Prenatal maternal stress is associated with site-specific and age acceleration changes in maternal and newborn DNA methylation

ABSTRACT

Prenatal maternal stress has a negative impact on child health but the mechanisms through which maternal stress affects child health are unclear. Epigenetic variation, such as DNA methylation, is a likely mechanistic candidate as DNA methylation is sensitive to environmental insults and can regulate long-term changes in gene expression. We recruited 155 mother-newborn dyads in the Democratic Republic of Congo to investigate the effects of maternal stress on DNA methylation in mothers and newborns. We used four measures of maternal stress to capture a range of stressful experiences: general trauma, sexual trauma, war trauma, and chronic stress. We identified differentially methylated positions (DMPs) associated with general trauma, sexual trauma, and war trauma in both mothers and newborns. No DMPs were associated with chronic stress. Sexual trauma was positively associated with epigenetic age acceleration across several epigenetic clocks in mothers. General trauma and war trauma were positively associated with newborn epigenetic age acceleration using the extrinsic epigenetic age clock. We tested the top DMPs for enrichment of DNase I hypersensitive sites (DHS) and found no enrichment in mothers. In newborns, top DMPs associated with war trauma were enriched for DHS in embryonic and foetal cell types. Finally, one of the top DMPs associated with war trauma in newborns also predicted birthweight, completing the cycle from maternal stress to DNA methylation to newborn health outcome. Our results indicate that maternal stress is associated with site-specific changes in DNAm and epigenetic age acceleration in both mothers and newborns.

KEYWORDS:

• epigenetics
• DNA methylation
• maternal stress
• DOHaD

Introduction

Prenatal maternal stress is known to influence offspring health throughout childhood [1] and even into adulthood [2]. These effects are consistent with the Developmental Origins of Health and Disease Hypothesis (DOHaD), which posits that early life adversity is associated with poor adult health outcomes [3]. For example, prenatal maternal stress is associated with newborn health outcomes like low birthweight [4], which is in turn associated with adult disease and mortality [5]. In childhood, prenatal maternal stress predicts poorer cognitive functioning [6] and behavioural problems [7]. These findings have resulted in a focus on early development as a critical time period in which stressors can influence health outcomes over the life course [8].

The biological mechanisms for the long-lasting effects of prenatal maternal stress are unclear. One candidate mechanism for the biological embedding of prenatal maternal stress is epigenetic variation [9], and in particular DNA methylation (DNAm) at cytosine-guanine (CpG) dinucleotide sites. DNAm helps regulate gene expression and plays an important role in human development and disease [10] and is sensitive to social experiences, such as trauma [11–13]. Several studies have reported associations between prenatal maternal stress and DNAm in epigenome-wide association studies (EWAS) [14,15]. These findings hint at the possibility that prenatal maternal stress may be biologically embedded through altered DNAm patterns, which then alter health outcomes.

DNAm may mediate the impact of maternal stress on health outcomes through site-specific changes, such as differentially methylated positions (DMPs), or more global measures, such as epigenetic age. Epigenetic age is an estimate of biological age based on age-associated CpG sites [16]. If the estimated epigenetic age is greater than predicted based on chronological age, that individual is said to demonstrate accelerated epigenetic age. Psychosocial stress has consistently been found to associate with accelerated epigenetic ageing in adults and children [17]. Lifetime stress [18] and exposure to violence [19] have been associated with accelerated epigenetic ageing in adults and children, respectively. Prenatal maternal stress has been found to predict epigenetic age in children [20]. Given associations between epigenetic age acceleration and morbidity and mortality [21], these results suggest that accelerated epigenetic ageing could be a mechanism linking early life stress with adult health outcomes.

Previous studies from our research group have reported associations between prenatal maternal stress and DNA methylation in a small sample of mothers and offspring in the Democratic Republic of Congo (DRC). Specifically, our studies have revealed associations with maternal stress in epigenome-wide methylation contexts [22,23], and at candidate genes involved in the stress response [24,25], brain development [26], growth [27], and the regulation of DNA methylation [28]. In the current study, we tested for the effect of maternal stress on maternal and newborn DNAm in an expanded sample of mothers and newborns from the DRC. We hypothesized that prenatal maternal stress would associate with site-specific and global measures of DNA methylation. Specifically, we tested for associations of maternal stress on mother and newborn DNAm by investigating 1) individual CpG sites through EWAS, 2) global DNAm through principal component analysis (PCA) and general mean methylation (GMM), and 3) epigenetic age acceleration using a range of epigenetic clocks. We also tested for association of maternal stress and DNAm with birthweight as an exemplar health outcome.

