Placental DNA methylation at term reflects maternal serum levels of INHA and FN1, but not PAPPA, early in pregnancy

Background Early detection of pregnancies at risk of complications, such as intrauterine growth restriction (IUGR) and preeclampsia (PE), is critical for improved monitoring and preventative treatment to optimize health outcomes. We predict that levels of placental-derived proteins circulating in maternal blood reflect placental gene expression, which is associated with placental DNA methylation (DNAm) profiles. As such, placental DNAm profiling may be useful to distinguish pregnancies at risk of developing complications and correlation between DNAm and protein levels in maternal blood may give further evidence for a protein’s use as a biomarker. However, few studies investigate all clinical parameters that may influence DNAm and/or protein expression, which can significantly affect the relationship between these measures. Results Candidate genes were chosen based on i) reported alterations of protein levels in maternal blood and ii) observed changes in placental DNAm (∆β > 0.05 and False Discovery Rate (FDR) <0.05) in pregnancies complicated by PE/IUGR. Fibronectin (FN1) enhancer DNAm and placental gene expression were inversely correlated (r = −0.88 p < 0.01). The same trend was observed between promoter DNAm and gene expression for INHBA and PAPPA, though not significant. INHBA and FN1 DNAm was associated with gestational–age corrected birth weight, while INHA levels were associated with fetal: placental weight ratio and FN1 level was associated with maternal body mass index (BMI). DNAm at the INHBA promoter in the term placenta was negatively correlated with second trimester maternal serum levels (r = −0.50 p = 0.01) and DNAm at the FN1 enhancer was negatively associated with third trimester maternal serum levels (r = −0.38, p = 0.009). However, a similar correlation was not found for PAPPA. Conclusions These results show that establishing a correlation between altered DNAm in the term placenta and altered maternal serum levels of the corresponding protein, is affected by a number of factors. Nonetheless, the correlation between placental DNAm of INHBA/FN1 and maternal serum INHA/FN1 levels indicate that DNAm may be a useful tool to identify novel biomarkers for adverse pregnancy outcomes in some cases. Electronic supplementary material The online version of this article (doi:10.1186/s12881-015-0257-z) contains supplementary material, which is available to authorized users.


Background
Placental insufficiency is the inability of the placenta to provide an adequate supply of nutrients to the growing fetus. This can lead to a number of pregnancy complications including intrauterine growth restriction (IUGR) [1] and preeclampsia (PE), a maternal hypertensive disorder, which manifests as maternal hypertension and proteinuria after 20 weeks (wks) gestation [2]. Early diagnosis of PE and IUGR before clinical signs of disease can improve management and outcomes of affected pregnancies. Placental-derived proteins may be released into the maternal circulation where they can be quantified and used to assess placental function during pregnancy [3][4][5][6]. Such protein markers have been investigated for the prediction of PE and/or IUGR with varying success [7][8][9]. Nicolaides et al. (2013) reported a detection rate of 95 % for earlyonset PE (EOPE, diagnosis <34 wks) using decreased levels of maternal serum markers, pregnancy associated plasma protein A (PAPPA) and placental growth factor (PlGF), in combination with maternal factors [7]. However, these measures might not be generalizable, as the etiology and confounding environmental factors vary between populations [8]. Moreover, the ability to predict women at risk of late-onset PE (LOPE, diagnosis >34 wks) and IUGR is limited using these markers.
Differential gene expression between placentas from PE and/or IUGR pregnancies [9][10][11][12] may be utilized to identify additional biomarkers to distinguish women at high risk of these complications early in gestation. DNA methylation (DNAm) is associated with gene expression, but is more robust to variation in technical conditions and less subject to short-term biological change [13]. We previously reported numerous changes in DNAm in placentas from pregnancies complicated by EOPE [14]. Alterations of placental DNAm were noted in genes for which the expression of the encoded protein is altered in maternal blood in PE and/or IUGR pregnancies (e.g.: PAPPA, sENG, PAPPA2) [14]. Furthermore we found that sites of altered DNAm in PE frequently reflected changes in gene expression. While proteins produced in the placenta can be released into maternal circulation, their levels in maternal serum may be affected by many additional factors including size of the placenta, the cell type expressing the protein, and how such proteins are transported and metabolized. The purpose of the present study was to delineate the relationship between changes we observed in DNAm at term and maternal protein levels in early pregnancy. We selected three genes for which there was evidence for both altered maternal protein levels and altered DNAm in PE; we then evaluated 1) the relationship between placental DNAm and gene expression; 2) the role of variables that might confound measurement of DNAm, mRNA or protein levels including gestational age, fetal sex, placental efficiency (fetal: placental weight ratio), fetal birth weight, placental breadth: width ratio and maternal body mass index (BMI); and 3) whether placental DNAm at term reflected protein levels in maternal blood during gestation after correcting for these variables.

