Background

Social support promotes mental and physical health in low stress environments [1, 2] and buffers the effects of stress in high stress environments [3, 4]. Maternal resilience factors such as prenatal social support have been linked to higher leukocyte telomere length in newborns [5] and lower adiposity during infancy [6]. Moreover, poor social support in early childhood may influence health outcomes later in life [7]. However, little is known about the biological mechanisms that underlie the relationship between prenatal social support and subsequent health outcomes.

The placenta undergoes dynamic DNA methylation changes throughout pregnancy in response to biological and environmental factors to provide an optimal environment for fetal development [8, 9]. Emerging evidence suggests that epigenetics may partly explain the link between prenatal psychosocial factors, such as maternal stress and depression, and child health outcomes [10]. Therefore, it is possible that social support during pregnancy may influence fetal development and long-term health outcomes by altering the placental epigenome. However, there is no published study on the association between social support and genome-wide DNA methylation of human placenta. Prenatal social support in humans has been associated with DNA methylation in maternal blood [11], and social rank in primates has been associated with placental DNA methylation [12]. Low social support has been linked to inflammation [13], and quality of prenatal social support has been linked to inflammation during pregnancy and early infancy [14, 15]. Therefore, identifying placental DNA methylation changes associated with prenatal social support in low-risk pregnancies may shed light on the molecular mechanisms underlying the effects of social support on fetal development, crucial information for developing interventions to promote fetal development and long-term health outcomes.

Using the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) Fetal Growth Studies (FGS) cohort data [16], we investigated the association between maternal social support during pregnancy and genome-wide DNA methylation in placenta at delivery. Given accumulating evidence on sex differences in placental methylation [17,18,19,20] and placental response to adverse prenatal environments [10, 21, 22], we also investigated the association separately in male and female fetuses. For cytosine-(phosphate)-guanine sites (CpGs) found to be significantly associated with social support, we examined whether methylation of CpGs was associated with expression of nearby genes in placenta.

Methods

Setting and subjects

We used data from the Eunice Kennedy Shriver NICHD FGS – Singletons. Among the total 2802 participants, 312 had placenta samples collected at delivery. Participants who provided placenta and those who did not provide placenta did not have significant differences in maternal age, fetal sex, job status, educational status, social support, or perceived stress scores (Additional file 1: Table S1). The study was approved by the Institutional Review Boards of NICHD and all respective participating clinical sites. All participants provided informed consent at enrollment into the study.

The participants were low risk pregnant women enrolled at gestational ages of 8 to 13 weeks from 12 clinics in the USA during the period between July 2009 and January 2013. The inclusion criteria were age 18–40 years, viable singleton pregnancy, and planning to give birth at the participating health facilities. Exclusion criteria were previous history of poor obstetric outcomes, pre-existing chronic medical and psychiatric conditions, smoking in the previous 6 months or use of illicit drugs during the previous 12 months, and consumption of ≥ 1 alcohol drink daily [16].

Main exposure variable

Maternal social support was assessed at enrollment using the self-report Enhancing Recovery in Coronary Heart Disease Social Support Instrument (ESSI) [23]. ESSI uses seven items for assessing the degree of social support an individual has. The higher total scores higher scores indicate greater degree of social support.

Covariates

Data on maternal age, parity, education, maternal job status, pre-pregnancy BMI, self-identified race/ethnicity, gestational age at delivery, and fetal sex were obtained through interviews and from medical records as described elsewhere [16]. Perceived stress was measured using the self-report ten-item Cohen’s Perceived Stress Scale (PSS-10). A higher PSS-10 score indicates greater level of perceived stress [24].

