Exposure to prenatal maternal stress is associated with epigenetic age acceleration and altered cell composition in the placenta: The QF2011 Queensland Flood Study

Abstract

Background:

Exposure to prenatal maternal stress (PNMS) in utero has been associated with several adverse perinatal outcomes, such as pre-term birth and perturbed cognitive development. As the interface between the fetal and maternal compartments during pregnancy, the placenta has been postulated to play a role in this process. We hypothesized that placental DNA methylation (DNAme) may be altered in association with natural disaster-mediated PNMS.

Methods:

Pooled placental samples from the Queensland Flood Study, or QF2011, cohort (n = 105) were processed for assessment of DNAme using the Illumina Infinium MethylationEPIC BeadChip array.

Results:

Overall, we did not find significant associations between placental DNAme and several stress measurements using linear modelling (FDR<0.05 and Δβ>0.03). While we found that XX placentas had slightly higher predicted cytotrophoblast to syncytiotrophoblast cell ratios than XY placentas (p = 0.01), this difference in cell ratio was not associated with PNMS exposure. However, we did observe associations between placental epigenetic age acceleration and all three types of PNMS investigated (objective hardship (QFOSS) (p = 0.03), subjective distress (COSMOSS) (p = 0.03), and maternal cognitive appraisal (CONSEQ) (p = 0.0005) scores).

Conclusion:

The lack of large global impacts of PNMS on placental DNAme possibly indicates that the placenta can buffer moderate levels of maternal stress during pregnancy. It remains unclear what the impact of increased placental epigenetic age acceleration is on fetal development or perinatal outcomes and will require further investigation.

1. Introduction

The Developmental Origins of Health and Disease framework suggests that in utero and early-life exposures can alter health outcomes for offspring later in life [1,2]. One such exposure, prenatal maternal stress (PNMS), has been associated with adverse outcomes, such as pre-term birth [3,4], altered fetal growth [5], and disrupted fetal brain development [6,7]. The mechanisms by which PNMS affects pregnancy and later-life outcomes are unclear, but may in part reflect the effects of increased maternal cortisol on the activity of the fetal hypothalamic-pituitary-adrenal (HPA) axis [8–10]. The adverse effects of PNMS may also differ by fetal sex [11,12] and gestational age at stress exposure [13–17]. PNMS can have diverse effects on pregnancy and fetal outcomes, as the human stress response is individual, and two people experiencing the same (objective) stressor may have completely different (subjective) responses [13,18].

The placenta plays an important role in fetal cortisol metabolism, in part through the expression of 11-βHSD2, an enzyme responsible for inactivating maternal cortisol to limit fetal exposure [13,19]. It has been proposed that increased levels of maternal cortisol (as a result of PNMS) affect 11-βHSD2 expression in the placenta [20–23]. Adverse outcomes such as pre-term birth and low birth weight are more strongly associated with exposure to stress in early-to-mid pregnancy when compared to third trimester exposure, suggesting an early window of vulnerability during pregnancy [13,17,24], and implying that early placentation may be particularly affected by maternal stress.

DNAme refers to the addition of a methyl group onto the 5’ carbon of cytosine and is a well-studied epigenetic mark associated with both gene expression patterns and some societal and environmental exposures, such as air pollution [25] and a variety of gestational stress exposures [26–28]. However, the impact of PNMS on the placental DNAme landscape is not well understood. In the placenta, altered DNAme patterns have been associated with several adverse pregnancy and fetal outcomes [28]. Placental DNAme is also useful for epiphenotyping of the placenta, whereby DNAme is used to infer variables such as cell composition and epigenetic age [29–31]. Epiphenotypes can provide a broad view of placental biology but these relatively new measures and their biological meaning are yet to be fully defined. For example, differences in placental cell type ratios have been associated with variation in placental sampling and processing procedures, as well as gestational age, sex, and ancestry [29]. Accelerated placental epigenetic aging has been associated with environmental exposures such as maternal smoking [32], and potential perturbations in placental function including early onset preeclampsia [33] and reduced fetal weight and growth in males [34,35]. Furthermore, placental epigenetic age acceleration amongst extremely preterm births was linked to systolic blood pressure at 15–18 years of age in males, but not females [36]. Although the biological significance of these associations is unclear, there is a need to better understand the factors that affect placental epigenetic aging and consider a sex stratified analysis when exploring questions about it.