Candidate site selection and characteristics
To isolate loci for which altered DNAm might reflect maternal serum levels early in pregnancy, we chose candidate genes that not only had sites showing altered DNAm in EOPE, but also encode for proteins previously reported to show altered maternal serum protein levels in pregnancies that subsequently developed PE and/or IUGR. Previous studies have shown upregulation of both PAPPA and INBHA in the placentas of pregnancies complicated by PE and IUGR [12,[14][15][16]. In addition, several studies have reported DNAm alterations in placentas from pregnancies complicated by PE and/or IUGR [14,17,18]. FN1 [19] was selected due to the large magnitude of change in DNAm between EOPE and control placentas (Δβ = −0.24, FDR < 0.05) (see methods). INHBA (Δβ = −0.16, FDR < 0.05) and PAPPA (Δβ = −0.074, FDR < 0.05) were selected because they additionally encode for proteins for which first (PAPPA) or second trimester (INHA) maternal serum measures were available from clinical prenatal serum screening testing. We also focused on DNAm alterations in gene regulatory elements. The CpGs of interest for INHBA and PAPPA were 76 base pairs (bp) and 163 bp upstream of the transcriptional start sites, respectively. In relation to FN1, the CpG site was~100 kb upstream of the transcriptional start site, within an enhancer region.
Although these sites were selected based on a significant association with EOPE, we also wanted to know if these changes were conserved in other clinical groups (Fig. 1). In addition to hypomethylation of these sites in EOPE, the LOPE + IUGR group was hypomethylated for the INHBA (promoter) (Δβ = −0.18, p < 0.001) (Fig. 1a) and the FN1 upstream enhancer (Δβ = −0.25, p < 0.01) (Fig. 1c). While reduced methylation at the PAPPA promoter was only found in EOPE (Fig. 1b). As differences in DNAm were only found in the EOPE and LOPE + IUGR groups, as potential biomarkers, these candidate genes would presumably only be useful in identifying this subset of pregnancies [20]. Markers useful to detect LOPE or normotensive IUGR may be more challenging to identify due to their weak association with placental pathology.
Is DNAm at candidate sites inversely correlated with gene expression?
To confirm that the DNAm change resulted in a change in gene expression, we assessed the relationship between placental DNAm (measured by Illumina 450 k array) and gene expression at these three candidate sites. FN1 showed an inverse correlation between DNAm of an upstream enhancer and gene expression at term (r = −0.88, p < 0.0001). INHBA and PAPPA, showed a non-significant trend with increasing DNAm being associated with decreased gene expression in the placenta (Fig. 2). This phenomenon may be due to alterations in cell composition between pathogenic and healthy placentas related to the pathology of PE/IUGR [14]. For all candidate genes, there was an observable divide between the controls and EOPE cases, where cases had decreased DNAm corresponding to increased gene expression in the placenta.
What clinical factors are associated with DNAm at candidate sites?
To better understand what factors might affect the measurement of DNAm and therefore the relationship with protein expression levels in maternal blood, we also evaluated several potential confounding factors including gestational age at delivery [21], fetal sex [22,23], fetal birth weight [24], placental dimensions and maternal BMI. Bisulfite pyrosequencing was used to extend our assessment of DNAm at the candidate sites into a larger cohort of controls for which clinical serum measurements (INHBA N = 36, PAPPA N = 33) or serum samples for assaying FN1 (N = 76) were available.
Birth-weight standard deviation (SD) was associated with DNAm at the INHBA promoter (p = 0.05) and the upstream enhancer of FN1 (p = 0.02). Gestational age was only associated with FN1 DNAm (p = 0.03). None of the clinical factors assessed was associated with DNAm at the PAPPA site ( Table 1). The observation that birth weight (SD) was associated with INHBA DNAm, without an association with gestational age, emphasizes the importance of including both gestational age and birth weight when considering the relationship between DNAm and other variables.
What clinical factors are associated with protein concentration in maternal blood?
We also assessed the same clinical parameters for association to protein concentration in maternal blood (Table 2). Gestational age at blood draw was only assessed as a covariate for FN1 as clinical values for INHA and PAPPA were given in multiples of the median (MoM), which was already corrected for GA at blood draw. Placental efficiency (fetal: placental weight-ratio, at birth) was associated with increased second trimester INHA levels in maternal blood. FN1 level was not associated with maternal BMI in the controls for which we had this information (N = 37), though it was significant when evaluating all clinical groups together (EOPE, LOPE, IUGR, Controls) (N = 75). It was therefore included in subsequent analyses. None of the assessed factors were associated with PAPPA maternal blood levels during pregnancy (Table 3).