Methods

Study sample

Participants were recruited at HEAL Africa Hospital in Goma, DRC from 2013 to 2017. Mothers were recruited from the maternity ward which served both the general population as well as victims of sexual violence. The inclusion criterion for the present study was singleton birth at HEAL Africa Hospital. After quality control procedures, 310 participants (n = 155 mothers and n = 155 newborns) were included in epigenetic age analyses. Four dyads were excluded in EWAS analyses (n = 151 mothers and n = 151 newborns) to improve protection against batch effects.

Participant recruitment began with the informed consent process. Upon arrival at HEAL Africa, mothers were asked by a staff midwife about their interest in participating in the study. If participants wanted to learn more about the study, they were given the option of remaining in the maternity ward or moving to a private room. Most participants elected to remain in the maternity ward because they felt comfortable there. An explanation of the study was given in Congolese Swahili. Mothers were asked if they had any questions about the study and were told they could withdraw from the study at any time. Oral informed consent was considered most culturally appropriate given high rates of illiteracy and a history of colonial exploitation. In cultures of the eastern DRC, pregnancy initiates a girl’s entry into adulthood. Young mothers typically move out of their parents’ homes and live outside their parents’ responsibilities. Consistent with cultural norms, all mothers were considered capable of consenting to enrolment in the study, even if they were under 18 years of age. Following informed consent, semi-structured interviews and surveys were conducted and blood samples were collected from mothers and babies within one day of delivery. Study procedures were approved by the Western Institutional Review Board, Olympia, WA (www.wirb.com, WIRB Project #20100993) and the University of Florida (Project #IRB202001503). The study was also approved by local ethics committees at the University of Goma and HEAL Africa Hospital.

Maternal stress measures

Interviews were conducted by two local female members of the research team in Congolese Swahili. The interviews followed a semi-structured life course format, starting from childhood and ending in the present. Next, the general trauma and sexual abuse parts of the Early Trauma Inventory-Self Report [29] were administered.

Four different measures were used to capture experiences of maternal stress: general trauma, sexual trauma, war trauma, and chronic stress (Table S1). Both the general trauma and sexual trauma measures come from the validated Early Trauma Inventory-Self Report [29]. General trauma was a sum of endorsed items such as witnessing violence and serious injuries in the family. One item loaded negatively (separation of parents) and was removed. The theoretical range for the general trauma scale was zero to ten. The sexual trauma measure was a sum of the sexual abuse subscale in the Early Trauma Inventory-Self Report [29], which contains six items such as being forced to touch intimate parts and forced genital sex. The theoretical range for the sexual abuse scale was zero to six.

War trauma and chronic stress were both ethnographically derived measures of stress that were specific to the local cultural context of eastern DRC. The development of these two scales in a pilot sample from HEAL Africa Hospital has been previously described [25]. Briefly, a combination of items relating to stress and trauma from previously described constructs, such as the Trauma History Questionnaire [30] and the Hassles Scale [31], were used to construct a list of sources of stress. More items were subsequently added to this list that were relevant stressors in the eastern Democratic Republic of Congo, such as war-related stressors. Factor analyses were then used to identify structure in the responses to these items, resulting in two measures: war trauma and chronic stress [25]. In the present study, items related to sexual violence were removed from the war trauma measure because they were captured in the sexual trauma measure described above. The theoretical range for war trauma was zero to six. The chronic stress measure for this study was modified by combining ‘No help cleaning during pregnancy’ and ‘No help at home’ into a single item. Two questions that were used in earlier iterations of the chronic stress measure were not asked in the present study and were therefore not included in the measure (‘travelled alone to the hospital’ and ‘choice in birth location’). The theoretical range for chronic stress was zero to 16.

Anthropometric and demographic data

Anthropometric and demographic data were collected from mothers and newborns. For mothers, height, weight, age, delivery method (vaginal birth or caesarean section), alcohol use in pregnancy (no use of alcohol = 0, use of alcohol = 1), parity (primigravida = 0, multigravida = 1), and recruitment population were recorded. For newborns, sex, birthweight, and gestational age were recorded. Gestational age was estimated from DNAm data using a validated epigenetic clock of gestational age [32]. Smoking was not included in the analyses because only one mother reported this exposure.