Most research on PNMS in humans has been focused on stressful maternal life events, which may be confounded with variables such as low socioeconomic status, prior mental illness, or lack of established support systems [13,24]. Stress induced by sudden environmental events offers the unique opportunity to study PNMS while avoiding the obstacles to random assignment and confounding variables [13]. Similarly to other types of PNMS, natural disasters have been associated with adverse fetal and child outcomes [13,24,37,38]. Queensland, Australia experienced severe flooding in early January of 2011, affecting over 200,000 people and causing more than $1 billion AUD in damages. The flooding had significant impacts for residents, with two-thirds of individuals polled in random telephone surveys 6 months later reporting being affected by the floods [39]. These effects included evacuation from their homes, property damage, relocation to other temporary housing, loss of income, as well as reported fear for their lives and bodily harm [13,39]. The Queensland Flood Study, or QF2011, was initiated to study the consequences of the flood, and recruited women who were pregnant at the time of the flooding and residing in the general vicinity of Brisbane, Queensland [13]. In the present study, we tested the hypothesis that placental DNAme would be altered in association with differing types of environmentally-mediated PNMS during the Queensland Flood. We investigated DNAme alterations at >750,000 CpGs genome-wide, as well as epigenetic age acceleration and estimated cell composition, in both sex-stratified and sex-agnostic models. We further assessed whether any DNAme-PNMS associations varied depending on gestational age at time of flood exposure or by fetal sex.

2. Materials & methods

2.1. Cohort

The QF2011 cohort consists of over 200 women exposed while pregnant to the flooding [13]. Written informed consent was obtained from all participants, and all procedures complied with the ethical standards on human experimentation and with the Helsinki Declaration of 1975 (revised in 2008). The QF2011 study received ethics approval for the initial and follow-up protocols from the Mater Hospital Human Research Ethics Committee (1709M, 1844M) and the University of Queensland Human Research Ethics Committee (2013001236). Ethics approval for the present placental sub-study, which included 105 placentas from singleton pregnancies, was obtained by the University of British Columbia/Children’s and Women’s Health Centre of British Columbia Research Ethics Board (H16–02280).

2.2. Stress measures

For our sample, subjects were recruited into the study prior to 24 weeks GA, at which time each participant was surveyed to assess objective and subjective stress with the Queensland Flood Objective Stress Scale (QFOSS) and the Composite Scale of Maternal Subjective Stress (COSMOSS), as well as cognitive appraisal of the flood’s consequences (CONSEQ) scale, each of which have been described previously [13]. The assessments were re-administered 12 months post-flooding in order to obtain updated values of financial loss, property damage, etc. [13].

Briefly, QFOSS (objective hardship) was assessed using a questionnaire designed especially for the Queensland flood, as described previously [13]. The scores of each of the four dimensions (Threat, Loss, Scope, Change) ranged from 0 to 50 for a maximum potential score of 200; higher scores indicate higher levels of objective hardship. In the placental cohort, the same group labels for high (n = 26), medium (n = 30), and low (n = 49) objective hardship were retained. In order to deal with the non-normal distribution of the scores (Fig. S1A), which skewed in the direction of lower levels of objective hardship, only the high and low objective QFOSS groups were analyzed; the medium QFOSS group was left out of analysis as no prior threshold was established and we aimed to avoid placing medium QFOSS samples into the high or low groups in error.

COSMOSS (subjective distress) was computed using a PCA-derived algorithm described previously [13]. With mean of 0 and SD of 1, a positive score reflects a higher subjective distress level, whereas negative scores indicate lower levels of subjective distress. In the placental cohort, the group labels for high (n = 28), medium (n = 41), and low (n = 36) subjective distress were retained. Like with QFOSS, in order to deal with the non-normal distribution of the COSMOSS scores (Fig. S1B), which again skewed in the direction of lower levels of subjective distress, only the high and low objective COSMOSS groups were analyzed and the medium group was left out of analysis to avoid introducing error.

CONSEQ (cognitive appraisal) assessed the participants’ overall impression of the flood using the prompt “Overall, what were the consequences of the flood on you and your family?”. The five-point scale of responses included: “Very negative” (1), “Negative” (2), “Neutral” (3), “Positive” (4), and “Very positive” (5) [13]. To isolate the effects of negative cognitive appraisal, scores in the placental DNAme cohort were recoded into 0 (Very Negative, Negative) (n = 37) and 1 (Neutral, Positive, Very Positive) (n = 66) (Fig. S1C). Two participants had no reported CONSEQ score and were subsequently removed from analysis. Other studies have demonstrated that single-item measures are still reliable and valid tools as compared to multi-item assessment tools [40]. In particular, this same cognitive appraisal item from other PNMS disaster studies has been reported to be associated with offspring outcomes such as DNAme in T-cells in adolescence [41], adiposity [42], and cytokines [43].