What is the relationship between DNAm and maternal serum levels?
DNAm in the promoter of INHBA correlated with second trimester protein levels in maternal blood (r = −0.50,  36) promoter DNAm in the term placenta and second trimester INHA levels in maternal blood, plotted as residuals corrected for fetal birth weight (SD) and fetal: placental ratio, e PAPPA (N = 34) promoter DNAm in the term placenta and first trimester PAPPA levels in maternal blood, and f FN1(N = 76) enhancer DNAm in the term placenta and second/third trimester FN1 levels in maternal blood, plotted as residuals corrected for fetal birth weight (SD), gestational age, and maternal body mass index (BMI). MoM = multiple of the median p = 0.01) while modeling for both fetal birth weight (SD) and fetal: placental weight ratio (Fig. 2d). Additionally, DNAm in an upstream enhancer of FN1 correlated with third trimester protein levels in maternal blood (r = −0.38, p = 0.009) while adjusting for birth weight (SD), gestational age, and maternal BMI (Fig. 2f). This supported our prediction that DNAm changes observed in the placenta could explain some of the previous reports of altered INHA and FN1 levels in maternal blood in PE. It is remarkable that these serum measurements from the second and third trimesters of pregnancy reflected DNAm at term. This implies that this DNAm change may be an early alteration in PE. In contrast, a similar result was not observed for PAPPA/PAPPA (Fig. 2e).
We had predicted that protein levels in maternal blood would reflect placental DNAm and gene expression. While this may be true in some instances (e.g. INHA, FN1), in other cases establishing a relationship may be challenging (e.g. PAPPA). Establishing such a relationship may be complicated by several factors. Protein level depends not only on the level of gene expression, but also on the total number of cells expressing that protein, the number of mRNA transcripts being translated into protein in those cells, and the rate and mode of release of the protein into maternal blood. These factors may be influenced by the underlying pathology (i.e. more protein may be released with increased apoptosis) and placental size; which, in turn may be associated with fetal weight and/or fetal: placental weight ratio. Other factors such as expression of the same protein from maternal tissues, and the metabolism of proteins by the placenta, reducing the amount of protein being secreted into the maternal circulation may have a substantial influence of the total protein concentration in maternal blood (Fig. 3). PAPPA has been found to be expressed from other maternal sources (e.g. ovary, some epithelial and endometrial cells, and breast)   Mean GA at blood draw(weeks ± SD) 31.6 (±6. besides the placenta, and it is possible be that these sources mask any relationship between placental derived protein and DNAm in the placenta [25][26][27][28]. It is also important in the case of PAPPA to note that maternal protein levels were measured in the first trimester and additional variation may arise over gestation affecting correlation with placental DNAm at term.
Are there any differences in protein levels between case and control placentas?
To confirm a previous report of altered maternal FN1 in association with PE and/or IUGR [19], FN1 levels were measured in maternal blood samples from pregnancies which subsequently developed EOPE, LOPE + IUGR, LOPE without IUGR, or normotensive IUGR, in addition  Fig. 3 Processes that may influence the relationship between DNAm, gene expression and protein expression. Outlines reasons why we may not see a correlation between placental DNAm and gene expression or between placental gene expression and circulating levels of placental-specific proteins in maternal blood to our control cohort (Table 2). Similar to the alterations in DNAm, changes in FN1 levels were found to be significantly different from controls only in the EOPE group (Mann U Whitney test), although there was a trend of increased FN1 levels between LOPE + IUGR and controls (p = 0.08) (Fig. 4). Our results were in concordance to Auer et al. (2010) who also reported increased levels of maternal FN1 in pregnancies complicated by EOPE and LOPE + IUGR. We did not confirm their observation of a decrease of FN1 in pregnancies complicated by IUGR; however, we may have been under-powered to observe this small difference. Furthermore, although we observe a difference in EOPE and LOPE + IUGR compared to controls, the range of FN1 levels completely overlap between the groups, hindering FN1 to be an adequate biomarker used alone.

