UBC Faculty Research and Publications

Epigenetic regulation of placental gene expression in transcriptional subtypes of preeclampsia Leavey, Katherine; Wilson, Samantha L; Bainbridge, Shannon A; Robinson, Wendy P; Cox, Brian J Mar 2, 2018

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
52383-13148_2018_Article_463.pdf [ 1.16MB ]
Metadata
JSON: 52383-1.0364135.json
JSON-LD: 52383-1.0364135-ld.json
RDF/XML (Pretty): 52383-1.0364135-rdf.xml
RDF/JSON: 52383-1.0364135-rdf.json
Turtle: 52383-1.0364135-turtle.txt
N-Triples: 52383-1.0364135-rdf-ntriples.txt
Original Record: 52383-1.0364135-source.json
Full Text
52383-1.0364135-fulltext.txt
Citation
52383-1.0364135.ris

Full Text

RESEARCH Open AccessEpigenetic regulation of placental geneexpression in transcriptional subtypes ofpreeclampsiaKatherine Leavey1, Samantha L. Wilson2,3, Shannon A. Bainbridge4,5, Wendy P. Robinson2,3 and Brian J. Cox1,6*AbstractBackground: Preeclampsia (PE) is a heterogeneous, hypertensive disorder of pregnancy, with no robust biomarkersor effective treatments. We hypothesized that this heterogeneity is due to the existence of multiple subtypes of PEand, in support of this hypothesis, we recently identified five clusters of placentas within a large gene expressionmicroarray dataset (N = 330), of which four (clusters 1, 2, 3, and 5) contained a substantial number of PE samples.However, while transcriptional analysis of placentas can subtype patients, we propose that the addition of epigeneticinformation could discern gene regulatory mechanisms behind the distinct PE pathologies, as well as identify clinicallyuseful potential biomarkers.Results: We subjected 48 of our samples from transcriptional clusters 1, 2, 3, and 5 to Infinium HumanMethylation450arrays. Samples belonging to transcriptional clusters 1–3 still showed visible relationships to each other by methylation,but cluster 5, with known chromosomal abnormalities, no longer formed a cohesive group. Within transcriptional clusters2 and 3, controlling for fetal sex and gestational age in the identification of differentially methylated sites, compared tothe healthier cluster 1, dramatically reduced the number of significant sites, but increased the percentage thatdemonstrated a strong linear correlation with gene expression (from 5% and 2% to 9% and 8%, respectively). Locationsexhibiting a positive relationship between methylation and gene expression were most frequently found in CpG opensea enhancer regions within the gene body, while those with a significant negative correlation were often annotatedto the promoter in a CpG shore region. Integrated transcriptome and epigenome analysis revealed modifications inTGF-beta signaling, cell adhesion, oxidative phosphorylation, and metabolism pathways in cluster 2 placentas, andaberrations in antigen presentation, allograft rejection, and cytokine-cytokine receptor interaction in cluster 3 samples.Conclusions: Overall, we have established DNA methylation alterations underlying a portion of the transcriptionaldevelopment of “canonical” PE in cluster 2 and “immunological” PE in cluster 3. However, a significant number of theobserved methylation changes were not associated with corresponding changes in gene expression, and vice versa,indicating that alternate methods of gene regulation will need to be explored to fully comprehend these PE subtypes.Keywords: Preeclampsia, Placenta, DNA methylation, Gene expression, Clustering, SubtypesBackgroundPreeclampsia (PE) is a complex, heterogeneous disorderof pregnancy, diagnosed by the onset of maternal hyper-tension after the 20th week of gestation, with signs ofmaternal multi-organ dysfunction [1]. As with manypathologies of pregnancy, PE has no cure, robustpredictive biomarkers, or effective treatments, other thanthe delivery of the infant to discontinue the pregnancy andremove what is thought to be the causative organ, theplacenta. Repeated attempts to characterize the placentalmolecular pathology and identify biomarkers of PE byapplying a binary approach (PE versus control) have notbeen clinically fruitful, and we hypothesized that this is dueto the existence of multiple molecular subtypes of PE [2].In support of this hypothesis, we recently published alarge unsupervised clustering analysis of microarray data* Correspondence: b.cox@utoronto.ca1Department of Physiology, University of Toronto, 1 King’s College Circle,Toronto, ON, Canada6Department of Obstetrics and Gynecology, University of Toronto, 23 EdwardStreet, Toronto, ON, CanadaFull list of author information is available at the end of the article© The Author(s). 2018 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, andreproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link tothe Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.Leavey et al. Clinical Epigenetics  (2018) 10:28 https://doi.org/10.1186/s13148-018-0463-6from a PE-focused placental cohort (N = 330), including157 highly annotated samples purchased from a singlebiobank [3]. This revealed five clusters of placental geneexpression containing at least three clinically significantetiological subtypes of PE: “maternal”, with term andnear-term deliveries of average-sized infants and placentasthat appear molecularly similar to normal healthy controlsamples; “canonical” with high placental expression ofknown PE markers, preterm deliveries, low fetal weights,and evidence of maternal malperfusion; and “immuno-logical” with severe fetal growth restriction, enrichment ofimmune response genes, and histological signs of maternalanti-fetal/placental rejection [3], belonging to transcrip-tional clusters 1, 2, and 3, respectively. An additionalsubtype of PE placentas with chromosomal abnormalitieswas also discovered within cluster 5 (and supported byarray-based comparative genomic hybridization (aCGH)analysis), but showed no strong clinical association [3].However, despite our considerable progress towardsunderstanding the molecular diversity observed amongstPE patients, RNA is relatively unstable, easily affected bytechnical variability [4], and rarely successful as atherapeutic target [5], limiting its clinical utility. We,therefore, propose that the integration of an additionallevel of molecular information in these placentas, suchas DNA methylation, will compensate for these restric-tions [4], as well as improve our understanding of themolecular pathology.DNA methylation is a mitotically heritable epigeneticmark employed by the cell to control gene expressionwithout altering the genetic sequence [6], although therelationship between the two data types is exceptionallycomplex [7–10]. Given the flexibility for modification inthe epigenome, these methylation events may also serveto provide insight into the environmental exposuressustained by the cell [11], and as potential biomarkers ofearly cellular transformations [12]. In fact, many examplesexist, particularly in the cancer field, for the exploitationof DNA methylation in the diagnosis, prognosis, andprediction of drug response in disease [12, 13], and aspossible therapeutic targets [14, 15].Here, we subject a subset of our highly annotated cohortsamples to DNA methylation arrays and investigatedifferences in the placental methylome between ourpreviously identified transcriptional clusters, as well asrelationships between the two data types. Furthermore, byassessing epigenetic changes associated with the observedpathological gene expression, we also attempt to discovernovel therapeutic targets for the various PE subtypes.MethodsSample selectionA total of 48 (out of 157) placentas from our highly an-notated cohort purchased from the Research Centre forWomen’s and Infants’ Health (RCWIH) BioBank [3]were selected for DNA methylation analysis (19 fromtranscriptional cluster 1, 19 from transcriptional cluster2, 5 from transcriptional cluster 3, and 5 from transcrip-tional cluster 5), using the sample function in R 3.1.3(Additional file 1: Figure S1). The selected number ofsamples per cluster is