Placenta sample collection and DNA methylation quantification

Placentas obtained at delivery were rinsed with sterile saline, pat dried with paper towel, and had nonadherent clots removed. The placental membrane and umbilical cord were trimmed before biopsies were taken. Four biopsies measuring 0.5 cm × 0.5 cm × 0.5 cm were collected directly below the fetal surface of each placenta within 1 h of delivery. The samples were placed in RNALater and frozen at – 80 °C for molecular analysis. The placental biopsy samples were processed at the Columbia University Irving Medical Center as described previously [25]. DNA was extracted from the samples and assayed using the Illumina Infinium Human Methylation450 Beadchip (Illumina Inc., San Diego, CA) array. A total of 301 placental samples that passed quality control were included in the analysis [26]. Eleven samples were excluded because they were outliers from the distribution of genetic clusters of the sample (n = 6), genotype sex mismatch between fetus and placenta (n = 4), and mismatch of sample identifiers (n = 1).

Standard Illumina protocols were followed for background correction, normalization to internal control probes, and quantile normalization. The Illumina 450k array’s plating scheme was adjusted according to the assay’s internal QC design. The GenomeStudio QC standard was implemented during data preprocessing, and the internal probes have been used for background correction, dye bias correction, normalization, probe-design bias correction, and an offset for Infinium I and II probe intensity. The assay quality controls comprised of controls for measuring staining sensitivity and controls for testing efficiency of bisulfite conversion. Bisulfite modification was performed using the EZ Methylation kit (Zymo Research, CA). Bisulfite-converted sequences without CpGs served as negative controls; the mean of the negative control probes was used as the system background. The resulting intensity files were processed with Illumina’s Genome Studio which generated average beta values for each CpG site (i.e., the fraction of methylated sites per sample by calculating the ratio of methylated and unmethylated fluorescent signals) and detection P-values which characterized the chance that the target sequence signal was distinguishable from the negative controls. The method was corrected for probe design bias in the Illumina Infinium Human Methylation450 BeadChip and achieved between-sample normalization. Normalization was performed using the modified Beta Mixture Quantile dilation (BMIQ) method to correct the probe design bias in the Illumina Infinium Human Methylation450 BeadChip and achieve between-sample normalization [27].

Missing CpGs were imputed by the k-nearest neighbors method, setting k = 10. Beta values with an associated detection P ≥ 0.05 were set to missing. Probes with mean detection P ≥ 0.05 (n = 36), cross-reactive (n = 24,491), non-autosomal (n = 14,589), and CpG sites within 20 bp from a known single nucleotide polymorphism (SNP) (n = 37,360) were removed [20]. Consequently, methylation data for 409,101 were obtained for analysis. We transformed the beta values to M value scale before analysis as recommended using the formula M = log2(Beta/(1-Beta)) [28].

RNA extraction and quantification

RNA from 80 placenta samples was isolated using TRIZOL reagent (Invitrogen, MA, USA). The mRNA libraries were sequenced on an Illumina HiSeq2000 machine with 100 bp paired-end reads as described elsewhere [25]. Data from 75 participants who had both DNA methylation and RNA-seq data were used for the methylation and gene expression association tests.

Statistical analysis

Association between CpG sites and social support

We performed epigenome-wide analyses using multiple linear regression models with the DNA methylation CpG site as response variable on the M value scale and maternal social support scores as predictor. We also performed similar analyses by subgroups based on fetal sex. All regression analysis models were adjusted for maternal age, parity, education, maternal job status, pre-pregnancy BMI, self-identified race/ethnicity, gestational age at delivery, fetal sex, maternal perceived stress scores measured at recruitment, 10 genetic principal components computed from genome-wide autosomal SNP genotypes of placenta from HumanOmni2.5 Beadchips (Illumina Inc., San Diego, CA) to adjust for population structure, three methylation-based principal components, methylation sample plate, and components based on putative cell-mixture estimates using surrogate variable analysis (SVA) to account for confounding by variation in cell composition [29]. In sensitivity analyses, linear regression models were additionally adjusted for cell composition variables created using methods developed by Yuan et al. [30]. Further sensitivity analysis was performed to assess whether the significant associations remain in a statistical model that did not include maternal sociodemographic factors (i.e., by excluding maternal age, parity, education, maternal job status, pre-pregnancy BMI, self-identified race/ethnicity, and gestational age at delivery from list of adjusted covariates). We assessed the direction of association and correlation of methylation fold-changes (logFC) between the fully adjusted model and the model without sociodemographic covariates.