2.3. Placental sampling

Full details of placental processing have been previously described [29]. Briefly, placentas were sampled in Brisbane, Australia within 60 min of birth. Eight sites (1 cm³ each) representing different cotyledons were sampled across the fetal-facing side of each placenta. Samples were washed and snap-frozen in liquid nitrogen and subsequently shipped to Laval, Canada. Pools of five sites were ground over dry ice, and DNA was extracted using the DNeasy Blood & Tissue Kit (Qiagen, Valencia, CA, USA) in Laval, Canada before being shipped to Vancouver, Canada on dry ice for DNAme processing.

2.4. DNAme arrays and data quality checks

DNA samples were purified and extracted using the DNeasy Blood & Tissue Kit (Qiagen, Valencia, CA, USA), bisulfite converted using the EZ DNAme Kit (Zymo Research, Orange, CA, USA), and hybridized to and processed on the Illumina Infinium MethylationEPIC BeadChip arrays according to the manufacturer’s protocol (Illumina, San Diego, CA, USA) in Vancouver, as previously described [29]. Samples were distributed across array chips and rows with respect to objective hardship (QFOSS high/low) and fetal sex variables, to minimize potential batch effects. The full sample plate map has been published previously [29].

2.5. DNAme data processing

DNAme data were read into R v 4.2.2 for processing. Epiphenotyping variables for gestational age, ancestry, and cell composition (all estimated from the DNAme data itself) were calculated using the PlaNET R package, as described in Ref. [29]. The raw data were normalized for analysis using the adjustedDasen normalization method [44]. Dasen was previously shown to perform well in this dataset [29], and adjustedDasen is an updated version which avoids introducing sex bias into the autosomal data when normalizing data from a mixed-sex cohort that includes sex chromosome data [44]. After normalization, we excluded poor-quality probes (bead count < 3 or detection p value > 0.01 in > 5 % of samples, n = 1953), as well as previously identified cross-hybridizing probes (n = 98,586) [45]. After data processing, a total of 765,320 CpGs (n = 748,167 autosomal; n = 16,833 chrX; n = 320 chrY) in 105 samples remained for analysis. More details on processing for this cohort have been previously described [29].

2.6. Statistical analysis

DNAme data (β values) at all filtered autosomal CpGs (n = 748,167) were converted to M values prior to linear modeling, to test for DNAme differences between high versus low QFOSS/COSMOSS, and negative versus positive/neutral CONSEQ scores. We chose to focus on all CpGs rather than limit to co-methylated regions (an approach that can increase power by reducing tests) aŝ300,000 CpGs would be eliminated in doing so, and these CpGs tend to be in non-promoter regions where most associations are identified. Fetal sex, gestational age at delivery, and two of three PlaNET ancestry estimates were included as additive covariates (probability of East Asian ancestry and probability of European ancestry were selected for inclusion, given that the probability of African ancestry estimate was near zero in this cohort). All covariates included have been shown to drive significant levels of placental DNAme variation [46,47]. Linear models were run using the limma R package [48,49]. The absolute delta beta (|Δβ|) effect size cut-off for all models was established based on the average root mean squared error of the technical replicate pairs, as described in our previous study [50]. Anything less than |Δβ| = 0.03 could represent technical noise and is unlikely to be biologically meaningful [50].

PlaNET cell composition estimates were used to calculate the ratio of cytotrophoblast to syncytiotrophoblast cell content (cyt:syn) in the bulk placental tissue samples [30]. Wilcoxon Rank Sum Tests were used to assess whether the mean cyt:syn ratio differed by sex, trimester of flood exposure, or gestational age at delivery (in weeks), as well as QFOSS, COSMOSS, and CONSEQ levels.