approximately representative ofthe sample distribution in the full placental dataset, withthe condition of a minimum of five samples per cluster.Our cohort selection and tissue sampling methods havebeen previously described [3]. Placentas demonstratingsigns of chorioamnionitis or belonging to thechorioamnionitis-associated transcriptional cluster 4 [3]were not included as these are a known entity, inde-pendent of preeclampsia (Additional file 1: Figure S1).Clinical differences between these 48 patients only wereassessed using Kruskal-Wallis rank sum, Wilcoxon ranksum, and Fisher’s exact tests, as appropriate.Methylation arrays and data processingDNA was isolated from the 48 placentas by ethanol pre-cipitation with the Wizard® Genomic DNA PurificationKit from Promega and quantified by a NanoDrop 1000spectrophotometer. A total of 750 ng of DNA persample was bisulfite converted using the EZ Gold DNAmethylation kit (Zymo) and assessed for methylationstatus with Infinium HumanMethylation450 arrays fromIllumina. This array covers CpG islands (tight clusters ofCpG sites) as well as shores (up to 2 kb from CpGislands), shelves (2–4 kb from CpG islands) and opensea (> 4 kb from CpG islands) [16]. Arrays were scannedby an Illumina HiScan 2000. This methylation data wasalso used as a validation cohort in [17].The resulting IDAT files were loaded into R using thechamp.load function (ChAMP library) [18], excludinglow quality probes with a detection p value above 0.01 inmore than one sample or a beadcount < 3 in at least 5%of samples (N = 1940). Probes known to bind sexchromosomes, cross-hybridize to multiple locations, ortarget a single-nucleotide polymorphism (SNP) wereremoved, based on previous annotation [19, 20]. Thisleft 410,664 probes for DNA methylation analysis. Thesamples underwent functional normalization with thepreprocessFunnorm function [21], which is an extensionof quantile normalization utilizing the control probes onthe array, applied separately to the methylated andunmethylated intensities, type I and type II signals, andthe male and female samples. The data was then batchcorrected for slide and array position using the ComBatfunction (sva library) [22] without accounting for anyoutcome of interest or other covariates to obtain themost unbiased results. All analysis was performed usingM values to improve the statistical calculation ofLeavey et al. Clinical Epigenetics  (2018) 10:28 Page 2 of 13differential methylation [23, 24], although beta values arealso included in the tables for biological interpretation.Gene expression data processingOur entire 157 placenta dataset was previously hybridizedagainst Human Gene 1.0 ST Array chips from Affymetrix[3]. The resulting microarray CEL files for the 48 placentasassessed for methylation in the current study were loadedinto R, and normalized and converted to log2 values usingthe affy library [25]. Expression values annotated to thesame gene symbol were merged to a mean value, and geneswith expression in the lowest quartile were filtered out toreduce confounding by background noise, using thevarFilter function.Identification of differentially methylated positionsThe global relationships between the 48 samples based onthe DNA methylation information alone were visualizedusing t-distributed stochastic neighbor embedding (t-SNE;tsne library) [26] with a perplexity of 10. Samples belongingto our previously described transcriptional clusters 2, 3,and 5 were compared to cluster 1 placentas (with a“healthy” transcriptional profile) to identify differentiallymethylated positions, using the limma library [27]. Theentire cluster 1 was employed as the comparison groupafter confirming that no significant differentially methylatedpositions exist between the PE and normotensive controlswithin cluster 1 by limma, and no segregation of thesephenotype groups within cluster 1 were observed by t-SNE(Fig. 1). Linear modeling, compared to cluster 1, wasperformed both with and without controlling for fetal sex(male or female) and/or gestational age (GA) at delivery(26–40 weeks) to investigate the impact of these variableson each cluster. Fetal sex was still considered despite theremoval of the sex chromosomes from the analysis due tolikely persistent differences on the autosomes [28, 29]. Siteswere considered differentially methylated at a false discoveryrate (FDR) corrected q value < 0.05, and groups of signifi-cant positions were noted when at least three significantsites were identified within 1000 base pairs of each other.Probe annotation and epigenetic regulation of geneexpressionAll DNA methylation probes were assigned to enhancerregions, CpG regions (island, shore, shelf, or open sea),and gene-centric locations (TSS1500: 200-1500 nucleotidesupstream of the transcriptional start site (TSS); TSS200:TSS to 200 nucleotides upstream of the TSS; 5′ untrans-lated region (UTR); 1st exon; gene body; 3′UTR; andintergenic region (IGR)) based on the IlluminaHuman-Methylation450kanno.ilmn12.hg19 library. A number ofsites (N = 45,354) were linked to multiple genes or genealiases, and all possible associations were maintained.Probes found in the IGR were assigned to the gene withthe closest TSS. Trends in significant probe positions wereassessed by Fisher’s exact tests.Sites identified as significantly differentially methylatedin transcriptional cluster 2 or 3 placentas, compared tocluster 1 samples, were investigated for linear correlationsbetween the M values and the corresponding log2 geneexpression values within the relevant two clusters. Corre-lations were considered significant at a FDR < 0.05 andcorrelation groups were compared by Fisher’s exact tests.Fig. 1 t-Distributed Stochastic Neighbor Embedding (t-SNE) visualization of the methylation data in the 48 placenta samples. a Transcriptional clusters1 (black, N = 19), 2 (red, N = 19), and 3 (green, N = 5) continued to display molecular similarity to each other based on the DNA methylation data alone,indicating that methylation plays a significant role in the development of these three clusters. Cluster 5 samples (cyan, N = 5), however, were founddispersed across the methylation plot, no longer forming a united group. b In general, preeclamptic (PE) placentas (pink) were found on the bottomhalf of the t-SNE plot, while the non-PE samples (blue) were predominately observed on the top half. However, the cluster 1 PE patients fully integratedwith their co-clustering controls by methylationLeavey et al. Clinical Epigenetics  (2018) 10:28 Page 3 of 13Significance-based modules integrating the transcriptomeand epigenome (SMITE)Differential gene expression between the current subsetof transcriptional cluster 2 and 3 samples, compared tocluster 1 placentas, was obtained using the limma library[27], controlling for fetal sex and gestational age. Usingthe hg19 genome build within the SMITE library [30] inR 3.3.2, a framework was constructed where each genewas associated with a promoter region (+/− 1500 bpfrom the TSS) and a gene body region (TSS + 1500 bp toTES). The fetal sex and GA-corrected gene expressionand methylation results for clusters 2 and 3 (comparedto cluster 1) were then separately integrated into theframework, and the adjusted and combined methylationp values in the promoter and body gene regions wereobtained using Stouffer’s method, weighted by effectstrength. The relationship between expression andmethylation was set to “bidirectional” in both generegions to avoid biasing the results, and genes werescored based on a weighted significance value (0.4 forexpression, 0.4 for promoter methylation, and 0.2 forbody methylation). Gene scores were considered signifi-cant at a nominal p value < 0.05. Functional modules ofgenes in transcriptional clusters 2 and 3 were then iden-tified based on these gene scores, a Reactome protein-protein interaction graph [31], and the spin-glass networkalgorithm. Significant modules (nominal p < 0.05 and10–500 genes) were subjected to KEGG pathwayenrichment analysis within the SMITE