressor, such as environmental pollutants, nutritional factors, psychosocial stress, or maternal depression during pregnancy, can increase the fetus’ risk for suboptimal growth in a wide range of somatic systems, and for developing a variety of disorders in adulthood (reviewed in [18–20]). Effects of PNMS on fetal and child development, such as immune function [21], brain development [22], and behavior outcomes [23•], have been largely discussed, although most of the research has been in laboratory animals. In humans, as depression and anxiety disorders are common mental health problems during pregnancy [24, 25], their impact on epigenetic changes has also been widely investigated.

DNA Methylation

Epigenetic mechanisms involve chromatin structures and their modifications such as DNA methylation [26]. The covalent modification of the DNA molecule itself by methylation [27] at the 5’ position of the cytosine ring catalyzed by DNA methyltransferases (DNMT) occurs without changes to the DNA sequence. It is well established that DNA methylation plays a critical role in cellular differentiation during development [28]. DNA methylation is involved in gene regulation mostly by silencing promoter activity, either through interfering with the binding of transcription factors [29] or through recruiting methylated DNA binding proteins (MBD) that recruit histone deacetylases (HDACs) and histone methyltransferases (HMTase) which trigger chromatin inactivation [30].

The impact of epigenetics, especially DNA methylation, on downstream gene expression has been well documented and plays a key role in mediating the effect of intrauterinal environmental factors on offspring health outcoms later in life (reviewed in [31]). A better understanding of epigenetic processes in mediating exposure–phenotype relationships would be beneficial for future human prenatal research. The model of the mediating role of epigenetics in fetal programming is illustrated in Fig. 1.

Fig. 1
figure 1

The key role of epigenetic mechanisms in mediating the long-term effects of exposure to intrauterine factors on offspring’s health outcomes

Full size image

Environmental Factors and DNA Methylation of Candidate Genes

The NR3C1gene, coding for the glucocorticoid receptor (GR), is the prime candidate gene for epigenetic influence on outcomes as it plays a key regulator role in HPA axis functioning (see Table 1 for detailed description of candidate gene studies). Oberlander et al. [32] were the first to investigate the association between prenatal exposure to maternal depressive symptoms and the methylation status in the NR3C1 promoter region in newborn infants. Results showed that increased maternal depressed mood in the third trimester was associated with increased methylation of the NGFI-A binding site in NR3C1 in the infants, which then predicted increases in their HPA stress reactivity. However, levels of maternal depressed mood and use of serotonin reuptake inhibitor (SSRI) antidepressants during the third trimester were not associated with their own material methylation levels. Another study [38•] explored the association between maternal psychological well-being (prenatal depression, state-anxiety, pregnancy-related anxiety, and mother–fetus relationship) and methylation of promoter 1D, 1B, and 1F of NR3C1 in cord blood at birth. Results indicated that, among the psychological variables, pregnancy-related anxiety was significantly associated with 1F methylation, particularly in CpG site 9, showing that pregnancy-related anxiety explained 7.8 % of variance in methylation level of this CpG site.

Table 1 Summary of studies analyzing DNA methylation of candidate genes resulting from prenatal maternal stress in humans (in chronological order)
Full size table

Mulligan et al. [36] examined the mediating effects of methylation levels of the NR3C1 promoter on the relationship between prenatal maternal material deprivation (intended to reflect availability of financial resources), daily mundane psychosocial stress, and war-related stress and newborn birth weight in a small cohort (25 mother–newborn dyads) from the Democratic Republic of Congo. Their results revealed a significant negative correlation between these culturally relevant measures of PNMS and newborn birth weight. Furthermore, methylation in NR3C1 promoter was found to interact with this correlation. In another study of extreme maternal psychosocial stressors derived from the Tutsi genocide [39•], traumatized widows and their 17–18-year-old children were recruited to explore the transgenerational transmission of maternal stress. It was observed that methylation levels of the NR3C1 exon 1F in in utero exposed children were 1.8 % higher than those observed in non-exposed children. A study by Radtke et al. [34] has identified the prenatal period as a critical window for this epigenetic effect. They analyzed the NR3C1 methylation status in 10- to 19-year-old children of mothers who experienced intimate partner violence (IPV) before, during, and after pregnancy. The methylation levels of NR3C1 were significantly higher in children whose mothers experienced IPV while pregnant relative to children whose mothers did not experience IPV during pregnancy. Interestingly, this relationship was not seen when the IPV occurred before or after pregnancy. That such a strong association was observed after more than a decade suggests that transgenerational effects can be maintained and may potentially have lifelong influences on HPA axis via NR3C1 regulation.