The control placental clock (CPC) was used to estimate epigenetic gestational age, to assess whether PNMS was associated with epigenetic age acceleration. This clock was trained on uncomplicated pregnancies without known pathologies [31]. The gestational ages predicted by the CPC clock are referred to herein as “epi-GA”. Extrinsic epigenetic age acceleration was calculated as the residuals of a linear regression model with chronological gestational age at birth as the independent variable and epi-GA as the dependent variable to provide an overall picture of epigenetic aging. Intrinsic epigenetic age acceleration was calculated in the same manner, but was also adjusted for PlaNET-estimated cell proportions to account for potential cell composition effects. The association between epigenetic age acceleration and QFOSS or COSMOSS scores was evaluated using linear models, both in whole-cohort and sex-stratified analyses. Wilcoxon Rank Sum Tests were used to evaluate whether epigenetic age acceleration varied by CONSEQ level.

3. Results

3.1. Cohort characteristics

DNAme data was obtained from 105 QF2011 cohort placentas and comprised 46 female (XX) and 59 male (XY) placentas (Table 1) and is available as GSE232778 [29]. In the present study we define placental sex by DNAme data-derived sex chromosome complement (see below), and ensured it was concordant in all cases with reported newborn sex. Except for one male delivered at 36 weeks, all births occurred at term (>37 weeks) and the majority were delivered vaginally (n = 81, 77%). To our knowledge, no preeclampsia diagnosis was present for any of the pregnancies in the placental cohort studied. Gestational age at birth was available to the nearest week for all samples, with 3 missing values that were imputed to the median of all other measurements (39 weeks). Most pregnant participants self-identified as ‘white’ (n = 102, 97.1 %) (Table 1). We next looked at demographic and DNAme-derived variable relationships, and found that only expected variables (i.e., trophoblast cell population estimates, PlaNET R package derived ancestry estimates, trimester/gestational age in days of exposure) were strongly associated with each other (R² > 0.5) (Fig. S2). No concerning confounding relationships were identified, and no demographic variables differed significantly by sex.

Table 1.

Demographic and clinical characteristics of the QF2011 cohort.

	Males (XY) N = 59	Females (XX) N = 46
Mean weeks gestational age at Delivery (±SD)	39.25 ± 1.23	39.52 ± 1.11
Method of Birth, n (%)
Emergency C-section	6 (10.2)	9 (19.6)
Planned C-section	8 (13.6)	1 (2.2)
Vacuum extraction	8 (13.6)	6 (13.0)
Vaginal birth	33 (55.9)	28 (60.9)
Vaginal birth - instrumental	4 (6.8)	2 (4.3)
Intrauterine growth restriction, n (%)
Yes	0 (0)	0 (0)
No	10 (16.9)	9 (19.6)
Unknown	49 (83.1)	37 (80.4)
Gestational Hypertension, n (%)
Yes	3 (5.1)	4 (8.7)
No	56 (94.9)	42 (91.3)
Diabetes Unrelated to Pregnancy, n (%)
Yes	3 (5.1)	0 (0)
No	56 (94.9)	46 (100.0)
Gestational Diabetes, n (%)
Yes	5 (8.5)	2 (4.3)
No	14 (23.7)	7 (15.2)
Unknown	40 (67.8)	37 (80.4)
Smoking During Pregnancy, n (%)
Yes	2 (3.4)	2 (4.3)
No	57 (96.6)	44 (95.7)
Alcohol During Pregnancy, n (%)
Yes	1 (1.7)	0 (0)
No	40 (67.8)	37 (80.4)
Unknown	18 (30.5)	9 (19.6)
Self-Reported Maternal Ethnicity, n (%)
Black	0 (0)	1 (2.2)
East Asian	1 (1.7)	0 (0)
European/white	57 (96.6)	45 (97.8)
Other	1 (1.7)	0 (0)
Socioeconomic Class, n (%)
Lower	6 (10.2)	1 (2.2)
Lower Middle	1 (1.7)	0 (0)
Middle	5 (8.5)	4 (8.7)
Upper Middle	16 (27.1)	11 (23.9)
Upper	31 (52.5)	30 (65.2)
Trimester at Flood Exposure, n (%)
1st trimester	36 (61.0)	25 (54.3)
2nd trimester	23 (39.0)	21 (45.7)
3rd trimester	0 (0)	0 (0)
Mean COSMOSS Score, (±SD)	−0.03 ± 1.06	−0.22 ± 0.61
Mean QFOSS Score, (±SD)	17.80 ± 16.55	18.37 ± 14.60
CONSEQ Level, n (%)
Very Negative or Negative	19 (32.2)	18 (39.1)
No Consequence, Positive, Very Positive	38 (64.4)	28 (60.9)
Not Reported	2 (3.4)	0 (0)

3.2. Differential DNAme associated with objective, subjective, and cognitive appraisal of stress

We first assessed autosomal CpGs in three linear models to evaluate whether placental DNAme was associated with objective hardship (high QFOSS vs. low QFOSS, Model A), subjective distress (high COSMOSS vs. low COSMOSS, Model B), and/or cognitive appraisal (negative CONSEQ vs. positive/neutral COSMOSS, Model C) (Fig. 1). Although we treated the stress scores as categorical variables to address their non-normal distribution, we did also test them as continuous variables; again no differential DNAme was observed for any of the PNMS scores.