DNA methylation

DNA was extracted from venous blood using the QIAmp DNA Blood Mini Kit (Qiagen, Germantown, Maryland). The protocol was performed in two parts. Steps one through seven were performed at HEAL Africa Hospital to kill infectious agents and stabilize DNA prior to shipment to the University of Florida (UF). The remaining steps were performed at UF and samples were cleaned and concentrated as needed [33]. Two hundred fifty grams of DNA were hybridized to the Infinium Methylation EPIC BeadChip (Illumina) at the Hussman Institute for Human Genomics, University of Miami. Data processing was performed in R version 4.2.0 [34]. Quality control of raw methylation data was performed at UF using the R packages meffil [35] and ewastools [36]. Immune cell type proportions were estimated using the IDOL algorithm [37,38]. The SeSame package was used for non-linear dye-bias correction, p-value detection, and noob background correction [39]. Batch correction was performed using ComBat [40] in the sva package [41]. Probes known to map to multiple regions of the genome were masked, as well as probes that could be affected by SNPs prevalent in the African superpopulation using a published probe annotation [42]. X chromosome probes were retained in analyses of mothers but not newborns. Y chromosome probes were masked in all analyses. For mothers, 706,981 probes were analysed. For newborns, 691,867 probes were analysed.

Epigenome-wide association study of maternal stress

Eight EWAS were performed that regressed DNAm at each probe on one of the four maternal stress measures (general trauma, sexual trauma, war trauma, and chronic stress) within each generation (mothers or newborns). For EWAS in mothers, robust linear models controlled for maternal body mass index (BMI), age, parity, delivery mode, alcohol use in pregnancy, recruitment site, and the first principal component (PC) of cell type variation, which accounted for 89% of the variance in cell type proportions. For newborns, covariates in EWAS included maternal BMI, maternal age, parity, delivery mode, alcohol use in pregnancy, recruitment site, gestational age, and the first two principal PCs of cell type variation, which accounted for 90% of the variance in cell type proportions. Covariates were chosen based on prior literature suggesting associations with DNA methylation [32,43–48]

Epigenetic clock analyses

Six different epigenetic clocks were used in analyses of maternal stress in mothers and the first three were tested in newborns. 1) Horvath’s pan-tissue clock [16] was trained on chronological age. 2) Intrinsic epigenetic age was similar to Horvath’s clock, but the effect of variation in immune cell type was removed [49]. 3) Extrinsic epigenetic age was also similar to Horvath’s clock, but the contribution of variation in immune cell type proportion to estimated biological age was upweighted [49]. 4) The telomere epigenetic clock was trained on telomere length and outperforms telomere length in predicting chronological age and mortality [50]. 5) The PhenoAge epigenetic clock was trained on a composite score of nine clinical biomarkers associated with ageing and chronological age, and was strongly associated with morbidity and mortality [51]. 6) The GrimAge epigenetic clock was trained on seven clinical biomarkers associated with ageing and smoking pack years. The GrimAge clock produced a measure of biological age strongly associated with morbidity and mortality [52]. In mothers, maternal BMI, delivery mode, alcohol use in pregnancy, parity, and recruitment site in epigenetic age acceleration analyses were controlled. In newborns, the same covariates plus newborn sex and gestational age were controlled. We restricted epigenetic clocks used in analyses of newborns to clocks which included newborns in the original publications describing development of the respective epigenetic clocks [16,49]. Covariates were chosen based on prior literature suggesting associations with DNA methylation [32,43–48].

DNase I hypersensitivity analyses

eFORGE was used for enrichment analyses [53]. Enrichment of DNase I hypersensitivity sites was tested in the top 125 differentially methylated positions (DMPs), and the top 125 hyper- and the top 125 hypo-methylated probes identified in the EWAS analyses. Recommendations for eFORGE suggest using between 100 and 1000 DMPs [54]; 125 sites were chosen here to focus on the most significant findings while also ensuring that minimum number of sites were available for analysis after filtering input DMPs as part of the eFORGE analysis. Enrichment analyses were controlled for multiple testing within maternal stress measures and analysis method (top 125 DMPs, top 125 hyper methylated DMPs, top 125 hypo methylated DMPs) using the false discovery rate. Results were visualized using the complexHeatmap package [55].

Statistical analyses

Additional statistical tests related to maternal stress and global measures of DNAm. The top ten PCs of a PCA to describe variation in DNAm and general mean methylation (GMM), which is the mean of DNAm across all probes for a given participant, were tested for associations with maternal stress. In both analyses, covariates included maternal BMI, age, delivery method, alcohol use in pregnancy, parity, recruitment site, and immune cell type proportion PCs.

Bonferroni correction for multiple testing within maternal stress measures was used in the EWAS and tests of DNAm PCs. Multiple testing correction was not used for tests of epigenetic age acceleration because of the exploratory nature of association tests between maternal stress and recently developed epigenetic clocks.