Differentially methylated CpG sites were mapped to genes within 250kb using R/Bioconductor package (IlluminaHumanMethylation450kmanifest) with a reference consisting of all genes present in the Illumina 450k platform. P-values were corrected for false discovery rate (FDR) using the Benjamini-Hochberg method. P-values were further corrected for genomic inflation (λ) by applying a Bayesian method in R/Bioconductor package (BACON) [31]. Quantile-Quantile (QQ) plots were generated for the regression models before and after BACON correction. The QQ plots do not exhibit significant inflation of the p-values with λ = 1.0, λ = 1.03, and λ = 0.97 after BACON correction for the overall, male-specific, and female-specific results, respectively (Additional file 1: Figure S1–S3). For sex-stratified analyses, we followed the approach described by Randall et al. which implements Welch’s t-test [32] to categorize the associations into one of three groups: (i) concordant effect direction (CED) defined, for effects sizes in the same direction, as association that is significant at PFDR < 0.05 in one fetal sex and at least nominally significant in the other fetal sex; (ii) single sex effect (SSE) when significant association is present in one fetal sex (PFDR < 0.05) and no association observed in the other fetal sex; or (iii) opposite effect direction (OED) defined, for effect sizes in opposite direction, as association that is significant in one fetal sex (PFDR < 0.05) and at least nominally significant in the other fetal sex. Post hoc statistical power analysis was performed using two-tailed tests assuming probability of error (α) = 0.05 and demonstrated that the study power was ≥ 90% for detecting the effect sizes of 82% of the CpGs found to be associated with social support in the overall as well as sex-stratified analyses (Additional file 1: Figure S4).

We employed the R package dmrff to identify differentially methylated regions (DMR) in placenta associated with maternal social support at 5% FDR [33]. A DMR was defined to have a maximum length of 500 base pairs harboring a set of CpGs with EWAS P < 0.05 and identical effect direction.

Association between DNA methylation and gene expression

We analyzed association between DNA methylation at differentially methylated CpG sites and placental expression of protein-coding genes located within 250kb up- and downstream from the CpG sites using linear regression. Correlations between expression of the genes and social support scores were assessed using Pearson’s correlation test.

Functional annotations and regulatory enrichment

We examined whether genetic variants influence DNA methylation levels of the CpGs associated with social support. For this, we explored the CpGs in the list of known placental methylation quantitative trait loci (mQTLs) [25].

Using eFORGE version 2.0 [34], we examined enrichment and depletion of the CpGs significantly associated with social support (PFDR < 0.05) for tissue or cell-type specific regulatory features. The CpGs identified in the total, male, and female samples were submitted to eFORGE and evaluated separately for overlap with DNase I hypersensitive sites, all 15-state chromatin marks, and all five H3 histone marks (i.e., H3K27me3, H3K4me1, H3K4me3, H3K36me3, H3K9me3).

Pathway enrichment analysis

We examined the biological functions of genes annotated to the top 100 CpG sites associated with social support using Ingenuity Pathway Analysis (IPA, Qiagen, Redwood City, CA, USA), separately for the overall and sex-stratified analysis results. Enriched biological pathways which contain at least two of the query genes and with P-values less than 0.05 were considered significantly enriched.

Results

The characteristics of the 301 participants have been described previously [35]. Briefly, the mean age (SD) of the women was 27.7 (5.3) years; 50.5% of the fetuses were male. The median (interquartile range, IQR) of ESSI scores was 24 (22–25). The ESSI scores were relatively low with the 75th centile being equivalent to the 25th centile of the ESSI tool development study where the participants were individuals who had recent myocardial infarction [36]. The median (IQR) perceived stress score was 11 (6–14) as described elsewhere [21], which is lower than the corresponding figures in a US cohort of pregnant women during the first trimester [37] and normative data of Swedish women 14 (10–19) [38]. ESSI scores were positively correlated with having high school or higher educational status (r = 0.16, P = 0.007) and being employed (r = 0.12, P = 0.046) and inversely correlated with higher PSS-10 scores (r = − 0.34, P = 2.2 × 10−9).