Fig. 1.

Linear models run to determine differential DNAme associated with PNMS exposure in the whole cohort (A, B, C) and sex stratified cohorts (D,E,F). The QFOSS models (A,D) investigated the effects of high vs. low objective hardship. The COSMOSS models (B,E) investigated the effects of high vs. low subjective distress. The CONSEQ models (C,F) investigated the effects of negative vs. neutral or positive cognitive appraisal of the flood.

No CpGs were differentially methylated (FDR < 0.05, |Δβ| > 0.03) between high and low stress groups in any of the three whole-cohort models tested (A-C). Even at relaxed statistical thresholds of “moderate-confidence” (FDR < 0.15, |Δβ| > 0.03) or “low-confidence” (FDR < 0.25, |Δβ| > 0.03) utilized in other placental studies [50], no CpGs were differentially methylated (Fig. 2).

Fig. 2.

Volcano plots for differential DNAme in association with PNMS severity. False discovery rate (FDR) is depicted along the Y axis. More significant (lower FDR) values are shown at the top of the plot. Vertical dashed lines outline |Δβ| = 0.03, and horizontal dashed lined indicate FDR = 0.05, FDR = 0.15, and FDR = 0.25. (A) Volcano plot for Model A (n = 75). (B) Volcano plot for Model B (n = 64). (C) Volcano plot for Model C (n = 103).

Although we observed no differential DNAme associated with stress in the full cohort, stratification by sex can sometimes reveal sex-differential effects. However, no CpGs were differentially methylated in any of the six sex stratified models run (Fig. 1D–F), at standard (FDR < 0.05) or relaxed significance thresholds (FDR < 0.15, FDR < 0.25) (Fig. S3). To fully explore if the effects of PNMS manifested differently by sex, we analyzed X and Y chromosome data from each of the sexes (i.e., female X chromosome, male X chromosome, male Y chromosome) data separately (Fig. S4). Again, no CpGs met significance (FDR < 0.05, |Δβ| > 0.03).

3.3. Cell composition is associated with sex but not PNMS

As cell composition is a major driver of DNAme variation amongst placental samples and can reflect pathology, we next assessed whether cell composition was associated with severity of PNMS. We focused specifically on the ratio of the two major trophoblast cell populations (cytotrophoblasts (cyt) and syncytiotrophoblasts (syn)), as we previously identified that sex differences in cyt:syn ratios were present in the QF2011 cohort, as well as observed that QF2011 had slightly lower cyt:syn ratios as compared to two other comparable cohorts [29]. We aimed to investigate if there was a potential association between the cyto:syn ratio and PNMS severity. We first looked at the association with cyt:syn ratios in the whole cohort and saw no significant associations with QFOSS, COSMOSS, or CONSEQ (Fig. 3). There were also no significant relationships with any other cell types and any of the PNMS measures, in the whole cohort (Fig. S5).

Fig. 3.

Relationship of stress measures to predicted cytotrophoblast to syncytiotrophoblast (cyt:syn) ratio. Objective hardship (QFOSS) (A), subjective distress (COSMOSS) (B), and cognitive appraisal (CONSEQ) (C). Spearman’s correlation and significance values (A,B) and statistically significant comparisons (p < 0.05) (C) are indicated if present.

We observed that XX placentas overall had higher average cyt:syn ratios than XY placentas (Fig. 4A). Interestingly this was unique to the cyt:syn ratio as upon investigation, no other cell-type ratios had significant differences between the sexes (Fig. S6). As the cyto:syn ratio is known to decrease with increasing gestational age [29], we wanted to ensure that the observed sex difference in cyto:syn ratio was not simply explained by sex differences in gestational age at birth. Reassuringly, there was no association between the cyt:syn ratio and gestational age at delivery by fetal sex, and in fact XX placentas were slightly older on average (39.52 weeks) compared to XY placentas (39.25 weeks) (Table 1, Fig. 4B). Furthermore, the sex difference was not associated with gestational age at exposure to the flooding (Fig. 4C), or trimester of pregnancy at exposure to the flooding (Fig. S7). There were no significantly different trends across QFOSS, COSMOSS, or CONSEQ PNMS levels (Fig. 4D–F), as the interaction term was not significant for the CONSEQ analysis when further investigated, thus could not be concluded to be different by sex [51].