Finally, associations between maternal stress, DNA methylation, and birthweight were tested as follow-up to pilot study results [22–25,27,28]. Linear regression models were used to test for relationships between maternal stress measures and birthweight. Regression models were fit to test for associations between Bonferroni-significant DMPs and birthweight. Both analyses controlled for maternal BMI, maternal age, delivery method, alcohol use in pregnancy, parity, recruitment site, gestational age, and newborn sex. Tests of relationships between DNAm and birthweight additionally controlled for immune cell type proportion. All analyses were completed in R [34] (version 4.2.0).

Results

Sample characteristics

Demographic and maternal stress data for participants are described in Table 1. A higher number of newborns were recruited in the general maternity population (n = 90) compared to the sexual violence population (n = 65). Mothers recruited from the sexual violence population tended to be younger than mothers recruited as part of the general hospital population because younger women were more likely to be raped and were more likely to seek help through services offered by the hospital. In general, newborns recruited in the sexual violence population were born to younger, lower BMI, and more frequently primigravida (all p < 0.001) mothers. Of the four maternal stress measures, general trauma and war trauma were comparable across recruitment populations (both p > 0.05). Mothers recruited from the sexual violence population had higher scores for chronic stress (p < 0.001) and sexual violence (p < 0.003). In some cases, there are small differences in summary scores for covariate and maternal stress measures because blood samples were not collected for mother and newborn in some dyads, e.g., see Delivery mode. A complete list of items and endorsement frequencies in the stress measures for all four maternal stress measures is available in Table S1.

Table 1. Descriptive statistics of newborns and mothers included in the study.

Variable	General maternity population (N = 90)	Sexual violence population (N = 65)	Total (N = 155)§	General maternity population (N = 90)	Sexual violence population (N = 65)	Total (N = 155)§
		Newborns			Mothers
Sex
Female	43 (47.8%)	33 (5.8%)	76 (49.0%)	90 (1.0%)	65 (1.0%)	155 (100.0%)
Male	47 (52.2%)	32 (49.2%)	79 (51.0%)	0 (.0%)	0 (.0%)	0 (0.0%)
Maternal age (years)
Mean (SD)	26.2 (5.92)	17.5 (3.97)***	22.6 (6.75)	26.3 (5.95)	17.3 (3.99)***	22.6 (6.83)
Maternal BMI
Mean (SD)	28.1 (4.09)	25.9 (2.80)***	27.2 (3.76)	28.1 (4.08)	25.8 (2.78)***	27.2 (3.76)
Delivery mode
Vaginal	73 (81.1%)	46 (7.8%)	119 (76.8%)	73 (81.1%)	48 (73.8%)	121 (78.1%)
Caesarean section	17 (18.9%)	19 (29.2%)	36 (23.2%)	17 (18.9%)	17 (26.2%)	34 (21.9%)
Alcohol in pregnancy
No	75 (83.3%)	55 (84.6%)	130 (83.9%)	75 (83.3%)	56 (86.2%)	131 (84.5%)
Yes	15 (16.7%)	7 (1.8%)	22 (14.2%)	15 (16.7%)	7 (1.8%)	22 (14.2%)
Parity
Multigravida	63 (7.0%)	8 (12.3%)***	71 (45.8%)	63 (7.0%)	8 (12.3%)***	71 (45.8%)
Primigravida	24 (26.7%)	54 (83.1%)	78 (50.3%)	24 (26.7%)	55 (84.6%)	79 (51.0%)
Gestational age† (weeks)
Mean (SD)	40.1 (1.00)	39.6 (1.44)*	39.9 (1.23)	40.1 (1.00)	39.5 (1.43)**	39.9 (1.23)
Birthweight (grams) Mean (SD)	3220 (449)	2900 (553)	3080 (519)	3220 (448)	2870 (536)	3080 (516)
General trauma
Mean (SD)	2.42 (1.90)	2.22 (1.46)	2.34 (1.73)	2.46 (1.88)	2.22 (1.45)	2.35 (1.71)
Sexual trauma
Mean (SD)	.911 (1.63)	1.69 (1.60)**	1.24 (1.66)	.956 (1.63)	1.72 (1.60)**	1.28 (1.66)
War trauma
Mean (SD)	1.04 (.886)	.969 (.968)	1.01 (0.919)	1.04 (.886)	.954 (.959)	1.01 (0.915)
Chronic stress
Mean (SD)	4.57 (3.58)	8.40 (3.10)***	6.17 (3.87)	4.63 (3.60)	8.43 (2.87)***	6.23 (3.80)

Note: Differences between populations were tested with Fischer’s exact tests for categorical variables and ANOVA for continuous variables. The small differences in descriptive statistics between newborns and mothers were due to some dyads being represented by only a newborn or only a mother.

*p < 0.05, **p < 0.01, ***p < 0.001.