Maternal social support and DNA methylation in placenta at delivery

Higher maternal social support during the first trimester of pregnancy was associated with higher methylation at seven CpGs (located within/near genes HAUS3, ARHGEF7, VGF, FAM210B, SBF1, ILVBL and EIF3F) (BACON-corrected PFDR ≤ 0.05). Most of these CpGs were either in promoter regions or gene bodies of the annotated genes. Also, the majority (6/7) loci were located in CpG islands (Table 1). In sensitivity analysis using a model additionally adjusted for cell composition variables, the methylation at these CpGs was associated with social support at PFDR < 0.001 (Additional file 1: Table S2). In sensitivity analysis without maternal sociodemographic factors, all seven association directions remained the same and the correlation in logFC between the fully adjusted model and the model without sociodemographic covariates was perfect (r = 1, P = 2.8 × 10−6) (Additional file 1: Table S2).

Table 1 Methylation sites in placenta associated with level of social support during pregnancy (n = 301)
Full size table

Maternal social support and fetal sex-specific DNA methylation in placenta

In analyses grouped by fetal sex, maternal social support was associated with higher methylation at nine CpGs in males (all exhibiting SSE, PFDR < 0.05) and with higher methylation at 32 CpGs and lower methylation at three CpGs in females (32 exhibiting SSE, 2 exhibiting OED, PFDR < 0.05) (Table 2; Additional file 1: Tables S3 & S4). In sex-stratified sensitivity analyses where the model additionally included cell composition variables, methylation at the 44 CpGs were associated with social support at PFDR < 0.001 (Additional file 1: Tables S5 & S6). In sensitivity analysis without maternal sociodemographic factors, all sex-specific association directions remained the same and the correlation in logFC between the fully adjusted model and the model without sociodemographic covariates was nearly perfect (male r = 0.99, P = 1.3 × 10−6; female r = 0.99, P < 2.2 × 10−16) (Additional file 1: Tables S5 & S6). Only two social support-associated CpGs in the overall sample, cg11364468 [VGF] and cg02672368 [ARHGEF7], were significant in male- and female-stratified analysis, respectively (Fig. 1). None of the CpGs associated with social support demonstrated concordant effects by fetal sex (Table 2).

Table 2 Comparison of effect sizes of social support-associated methylation sites between male fetus- and female fetus-bearing pregnancies
Full size table
Fig. 1
figure 1

Placental methylation sites associated with social support during pregnancy by sex of the fetus. All models are adjusted to maternal age, ethnicity, pre-pregnancy body mass index (BMI), education, job status, gestational age, parity, perceived stress, methylation principal components (PCs), genotype PCs, and surrogate variable. The model for the total sample is additionally adjusted for sex of the fetus

Full size image

Correlation between methylation of CpGs and expression of nearby genes in placenta

Higher methylation at cg11364468 (found to be associated with higher social support in the overall sample and male sample) was associated with lower expression of VGF. Higher methylation at cg16763895 (found to be associated with higher social support in the overall sample) was associated with lower expression of ILVBL (Table 3). VGF is a protein-coding gene known to be highly expressed in parts of the brain and neuroendocrine cells (Additional file 1: Figure S5). Several peptide proteins encoded by VGF have important roles in brain development and behavioral phenotypes [39] and regulation of energy metabolism [40]. Gene ontologies indicate that the protein encoded by ILVBL, which is widely expressed across different tissues (Additional file 1: Figure S6), is involved in fatty acid alpha-oxidation in the endoplasmic reticulum [41] and biosynthesis of isoleucine and valine [42].