Epigenetic effects from other candidate genes have also been investigated. HSD11B2, the cortisol catalyzing gene, plays a critical role in the regulation of fetal cortisol exposure. Evidence from animal research suggests that an association exists between prenatal stress and HSD11B2 methylation levels [40]. In humans, the epigenetic role of HSD11B2 was revealed through a study of prenatal socioeconomic adversity by Appleton et al. [37•]. The authors observed that maternal socioeconomic adversity during pregnancy was associated with lower levels of placental HSD11B2 methylation in children, particularly in males. In the same study, the authors investigated whether methylation status of SLC6A4 and of BDNF (brain-derived neurotrophic factor) in neonatal cord blood was associated with maternal depressed mood in the second and third trimesters [33]. Their results showed that increased second (but not third) trimester maternal depressed mood was associated with decreased maternal and infant SLC6A4 promoter methylation; the BDNF promoter methylation status in both mothers and infants was not associated with maternal depressed mood in pregnancy.

Potential epigenetic markers believed to play a role in the relationship between maternal depressed mood and birth outcomes have also been investigated [35]. In a study of more than 500 pregnant women which investigated the role of imprinted genes, the authors examined the association between maternal depressed mood during pregnancy and birth weight. Their goal was to determine the extent to which this association was influenced by DNA methylation differences for differentially methylated regions (DMRs) regulating nine imprinted genes (IGF2/H19, DLK1/MEG3, MEST, PEG3, PEG10/SGCE, NNAT, and PLAGL1). The results indicated that severe maternal depressed mood was associated with a threefold increase in the risk of low birth weight and DNA methylation levels were 2.4 % higher in neonates exposed to severe maternal depressed mood during pregnancy. The authors concluded that MEG3 IGF2 plasticity may play important roles in regulating the association between prenatal maternal depressed mood and low birth weight.

Environmental Factors and Genome-Wide DNA Methylation

Although many candidate gene studies have shed light on the epigenetic mechanisms of PNMS, genome-wide approaches offer additional insights. Table 2 presents summaries of the human genome-wide studies described in the following sections. For example, Schroeder et al. [41] examined the genome-wide methylation patterns in umbilical cord blood obtained from the neonates of women treated for mood disorders during the perinatal period, and from women without current depression. No significant DNA methylation changes were observed in the cord blood of neonates exposed to maternal depressive symptoms. However, fetal exposure to antidepressants was associated with differential methylation of TNFRSF21 and CHRNA2 CpG sites, although the average group differences in methylation for these two CpG sites were less than 3 %. These weak findings probably stem from the fact that all women in the control group had previously been diagnosed with a mood disorder and may have had residual symptoms or other vulnerability factors that rendered them more similar to the depressed group than a “supernormal” control group would have been. In another similar study [42] on the effect of antidepressant medication during pregnancy, the authors investigated if different epigenomic patterns can be detected in umbilical cord blood of neonates of mothers with non-medicated depression or anxiety or mothers using selective serotonin reuptake inhibitors (SSRIs) during pregnancy compared to neonates of healthy non-medicated controls. The results indicated that 42 CpG sites were significantly differentially methylated in neonates whose mothers had non-medicated depression or anxiety relative to controls. The largest DNA methylation difference was observed in COL7A1. Furthermore, using a gene ontology analysis, the authors reported that significant clustering of the top genes associated with non-medicated maternal depression/anxiety was related to regulation of transcription, translation, and cell division processes. Interestingly, CpG methylation differences were not observed in neonates whose mothers had taken SSRIs relative to the controls. Both findings suggest that prenatal exposure to some (but not all) psychotropic medications, maternal psychiatric symptoms, or the combination of the two could impact on offspring outcomes by a mediating effect of DNA methylation.