§Four participants were excluded from the EWAS analyses to improve protection against batch effects.

†Gestational age was measured using DNA methylation gestational age clock.

SD: standard deviation.

BMI: Body mass index.

Associations between maternal stress and DNA methylation

Epigenome-wide association studies (EWAS)

Fifteen DMPs were identified in mothers that associated with one of the maternal stress measures and met Bonferroni-corrected significance (p < 7.70 × 10⁻⁸; Figure 1a-d). Four DMPs associated with general trauma (Figure 1a), nine with sexual trauma (Figure 1b), and two with war trauma (Figure 1c). No DMPs were significantly associated with chronic stress in mothers (Figure 1d). All Bonferroni-corrected DMPs and gene annotations are listed in Table 2.

Figure 1. Manhattan plots of eight epigenome-wide association tests of maternal stress measures among mothers (a–d) and newborns (e–h). Dashed lines indicate the Bonferroni level of significance (p = 7.07 × 10⁻⁸ for mothers and 7.23 × 10⁻⁸ for newborns). A total of 706,981 probes for mothers and 691,867 for newborns were included in analyses. N = 145.

Table 2. Bonferroni-corrected differentially methylated positions in EWAS in mothers and newborns.

Maternal stress measure	Probe	Unstandardized beta coefficient	p	Gene(s)
Mothers
General trauma
	cg11408019	0.003	1.10e − 10	NA
	cg14519777	0.006	6.87e − 08	CTA−339C12.1;CUX1
	cg14282695	0.009	1.48e − 08	SAMD4A
	cg16543391	−0.001	6.25e − 08	EML2;MIR330
Sexual trauma
	cg21219607	−0.002	5.50e − 08	PARP15
	cg06308131	−0.004	9.71e − 10	MUC4
	cg04358942	−0.001	2.86e − 08	RP11-64D24.2
	cg23527517	−0.008	1.23e − 08	OTOP3
	cg14859642	0.007	6.99e − 08	NA
	cg00489624	−0.003	1.10e − 08	DLGAP4
	cg24308336	−0.007	3.79e − 08	ARHGAP40
	cg16765764	−0.002	9.46e − 09	DHX35
	cg10897169	−0.003	6.81e − 08	BCAS4
War trauma
	cg13740840	−0.002	5.99e − 08	KIF15;MIR564;TMEM42
	cg26486174	0.013	1.61e − 08	NA
Newborns
General trauma
	cg24590750	0.000	4.25e − 08	TBPL1
	cg10783680	0.000	6.23e − 08	EXOC7
Sexual trauma
	cg10338475	0.006	4.74e − 11	CTC−436P18.1;SMIM15
	cg11386818	0.002	3.97e − 09	SYNCRIP
	cg02176407	0.003	5.69e − 08	POU3F2
	cg20807701	0.003	2.80e − 11	DNAJB9;PNPLA8;THAP5
	cg09631059	−0.005	8.48e − 09	RP11-1391J7.1;TSPAN4
	cg06873316	0.001	4.03e − 08	NELL1
War trauma
	cg08985979	0.006	2.10e − 08	AC108142.1
	cg21172322	0.018	5.88e − 10	BCAT1;RP11-662I13.3
	cg00741900	0.007	2.98e − 08	DIO3;DIO3OS;MIR1247

Eleven DMPs were identified in newborns that associated with one of the maternal stress measures and met Bonferroni-corrected significance (p < 7.23 × 10⁻⁸; Figure 1e-h). Two DMPs associated with general trauma (Figure 1e), six DMPs with sexual trauma (Figure 1f), and three DMPs with war trauma (Figure 1g). No DMPs were significantly associated with chronic stress (Figure 1h). All Bonferroni-corrected DMPs and gene annotations are listed in Table 2. There was no overlap between significant DMPs in mothers and DMPs in newborns.

Maternal stress and principal components of DNA methylation

No maternal stress measures were significantly associated with the top ten PCs of DNAm in linear multiple regression analyses after Bonferroni correction for multiple testing.

Maternal stress and GMM

No maternal stress measures were significantly associated with GMM in linear multiple regression analyses.

Maternal stress and epigenetic age

A range of epigenetic clocks were used to test for relationships between biological ageing and maternal stress in mothers and newborns. Among mothers, significant associations were identified between sexual trauma and four different epigenetic clocks (Figure 2). Sexual trauma was positively associated with epigenetic age acceleration as measured by the extrinsic epigenetic age clock (b = 1.09, p = 0.004), the PhenoAge clock (b = 1.36, p = 0.03), and the GrimAge clock (b = 0.42, p = 0.01) and was negatively associated with the telomere epigenetic clock (b = −0.02, p = 0.02). No other maternal stress measures in mothers were associated with measures of epigenetic age acceleration.