Conclusion
This study provides a link between changes in placental DNAm at term and protein biomarkers present in the mother's circulation earlier in pregnancy. It emphasizes the many confounding factors that may influence this relationship, explaining why this linkage may not be observed for all loci. We chose three genomic sites with significantly altered DNAm in term placenta associated with PE and that were associated with genes for which the protein product is altered in PE/IGUR. Despite this, for only two of the three loci (INHA and FN1) did we find a correlation between placental DNAm and second and third trimester maternal serum protein expression in control samples. Nonetheless, this does suggest that other DNAm marks may be associated with early differences in gene expression. Furthermore, with the advent of techniques to quantify placental nucleic acids in maternal serum [29], DNAm changes may be more directly linked to measurable miRNA and RNA in maternal blood. Factors such as placental surface area and mechanisms for release into maternal blood, will also affect serum levels of placental nucleic acids [30]. Future studies measuring protein levels directly in placental tissue, correlating with maternal levels and investigating the factors affecting rate of release are needed to help translate findings measured in the term placenta into maternal biomarkers of pregnancy outcomes in early gestation.

Sample information
Ethics approval was obtained from both the University of British Columbia and BC Women's and Children's Hospital ethics committees in Vancouver, BC, Canada (H04-70488). Placental samples were obtained with consent via recruitment through the Medical Genetics and Obstetrics and Gynecology departments. Case information such as: maternal age, maternal BMI, mode of delivery, gestational age at delivery, fetal sex, birth weight, gestational age at blood draw, results on any molecular testing, and placental dimensions were recorded. Preeclampsia (PE) was defined according to Society of Obstetricians and Gynecologists of Canada (SOGC) criteria as one of i) hypertension (BP > 140/90 mm Hg) and proteinuria (>300 g/day) arising after 20 weeks gestation [2]; ii) HELLP syndrome without hypertension or proteinuria [31]; or iii) eclamptic seizure without previous hypertension or proteinuria [32]. EOPE was defined by a diagnosis of PE prior to 34 weeks gestation, and LOPE was defined as a diagnosis after 34 weeks gestation [33]. Intrauterine growth restriction (IUGR) was also defined following SOGC criteria [34] as birth weight < 3rd percentile accounting for fetal sex and gestational age, or birth weight < 10th percentile with additional clinical findings indicative of poor growth such as: absent or reversed end diastolic velocity on Doppler ultrasound, or oligohydramnios. Criteria for exclusion were chronic/pre-existing maternal hypertension, gestational diabetes, multi-fetal pregnancies, and fetal chromosomal abnormalities. Controls were selected based on absence of any criteria listed above and a placenta with no observable pathology.
Whole chorionic villi were sampled from four sites, each from distinct cotelydons of the placenta [13]. Sampling from infarcts or other abnormal regions of the placenta was avoided. DNA was extracted from each sampled site and pooled together in equal proportions. DNA was assessed for quality on the Nanodrop 1000 spectrophotometer (ThermoScientific, Wilmington, DE, USA). Three hundred nanograms of each DNA sample was bisulfite converted for subsequent analyses. Additionally, RNA extracted from the placental villi with RNeasy kit (Qiagen, Heiden, Germany) and was stored in RNAlater at−80°C. While we used a total of 171 placentas for our studies, not all placentas were used in all studies as we were limited by samples run on the 450 K array (N = 66); samples run on the Illumina expression array (N = 16), maternal serum screening results (first trimester N = 34, second trimester N = 36), or maternal serum samples for FN1 testing (N = 114). Additional file 1: Table S1 outlines a list of all samples and which analyses they were used in.