Table 2 Summary of studies analyzing genome-wide DNA methylation of epigenome results from prenatal maternal stress in humans (in chronological order)
Full size table

Taken together, both candidate gene and genome-wide approaches provide evidence supporting the role of epigenetic mechanism in fetal programming.

Natural Disasters as Models of Human PNMS

Human research cannot match research with laboratory animals for its random assignment of animals to groups and its control of experimental variables. Retrospective psychosocial stress studies [44–48] query mothers directly about life events during the pregnancy and correlate these events to outcomes. However, life event studies are hampered by the rarity of serious events during pregnancy, and by the fact that many events (e.g., marital discord and job loss) may not have been “independent” of the women’s own characteristics. Thus, life event studies present challenges to internal validity because many such events are not randomly assigned and characteristics of a woman that increase her risk of experiencing these events (e.g., impulsivity) are likely to be heritable and influence the offspring genetically. Similarly, studies of maternal anxiety or mood in pregnancy find associations with developmental difficulties in children [49–51], but because anxiety is not randomly assigned, it is challenging to disentangle genetic effects from effects of the intrauterine and postnatal environments. Therefore, one approach to circumventing these methodological challenges is to study the effects of exposure to an independent random stressor, such as a natural disaster.

A disaster is any event that “causes disruption exceeding the adjustment capacity of the affected community” [52]. Many disasters act as “natural experiments”, distributing hardship in quasi-random fashion such that the degree of objective exposure is independent of personal control or psychosocial circumstances. Disasters also affect large numbers of pregnant women, which is important for obtaining sample sizes needed for advanced analyses. They also have sudden onsets that allow the researcher to accurately time the event in pregnancy, which is important for dating PNMS effects with respect to fetal development. Furthermore, disasters allow researchers to distinguish between pregnant women’s objective exposure to an event and their subjective distress, since these may not be highly correlated, may have different effects on the fetus, and may have different implications for prevention and intervention. Project Ice Storm is the first human prospective study using a natural disaster to explore the effects of different components of PNMS on developmental outcomes in offspring.

Project Ice Storm

In January 1998, 5 days of freezing rain collapsed the power grid of southern Quebec leaving three million people without electricity for as long as 45 days in the coldest month of the year. The Canadian Insurance Bureau counts the 1998 ice storm as Canada’s worst natural disaster in history. Five months after the ice storm, King recruited 224 women who were pregnant during the crisis. She sent them questionnaires to assess their levels of objective hardship (threat, loss, scope, and change), their cognitive appraisal of the crisis (negative, neutral, positive), and their degree of subjective distress (post-traumatic stress symptoms). The women also provided saliva samples over 24 h to assess diurnal cortisol. Maternal psychopathology and child development have been monitored since then through questionnaires beginning at 6 months postpartum, and face-to-face assessments beginning at 2 years. Each child assessment includes cognitive, behavioral, motor, and physical assessments. Physical assessments in our protocols have included measures of cardiovascular function (resting heart rate and blood pressure, heart rate variability, stress response), HPA axis (cortisol awakening response, diurnal cortisol, stress reactivity), growth and metabolism (body mass index (BMI), skin folds, glucose challenge), immune function (cytokines, cell populations, history of asthma and allergies), and physical markers (dermatoglyphic asymmetry, brain structure, finger length ratios) (for details, see review in [23•]).

We recently reported changes in genome-wide DNA methylation levels triggered by in utero exposure to the ice storm [43•]. Using blood obtained from a subsample of Project Ice Storm teenagers, we assessed the DNA methylation status in T cells and correlated the severity of the mothers’ objective hardship with DNA methylation using Illumina Infinium 450 K. Our analysis revealed that the mothers’ objective hardship was significantly correlated with 1675 differentially methylated CpG sites affiliated with 957 genes that are predominantly related to immune function; the mothers’ subjective distress from the ice storm was completely uncorrelated with their children’s genome-wide methylation profiling in T cells. In addition, we compared methylation between children whose mothers had a negative cognitive appraisal of the ice storm with children whose mothers had either a neutral or positive appraisal of the ice storm [53]. We found that 2872 CpG sites were differentially methylated as a function of maternal cognitive appraisal, an even stronger result than that observed with objective stress [43•]. Moreover, there was a significant overlap, but also considerable non-overlap, in the differentially methylated CpGs or genes and biological pathways that were associated with cognitive appraisal and objective PNMS, suggesting that, indeed, different aspects of the maternal stress experience may be operating through distinct, yet overlapping, mechanisms.