library [30]and terms with a FDR < 0.05 were held as significant.ResultsClinical characteristics and global methylation patternsWithin this subset of 48 cases, transcriptional cluster 1patients (N = 19) remained the healthiest clinically, withthe latest gestational ages at delivery and the highestrates of average-for-gestational-age (AGA) infants (95%)(Additional file 2: Table S1 and Additional file 3: TableS2). Of these cluster 1 patients, 32% (6/19) were associatedwith a diagnosis of PE, though none had co-occurring fetalgrowth restriction. Cluster 2 (N = 19) and cluster 3 (N = 5)samples demonstrated substantially worse clinical out-comes, with abnormal Doppler ultrasound results, earlydeliveries (mean = 31 weeks), and low placental andnewborn weights (mean z-scores < − 1.4) (Additional file 2:Table S1 and Additional file 3: Table S2). In cluster 2, 89%(17/19) were diagnosed with PE and exhibited the highestmaternal blood pressures (average maximum systolicpressure = 175 mmHg) and proteinuria levels (averagemaximum= + 3.5). Cluster 3 (60% PE (3/5)) was morestrongly associated with poor fetal growth, with the largestportion of small-for-gestational-age (SGA) infants (80%)and NICU transfers after delivery (80%). Cluster 5 patients(N = 5, 80% PE) continued to display no unique clinicalassociation (Additional file 2: Table S1 and Additional file 3:Table S2). These results are consistent with our previousobservations in the full transcriptional clusters [3].When the global relationships between these 48patients were visualized using t-SNE of the DNAmethylation data only, transcriptional cluster 1, 2, and 3samples continued to demonstrate molecular similarityto each other (Fig. 1a), indicating that methylation playsan important role in the development of these threeclusters. Cluster 5 samples, however, were founddispersed across the methylation plot, no longer forminga united group (Fig. 1a).Differentially methylated positions between transcriptionalclustersTo identify potential epigenetic markers related to ourtranscriptional clusters, placentas belonging to clusters2, 3, and 5 were independently assessed for differentiallymethylated positions (CpG sites) compared to thehealthier cluster 1. When fetal sex and gestational agewere not considered, this revealed a total of 66,837 positions(53,635 hypo- and 13,202 hyper-) with significantly divergentmethylation in transcriptional cluster 2 samples comparedto cluster 1 (FDR < 0.05; Additional file 4: Table S3). Whenfetal sex (p= 0.51 between clusters 1 and 2) was integratedinto the model, this number was reduced to 64,025, whereaswhen gestational age (p < 0.01 between clusters 1 and 2)alone was incorporated, only 8711 significant positions wereobserved. However, when these two covariates were simul-taneously included in the model, the number of significantsites was 8763 (3310 hypo- and 5453 hyper-) (Table 1 andAdditional file 4: Table S3). Similar to the reference distribu-tion across the full set of possible probes, the majority ofthese (fetal sex and gestational age controlled) significantsites were located in a gene body or an intergenic region (allp > 0.05; Additional file 5: Figure S2a). Conversely, substan-tially fewer significant positions were annotated to a CpGisland (12% versus 34%; p < 0.01) and considerably more tothe CpG open sea (49% versus 33%; p= 0.03) than the distri-bution of the array as a whole (Additional file 5: Figure S2b).Furthermore, 8% (735/8763) of these significant cluster 2sites were found in a group of at least three significantpositions within a span of 1000 base pairs, which were,unsurprisingly, often associated with a CpG island or shoreregion (p < 0.01) (Table 1 and Additional file 4: Table S3).In cluster 3 placentas, 13,348 positions were differentiallymethylated (9084 hypo- and 4264 hyper-) compared tocluster 1 (FDR < 0.05) without accounting for fetal sex andGA (Additional file 6: Table S4). The inclusion of fetal sex(p = 0.12 between clusters 1 and 3) dropped this number to4343, while accounting for gestational age (p = 0.02between clusters 1 and 3) only in the model reduced thesignificant positions to 1749. When differences in boththese variables were considered, the number of significantlyLeavey et al. Clinical Epigenetics  (2018) 10:28 Page 4 of 13altered sites in transcriptional cluster 3 furtherdecreased to 340 (164 hypo- and 176 hyper-) (Table 2and Additional file 6: Table S4). The dispersion of theseprobes was very similar to the results observed incluster 2: within the gene-based regions, the (fetal sexand GA corrected) significant sites were randomlydistributed (all p > 0.05; Additional file 7: Figure S3a);however, the number of probes annotated to CpG islandswas lower than random (14% versus 34%; p < 0.01) andthose located in the CpG open sea was higher (51% versus33%; p = 0.01) (Additional file 7: Figure S3b). Additionally,5% (16/340) of these cluster 3 sites were involved in agroup of significant positions that were again more likelyto be associated with a CpG island region (p < 0.01)(Table 2 and Additional file 6: Table S4).Compared to transcriptional cluster 1, only four CpGsites were initially identified as differentially methylatedin cluster 5 placentas (one hypo- and three hyper-)(FDR < 0.05; Additional file 8: Table S5), and thisnumber became zero when fetal sex and gestational agewere included. This indicates that the gene expressionchanges that define this cluster are not associated withconsistent DNA methylation differences. As such,cluster 5 samples were not investigated further forepigenetic regulation.Specific functional epigenetic modificationsIn order to identify individual epigenetic changes involvedin the transcriptional formation of clusters 2 and 3,significantly differentially methylated sites in these sam-ples compared to cluster 1 were assessed for correlatingchanges in placental gene expression. Of the 66,837identified significant positions in transcriptional cluster 2(before correction for fetal sex and GA), correlative analysiswith the expression of all available associated genes revealedonly 5% with a strong linear relationship (FDR < 0.05;Additional file 9: Table S6). When restricted to the 8763sites that maintained a significant difference betweenclusters 1 and 2 after correction for both fetal sex and GA,9% of potential DNA methylation values exhibited a signifi-cant linear relationship with gene expression (FDR < 0.05;Table 3 and Additional file 9: Table S6). Positively correlat-ing positions were more frequently found in a CpG islandwithin a gene body (p < 0.01) or in the CpG open sea in agene body (p < 0.01) or intergenic region (p = 0.01) (Fig. 2a).Sites with a negative relationship to gene expression wereTable 1 Top 20 significantly differentially methylated sites in transcriptional cluster 2 placentas (N = 19) compared to transcriptionalcluster 1 placentas (N = 19), corrected for fetal sex and gestational age at deliveryProbe Delta M Average M Delta β Average β FDR q value Gene(s) Location(s)a Enhancer Groupbcg10900537 0.41 2.75 0.03 0.87 9.93E-04 FOXN3 Body-open sea True Nocg18498598 0.35 1.80 0.03 0.78 2.15E-03 CUX1 Body-open sea False Nocg11235787 0.36 1.72 0.04 0.77 2.15E-03 MIR195 Body-shelf False Yescg17850498c − 0.93 1.58 − 0.14 0.74 2.15E-03 ECE1 Body-open sea True Nocg01938025 0.62 2.62 0.05 0.86 2.15E-03 SKI Body-shelf False Nocg22807822 0.61 3.49 0.03 0.92 2.15E-03 KANK2 Body-shore False Yescg14601621 0.54 1.79 0.08 0.77 2.15E-03 C9orf3 3′UTR-island False Yescg00483891 − 0.55 1.90 − 0.06 0.79 2.15E-03 CCDC115 Body-shore False Nocg10994126 − 0.63 − 0.23 − 0.13 0.46 2.92E-03 PAPPA2 1stExon-open sea False Nocg17107691 0.76 3.14 0.05 0.89 2.92E-03 KANK2 Body-shore False Yescg01412654 −0.56 −0.38 −0.11 0.44 2.92E-03 PPARG TSS1500-shore False Nocg05359207 −0.52 −2.62 − 0.05 0.14 2.92E-03 ZNF217 Body-shore False Nocg06917772 −0.35 −1.33 − 0.06 0.29 2.92E-03 MIR3167 IGR-shore True Nocg24787238 0.34 1.29 0.04 0.71 2.92E-03 MAD1L1 Body-open sea True