Fig. 4.

Sex stratified associations of predicted cytotrophoblast to syncytiotrophoblast (cyt:syn) ratio. Cyt:syn ratio versus fetal sex (A), gestational age at delivery (B), gestational age at flood exposure (C), objective hardship (QFOSS) (D), subjective distress (COSMOSS) (E), and cognitive appraisal (CONSEQ) (F). For all plots, points are coloured by fetal sex (pink = XX, blue = XY). Spearman’s correlation and significance values (B–E) and statistically significant comparisons (p < 0.05) (A,F) are indicated if present.

3.4. Epigenetic age acceleration associated with PNMS

In adult tissues, stress has been associated with epigenetic age acceleration [52]. Accordingly, we investigated whether PNMS was also associated with epigenetic aging in the placenta. Epigenetic age acceleration was evaluated both without accounting for cell composition (extrinsic) and after adjustment for placental cell types (intrinsic). In the full cohort, intrinsic (Fig. 5B–D), but not extrinsic (Fig. 5A–C) epigenetic age acceleration was significantly associated with increasing QFOSS (p = 0.027) and COSMOSS (p = 0.026) scores. Both intrinsic (p = 0.0027) and extrinsic (p = 0.015) placental epigenetic age were accelerated in cases with negative CONSEQ, when compared to those cases with positive or neutral scores (Fig. 5E and F). We did not observe any sex difference in these associations with epigenetic age acceleration. Specifically, QFOSS and COSMOSS and the age acceleration metrics did not differ by sex and the interaction term between CONSEQ and sex was not significant with either extrinsic (p = 0.60) nor intrinsic (p = 0.46) age acceleration (Fig. S8). Additionally, trimester of exposure stratified analyses were run to investigate the timing of stress exposure during gestation in the context of epigenetic age acceleration, but yielded no significant differences in placental age acceleration between first or second trimester exposure with any of the three PNMS variables. Lastly, epigenetic age acceleration and cell type composition analyses were analyzed separately across QFOSS, COSMOSS, and CONSEQ groupings, as extrinsic, but not intrinsic, epigenetic age acceleration inversely correlated with cyt:syn ratio (R = −0.26) at a similar rate in both sexes (Fig. S9). Across all maternal stress variables and both epigenetic age acceleration metrics, there was no clear pattern to cell type composition changes in the whole cohort or in the sex stratified cohorts (Fig. S10).

Fig. 5.

Relationships between PlaNET-estimated extrinsic (A,C,E) and intrinsic (B,D,F) epigenetic age acceleration and QFOSS (A,B), COSMOSS (C,D), and CONSEQ (E,F). Statistically significant (p < 0.05) Spearman’s correlation values (A–D) and Wilcoxon Rank Sum Test values (E–F) are shown above respective plots if present.

4. Discussion

In this study, we examined whether placental DNAme was altered in association with differing types of PNMS induced by the 2011 Queensland, Australia flooding. We did not detect significant DNAme differences at individual CpG sites associated with either QFOSS, COSMOSS, or CONSEQ in the whole cohort, or sex-stratified subsets. Interestingly, we observed that higher QFOSS and COSMOSS were moderately correlated with accelerated intrinsic (cell-type adjusted) epigenetic age, while both intrinsic and extrinsic epigenetic age acceleration were significantly higher in association with negative CONSEQ scores compared to cases with neutral or positive CONSEQ scores.

Previous work has demonstrated in utero psychosocial maternal stress exposure and changes in placental methylation [53–55]. Tesfaye et al. reported that elevated perceived stress during the third trimester was associated with differential methylation in the placenta at 2 CpGs [53]. We specifically considered these loci, but one of the two CpGs was not present on the EPIC array, while the other showed no association in our data even with nominal p value. The same group reported 16 CpGs associated with maternal depression during the second and third trimesters (16 CpGs), while Lund et al. (2021) reported changes in DNAme at 2833 CpG sites with maternal depressive symptoms in early pregnancy (gestational week 14), which were over-represented in genomic enhancers and genes overlapping or nearby functionally enriched for neurological development [53,54]. Differential methylation at 7 CpGs was reported in association to maternal access to prenatal social supports, again with methylation at 2 of the 7 CpGs having been correlated with placental expression genes involved in neurodevelopment and energy metabolism [55]. Although we found no associations with any of these sites, analysis of different exposures in differing populations make it difficult to compare results across studies.