Figure 2. Epigenetic age acceleration and maternal stress measures in mothers and newborns. Colors correspond to the unstandardized beta coefficient of a regression of epigenetic age acceleration measures on maternal stress measures. N = 149 for mothers and N = 145 for newborns.

NOTE: IEAA: Intrinsic Epigenetic Age Acceleration.

EEAA: Extrinsic Epigenetic Age Acceleration.

*p < 0.05.

**p < 0.01.

Among newborns, two maternal stress measures were associated with extrinsic epigenetic age acceleration (Figure 2). General trauma (b = 0.70, p = 0.02) and war trauma (b = 1.12, p = 0.048) were positively associated with extrinsic epigenetic age acceleration. No other maternal stress measures in newborns were associated with measures of epigenetic age acceleration. Plots of the significant associations between maternal stress measures and epigenetic age acceleration in mothers and newborns are shown in Figure 3.

Figure 3. Maternal stress and epigenetic age acceleration in mothers (a–d) and newborns (e–f). Darker colour corresponds to greater biological ageing. Sample sizes were N = 149 for mothers and N = 145 for newborns.

DNase I enrichment analyses

Top DMPs identified in EWAS were tested for enrichment of DNase I hypersensitive sites. In newborns, enrichment of DNase I hypersensitive sites was significant in three analyses (Figure 4). The top 125 hypermethylated probes from the general trauma EWAS were enriched for DNase I hypersensitive sites across virtually all cell types and tissues tested (q < 0.05). From the war trauma EWAS, the top 125 DMPs were enriched for DNase I hypersensitive sites in derived neuronal progenitor cultured cells and foetal stomach cells (q < 0.05) and the top 125 hypermethylated DMPs were enriched in seven embryonic, foetal, and stem cell types (all q < 0.05). No evidence for enrichment was identified in mothers with any of the top DMPs for general trauma, sexual trauma, and war trauma (data not shown).

Figure 4. eFORGE analysis for enrichment of DNase I hypersensitive sites in newborns. Analyses were performed with the top 125 differentially methylated, hyper-methylated, and hypo-methylated CpG sites for the three maternal stress measures with Bonferroni-corrected significant sites in the EWAS. The top 125 DMPs were compared to the EPIC array distribution of probes to test whether the top 125 DMPs were significantly associated with DNase I hypersensitive locations in genome.

NOTE: * q < 0.05.

** q < 0.01.

Associations between maternal stress, DNA methylation, and birthweight

Associations between the four maternal stress measures, DNAm, and birthweight were tested. War trauma was negatively associated with birthweight (b = −126.0, p = 0.001; Figure 5a). No other maternal stress measure was associated with birthweight. Next, the 26 Bonferroni-significant DMPs from the EWAS analyses were tested for associations with birthweight. Methylation at one DMP (cg08985979) from the newborn war trauma EWAS was negatively associated with birthweight (b = −4770.2, p = 0.04; Figure 5b). Methylation at cg08985979 was positively associated with war trauma (b = 0.006, p = 2.10e−8; Figure 5c). Thus, greater war trauma was associated with higher methylation at cg08985979 in newborns and lower birthweight. A model linking war trauma as the exposure, birthweight as the outcome, and DNAm as a possible linking mechanism is depicted in Figure 5d.

Figure 5. War trauma was associated with DNA methylation and birthweight in newborns. (a) War trauma was negatively associated with birthweight. (b) DNA methylation at cg08985979 was negatively associated with birthweight. (c) War trauma was positively associated with DNA methylation at cg08985979 in an EWAS. (d) Model linking war trauma with DNA methylation and birthweight.

Discussion

We report novel associations between multiple measures of maternal stress with maternal and newborn DNAm. It is noteworthy that not only the direct victim, but also the indirect recipient of maternal stress demonstrated epigenetic impacts, consistent with the tenets of DOHaD that predict biological impacts of adverse early life exposures. Furthermore, more than 50% of the DMPs identified in the EWAS were associated with maternal sexual trauma, in both mothers and newborns (Table 2), highlighting the devastating and intergenerational impact of rape and related sexual trauma. Interestingly, only mothers showed epigenetic age acceleration in association with sexual trauma and this ageing was manifest as increased ageing of the mother’s immune system (Figure 2, EEAA), increased mortality risk (PhenoAge and GrimAge), and increased telomere shortening (Telomere), all consistent with accelerated biological ageing in the mother. In contrast, newborns showed increased epigenetic ageing (EEAA) that was associated with general trauma and war trauma, suggesting that trauma-related ageing may start very early in life. Finally, of the 26 stress-associated DMPs we identified, we predicted that just a small number of those would also show phenotypic effects. In fact, one DMP that showed altered newborn DNAm was associated with both increased maternal war trauma and decreased newborn birthweight (Figure 5), thus providing the foundation for a model where DNAm links maternal trauma with newborn health outcomes. Additional details on our results are presented below.