Table 3 Association between methylation levels of social support-related placental methylation sites and placental expression level of nearby genes (n = 75)a
Full size table

Functional annotations and regulatory enrichment

CpGs associated with social support in the female sample showed enrichment for DNase 1 hypersensitive sites in fetal brain (PFDR < 0.05), but no enrichment was found for the overall or male-specific CpGs associated with social support (Additional file 2: Tables S7–S24). None of the social support-associated CpGs has previously been identified as cis-mQTL in placenta [25] which further suggests the observed methylation differences are likely to be the effect of social support rather than that of genetic variants.

Differentially methylated regions

Analyses of DMRs identified 18, 28, and 22 DMRs associated with social support in the overall, male, and female samples, respectively. Two genes (KNDC1 and KIAA0664) annotating DMRs overlapped with genes annotating CpGs identified in the male sample (Additional file 3: Tables S25–S27).

Pathway analysis

The genes annotating the top 100 social support-associated CpGs in the overall sample showed enrichment of IPA canonical pathways related to fetal growth, coagulation system, energy metabolism, and neurodevelopment (Table 4). For male-specific CpGs, enrichment was found for pathways related to immune system, cell cycle, tissue growth, and endocrine receptors signaling (Additional file 4: Table S28). For female-specific CpGs, enrichment was found for pathways relevant for immune system, neurodevelopment, and endocrine receptors signaling as well as processes important in placental development and maturation such as cell proliferation and cellular migration (Additional file 4: Table S29).

Table 4 Ingenuity pathway analysis canonical pathways of genes annotated to the top 100 social support associated methylation sites in placenta (total sample, n = 301)
Full size table

Discussion

In this first report of epigenetic signatures of social support in human placentas, we found that the level of prenatal social support during the first trimester of pregnancy is associated with differential methylation of seven CpGs in placenta at delivery. We also identified an additional 42 social support-associated CpGs in placenta dependent on fetal sex. The social support-associated epigenetic signatures in placenta are independent of prenatal stress; hence, social support may have impact on placental methylation even when maternal stress levels are not high. The association between placental expressions of VGF, ILVBL and MUC17, and DNA methylation at two of the social support-associated CpGs hints at the potential gene regulatory roles of the DNA methylation changes. Studies have previously demonstrated the epigenetic regulation of VGF [43, 44] and MUC17 [45, 46] expressions in different tissues. Genes annotated to social support-associated CpGs were enriched for pathways related to the immune system among others. Collectively, our findings support the biological effects of prenatal social support on the in-utero environment which may potentially have fetal programming effects [47], extending previous reports on the relations between social factors during pregnancy and methylation in maternal blood [11] and in placenta of Rhesus monkeys [12].

A positive effect of social support on health and well-being even under low stress environment has long been recognized [2]. While social support may mitigate the negative effects of stress on health outcomes, it is possible that social support independently promotes health and pregnancy outcomes. For example, prenatal social support has been linked to higher newborn leukocyte telomere length [5] and higher birth weight [48,49,50,51]—a marker of fetal growth and a predictor of adulthood health outcomes. The enrichment of FGF signaling and Hippo signaling pathways, which are reportedly involved in regulation of telomerase activity [52, 53], also suggests a potential mechanism for the effect of prenatal social support on fetal outcomes.

The enrichment of pathways related to the immune system and cytokines supports shared mechanisms for the potential effects of social support, stress, infections, and other factors. A meta-analytic review has found evidence supporting the link between low social support and inflammation [13]. The quality of social support during pregnancy has also been associated with inflammation during pregnancy and early infancy [14, 15]. Given the link between MUC17 expression level in different tissues and inflammatory activation [54, 55], our finding of decreased MUC17 expression with increased methylation at cg11364468 which in turn is associated with higher social support suggests involvement of inflammatory pathways. Therefore, we speculate that prenatal social support may promote fetal outcomes through attenuation of excessive inflammatory activation in placenta in response to various environmental and biological factors. Since psychosocial stress is only one of many proinflammatory environmental factors [56], the positive effect of social support on fetal outcomes may extend beyond pregnancies with high levels of stress.