Positive epigenetic age acceleration in adults has been associated with a variety of stress exposures, including psychosocial stress and low socioeconomic status; this age acceleration has been associated with adverse health outcomes, shorter lifespan and overall mortality risk in several different tissues [56–60]. In the placenta, modest increases in epigenetic age acceleration have been associated with improper placental function and pregnancy complications, such as early-onset preeclampsia and sex differences in fetal weight and growth [33,34, 61]. A greater association with epigenetic age acceleration has been observed in males as compared to females both in the context of environmental exposures [35], fetal growth [33], and long-term health outcomes [36]. Furthermore, accelerated placental and cord blood epigenetic age was linked in one study to male sex generally [62]. Additionally, an association with accelerated placental epigenetic aging with maternal antenatal depressive symptoms in the second trimester was reported; this association was also significant in males but in females [63]. In the present study, we found that intrinsic (cell-type adjusted) but not extrinsic, epigenetic age acceleration was associated higher QFOSS and COSMOSS, and both measures were significantly higher with negative CONSEQ. Although when stratified on sex the association between intrinsic age acceleration and CONSEQ reached significance in XX (female) but not XY (male) samples, it trended in the same direction in both, and the sex*epigenetic age acceleration interaction term was not significant in either measure, thus failing to support a difference in epigenetic age acceleration by sex. The consequences of age acceleration in the context of the placenta are unclear as this clocks were built in a manner to be robust to pathology and are based largely on intragenic CpGs that change consistently with gestational age, but do not have functional significance [31]. Epigenetic age acceleration in the placenta has been correlated with long term health outcomes (e.g. systolic blood pressure in teenagers), but if maternal stress contributes to placental epigenetic aging, then such stress might be independently linked to such outcomes (i.e. due to shared genetic or environmental factors) rather than resulting from altered placental function directly.

Additionally, we observed that XX placentas had higher cyt:syn ratios than XY placentas, confirming previous reports of sex differences in cyt:syn ratios/cell composition [29]. While the cyt:syn ratio decreases across gestation [29], this observation was not explained by gestational age differences between XX and XY placentas at delivery. Additionally, a recent meta-analysis aiming to examine the effect of in vitro fertilization (IVF) on the placental DNAme landscape observed this same association, in both spontaneous and IVF placentas compiled from multiple public DNAme datasets [64]. The causes and implications of this apparent sex difference in estimated cyt:syn cell type proportions remain unclear, although we found no evidence that this ratio differed by PNMS exposures in utero. It remains unclear if cell composition sex differences are associated with the differences we see in male and female vulnerability to exposures and prevalence of placental complications [14,65,66], but future investigation is warrented.

Other studies conducted on this QF2011 cohort include studies of cognitive development and urinary metabolomes in children exposed to the flood in utero, as well as placental mRNA expression analysis [67–70]. Moss et al. (2017) reported that increased levels of subjective distress (COSMOSS) and negative CONSEQ were associated with poorer child motor development at 16 months of age [67]. St. Pierre et al. (2018) reported that subjective distress (COSMOSS) was significantly associated with a reduction in specific placental mRNA expression levels involved in the glucocorticoid system and glucose transport [69]. Heynen et al. (2023) examined the urinary metabolomic profile of children (at 4 years old) exposed prenatally to the flooding and saw significant alterations in metabolic pathways linked to oxidative stress in association with high and low QFOSS and COSMOSS scores; oxidative stress pathways have previously been linked to PNMS via the HPA axis [70]. These studies are in line with our results, which suggest different types of PNMS can have differential molecular effects, and taken together the QF2011 studies may suggest that different forms of PNMS affect distinct biological pathways.

