Background
Despite decades of research, the underlying cause of preterm birth remains enigmatic. It is a leading cause of newborn morbidity, hospitalization, and developmental delays [
1]. In addition, preterm birth is associated with health risks across the later life course of the newborn, including cardiovascular disease, metabolic syndromes, psychiatric conditions, obesity and cognitive disabilities [
1,
2]. The “fetal origins” or Developmental Origins and Health and Disease (DOHaD) hypothesis, developed from a series of epidemiologic observations, demonstrated that measures of birth size were associated with long-term chronic disease risk [
3]. Numerous investigations have shown that antenatal maternal environmental factors, including diet, xenobiotic exposure, stress, and lifestyle factors can alter fetal growth and result in permanent biological and physiologic changes of the offspring [
3]. Environmental factors like race, diet, smoking, socioeconomic status may also increase the risk of spontaneous preterm birth [
1,
4,
5] and are associated with epigenetic alterations [
6].
DNA methylation is the most well studied epigenetic mechanism of gene regulation, often associated with transcriptional silencing of downstream gene(s). The presence of the methyl group(s) alone is not sufficient for transcriptional silencing, but instead alters recruitment of component proteins related to gene repression and results in a silenced chromatin conformation. DNA methylation is an essential epigenetic mechanism in fetal development [
7].
The placenta facilitates the exchange of gas, nutrients, and waste between the mother and the fetus, and modulates effects on the fetus from the mother’s immune system, thus playing an essential role in fetal growth and development. It is also essential in understanding the long-term effects of in-utero development on post-natal disease. The placenta undergoes many changes throughout gestation and the mechanisms behind these changes need to be better understood. In an attempt to do so, several studies have examined genome wide expression differences in placentas at different time points during gestation, comparing first, second and third trimester placental methylation [
8,
9]. Changes in expression with increasing gestational age were found in common between the studies. Others are attempting to better understand placental development and fetal programming through the study of epigenetic factors, including DNA methylation of placental tissue and umbilical cord blood. Studies of umbilical cord blood from preterm and term pregnancies have releaved differences in methylation associated with gestational age [
10,
11]. Novakovic et al. have studied genome scale placental promoter methylation from the three trimesters of pregnancy, revealing a progressive increase in methylation from first to third trimester. They also identified increased inter-individual variability in third trimester samples [
12]. Other studies have alsofound varied methylation differences associated with gestational age comparing placentas in the third trimester, as well as a global increase in methylation with gestational age (28–40 weeks) [
13‐
15]. In addition, the placenta has the highest overall variability in DNA methylation when compared to other tissues [
16]. These studies all support the emerging paradigm that the placenta is an active mediator of fetal well-being and neurodevelopmental outcome and can serve as a blueprint for intrauterine life [
17]. This is an exploratory study seeking to investigate genome wide placental DNA methylation across a wide range of preterm gestational ages and compared it to that of placenta from term deliveries. In order to generate genome-wide information, we employed immunoprecipitation of methylated DNA followed by whole-genome sequencing, so called MeDIP-seq [
18]. We hypothesize that using this approach, we would be able to identify potential regions of interest and pathways involved in and influenced by changes in placental methylation associated with preterm birth and gestational age. Our objectives were to demonstrate the feasibility of this approach and to generate placental methylation data that would be useful to our own studies and to those of others.
Discussion
We used methylation-dependent immunoprecipitation followed by high throughput sequencing to generate non-biased, genome-wide map of DNA methylation in placenta from a wide range of gestational ages. We investigated regions for which there was differential methylation between preterm (
< 34 weeks) and term placentas (> 37 weeks), as well as regions for which the differences in methylation were associated with the continuous variable gestational age. Our results demonstrate significant differences in DNA methylation in preterm versus term placenta. Approximately half of the DMRs associated with preterm birth were not significantly associated with changes in gestational age. There were more hypo-methylated regions in preterm patients compared to term patients. The highest percentages of differentially methylated regions mapped to distal intergenic regions followed by introns, exons and then promoter regions. Mapping of these significant DMRs to the nearest genes demonstrated some overlap with patterns of differential gene expression in placentas from preterm and term patients [
20]. There was also overlap with genes shown to be evolutionarily linked to preterm birth and to networks and pathways associated with preterm birth [
21,
22].