Nocg21564965 0.53 4.73 0.01 0.96 2.92E-03 ARHGAP23 Body-open sea False Nocg13562353 −0.53 1.40 −0.09 0.72 2.92E-03 CCL27 TSS200-shelf False Nocg26897909 0.45 1.85 0.04 0.78 2.92E-03 SRGAP2 Body-open sea True Nocg09106999 −0.44 −2.01 −0.05 0.20 2.92E-03 CDK2SILVPMELTSS1500-shoreTSS1500-shoreTSS1500-shoreFalse Nocg02006426 −0.39 −0.23 −0.05 0.46 2.92E-03 DYSF IGR-shore False Nocg19431235 0.36 1.79 0.04 0.78 2.92E-03 DIAPH3 Body-open sea True NoaTSS transcription start site, IGR intergenic region, UTR untranslated regionbIncluded in a group of at least three significantly differentially methylated positions within the span of 1000 base pairscAlso significantly differentially methylated in cluster 3 compared to cluster 1Leavey et al. Clinical Epigenetics  (2018) 10:28 Page 5 of 13commonly annotated to a CpG shelf region in a 5′UTR(p = 0.02) or a CpG shore region in a 5′UTR (p < 0.01),TSS1500 (p = 0.05), or TSS200 (p = 0.01) (Fig. 2a). Mostsignificantly correlating positions within the CpG opensea of a gene body or intergenic region were alsoassociated with an enhancer region (72%; p < 0.01compared to the other CpG/gene regions).In transcriptional cluster 3, the 13,348 significant sitescompared to cluster 1 (before correction for fetal sex andGA) showed a strong linear relationship to gene expressiononly 2% of the time (Additional file 10: Table S7). Thisvalue increased to 8% when the analysis was restricted tothe 340 positions that were significantly differentiallymethylated between clusters 1 and 3 when simultaneouslycontrolling for fetal sex and GA (Table 4). Only three sitesdemonstrated a strong positive relationship with expression:one was in the CpG open sea of AFF3’s 1st exon (p = 0.01),which was not annotated as an enhancer region, and theother two were in the CpG open sea of the MGST1 genebody (p = 0.11) in an enhancer (Fig. 2b). Negativelycorrelating positions were more frequently associated withan open sea region in a TSS200 (p = 0.02), although severalwere also in gene bodies or the IGR (Fig. 2b).Integrated functional gene modulesLastly, in order to reveal significant functional modules ofgenes within clusters 2 and 3, their differential geneexpression and differential gene promoter and bodymethylation information, compared to cluster 1 andcorrected for fetal sex and GA, were subjected to Signifi-cance-based Modules Integrating the Transcriptome andEpigenome (SMITE) analysis [30]. Transcriptional cluster 2contained 9 significant integrated gene modules (p < 0.05),consisting of 18–149 genes each (Fig. 3a and Add-itional file 11: Figure S4). Modules in this cluster withunique genes (1, 4, and 6) were associated with TGF-beta signaling, cell adhesion, endocytosis, leukocytetransendothelial migration, and carbohydrate metabolism(Additional file 12: Table S8). Module 3 genes werecontained within module 2, and these were involved infocal adhesion and regulation of the actin cytoskeleton.Modules 5 and 9 were associated with lipid metabolism,while modules 7 and 8 were linked to oxidative phosphoryl-ation and the citrate cycle. The significantly deregulatedgenes in cluster 2, based on the integrated epigenetic andtranscriptional scores, and their module inclusions, areshown in Additional file 13: Table S9.Table 2 Top 20 significantly differentially methylated sites in transcriptional cluster 3 placentas (N = 5) compared to transcriptionalcluster 1 placentas (N = 19), corrected for fetal sex and gestational age at deliveryProbe Delta M Average M Delta β Average β FDR q value Gene(s) Location(s)a Enhancer Groupbcg22131172 0.79 −3.59 0.04 0.08 1.58E-02 C13orf29LINC00346TSS1500-open seaTSS1500-open seaFalse Yescg24079702 0.50 −5.87 0.01 0.02 1.94E-02 FHL2 TSS200-island False Nocg10959820 0.75 −3.98 0.03 0.06 1.94E-02 RGS12 IGR-shelf True Nocg10319331 0.63 −3.84 0.02 0.07 1.94E-02 TMEM132B Body-open sea False Nocg05929019 −0.80 −2.79 −0.05 0.13 1.94E-02 LAMC2 TSS200-open sea False Nocg22790835 0.60 −2.93 0.04 0.12 1.94E-02 C13orf29LINC00346TSS1500-open seaTSS1500-open seaFalse Yescg14557185 0.78 −5.55 0.01 0.02 1.94E-02 WWTR1 Body-island False Yescg00296578 0.74 −3.60 0.04 0.08 1.94E-02 CRIM1 Body-open sea True Nocg21834463 −0.47 3.31 −0.03 0.91 1.94E-02 SGK1 Body-open sea True Nocg12634306 1.44 −4.70 0.05 0.04 1.94E-02 HEYL Body-open sea True Nocg22342100 0.83 −4.06 0.03 0.06 1.94E-02 KLHL38 TSS1500-open sea False Nocg27570256 −0.67 1.50 −0.10 0.74 1.94E-02 LOC100270710 TSS200-shelf True Nocg24741430 0.66 −3.09 0.05 0.11 1.94E-02 SMAD6 IGR-open sea True Nocg04082512 −0.43 0.77 −0.07 0.63 1.94E-02 GSE1 IGR-open sea True Nocg19458020 0.89 −4.52 0.03 0.04 2.14E-02 RARA TSS1500-island True Nocg25892587 0.67 −3.41 0.04 0.09 2.14E-02 KLF6 IGR-open sea True Nocg07605236c −1.22 2.60 −0.11 0.85 2.14E-02 SFXN5 TSS1500-shore False Nocg19478410 −0.61 2.34 −0.07 0.83 2.16E-02 SYT13 IGR-open sea False Nocg13250752 0.57 −1.98 0.06 0.20 2.16E-02 PCDH18 IGR-open sea True Nocg20669292 −1.03 3.93 −0.05 0.94 2.16E-02 PLEKHH3 Body-island True NoaTSS transcription start site, IGR intergenic regionbIncluded in a group of at least three significantly differentially methylated positions within the span of 1000 base pairscAlso significantly differentially methylated in cluster 2 compared to cluster 1Leavey et al. Clinical Epigenetics  (2018) 10:28 Page 6 of 13In contrast, cluster 3 consisted of 11 significant genemodules (p < 0.05), made up of 24–293 genes each(Fig. 3b and Additional file 14: Figure S5). Modules 1, 4,5, 6, and 7 displayed varying degrees of gene overlap andwere all involved in TGF-beta signaling, focal adhesion,and glycosaminoglycan biosynthesis (Additional file 12:Table S8). Modules 2 and 3 were linked to antigenprocessing/presentation and allograft rejection, whilemodules 8 and 9, with ~ 74% gene overlap, were associatedwith cytokine-cytokine receptor interaction and Jak-STATsignaling. Modules in this cluster with unique genes (10and 11) were involved in purine, amino acid, and biotinmetabolism. The significantly deregulated genes intranscriptional cluster 3, based on both gene expressionand methylation, and their module inclusions, are alsoshown in Additional file 13: Table S9.DiscussionOur previous work unbiasedly investigating the placentalheterogeneity observed in preeclampsia [3] revealed fivetranscriptional clusters, including four subtypes of PEplacentas. However, while gene expression microarraysare an invaluable tool for understanding disease, it isalso possible that, in some cases, an alternate level ofmolecular information is highly involved in thedevelopment of the pathology. Combined epigenetic andexpression analysis of the same preeclamptic placentashas only ever been performed for a small number ofsamples [32, 33] or genes [34]. We, therefore, predictedthat the integration of matched genome-wide DNAmethylation data would further improve our understandingof these placentas, and allow us to investigate both themechanisms underlying the formation of the transcrip-tional clusters and the associations between the multi-molecular data.Overall, we found that the relationships between thetranscriptional cluster 1–3 samples were still visible withinthe DNA methylation information, indicating a significantglobal relationship between the two data types in thesesamples. Cluster 5 placentas, on the other hand, no longerformed a distinct group by methylation. This is unsurprisinggiven that this data type is known to be fairly robust to copynumber abnormalities [35], the driving force behind themolecular formation of this cluster.Within transcriptional clusters 2 and 3, controlling forfetal sex and gestational age in the identification of differen-tially methylated sites, compared to the healthier cluster 1,dramatically reduced the number of significant sites (66,837to 8763 in cluster 2; 13,348 to 340 in cluster 3). However, itpredominately corrected the observed imbalance in theTable 3 Top 20 significant gene expression correlations associated with the 8763 significantly differentially methylated