EWAS results

Associations were detected between general trauma, sexual trauma, and war trauma in separate EWAS in both mothers and newborns. In total, 26 CpG sites met Bonferroni-corrected levels of significance across eight EWAS (four EWAS in mothers and four EWAS in newborns). The top general trauma DMP (cg11408019) is unannotated, but the nearest gene is RAD54L2. This gene interacts with androgen receptors to regulate transcription and is ubiquitously expressed. Our results with RAD54L2 suggest possible effects of maternal stress on the regulation of androgen-driven transcription, which is associated with growth and reproduction in both males and females [56].

In mothers, the top DMP for sexual trauma (cg06308131) is annotated to MUC4. This gene encodes mucin proteins, which protect epithelial surfaces and may be expressed in the endometrium [57,58]. MUC4 polymorphisms have been associated with endometriosis [59] and recurrent pregnancy loss [60]. Furthermore, sexual abuse has been associated with chronic pelvic pain as well as laparoscopically confirmed endometriosis [61], reflecting our association of sexual trauma and DNAm at MUC4.

Finally, the top DMP for war trauma in mothers (cg26486174) is unannotated, but the nearest gene is HMGA1. HMGA1 is reported to bind DNA and regulate transcription through structural modification and transcription factor interaction. HMGA1 has been associated with proliferative embryonic and tumour tissues [62] as well as placental expression [63] and has been implicated in preeclampsia through interference with trophoblast invasion [64]. DNAm at this gene may associate with foetal health outcomes by affecting trophoblast invasion success.

In newborns, the top DMP for general trauma (cg24590750) is annotated to TBPL1. This gene is a transcriptional regulator highly expressed in testis. Early mouse studies found evidence for a role of TBPL1 in spermiogenesis [65]. A candidate gene study discovered nonsynonymous SNPs in infertile men in TBPL1 [66]. Our results suggest that methylation of TBPL1 may represent a possible mechanism for intergenerational effects of prenatal maternal stress on males.

The top DMP for sexual trauma exposure in newborns (cg20807701) is annotated to several genes, the nearest of which is THAP5. THAP5 has been associated with the cell cycle and cell death [67]. Balakrishnan et al. [68] found that THAP5 was primarily expressed in heart tissue, but also in brain and muscle tissue. These authors noted a decrease in THAP5 protein following myocardial infarction, hinting at a role for THAP5 in heart disease. Other studies have found in utero exposures to be associated with adult heart disease [69], and our results hint at DNAm as a possible mechanism for such associations.

Finally, the top DMP for war trauma in newborns (cg21172322) is annotated to BCAT1. In a rat model, BCAT1 expression was upregulated in foetal brain tissue in a maternal immune activation paradigm [70]. Researchers linked BCAT1 expression to changes in foetal brain metabolism, with implications for brain growth and development. Given the similarity in biological pathways for transfer of stress to the foetus between maternal immune activation and psychosocial stress [71], these findings may be relevant to results reported here. Our DNase I hypersensitivity results show that top DMPs in a war trauma EWAS with newborns were enriched for regulatory elements in derived neuronal progenitor cells. The association we report suggests that brain development in newborns is sensitive to prenatal maternal experiences of war trauma.

Epigenetic age acceleration

Accelerated epigenetic ageing, calculated using several different epigenetic clocks, associated with sexual trauma in mothers. These associations controlled for age and represent biological ageing in excess of that expected by chronological age alone. The directionality of effects of maternal stress on biological ageing was consistent across four different epigenetic clocks. The positive association between sexual trauma and extrinsic epigenetic age acceleration suggests that sexual trauma leads to increased ageing of the immune system as measured by the extrinsic epigenetic ageing clock [49]. This finding is consistent with research describing the proinflammatory effects of intimate partner trauma [72]. Furthermore, the positive associations between sexual trauma and the PhenoAge and GrimAge clocks suggest that maternal sexual trauma is associated with increased mortality risk [51,52] due to accelerated biological ageing. Finally, the negative association between sexual trauma and estimated telomere length is consistent with the idea that telomere shortening can be accelerated by stress [73].