The placenta has functional roles in fetal neurodevelopment via the “placenta-brain axis,” with potential programming for future mental health outcomes [57]. VGF is a protein-coding gene with biased expression in the brain (Figure S5), and its dysregulation has been linked to abnormalities in neural progenitor cell differentiation [58]. In animal studies, dysregulation of VGF had effect on brain development and behavioral phenotypes [39], depression-like behaviors [59], and memory consolidation and stress resilience [60, 61]. In humans, VGF has been suggested as a biomarker of different neuropsychiatric diseases [62]. The decreased expression of VGF associated with hypermethylation of cg11364468, enrichment of CpGs for fetal brain cells, and enrichment of annotated genes for pathways involved in brain development suggest that prenatal social environment may be involved in fetal programming for neuropsychiatric outcomes.

On the other hand, research suggests that VGF-derived peptides have an important role in the regulation of energy balance [40]. Although different mechanisms may exist, VGF activity in the hypothalamus, which is key in the regulation of feeding and energy metabolism, has been implicated [63, 64]. Increased methylation at cg16763895 associated with decreased expression of ILVBL which is involved in oxidation of fatty acids, suggesting fetal programming effect of social support on pathways relevant to energy metabolism. However, further research is needed to elucidate whether the epigenetic changes associated with prenatal social support in placenta are associated with later health outcomes in the offspring.

Our findings indicate sex-specific responses of placental epigenome to prenatal social environment. Nevertheless, pathway analyses revealed convergence in enrichment of canonical pathways such as those related to the immune system for the genes annotated to the top 100 social support associated CpGs in pregnancies with male and female fetuses. Studies have previously demonstrated that epigenetic programming of placenta occurs in a sex-dependent manner [65, 66], and in the case of social support, both converge at immune response and inflammation pathways, despite involvement of different CpG sites. We found hypermethylation of cg00140191 (FKBP5) with higher social support in only male pregnancies. Prenatal stress-associated differential methylation of FKBP5 in placenta has previously been linked to infant neurobehavioral outcomes [67]. Hypomethylation of cg00140191 was reported in peripheral blood of adolescents who had childhood victimization [68]. Overall, our findings indicate that sex-specific analyses offer the opportunity for better understanding the effects of social support and perhaps other environmental factors on placental epigenome. The potential implications of these sex differences on long term health outcomes may be crucial for understanding health disparities in men and women.

We acknowledge the following limitations arising from our design. First, our study may have been underpowered to detect additional associations because of relatively small sample size, particularly for subgroup and gene expression analyses. However, the post hoc power estimates indicate that most of the DNA methylation effect sizes were adequately powered. Second, the participants were selected to study low risk pregnancy, and this may have led to exclusion of individuals with low social support, e.g., individuals with drug addiction or psychiatric disorders. Finally, the level of social support may have changed later during pregnancy. Despite these limitations, we found novel CpGs in placenta associated with social support which withstood correction for multiple testing and adjustment for several important confounders, including estimates of placental cell composition and genetic ancestry. Our data support placental epigenetic programming effect of social support in racially diverse pregnant women with implications for offspring neuropsychiatric and cardiometabolic health. These findings need to be interpreted in the light of the shared genetic risk between loneliness, neuropsychiatric and cardiovascular morbidities [69].

Conclusions

We identified placental DNA methylation changes associated with prenatal social support independent of the level of prenatal stress during pregnancy. Some of these placental DNA methylation changes varied by fetal sex. The genes annotated to the DNA methylation loci are enriched for pathways involved in the immune system, placental growth and maturation, brain development, and energy metabolism. Research in molecular mechanisms of effect of social support on health outcomes may provide useful insight for developing interventions that promote fetal neurodevelopment. Further research is needed to replicate the findings and identify molecular mechanisms of effect of the broader social environment on pregnancy and fetal outcomes.