Exploring the effect of antenatal depression treatment on children’s epigenetic profiles: findings from a pilot randomized controlled trial

Many pregnant women experience clinically significant depressive symptoms before delivery, with an estimated prevalence of 7.4 to 12.8% [1]. Mounting evidence demonstrates that children prenatally exposed to maternal depression more often have a difficult temperament [2], are more prone to develop internalizing and externalizing behavioral problems [3–7], show poorer performance on cognitive tasks [8, 9], and more often develop depression and anxiety symptoms themselves in (pre)adolescence [10–12]. One mechanism by which antenatal depression might influence susceptibility for psychopathology is by epigenetic regulation of gene expression [13, 14]. Epigenetic mechanisms regulate the activity of DNA and include post-translational histone modification, micro-RNAs, and DNA methylation [15]. In contrast to the fixed genotype, the epigenome has shown to be highly variable early in development under the influence of environmental factors [16, 17].

Animal studies have provided evidence that antenatal stress alters methylation of offspring genes involved in neurodevelopment and is associated with behavioral changes. For example, exposure to chronic stress in early gestation in mice resulted in a stress-sensitive phenotype in male offspring, showing increased immobility in the tail suspension and forced swim test and heightened hypothalamic pituitary adrenal (HPA) axis responsivity, which was accompanied by increased DNA methylation and decreased gene expression of the glucocorticoid receptor in the hippocampus and amygdala [18]. Moreover, alterations in epigenetic profiles have been shown to remain stable across generations, passing on susceptibility for emotional and behavioral disorders from one generation to the next [19].

Since 2008, many human studies have investigated associations between prenatal stress exposure and offspring gene methylation, with a special focus on NR3C1, coding for the glucocorticoid receptor [20]. While the reported effect sizes are usually small, increased methylation status of NR3C1 has been linked to an increased HPA axis stress-response [21]. All studies to date are, however, observational and therefore susceptible to confounding by factors that are both associated with antenatal stress and with methylation patterns, such as maternal smoking during pregnancy [22]. Experimental designs including follow-up of children are currently scarce and urgently needed to establish causality between intrauterine exposures and later life outcomes [23].

The current study investigated effects of maternal depression treatment during pregnancy on DNA methylation profiles in the children. In the Beating the Blues before Birth (BBB) study, pregnant women with a confirmed Diagnostic and Statistical Manual of Mental Disorders, 4th Edition (DSM-IV) depressive disorder were randomized to either the intervention group, consisting of eight cognitive behavioral therapy (CBT) sessions, or to a control group, consisting of treatment as usual (TAU), which comprised case-managing by a midwife or referral to a general practitioner. Beneficial treatment effects favoring the intervention were found on maternal depression and anxiety. Anxiety symptoms significantly decreased, and depressive symptoms showed a decreasing trend nearly reaching significance, in the intervention versus the control group [24].

We hypothesized that compared to the control group, the intervention would be associated with a change in DNA methylation profiles of buccal swab DNA from the children, (1) at an epigenome-wide level, (2) at 16 a priori selected candidate genes, and (3) at promoter-associated glucocorticoid receptor (NR3C1) probes. We additionally explored whether severity of maternal symptoms of depression and anxiety at baseline would be associated with DNA methylation profiles in the children, regardless of treatment.

In this follow-up of one of the first randomized controlled trials on the effect of antenatal psychological depression treatment (CBT) on children’s DNA methylation patterns, we found no robust evidence of widespread methylation differences between children of women in the control or intervention group. However, at a pre-specified nominal significance level of p < 0.01, 4780 differentially methylated probes according to allocation pointed to an overall 2.7% lower DNA methylation level of probes in children from the intervention group. Applying a candidate approach, non-significant trends toward lower DNA methylation in the intervention group were seen in OXTR, MEST, MEG3, H19, and CRHR2. We did not find a significant difference in mean DNA methylation of 34 NR3C1 promoter-associated probes between the intervention and control groups. Nevertheless, the majority of probes (68%) showed lower DNA methylation in the intervention group compared to the control group, with cg26464411 as topmost differentially methylated probe, a CpG site that has been associated with depression in earlier studies [25, 26]. Whether these trends are persistent and clinically relevant remains to be determined in future studies with larger sample size and longer follow-up.

Of the top five probes that were most differentially methylated between the intervention and the control group, three corresponded to annotated genes: cg15495292 on the AIG1 gene, which is a gene involved in androgen regulation; cg18818484 on the PTCHD2 gene, which is involved in neuronal proliferation and differentiation; and cg05155812 on SUN1, a gene that potentially plays a role in neuronal migration and cerebellar development. These findings may be relevant as the desired effect of a prenatal intervention would be to target genes that mediate the associations of prenatal stress, depression or anxiety with adverse neurodevelopmental disorders in children [27, 28]. Our results are promising, but evidently replication in larger studies is necessary.

Additionally, we revealed trends toward lower DNA methylation in children from the intervention group compared to the control group in 5 out of 16 candidate genes that have previously been associated with prenatal exposure to maternal stress, depression, or anxiety. These trends were observed in OXTR, the gene coding for the Oxytocin receptor; the MEST gene, a gene involved in metabolism; MEG3, a long noncoding RNA; H19, an imprinted gene; and CRHR1, a gene for corticotrophin releasing hormone receptors. We did not find a significant difference in mean DNA methylation between the intervention and control group on the promoter region of the NR3C1 gene, coding for the glucocorticoid receptor. Nevertheless, cg26464411 showed a trend toward lower DNA methylation in the intervention group. This CpG site has been positively correlated with depressive symptoms or hypercortisolism in earlier studies [25, 26]. Although our results were not significant, the trends we have observed were in line with our expectations, based on earlier findings from observational studies showing increased methylation of NR3C1 in newborns and young children of antenatally stressed, depressed, or anxious women [20, 29], which was associated with increased stress responses [21, 30].