sites intranscriptional cluster 2 placentas (N = 19) compared to transcriptional cluster 1 placentas (N = 19), corrected for fetal sex andgestational age at deliveryProbe Gene Locationa Enhancer Pearson r FDR q valuecg23677911 GALNT2 Body-open sea True − 0.81 3.53E-06cg26333638 HEXB Body-shore False − 0.81 3.53E-06cg04858987 SH3BP5 Body-open sea True − 0.78 1.34E-05cg13553455 COL17A1 TSS1500-open sea False − 0.78 1.34E-05cg16557964 TMEM45A 5′UTR-open sea True − 0.77 1.68E-05cg19140548 SH3PXD2A Body-open sea True − 0.77 2.50E-05cg15700009 LDHA TSS1500-shore False − 0.76 2.70E-05cg23730027 FLNB Body-island False − 0.76 2.93E-05cg18444702 SH3BP5 Body-open sea True − 0.75 4.27E-05cg17338821 FLNB Body-shore True − 0.75 5.14E-05cg25549791 GALE TSS200-shore False − 0.74 5.49E-05cg14019050 ABCA1 TSS1500-island False − 0.74 5.49E-05cg19512693 FLT1 Body-open sea True 0.74 6.15E-05cg04704064 SCARB1 IGR-island False 0.74 6.29E-05cg00411097 TMEM184A 1stExon-open sea True − 0.73 7.24E-05cg11079619 INHBA 5′UTR-shelf False − 0.73 7.32E-05cg00513984 SCARB1 IGR-island False 0.73 7.32E-05cg26509870 PHYHIP IGR-shelf False − 0.73 8.19E-05cg06531595 PDE5A Body-open sea True − 0.72 9.89E-05cg18874575 ZNF559 3′UTR-open sea False 0.72 9.89E-05aTSS transcription start site, IGR intergenic region, UTR untranslated regionLeavey et al. Clinical Epigenetics  (2018) 10:28 Page 7 of 13direction of change (80% hypomethylated to 38% hypo-methylated in cluster 2; 68% hypomethylated to 48% hypo-methylated in cluster 3). Since both clusters 2 and 3 aresignificantly younger than cluster 1 (p < 0.01 and p = 0.02,respectively), this fits with the knowledge that placentasbecome progressively more methylated with time [36],while in cluster 3, a moderate bias in fetal sex (p = 0.12)may have also been involved. Additionally, controlling forfetal sex and GA substantially increased the proportion ofsignificant sites that showed a strong linear relationshipwith gene expression (5% to 9% in cluster 2; 2% to8% in cluster 3), thereby confirming that a large num-ber of sites in the genome undergo DNA methylationchanges in response to differences in these twofactors that are independent of epigenetic regulationand gene expression [9, 36, 37].An additional result of interest was the CpG distributionof significant positions found in transcriptional clusters 2and 3. CpG islands are most commonly associated withthe regulation of gene expression, especially when locatedin the gene’s promoter region [8, 38]. We discovered thatsubstantially fewer of the significant sites were mappedinto CpG islands than anticipated, based on the referencedistribution of all potential CpG sites, although those thatwere annotated to islands were, unsurprisingly, oftenfound in close proximity to each other. Instead, the majorityof significant positions were associated with CpG open seaenhancer regions. This is consistent with a previous reportof enrichment of altered DNA methylation at enhancersand low CpG density regions in early-onset preeclampticplacentas [33]. These open sea enhancer regions, whensignificantly associated with gene expression, were generallyFig. 2 Distribution of the significantly differentially methylated positions in transcriptional cluster 2 and 3 placentas, compared to cluster 1, in terms of theirlinear relationships to gene expression. a Within the 8763 (fetal sex and gestational age corrected) significant sites identified in transcriptional cluster 2,positively correlating positions were more frequently found in a CpG island within a gene body or in the CpG open sea in a gene body or intergenicregion (IGR). Significant methylation sites with a negative relationship to gene expression were commonly annotated to a CpG shelf region in a 5′untranslated region (UTR) or a CpG shore region in a 5′UTR, transcription start site (TSS)1500, or TSS200. b Within the 340 cluster 3 (fetal sex and gestationalage corrected) significant sites, only three demonstrated a strong positive relationship with expression: one was in the CpG open sea of AFF3′s 1st exonand the other two were in the CpG open sea of the MGST1 gene body. Negatively correlating positions were more frequently annotated to an open searegion in a TSS200, although several were also in gene bodies or the IGR. Non-significant correlations are shown in light gray, positive correlations are inmedium gray, and negative correlations are in dark gray. Nominal p values were obtained from Fisher’s exact tests. P values > 0.05 are not shownLeavey et al. Clinical Epigenetics  (2018) 10:28 Page 8 of 13Table 4 All significant gene expression correlations associated with the 340 significantly differentially methylated sites in transcriptionalcluster 3 placentas (N = 5) compared to transcriptional cluster 1 placentas (N = 19), corrected for fetal sex and gestational age at deliveryProbe Gene Locationa Enhancer Pearson r FDR q valuecg03983223 WIPF1 1stExon-open sea FALSE − 0.74 4.45E-03cg05544807 DNMT3A Body-island FALSE − 0.73 4.45E-03cg22462240 LGALS3BP IGR-open sea FALSE − 0.74 4.45E-03cg18275589 DAB2 IGR-shelf FALSE − 0.72 5.02E-03cg09258479 PDZK1IP1 TSS200-open sea FALSE − 0.69 7.40E-03cg07593977 CTSB IGR-open sea TRUE − 0.69 7.40E-03cg24506086 TEAD1 Body-open sea TRUE − 0.66 1.40E-02cg17850498 ECE1 Body-open sea TRUE − 0.65 1.65E-02cg07349094 AFF3 1stExon-open sea FALSE 0.64 1.65E-02cg03821121 MICAL2 5′UTR-open sea TRUE − 0.65 1.65E-02cg04885072 MGST1 Body-open sea TRUE 0.63 1.94E-02cg00874480 MGST1 Body-open sea TRUE 0.62 2.45E-02cg11535839 FOSL2 IGR-open sea FALSE − 0.61 2.59E-02cg05305434 LSP1 TSS200-open sea FALSE − 0.61 2.72E-02cg23170988 SNCG Body-open sea FALSE − 0.60 3.17E-02cg05929019 LAMC2 TSS200-open sea FALSE − 0.59 3.75E-02cg15300730 ZFP36L2 TSS1500-shore FALSE − 0.58 3.75E-02cg22234930 PKM 5′UTR-shelf FALSE − 0.58 3.83E-02aTSS transcription start site, IGR intergenic region, UTR untranslated regionFig. 3 Example SMITE modules. a Cluster 2 module 6 (N= 28 genes; p= 0.02) built around SLC2A1 and HK1 genes and involved in carbohydrate metabolism.b Cluster 3 module 11 (N= 25 genes; p= 0.04) built around MCCC1 and ACACA genes and involved in amino acid and biotin metabolism. Expression isdisplayed on the top left edge of each gene circle (upregulated: dark pink; downregulated: light pink; gray: not significant; white: no data), and combinedpromoter and body methylation are displayed on the bottom left and top right of each circle, respectively (hypermethylated: dark blue; hypomethylated:light blue; gray: not significant; white: no data), compared to cluster 1. The symbol text sizes and center node colors are based on the total gene score (low(gray) to high (red)) and the edge colors are representative of the strength of the associations between the genes (low (gray) to high (red)). The remainingmodules are shown in Additional file 11: Figure S4 and Additional file 14: Figure S5Leavey et al. Clinical Epigenetics  (2018) 10:28 Page 9 of 13located in the gene body and exhibited a positive relation-ship. Sites with a strong negative correlation, on the otherhand, were frequently located in the promoter region(TSS200, TSS1500, 5′UTR), as expected, but wereannotated to CpG shore regions, not islands. Relationshipsbetween CpG shores and gene expression are thoughtto be in response to the binding of transcription factorsand changes in the chromatin structure around thepromoter [39, 40].While the observed proportions of differentiallymethylated sites that were associated with correspondingchanges in gene expression (2–9%) are in line with priorstudies [9, 32, 41, 42], this indicates that a large numberof significant sites in clusters 2 and 3, compared tocluster 1, show no meaningful relationship to geneexpression. Some of these DNA methylation alterationscould be the consequence of changes in gene expressionor function [29, 43, 44], or an adaptive response tomaintain stable or rebalanced expression. They couldfurther be remnants of an earlier developmental process,or the result of environmental