DNA methylation analysis Illumina infinium HumanMethylation450 BeadChip array
To compare the DNAm differences between clinical groups for each of our candidate genes twenty EOPE, 11 LOPE, 8 LOPE + IUGR, 10 IUGR, and 37 control cases were run on the Illumina Infinium HumanMethylation450 BeadChip (450 k) array, which interrogates >480,000 CpG sites in >20,000 genes [35]. Some of these samples were previously analyzed in the study reported by Blair et al. (2013). To compare the association between DNAm and protein levels in maternal blood, 122 placental DNA samples (750 ng) bisulfite converted using the EZ DNA Methylation kit (Zymo Research, Irvine, USA). Hybridization of samples to the array was completed as per the manufacturer's protocol. The microarray chips were scanned by the HiScan 2000 or iScan (Illumina). Data was normalized and analyzed as per Blair et al. (2010) methods [14].

Bisulfite pyrosequencing
Candidate CpGs determined from the 450 k array data in Blair et al. (2013) were followed up with bisulfite pyrosequencing in control cohorts for each candidate gene (  Table S3).

Candidate DNAm selection
CpG sites chosen to investigate in the present study were selected on i) a significant change in placental DNAm, defined as a false discovery rate (FDR) < 0.05 a Δ β > 0.05 (i.e. at least 5 percentage points difference in DNAm), a cut-off that enriches for changes in DNAm that would likely have biological impact [36], in placentas associated with PE and ii) genes encoding for proteins reported to show altered levels in maternal blood in pregnancies complicated by PE and/or IUGR. In addition to meeting these criteria, INHBA and PAPPA were chosen as we had maternal serum measures available on INHA and PAPPA from the maternal serumscreening program. We chose FN1 since the difference in DNAm between EOPE and controls was Δβ = 0.24. We also took into account where the DNAm alteration was in the genome, taking interest in alterations in gene regulatory elements (Table 4).

Maternal blood protein measurements
Measurements of Pregnancy associated plasma protein A (PAPPA) and Inhibin alpha (INHA) were obtained from clinical maternal serum screening data for 36 and 33 women, respectively, and are measured in multiples of the median (MoM). Additionally, blood was drawn in EDTA tubes during the second trimester for a subset of 158 women (Table 2). Plasma was obtained via centrifugation at 3000 rpm for 10 min 4°C. Plasma Fibronectin (FN1) was measured using a FN1 ELISA kit (eBioscience, San Diego, CA, USA). FN1 measurements were run in duplicate and absorbance was measured at 450 nm. A 5 parameter asymmetrical logistic curve was generated from the standard data points which ranged from 0.31-20.0 ng/mL. Samples were diluted as per manufacturer's protocol; samples which FN1 concentration was over the standard curve were diluted to 1 in 80,000, and 4 samples which remained were further diluted to 1 in 100,000.