Both candidate gene studies and genome-wide studies of DNA methylation in the placenta have been performed to investigate the mechanism(s) of preterm birth. One study found a positive association between global methylation and gestational age but others found little variation amongst the partially methylated domains across all three trimesters [
11,
14]. Another study of promoter region methylation found overall differences in methylation between second and third trimester placentas, but not between first and second trimester [
12]. Several studies examining gestational age and DNA methylation used umbilical cord blood to gain understanding into fetal programming and methylation state at birth. In one study, among the 39 genes showing differential methylation, 29 showed a decrease in methylation with increase in gestational age while the remainder showed an increase and no relationship to type of delivery [
11]. Parets et al. studied methylation of cord blood leukocytes from 24 weeks to 41 weeks [
10]. Most sites showed lower degrees of methylation with shorter gestational age, suggesting that one mechanism regulating the extent of methylation is gestational timing. We and others have also found associations with the preterm birth process itself. The Norwegian Mother and Child Cohort Study (MoBa) compared cord blood methylation with birthweight and found both increased and decreased patterns of methylation associated with specific genes [
23]. Another study using the Illumina 450 k array found 1400 variably-methylated regions which correlated with significant variables in the intrauterine environment including maternal smoking, maternal depression, maternal BMI, infant birthweight and gestational age [
24]. Thus, while no unifying picture of the association between gestational age and DNA methylation has been demonstrated, we believe the mechanisms regulating the extent and pattern of placental DNA methylation include programmed changes linked to gestational timing as well as experiential changes. Our study, with a wide range of gestation ages, using a non-biased, genome-wide approach, shows a significant effect of both gestational age as a continuous predictor and PTB status as a categorical predictor of placental DNA methylation.
The site of methylation may be crucial to the effect on gene expression or a reflection of the impact of environment on gene expression. Clusters of CpG’s also known as CpG islands (CGI) are present in 5′ promoter regions of many genes. Methylation can also take place in shores and shelves, which are more distant to the promoter. Some studies have shown that tissue- and cancer-specific DMRs occur more frequently within CpG shores than CGIs themselves [
25]. The functional implications of alterations in methylation are context-specific. Methylation in the immediate vicinity of the transcription start site is believed to block initiation, whereas methylation in the gene body may stimulate transcription elongation and/or have an impact on splicing [
25]. We saw the greatest degree of differential methylation (almost 60%) in distal intergenic regions. Second greatest differential methylation was seen in introns other than the first intron. In addition, enrichment analysis showed that hypo-methylated DMRs were enriched for CpG Islands, while hyper-methylated DMR were enriched for CpG shores and shelves (Fig.
2). The annotation results, along with the later enrichment results, are consistent with the results from previous studies suggesting methylation is more dynamic outside of CpG islands in promoter regions. The enrichment of CpG islands amongst the hypo-methylated DMRs could be linked to chromosomal instability and imprinting [
26]. The implications of the intergenic and intragenic methylation, as well as in shores and shelves on preterm birth are significant, yet mechanistically still unclear.
The most significant pathway associated with the genes nearest to the 427 DMRs we observed was Citrulline-Nitric Oxide Cycle, which contains the NOS1 gene. Our results found a hypermethylated DMR associated with both PTB and gestational age proximal to NOS1. NO is secreted by placenta [
27] and known to modulate both fetal and utero placental blood flow [
28]. Bielecki et al. found a lower concentration of NO in a group of women with premature contractile activity in comparison with gestational age-matched healthy pregnant women [
29]. In another study the amniotic fluid concentration of NO was significantly higher in patients with intra-amniotic infection compared to those without intra-amniotic infection [
30]. A decrease in NO production may contribute to the initiation of labor and cervical ripening [
31]. A study suggests that NO produced by the placenta could play role in maintaining uterine quiescence by paracrine effect [
32]. These results suggest that increased methylation of NOS1 may play an important role in the production of NO and subsequently preterm birth.