exposures or treatments,where the transcriptional evidence is no longer measur-able [36]. Furthermore, methylation is involved in a rangeof functions outside of direct transcriptional regulation,such as genome stability [45], splicing [8, 46], and cellulardevelopment [47], while gene expression can be regulatedby a number of other factors, such as microRNAs [48, 49],transcription factors [43, 50], and/or histone modifications[51, 52]. Therefore, it is expected that these two data typeswould not fully agree at the individual gene level, althoughaltered methylation sites not associated with changes ingene expression could still provide important informationabout the overall status and gestational history of thesepathological placentas.When the transcriptome and epigenome data wasutilized simultaneously in an integrated analysis, thisrevealed modifications in TGF-beta signaling, celladhesion and migration, oxidative phosphorylation, andcarbohydrate and lipid metabolism pathways in cluster 2placentas, confirming that a significant global relationshipexists between the two data types. Placental dysfunctionencompassing dysregulation of these pathways has beenextensively described in the classical paradigm of PEpathophysiology and fits with our characterization of clus-ter 2 patients as demonstrating a “canonical” early-onsetform of PE [3, 53–59]. Additionally, a number of the topsignificant methylation and gene expression correlationsin this cluster (cg23730027 and FLNB, cg13553455 andCOL17A1, cg11079619 and INHBA, cg19140548 andSH3PXD2A, and cg26509870 and PHYHIP) have beenpreviously described in a smaller sample set of early-onsetPE placentas [33], thus validating these relationships. Wealso identified several methylation probes in the gene bodyof FLT1, one of the most frequently investigated markersof PE, with a strong positive correlation to expression, aswell as one associated site in the IGR with a strongnegative correlation. These methylation differences couldbe involved in or attempting to compensate for thepathologically elevated expression of this gene [2, 3], andare significant findings missed by prior studies that havefocused only on FLT1 promoter methylation in early-onset PE [34].In cluster 3 samples, integrated alterations were identifiedinvolving antigen presentation, allograft rejection, cytokine-cytokine receptor interaction, Jak-STAT and TGF-betasignaling, glycosaminoglycan biosynthesis, and metabolism.These are also in line with our prior transcriptional resultsin this “immunological” PE group [3], in which we charac-terized this cluster of patients as demonstrating evidence ofmaternal anti-fetal/placental rejection. While not as widelydiscussed in the literature, the primary involvement ofheightened immune activation has been described inseveral previous studies of PE pathophysiology, along withthese other metabolic pathways [60–66]. Interestingly, oneof the most significant methylation and expression relation-ships observed in this cluster involved DNMT3A (one ofthe DNA methyltransferase enzymes responsible for denovo methylation): a CpG island site (cg05544807) washypermethylated in the DNMT3A gene body, compared tocluster 1, and demonstrated a negative relationship toexpression. While this likely has global implications for theDNA methylation pattern observed in these cluster 3placentas, decreased expression of DNMT3A has beenspecifically implicated in immunological-associated disor-ders [67, 68] and abnormal placentation in preeclampsia[69]. Therefore, this CpG site may serve as a potentialtarget for the epigenetic modulation of pathological geneexpression in this PE subtype.Our study also has some inherent limitations. In ourprevious gene expression analysis, we utilized a largecohort of over 300 placentas to identify clusters and dys-regulated pathways between them. Despite our currentstudy being the largest to integrate methylation andtranscriptional information in PE, this analysis involvedonly 48 placentas. Therefore, it is likely to still be under-powered, thus restricting our discovery of epigeneticchanges in these samples to those with large effect sizes.As such, a future direction will be the validation of thesefindings, and perhaps the identification of new significantsites, in a larger cohort of samples. Additionally, as withall investigations of delivered placentas, it is impossible todetermine whether the observed molecular modificationsare part of the cause or the consequence of the diseaseprocess. Finally, our analysis is based on the assumptionthat the cell composition is the same across all samples.This is probably not the case, as differences in cell ratioscan occur for a range of reasons [4, 42, 70–72], includingplacental maturation or sampling variability. Therefore,Leavey et al. Clinical Epigenetics  (2018) 10:28 Page 10 of 13some of the epigenetic changes that we are interpreting asbeing reflective of gene regulation in all cells may insteadbe due to shifts in cell numbers [29]. However, unfortu-nately, until individual methylation patterns for allpossible placental cell types have been established, thislimitation cannot be resolved. This investigation iscurrently ongoing in our groups.ConclusionsOverall, we have improved our understanding of theportion of the divergent gene expression involved in thedevelopment of transcriptional clusters 2 and 3 that isassociated with changes in DNA methylation, as well asconfirmed the lack of true biological cohesion in cluster 5placentas. Differentially methylated sites in clusters 2 and3, compared to the healthier cluster 1, may have potentialas biomarkers of these patient groups for early screeningin maternal serum, whereas specific genes and sets ofgenes exhibiting a strong epigenetic and transcriptionalrelationship (either linear or integrated) may serve astherapeutic targets to modify or prevent pathologicalchanges in PE placental groups. However, a furtherincrease in sample size and an assessment of additionalmodes of gene regulation will be required to fully compre-hend the mechanisms underlying these subtypes.Additional filesAdditional file 1: Figure S1. Selected samples for methylation arrays.(PDF 199 kb)Additional file 2: Table S1. Continuous clinical characteristics of the 48samples across the transcriptional clusters (PDF 72 kb)Additional file 3: Table S2. Categorical clinical characteristics of the 48samples across the transcriptional clusters (PDF 79 kb)Additional file 4: Table S3. Significantly differentially methylated sitesin transcriptional cluster 2 placentas versus transcriptional cluster 1placentas. (XLSX 9040 kb)Additional file 5: Figure S2. Distribution of significantly differentiallymethylated positions in transcriptional cluster 2 (versus transcriptionalcluster 1) compared to the full set of possible methylation probes. (PDF559 kb)Additional file 6: Table S4. Significantly differentially methylated sitesin transcriptional cluster 3 placentas versus transcriptional cluster 1placentas. (XLSX 1669 kb)Additional file 7: Figure S3. Distribution of significantly differentiallymethylated positions in transcriptional cluster 3 (versus transcriptionalcluster 1) compared to the full set of possible methylation probes. (PDF560 kb)Additional file 8: Table S5. Significantly differentially methylated sitesin transcriptional cluster 5 placentas versus transcriptional cluster 1placentas. (XLSX 38 kb)Additional file 9: Table S6. Significant gene expression correlationsassociated with the significantly differentially methylated sites intranscriptional cluster 2 placentas versus transcriptional cluster 1placentas. (XLSX 259 kb)Additional file 10: Table S7. Significant gene expression correlationsassociated with the significantly differentially methylated sites intranscriptional cluster 3 placentas versus transcriptional cluster 1placentas. (XLSX 63 kb)Additional file 11: Figure S4. Remaining functional SMITE modulesidentified in cluster 2. (PDF 2447 kb)Additional file 12: Table S8. Significant