Results from this Project Ice Storm study also address the specificity of the results. Of importance, we also demonstrated the comparability of results across T cells, peripheral blood mononuclear cells (PBMCs), and saliva which provide evidence that the effects of objective PNMS on DNA methylation can be observed in peripheral tissues [43•].

Project Ice Storm data go beyond the association between an environmental event and DNA methylation levels to show that the epigenetic profile then predicts health outcomes in the offspring. We tested the hypothesis that DNA methylation mediates the effect of PNMS on immune and metabolic outcomes. We selected NF-κB signaling-related genes (immune) and type 1 and 2 diabetes-related genes (metabolic) to test whether their methylation status mediated the effect of objective and/or subjective PNMS on Th1- and Th2-cytokine production in blood in the 13 ½-year-old Project Ice Storm children. We found a negative mediation effect of DNA methylation on the relationship between objective PNMS and IFN-γ and a positive mediation of the relationship between objective PNMS and IL-2, IL-4, and IL-13. Similarly, DNA methylation of type 1 and 2 diabetes-related genes had a negative mediating effect on BMI and central adiposity. Given that greater objective stress predicts greater BMI and central adiposity, the negative mediation suggests that the alterations in DNA methylation profile are protective, and associated with lower levels of adiposity, rather than an explanatory mechanism for the positive effects of PNMS on obesity [54•]. These data provide evidence supporting the mediating role of DNA methylation on the association between objective aspects of PNMS and child immune and metabolic status.

Conclusions and Discussion

The prenatal period is a vulnerable time for fetal development in humans. In utero exposure to environmental stress seems to result in long-term epigenome alterations. In this review, we outlined evidence from human prenatal research supporting the view that PNMS could lead to lasting, broad, and functionally organized signatures in DNA methylation which, in turn, could mediate exposure–phenotype associations. We also emphasized the advantage of using disaster-related stress for studying prenatal maternal stress. Although considerable discoveries in this field have been made, many challenges still need to be addressed.

Since the seminal finding reported by Weaver et al. [55] that maternal behavior can program the epigenome of the promoter 17 of NR3C1 in rats, this gene has received considerably more attention. Although the critical role of NR3C1 in the stress response has been shown to influence fetal development leading to the dysregulation of the infant’s HPA axis, the role of other candidate genes in the regulation of the fetal stress response needs further investigation. For example, in the same cohort studied by Oberlander’s group, the authors [33] also revealed that the methylation status of SLC6A4 was altered by maternal depressive symptom during pregnancy. In addition, epigenetic reprogramming in response to early adverse environments is not limited to a few candidate genes but is an organized response involving multiple gene networks [56, 57]. The “genome-wide” approach is atheoretical and exploratory but provides opportunities to inform the researcher of patterns across much of the epigenome. This approach could result in the discovery of new functional gene pathways, leading to new potential mechanistic insights. In this regard, further research should not be limited to selected candidate genes, and it would be benefited by using new approaches with the aim of discovering new biomarkers for health and disease.

Increasing evidence supports the view that stable epigenetic markers could be triggered by exposure to prenatal stress. To date, the Tutsi genocide study provides the most convincing evidence of this stability; NR3C1 methylation changes were still observable in individuals almost 20 years after their mothers were exposed to these traumatic events while pregnant [39•]. Similarly, epigenetic patterns of NR3C1 of children prenatally exposed to the effects of intimate partner violence were detectable 10 to 19 years after birth [34]. In addition, the genome-wide epigenetic signatures of prenatally exposed children were still measurable in immune cells 13 years after the 1998 Quebec ice storm [43•]. These results suggest that continued longitudinal assessment of offsping exposed to PNMS will provide a more comprehensive understanding of the enduring impact of PNMS on the epigenome. The ability of prospective disaster-related PNMS studies to identify the active ingredient in these effects, both maternal objective exposure and cognitive appraisal in the case of Project Ice Storm, is a distinct advantage of this approach.