The women in the current study were treated at a mean of 18.6 weeks gestational age, and it may be possible that the effect of treatment on offspring DNA methylation would have been stronger if the women had been treated earlier in their pregnancies. Increased attention is currently focused on the period of early pregnancy, and even the preconception period, as an important time window for adverse environmental factors inducing prenatal programming, which has been shown in animal studies [18]. Further evidence in humans is derived from studies examining prenatal famine, in which the largest effect on offspring methylation was found after prenatal exposure to undernutrition in early pregnancy [31]. We did not test for an interaction between allocation status and gestational age on mean methylation in candidate genes because of the lack of significance in the initial analyses, but in larger future studies, exploring moderation through gestational age would be highly informative to identify treatment effects on DNA methylation during specific stages of pregnancy.

A limitation of the study was a lack of statistical power, as we were only able to include approximately half (23/54 = 43%) of the original sample in this follow-up. Nevertheless, associations between prenatal stress and methylation status of NR3C1 have been reported in studies with a similar sample size [30, 32]. It was of interest that women who participated in the current follow-up study had lower levels of depression and anxiety at baseline compared to the participants that were lost to follow-up (Table 1). Also, they were observed to have higher incomes and were more highly educated at baseline. However, attrition bias is not likely to have occurred as this was the case in both groups [33]. Despite no formal statistical tests being conducted [34], it was evident that the difference in anxiety (BAI) scores before and after treatment between the intervention and control group was twice as high in the non-responders compared to the responders (14.5 versus 7.5), indicating that women with greater response to treatment were relatively underrepresented in the current sample. Additionally, some women in the control group also reported accessing psychological or medical treatment outside the trial [24]. This, and the lower participation of those who responded better to treatment, might have led to an underestimation of the effect of therapy on methylation profiles in the children in the current study.

Although both groups were reasonably balanced in terms of psychological and sociodemographic factors at the time of follow-up, it is still possible that other, unmeasured factors are (partly) responsible for the trends observed in the children’s epigenetic profiles according to allocation status. Because of the small sample size of our study, we chose to include only those variables that were likely to attribute mostly to the variation in DNA methylation, such as child gender, age, birth weight, and income. We did not include educational attainment, although this also appeared to be somewhat higher in the intervention group (although not statistically significant, results not shown). In addition, maternal body composition in pregnancy, pregnancy complications, and mode of delivery were not recorded in the original study files, and hence, not included in the current study. As these factors may act as mediators in the causal path from improved mood in pregnancy to better child outcomes, in future studies these variables should be included as well. Nevertheless, we did have access to the children’s birth weight, an important marker for general health of the baby, which showed to be similar between both groups. Also, we were unable to control for PC5 in the analyses, as none of the variables included in the model was associated with PC5. Nevertheless, the contribution fraction of PC5 to the variation in DNA methylation was very marginal compared to the contribution fraction of PC1, PC2, PC3, and PC4, which were associated with known variables and therefore were controlled for in our analyses. Finally, we did not adjust for cellular heterogeneity in our study. The most widely applied method is the reference-based deconvolution method originally described by Houseman et al., which permits the estimation of the proportion of various cell types within a sample based on existing reference data sets [35]. For blood, several studies have analyzed the methylation profile of the specific cell- types present in whole blood, which can serve as reference data. However, for saliva, this has not been performed systematically, but studies that have applied the Houseman deconvolution method on salivary genome wide DNA methylation data (combining reference methylomes from leucocyte subtypes and buccal epithelial cells references methylomes) have shown that saliva is less heterogenic compared to blood [36].

The impact of the postnatal environment on methylation profiles in children also cannot be ignored. Exposure to stressful life events from birth to adolescence has been associated with higher NR3C1 methylation [37]. Although in both intervention and control group, more women were currently using antidepressant medication compared to when they were pregnant at enrollment of the original study, this was much more pronounced in the control group (relative increase of 43.3%) compared to the intervention group (relative increase of 16.7%). These observations may be consistent with a potential longer-term beneficial effect of treatment in the women, which in turn, might have positively affected child outcomes. Women from the intervention group also reported higher incomes compared to baseline, which was not the case in the control group, although including income as additional covariate did not significantly alter the results. To be able to isolate the effect of antenatal CBT on offspring DNA methylation in utero, prior to any postnatal confounding, evidence from trials that include cord blood and/or placenta samples for DNA methylation (and gene expression) are needed.

Finally, it has not yet been fully elucidated how maternal depression affects child adversity. Nevertheless, epigenetic modification of fetal genes in response to increased cortisol exposure, either directly or via a decrease in placental inactivation, has been widely accepted as a potential underlying mechanism. Although our study findings could not robustly support this hypothesis, the trends observed are in line with earlier evidence. The existing evidence is nearly exclusively based on findings from experiments in animals and observational human studies. The fact that the exploratory findings from this novel experimental study in humans are in line with the available evidence is therefore promising. It must be noted that we mostly looked at statistically significant results at an uncorrected p-value level. The results of our study should therefore be interpreted with caution. Although the observed effect sizes were small, with mean differences of 1–5% in methylation status, they are in line with earlier evidence [20]. Because of the lack of studies with a comparable study design, it is not yet possible to replicate our findings in a similar trial; however, plans for a larger trial are currently in progress.

We found preliminary evidence of a possible effect of cognitive behavioral therapy during pregnancy on widespread methylation and a non-significant trend towards lower methylation of a specific CpG site previously linked to depressive symptoms and child maltreatment in the intervention group. However, none of the effects survived correction for multiple testing. Larger studies are now warranted.