Another significant pathway was Fc-gamma Receptor Mediated Phagocytosis in Macrophages. There is abundant evidence for Fc gamma R mediated transcytosis of IgG in the placenta. The transfer of IgG from mother to fetus begins around 13 weeks of gestation and the total IgG concentrations in newborns is directly related to length of gestation. Infants born preterm have substantially lower IgG levels than full-term babies [
33]. We also identified a DMR whose nearest gene is mannose binding lectin (
MBL2), which has previously been identified by pathway and network analysis to be related to preterm birth and evolutionarily associated as well [
21,
22].
MBL2, found in amniotic fluid, is a serum protein involved in the activation of the complement system of the innate immune system and plays a role in fetal inflammatory response to infection and injury [
34,
35]. It activates complement system by binding to carbohydrates, present on a wide range of proteins [
36]. Moreover, fetal
MBL2 haplotypes and in utero exposure to viral infection increases the risk of preterm birth [
37].
When we compared our DMR results to data sets important in preterm birth, we identified a hyper-methylated peak whose nearest gene is transferrin receptor 1,
TFRC.
TFRC is expressed in the placenta and mediates cellular iron uptake. Iron deficiency during pregnancy increases the risk of preterm birth [
38]. While
TFRC was upregulated spontaneous preterm birth in the Chim et al. placental expression study, it was also upregulated in the Lynch evolution of mammalian pregnancy and found to be reduced placentas with intrauterine growth restriction and preeclampsia [
39]. Because prematurity, IUGR and preeclampsia have different pathogenic etiologies, the results suggest the importance of further investigation of the epigenetic regulation of TFRC with respect to pregnancy related disorders.
The current study demonstrates the feasibility of sample collection, technical analysis and data processing. Potential limitations of the study are the relatively small sample size and the diversity of patients. Nonetheless, in order to clearly define an effect of prematurity, we purposefully collected placental samples from a wide range of gestational ages. There was some variation in the mothers’ clinical features beyond prematurity that may have impacted DNA methylation. Nonetheless, these unbiased data provide a useful reference for future studies by us and others. In addition, we chose to study genome-wide methylation using MeDIP-Seq due to its feasibility and moderate expense as compared to other techniques such as Whole Genome Bisulfite Sequencing. The affinity-based approach coupled with deep sequencing has a resolution of 100-300 bp and is cost effective when single-base resolution is not necessary [
40,
41]. Previous research suggested that at 1x coverage, a majority of the methylated CpG can be studied [
40]. It is important to note that MeDIP-seq, similar to restriction enzyme digestion approaches, can only measure relative enrichment of methylated DNA rather than absolute methylation levels. Lastly, another advantage of MeDIP-seq over WGBS is its ability to detect both 5-Methylcytosine (5mC) and 5-hydroxymethylctyosine (5hmC) independently [
40,
41].
Methods
Placental samples
Placenta samples were collected by our research staff at Women & Infants Hospital of Rhode Island. They obtained shortly after delivery from births ranging from 25 weeks to 41 weeks of gestational age. Samples of villous parenchyma were taken from four quadrants between the chorionic and basal plate. Care was taken to avoid maternal decidua and areas of hemorrhage or calcification. Samples were placed immediately into RNAlater
Tm (Ambion, Inc., #AM7021) and stored at − 80 °C until DNA extraction. Preliminary studies have shown that macromolecules like RNA levels were similar from each sample site and that this approach was equal to or superior to immediate immersion in liquid nitrogen for prevention of RNA degradation [
43,
44].
Genomic DNA was extracted using the Qiagen DNeasy Blood and Tissue kit (Qiagen, # 69506) and quantified on a NanoDrop 1000. 5μg of DNA was digested to fragment size 200–300 base pairs using dsDNA Fragmentase enzyme at 37 °C for 30 min (New England Biolabs, #MO348L). Fragments were end-repaired, 3′-ends were adenylated, and appropriate adapter indexes were ligated using the Truseq protocol (Illumina). Between each reaction, fragments were cleaned using Agencourt AmPure magnetic beads (Beckman Coulter, # A63881). Fragments were then amplified by PCR at 98 °C/30 s; 10 cycles of 98 °C/10s, 60 °C/30s, 72 °C/30s; and 72 °C 5 min with a hold at 10 °C. Enriched fragments were then cleaned using Agencourt AmPure magnetic beads and quantified before methylation-dependent immunoprecipitation.