Environmental programming effects are often sex-specific [58], and there is increasing awareness of possible sexual dimorphisms in environmental epigenetic programming (reviewed in [59]). It was suggested that environmental factors could trigger, in a sex-specific manner, epigenetic biomarkers during developmental windows throughout the individual’s life under the control of sex chromosomes and/or sex hormones. Although the sex-specific epigenetic markers related to prenatal adversity have been better investigated in animals, a limited number of studies in humans have examined sex differences in methylation of candidate genes or genome-wide level. As mentioned above, the methylation pattern of placental HSD11B2 related to prenatal socioeconomic adversity showed methylation sex specificity, with effects seen particularly in males [37•]. More work is required to identify novel sex-associated CpGs or genes in further studies.

Although epigenetic patterns are considered to be cell- or tissue-specific [28, 60], emerging evidence has also shown that early life stressors, such as social-economic positioning in humans [57] and maternal deprivation in rhesus monkeys [61], are associated with broad DNA methylation signatures in blood [57] and in T cells [61]. Supporting this idea, findings from Project Ice Storm have revealed that the DNA methylation pattern in T cells is comparable with that observed in both PBMCs and saliva samples in children. Since most of the human prenatal research uses neonatal cord blood or placental tissue as biological materials, as such, these studies shared a common limitation of a heterogeneous mixture of different type of cells. Hence, it may be important in future studies to perform detailed analyses of epigenetic status with well-distinguished cell- or tissue-type samples.

It should be stressed that although epigenetic mechanisms, in particular DNA methylation, have been postulated to play a mediating role through which PNMS may have important lifespan health consequences, only a few studies have conducted empirical tests of these effects using mediation analyses. Using bootstrapping analysis, we found that objective hardship experienced by the pregnant women during the ice storm had a large impact on genome-wide DNA methylation level, which, in turn, mediated the effect on immune and metabolic outcomes. However, more replication is warranted in this field.

Although standard bisulfite-based sequencing approaches, such as pyrosequencing, are powerful methods to explore DNA methylation, it should be noted that they cannot distinguish between 5-methyl-cytosine (5-mC) and 5-hydroxymethyl-cytosine (5-hmC) which is considered as the “sixth base”. Recently, hydroxymethylation has been gaining more attention and revealing a new layer of epigenetic regulation. However, few studies have attempted to link hydroxymethylated DNA biomarkers to exposure to prenatal maternal stress.

A primary interest in epigenetic research is to find relevant epigenetic biomarkers. However, the magnitude of changes in DNA methylation of many biomarkers discussed above are very modest (rarely exceeds 10 %). It remains unclear of how these modest DNA methylation changes actually affect functional downstream factors which influence phenotypes. An explanation could be that the observed magnitude of change was somewhat deflated due to confounding factors. It is also possible that the association between DNA methylation and gene expression might be less straightforward or other additional layers of epigenetic regulation, above and beyond DNA methylation, are required to regulate gene expression. Regardless, in order to gain better insight into the biological function of these modest DNA methylation changes, gene expression experiments are required to determine the functional repercussions of observed alterations in DNA methylation due to prenatal maternal stress exposure. However, only a few studies [39•, 62] have examined gene expression by accessing mRNA and/or protein levels in relation to epigenetic modifications. Moreover, it is also important to identify whether upstream factors/pathways could direct differential DNA methylation to a small subset of CpGs as biomarkers in the epigenome, and furthermore, questions remain to be explored in understanding of the potential underlying mechamisms such as loss of methylation due to transcriptional activation or transient/sustained inhibition of DNMT containing complex recruitment. Taken together, there is a need for follow-up studies to more precisely address these issues.

In summary, we have reviewed evidence from human studies that prenatal maternal stress can have important epigenetic consequences in genes that are associated with environmental adversity, and that DNA methylation profiles appear to have long-term consequences for adaptation and development in the unborn child.