Both general trauma and war trauma were associated with accelerated extrinsic epigenetic ageing in newborns. This finding suggests that biological age at birth of newborns with exposure to increased general trauma and war trauma is already increased. These early exposures may shape later life immune responses, possibly associating with greater risk for psychopathology in adulthood [74]. If supported by additional research, trauma-associated DNAm could be used as a prognostic biomarker to ascertain psychopathology risk on an individual basis [75].

DNase I hypersensitivity site results

Enrichment analyses revealed that only newborn DMPs were enriched for DNase I hypersensitive sites, which mark active regulatory regions of the genome. The top 125 hypermethylated DMPs in the newborn EWAS for general trauma were associated with DNase I hypersensitivity sites across a broad range of tissues. Breeze [54] suggested that this type of general enrichment reflects possible enrichment for housekeeping genes, or genes affecting processes fundamental to cell survival in all tissues. These results suggest widespread effects of general trauma in key regulatory regions focused on hypermethylation in newborns. In contrast, enrichment for DNase I hypersensitivity sites from top war trauma DMPs was more focused on embryonic and foetal tissue types, suggesting that war trauma-associated changes in DNAm may be occurring in developmentally critical tissues during gestation.

Association of birthweight with war trauma and cg08985979

Associations were discovered between maternal war trauma, newborn DNAm at cg08985979, and birthweight. From the newborn war trauma EWAS, cg08985979 was found to be negatively associated with war trauma. This site is annotated to a long non-coding RNA (TENM3-AS1) just upstream of TENM3, which is involved in neuronal development [76]. In our study, increased prenatal maternal stress was associated with an increase in newborn cg08985979 methylation and decreased birthweight. These results suggest that DNAm is a possible mechanism by which prenatal maternal stress becomes biologically embedded to affect newborn health outcomes.

Strengths and limitations

Our study had several strengths and limitations. Studies of prenatal maternal stress outside of European and North American contexts are rare, and increasing the diversity of research participants in epigenomic studies fills an important gap in knowledge [77,78]. Relatedly, studies of prenatal maternal stressors in a Western context frequently focus on anxiety or depression. Our study tested a wider range of stressors and identified associations of DNAm with both standardized measures of general trauma and sexual trauma as well as the ethnographically informed measure of war trauma. One limitation of these measures was that they did not allow us to distinguish the effects of educational background and poverty more generally from the effects of psychosocial stress. Our finding of >50% of maternal and newborn DMPs, as well as maternal accelerated epigenetic ageing, that were associated with sexual trauma suggests that efforts to reduce sexual violence, and gender inequity in general, could have outsize effects on improved health for generations to come. Proposed efforts would include improved education for girls and increased support for hospitals like HEAL Africa that have broad agendas that include education and economic opportunities, conflict resolution initiatives, and programmes targeted to rape victims (such as the Sexual Violence Unit in this study) that provide health care as well as psychological and social support in order to improve maternal and infant health [79]. Due to problems associated with working in resource-limited countries, our sample size was small, which may have limited our ability to detect the typically small effect sizes that characterize EWAS. Effect sizes are generally larger in studies of epigenetic age acceleration, so the small sample size may have been less of a limitation in this part of our study. Another important limitation to consider is the use of blood to analyse epigenetic patterns since the relationship between DNAm in a peripheral tissue, such as blood, with DNAm in organs of relevance, such as brain, remains unclear [80].

Conclusion

We provide evidence that changes in DNAm may act as a mechanism to mediate the effects of maternal stress on newborn health outcomes. Specifically, we find evidence that multiple measures of maternal stress are associated with specific changes in DNAm in mothers and newborns, accelerated epigenetic ageing in mothers and newborns, and association with newborn birthweight. Importantly, we find associations between maternal war trauma, newborn DNAm at cg08985979, and birthweight, providing support for the role of DNAm as a mechanism to link maternal stress and offspring health outcomes.

Acknowledgments

We thank the mothers and children who participated in this study. The authors also acknowledge the doctors as well as the research team at HEAL Africa Hospital who made this study possible: Bisho Mushagalusa, Georgette Kamate, Chantal Nyaramugisha, Anne Marie Rutega, and Bernard Kitumaini. Sonile Peck processed survey data.

Disclosure statement

No potential conflict of interest was reported by the authors.

Data availability statement

DNA methylation data are available at gene expression omnibus under record GSE224365. R scripts are available at github.com/edward-quinn/congo_ewas.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/15592294.2023.2222473

Additional information

Funding

This work was supported by the National Science Foundation under BCS1231264 and BCS1719866. CJH was funded by the National Institutes of Health (NCATS Clinical and Translational Science Awards to the University of Florida, UL1TR001427, KL2TR001429 and TL1TR001428; NICHD F30HD097935).