MeDIP-seq
Methylated-DNA immunoprecipitation was performed using the Methylated-DNA IP kit (Zymo Research, # D5101). 320 ng of each sample was mixed with denaturation buffer and heated to 98 °C for 5 min. DNA is then mixed with MIB buffer, ZymoMag Protein A beads, and Mouse Anti-5-Methylcytosine from and incubated at 37 °C for one hour, with mixing every 15 min. The tubes were rocked, allowed to cluster, washed with reagent buffer and then eluted at 75 °C for 5 min. This was followed by a 2-min spin in a mini centrifuge at 18,000 g. The recovered DNA underwent 100 bp paired-end sequencing in the Brown University Genomics Core in triplicate on an Illumina HiSeq 2500.
Raw sequence reads were separated according to sample-specific barcodes and mapped to the NCBI Build UCSC Hg19 human genome using the Burrows-Wheeler Aligner (BWA v0.6.2) [
45]. The SAM files were converted to BAM files with SamTools (v0.1.18) [
46] and duplicate reads (reads with the same start location) were removed using Picard Tools (v1.77) (
https://github.com/broadinstitute/picard). We used Model-based Analysis for ChIP-Seq (MACS v1.4) [
47] to identify significantly enriched regions (peaks) using
p < 1 × 10
− 5 as the significance threshold for each individual and technical replicate independently.
Identification of differentially methylated regions
We used the R Bioconductor packages DiffBind (
http://bioconductor.org/packages/DiffBind/) and DESeq2 [
48] to identify Differentially Methylated Regions (DMRs). We used DiffBind to identify a peak set for the study cohort, requiring that each individual’s consensus peak set contain only peaks which were present in all three technical replicates. For each individual, the read count for each peak in the consensus peak set was merged by taking the sum over all three technical replicates. DMRs were identified using DESeq2.
P-values were corrected using FDR with independent filtering of overall low mean counts.
Genomic annotation and enrichment
DMRs with a
p-value < 0.01 were annotated using R Bioconductor package ChIPseeker [
19] to retrieve the nearest gene to the peaks of interest and annotate the genomic region of the peak. CpG islands and Refseq gene exons and introns were downloaded from the UCSC Genome Browser [
49]. CpG shores and shelves were defined 2 kb and 4 kb up and downstream from the CpG islands, respectively. The Hg19 reference genome was spilt into 500 bp windows and each window was annotated with the above genomic features if any overlap existed. The ChromHMM annotation of the Placenta Cell Line from the Roadmap Epigenome Project, obtained from the UCSC Genome Browser, was used to align the 500 bp windows with “promoter” and “enhancer” state annotation [
50]. The enrichment score for each genomic feature (CpG islands, shores, shelves, exons, introns, promoters, and enhancers) with respect to the DMRs was calculated via the method in Zhang et al. as the ratio between the fraction of DMRs overlapping widows with genomic feature and the fraction of total windows with the genomic feature [
51].
Comparative gene set analysis
In order to examine the potential role of DNA methylation in the regulation of preterm birth we compared our DMRs with previously published gene sets associated with preterm birth and pregnancy.
Chim et al. used an array based approach to study differential placental gene expression between spontaneous preterm birth and spontaneous term birth. “They reported 240 significantly upregulated and 186 significantly downregulated genes in the placenta associated with spontaneous preterm birth.” [
20]. We also compared the significant DMRs with a gene set identified in curated articles, networks and pathways important in the risk of preterm birth [
21]. This set was obtained via extensive literature curation and imputation. Lastly, we compared significant DMRs to a gene set linked evolutionarily to mammalian pregnancy [
22]. In this work Lynch et al. explore the evolution of pregnancy in placental mammals and identify 1532 gene that are uniquely expressed in the endometrium. Many of these genes were in close proximity to MER20, which regulate gene expression in response to progesterone and cAMP. These genes were broken down into gain and loss of expression in response to the stimuli.
Pathway analysis
Pathway analysis of the genes nearest to the DMRs with
p < 0.01 was performed using QIAGEN’s Ingenuity Pathway Analysis (
www.qiagen.com/ingenuity).
Statistical analysis
The Student’s t-test was used to evaluate significant differences between cases and controls. A two- tailed p < .05 was considered to indicate statistical significant difference.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.