Altered DNAme and/or gene expression has been reported in several different tissues in association with other environmental disaster studies of PNMS. For example, in a study of genome-wide DNAme in T cells from adolescents exposed in utero to the 1998 Québec Ice Storm, the largest number of differentially methylated CpGs were associated with maternal cognitive appraisal (CONSEQ), followed by objective hardship (Storm32, analogous to QFOSS), while no CpGs were differentially methylated in association with subjective distress (as assessed by the IES-R) [41,71]. These results from Project Ice Storm mirror our findings of epigenetic age acceleration being most strongly associated with CONSEQ scores. In a study profiling mRNA expression levels of 40 selected HPA-axis and neurodevelopmental genes in placentas exposed to Superstorm Sandy, genes vital for placental function and fetal development were downregulated across all trimesters of exposure [23]. The same group then performed whole-transcriptome analysis of Superstorm Sandy exposed-placentas; notably, many of the most differentially expressed genes were syncytiotrophoblast-specific, with glucocorticoid response elements in their promoter regions [72]. DNAme is quite a stable epigenetic mark that is strongly influenced by cell composition and genetic variation. In comparison, gene expression is in flux and can change with altered availability of transcription factors or chromatin remodelers [73]. Thus, gene expression may be more likely to be responsive to environmental influences, potentially explaining the absence of large-scale DNAme changes observed in the current study, and in other PNMS-DNAme studies as compared to PNMS gene expression analyses.

Our study is not without its limitations. Importantly, we were only able to analyze placentas from women who were exposed to the flood in their first or second trimester; women exposed to the flood in the third trimester gave birth before ethics approval for the study could be obtained. Additionally, the differing times of flood exposure during gestation can introduce heterogenous cumulative exposure, further complicating analysis. Our cohort also included a very large majority of women of upper or upper-middle socioeconomic status (83.8 %) and white self-reported race (97.1 %). Thus, efforts should be made to evaluate whether these trends replicate in more diverse cohorts. Lastly, levels of PNMS were not normally-distributed in the present cohort, with a significant majority of the cohort skewing towards lower to moderate levels of PNMS exposure, and thus we cannot comment on the extent to which our findings would extend to higher stress cohorts. It is important to note that epigenome-wide association studies are underpowered for detecting small effect size DNAme alterations, or when there is a heterogeneity in sample response [74]. While we may have been able to obtain greater power by identifying and testing only co-methylated regions, this approach would also have led to a significant loss of data.

The major strength in our study lies in the analysis of PNMS in a natural disaster context, as this acts as a natural experiment that occurs independently of confounding variables typically associated with maternal stress, such as socioeconomic status, maternal mental health or the availability of support systems. The QF2011 flooding also had a sudden onset, allowing fine-grained analysis of timing effects in gestation. This natural experiment with an independent stressor also allowed us to determine what aspects of the mothers’ stress experience were most important: degree of hardship, levels of distress, or cognitive appraisal. In addition, we considered the impact of sex, both in sex-stratified and interaction analyses, and in our direct examination of the sex chromosomes, which are often overlooked in biological studies. Additionally, we accounted for cell composition differences and evaluated cell composition distributions across the cohort for association with technical variables [29].

In conclusion, this study aimed to provide a better understanding of the impact of PNMS on the placental DNAme landscape, which we and others have hypothesized may be associated with adverse offspring outcomes observed in association with PNMS. Overall, we did not find large impacts of PNMS associated with the 2011 Queensland flood on the placental DNA methylome, which may imply the placenta’s ability to buffer moderate levels of PNMS in pregnancy. We did, however, observe associations between PNMS and advanced placental epigenetic age acceleration, particularly with negative maternal cognitive appraisal (CONSEQ). While this is intriguing as it parallels results of stress on epigenetics in somatic tissues, the long-term impacts of accelerated placental epigenetic age are unknown, providing a direction for further research. Our results would benefit from replication in more socioeconomically and genetically diverse cohorts, extension to cohorts with potentially higher reported levels of PNMS, and further understanding of the factors that affect placental epiphenotypes to fully elucidate the effects of PNMS on placental epigenetic mechanisms.

Abbreviations:

COSMOSS

Composite Scale of Maternal Subjective Stress

CONSEQ

Cognitive appraisal of the flood consequences

CPC

control placental clock

DNAme

DNA methylation

epi-GA

gestational ages predicted by the CPC clock

HPA

hypothalamic-pituitary-adrenal

PNMS

prenatal maternal stress

QFOSS

Queensland Flood Objective Stress Scale

QF2011

Queensland Flood Study

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.placenta.2025.09.003.

Availability of data

The DNAme data included in this study is available on the Gene Expression Omnibus under accession number GSE232778. Additional clinical data included in the present analyses are available from the authors upon reasonable request.