KEGG pathways associatedwith the significant SMITE modules in transcriptional clusters 2 and 3(XLSX 58 kb)Additional file 13: Table S9. Genes with significant integrated geneexpression and methylation scores by SMITE analysis in transcriptionalclusters 2 and 3. (XLSX 86 kb)Additional file 14: Figure S5. Remaining functional SMITE modulesidentified in cluster 3. (PDF 4125 kb)AbbreviationsaCGH: Array-based comparative genomic hybridization; AGA: Average-for-gestational-age; FDR: False discovery rate; GA: Gestational age; IGR: Intergenicregion; PE: Preeclampsia; RCWIH: Research Centre for Women’s and Infants’Health; SGA: Small-for-gestational-age; SMITE: Significance-based ModulesIntegrating the Transcriptome and Epigenome; SNP: Single-nucleotidepolymorphism; TES: Transcriptional end site; t-SNE: t-distributed stochasticneighbor embedding; TSS: Transcriptional start site; UTR: Untranslated regionAcknowledgementsWe thank the donors and the Research Centre for Women’s and Infants’Health (RCWIH) BioBank for the human samples used in this study. Wewould also like to thank Dr. Maria Peñaherrera for assistance in running the450K arrays.FundingThis work was funded by the Canadian Institutes of Health Research (CIHR)grant #49520 to WPR and the CIHR grant #128369 to SAB and BJC. KL issupported by an Ontario Graduate Scholarship, SLW is funded by aUniversity of British Columbia Four Year Doctoral Fellowship, WPR receivessalary support from the BC Children’s Hospital Research Institute, and BJCreceives salary support from a Tier 2 Canada Research Chair in PlacentalDevelopment and Maternal-Fetal Health. The funding bodies had no role inthe design of the study, the collection, analysis, and interpretation of thedata, and the writing of the manuscript.Availability of data and materialsThe gene expression microarray data for our full highly annotated sample set(N = 157) is available from the Gene Expression Omnibus database under theaccession number GSE75010. The matched gene expression and DNAmethylation data for the 48 placentas assessed in the current study isavailable under the accession number GSE98224.Authors’ contributionsKL, BJC, and WPR conceived of the study. SAB and BJC extracted RNA formicroarray analysis. KL extracted DNA for methylation analysis. SLW and WPRran the DNA methylation arrays. KL analyzed the data and drafted themanuscript. BJC, WPR, SLW, and SAB critically revised the manuscript. Allauthors approved the final manuscript.Ethics approval and consent to participateEthics approval was granted from the Research Ethics Boards of Mount SinaiHospital (#13-0211-E), the University of Toronto (#29435), and the OttawaHealth Science Network (#2011623-01H). All women provided writteninformed consent for the collection of biological specimens and medicalinformation.Consent for publicationNot applicable.Competing interestsThe authors declare that they have no competing interests.Leavey et al. Clinical Epigenetics  (2018) 10:28 Page 11 of 13Publisher’s NoteSpringer Nature remains neutral with regard to jurisdictional claims inpublished maps and institutional affiliations.Author details1Department of Physiology, University of Toronto, 1 King’s College Circle,Toronto, ON, Canada. 2BC Children’s Hospital Research Institute, 950 W 28thAve, Vancouver, BC, Canada. 3Department of Medical Genetics, University ofBritish Columbia, C201-4500 Oak St, Vancouver, BC, Canada. 4InterdisciplinarySchool of Health Sciences, University of Ottawa, 25 University Private, Ottawa,ON, Canada. 5Department of Cellular and Molecular Medicine, University ofOttawa, 451 Smyth Rd, Ottawa, ON, Canada. 6Department of Obstetrics andGynecology, University of Toronto, 23 Edward Street, Toronto, ON, Canada.Received: 1 November 2017 Accepted: 21 February 2018References1. Magee LA, Pels A, Helewa M, Rey E, von Dadelszen P. Diagnosis, evaluation,and management of the hypertensive disorders of pregnancy. PregnancyHypertens. 2014;4(2):105–45.2. Leavey K, Bainbridge SA, Cox BJ. Large scale aggregate microarray analysisreveals three distinct molecular subclasses of human preeclampsia. PLoSOne. 2015;10(2):e0116508.3. Leavey K, Benton SJ, Grynspan D, Kingdom JC, Bainbridge SA, Cox BJ.Unsupervised placental gene expression profiling identifies clinicallyrelevant subclasses of human preeclampsia. Hypertension. 2016;68:137–47.4. Avila L, Yuen RK, Diego-Alvarez D, Penaherrera MS, Jiang R, Robinson WP.Evaluating DNA methylation and gene expression variability in the humanterm placenta. Placenta. 2010;31(12):1070–7.5. Matsui M, Corey DR. Non-coding RNAs as drug targets. Nat Rev DrugDiscov. 2017;16(3):167–79.6. Robertson KD. DNA methylation and human disease. Nat Rev Genet. 2005;6(8):597–610.7. Bird AP. DNA methylation versus gene expression. Development. 1984;83(Supplement):31–40.8. Jones PA. Functions of DNA methylation: islands, start sites, gene bodiesand beyond. Nat Rev Genet. 2012;13(7):484–92.9. Lim YC, Li J, Ni Y, Liang Q, Zhang J, Yeo GS, et al. A complex associationbetween DNA methylation and gene expression in human placenta at firstand third trimesters. PLoS One. 2017;12(7):e0181155.10. Schultz MD, He Y, Whitaker JW, Hariharan M, Mukamel EA, Leung D, et al.Human body epigenome maps reveal noncanonical DNA methylationvariation. Nature. 2015;523(7559):212–6.11. Yuen RK, Chen B, Blair JD, Robinson WP, Nelson DM. Hypoxia alters theepigenetic profile in cultured human placental trophoblasts. Epigenetics.2013;8(2):192–202.12. Van Neste L, Herman JG, Otto G, Bigley JW, Epstein JI, Van Criekinge W. Theepigenetic promise for prostate cancer diagnosis. Prostate. 2012;72(11):1248–61.13. Heyn H, Esteller M. DNA methylation profiling in the clinic: applications andchallenges. Nat Rev Genet. 2012;13(10):679–92.14. Yang X, Lay F, Han H, Jones PA. Targeting DNA methylation for epigenetictherapy. Trends Pharmacol Sci. 2010;31(11):536–46.15. Issa JP. DNA methylation as a therapeutic target in cancer. Clin Cancer Res.2007;13(6):1634–7.16. Rechache NS, Wang Y, Stevenson HS, Killian JK, Edelman DC, Merino M, etal. DNA methylation profiling identifies global methylation differences andmarkers of adrenocortical tumors. J Clin Endocrinol Metab. 2012;97(6):E1004–13.17. Wilson SL, Leavey K, Cox B, Robinson WP. Mining DNA methylationalterations towards a classification of placental pathologies. Hum Mol Genet.2018;27(1):135–46.18. Morris TJ, Butcher LM, Feber A, Teschendorff AE, Chakravarthy AR, WojdaczTK, et al. ChAMP: 450k chip analysis methylation pipeline. Bioinformatics.2013;30(3):428–30.19. Nordlund J, Bäcklin CL, Wahlberg P, Busche S, Berglund EC, Eloranta ML, etal. Genome-wide signatures of differential DNA methylation in pediatricacute lymphoblastic leukemia. Genome Biol. 2013;14(9):r105.20. Price EM, Cotton AM, Lam LL, Farré P, Emberly E, Brown CJ, et al. Additionalannotation enhances potential for biologically-relevant analysis of theIllumina Infinium HumanMethylation450 BeadChip array. EpigeneticsChromatin. 2013;6(1):4.21. Fortin JP, Labbe A, Lemire M, Zanke BW, Hudson TJ, Fertig EJ, et al.Functional normalization of 450k methylation array data improvesreplication in large cancer studies. Genome Biol. 2014;15(11):503.22. Leek JT, Johnson WE, Parker HS, Jaffe AE, Storey JD. The sva package forremoving batch effects and other unwanted variation in high-throughputexperiments. Bioinformatics. 2012;28(6):882–3.23. Du P, Zhang X, Huang CC, Jafari N, Kibbe WA, Hou L, et al. Comparison ofBeta-value and M-value methods for quantifying methylation levels bymicroarray analysis. BMC Bioinformatics. 2010;11(1):587.24. Zhuang J, Widschwendter M, Teschendorff AE. A comparison of featureselection and classification methods in DNA methylation studies using theIllumina Infinium platform. BMC Bioinformatics. 2012;13(1):59.25. Gautier L, Cope L, Bolstad BM, Irizarry RA. affy—analysis of AffymetrixGeneChip data at the probe level. Bioinformatics. 2004;20(3):307–15.26. Maaten LV, Hinton G. Visualizing data using t-SNE. J Mach Learn Res. 2008;9:2579–605.27. Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W, et al. Limma powersdifferential expression analyses for RNA-sequencing and microarray studies.Nucleic Acids Res. 2015;43(7):e47.28. Singmann P, Shem-Tov D, Wahl S, Grallert H, Fiorito G, Shin SY, et al.Characterization of whole-genome autosomal differences of DNA methylationbetween men and women. Epigenetics Chromatin. 2015;8(1):43.29. Wilson SL, Robinson WP. Utility of DNA methylation to assess placentalhealth. Placenta. 2017; https://doi.org/10.1016/j.placenta.2017.12.013.30. Wijetunga NA, Johnston AD, Maekawa R, Delahaye F, Ulahannan N, Kim K,et al. SMITE: an R/Bioconductor package that identifies network modules byintegrating genomic and epigenomic information. BMC Bioinformatics.2017;18(1):41.31. Fabregat A, Sidiropoulos K, Garapati P, Gillespie M, Hausmann K, Haw R, etal. The reactome pathway knowledgebase. Nucleic Acids Res. 2015;44(D1):D481–7.32. Xuan J, Jing Z, Yuanfang Z, Xiaoju H, Pei L, Guiyin J, et al. Comprehensiveanalysis of DNA methylation and gene expression of placental tissue inpreeclampsia patients. Hypertens Pregnancy. 2016;35(1):129–38.33. Blair JD, Yuen RK, Lim BK, McFadden DE, von Dadelszen P, Robinson WP.Widespread DNA hypomethylation at gene enhancer regions in placentasassociated with early-onset pre-eclampsia. Mol Hum Reprod. 2013;19(10):697–708.34. Sundrani DP, Reddy US, Joshi AA, Mehendale SS, Chavan-Gautam PM,Hardikar AA, et al. Differential placental methylation and expression of VEGF,FLT-1 and KDR genes in human term and preterm preeclampsia. ClinEpigenetics. 2013;5(1):6.35. Blair JD, Langlois S, McFadden DE, Robinson WP. Overlapping DNAmethylation profile between placentas with trisomy 16 and early-onsetpreeclampsia. Placenta. 2014;35(3):216–22.36. Novakovic B, Yuen RK, Gordon L, Penaherrera MS, Sharkey A, Moffett A, etal. Evidence for widespread changes in promoter methylation profile inhuman placenta in response to increasing gestational age andenvironmental/stochastic factors. BMC Genomics. 2011;12(1):529.37. Martin E, Smeester L, Bommarito PA, Grace MR, Boggess K, Kuban K, et al.Sexual epigenetic dimorphism in the human placenta: implications forsusceptibility during the prenatal period. Epigenomics. 2017;9(3):267–78.38. Esteller M. CpG island hypermethylation and tumor suppressor genes: abooming present, a brighter future. Oncogene. 2002;21(35):5427.39. Irizarry RA, Ladd-Acosta C, Wen B, Wu Z, Montano C, Onyango P, et al.The human colon cancer methylome shows similar hypo-andhypermethylation at conserved tissue-specific CpG island shores. NatGenet. 2009;41(2):178–86.40. Robinson WP, Price EM. The human placental methylome. Cold Spring HarbPerspect Med. 2015;5(5):a023044.41. Edgar R, Tan PP, Portales-Casamar E, Pavlidis P. Meta-analysis of humanmethylomes reveals stably methylated sequences surrounding CpGislands associated with high gene expression. Epigenetics Chromatin.2014;7(1):28.42. Lam LL, Emberly E, Fraser HB, Neumann SM, Chen E, Miller GE, et al. Factorsunderlying variable DNA methylation in a human community cohort. ProcNatl Acad Sci. 2012;109(Suppl 2):17253–60.43. Moarii M, Boeva V, Vert JP, Reyal F. Changes in correlation between promotermethylation and gene expression in cancer. BMC Genomics. 2015;16(1):873.Leavey et al. Clinical Epigenetics  (2018) 10:28 Page 12 of 1344. Timp W, Feinberg AP. Cancer as a dysregulated epigenome allowingcellular growth advantage at the expense of the host. Nat Rev Cancer. 2013;13(7):497–510.45. Putiri EL, Robertson KD. Epigenetic mechanisms and genome stability. ClinEpigenetics. 2011;2(2):299.46. Luco RF, Allo M, Schor IE, Kornblihtt AR, Misteli T. Epigenetics in alternativepre-mRNA splicing. Cell. 2011;144(1):16–26.47. Smith ZD, Meissner A. DNA methylation: roles in mammalian development.Nat Rev Genet. 2013;14(3):204–20.48. Bartel DP. MicroRNAs: genomics, biogenesis, mechanism, and function. Cell.2004;116(2):281–97.49. Chen K, Rajewsky N. The evolution of gene regulation by transcriptionfactors and microRNAs. Nat Rev Genet. 2007;8(2):93–103.50. Karin M. Too many transcription factors: positive and negative interactions.New Biol. 1990;2(2):126–31.51. Berger SL. Histone modifications in transcriptional regulation. Curr OpinGenet Dev. 2002;12(2):142–8.52. Heintzman ND, Hon GC, Hawkins RD, Kheradpour P, Stark A, Harp LF, et al.Histone modifications at human enhancers reflect global cell-type-specificgene expression. Nature. 2009;459(7243):108–12.53. Caniggia I, Grisaru-Gravnosky S, Kuliszewsky M, Post M, Lye SJ. Inhibition ofTGF-β3 restores the invasive capability of extravillous trophoblasts inpreeclamptic pregnancies. J Clin Investig. 1999;103(12):1641.54. Zhou X, Li Q, Xu J, Zhang X, Zhang H, Xiang Y, et al. The aberrantlyexpressed miR-193b-3p contributes to preeclampsia through regulatingtransforming growth factor-β signaling. Sci Rep. 2016;6:19910.55. Kang JH, Song H, Yoon JA, Park DY, Kim SH, Lee KJ, et al. Preeclampsia leadsto dysregulation of various signaling pathways in placenta. J Hypertens.2011;29(5):928–36.56. Bloxam DL, Bullen BE, Walters BN, Lao TT. Placental glycolysis and energymetabolism in preeclampsia. Am J Obstet Gynecol. 1987;157(1):97–101.57. Korkes HA, Sass N, Moron AF, Câmara NO, Bonetti T, Cerdeira AS, et al.Lipidomic assessment of plasma and placenta of women with early-onsetpreeclampsia. PLoS One. 2014;9(10):e110747.58. Riquelme G, Vallejos C, De Gregorio N, Morales B, Godoy V, Berrios M, et al.Lipid rafts and cytoskeletal proteins in placental microvilli membranes frompreeclamptic and IUGR pregnancies. J Membr Biol. 2011;241(3):127.59. Kim MS, Yu JH, Lee MY, Kim AL, Jo MH, Kim M, et al. Differential expressionof extracellular matrix and adhesion molecules in fetal-origin amnioticepithelial cells of Preeclamptic pregnancy. PLoS One. 2016;11(5):e0156038.60. Laresgoiti-Servitje E. A leading role for the immune system in thepathophysiology of preeclampsia. J Leukoc Biol. 2013;94(2):247–57.61. Conrad KP, Benyo DF. Placental cytokines and the pathogenesis ofpreeclampsia. Am J Reprod Immunol. 1997;37(3):240–9.62. Kim CJ, Romero R, Chaemsaithong P, Kim JS. Chronic inflammation of theplacenta: definition, classification, pathogenesis, and clinical significance. AmJ Obstet Gynecol. 2015;213(4):S53–69.63. Wilczyński JR. Immunological analogy between allograft rejection, recurrentabortion and pre-eclampsia–the same basic mechanism? Hum Immunol.2006;67(7):492–511.64. Gleicher N. Why much of the pathophysiology of preeclampsia-eclampsiamust be of an autoimmune nature. Am J Obstet Gynecol. 2007;196(1):5–e1-7.65. Heyer-Chauhan N, Ovbude IJ, Hills AA, Sullivan MH, Hills FA. Placentalsyndecan-1 and sulphated glycosaminoglycans are decreased inpreeclampsia. J Perinat Med. 2014;42(3):329–38.66. Pérez-Sepúlveda A, España-Perrot PP, Fernández BX, Ahumada V, Bustos V,Arraztoa JA, et al. Levels of key enzymes of methionine-homocysteinemetabolism in preeclampsia. Biomed Res Int. 2013;2013:731962.67. Nawrocki MJ, Majewski D, Puszczewicz M, Jagodziński PP. Decreased mRNAexpression levels of DNA methyltransferases type 1 and 3A in systemiclupus erythematosus. Rheumatol Int. 2017;37(5):775–83.68. Liu Y, Chen Y, Richardson B. Decreased DNA methyltransferase levelscontribute to abnormal gene expression in “senescent” CD4+ CD28− T cells.Clin Immunol. 2009;132(2):257–65.69. Jia Y, Li T, Huang X, Xu X, Zhou X, Jia L, et al. Dysregulated DNAmethyltransferase 3A upregulates IGFBP5 to suppress trophoblast cellmigration and invasion in preeclampsia. Hypertension. 2017;69(2):356–66.70. Adalsteinsson BT, Gudnason H, Aspelund T, Harris TB, Launer LJ, EiriksdottirG, et al. Heterogeneity in white blood cells has potential to confound DNAmethylation measurements. PLoS One. 2012;7(10):e46705.71. Reinius LE, Acevedo N, Joerink M, Pershagen G, Dahlén SE, Greco D, et al.Differential DNA methylation in purified human blood cells: implications forcell lineage and studies on disease susceptibility. PLoS One. 2012;7(7):e41361.72. Varley KE, Gertz J, Bowling KM, Parker SL, Reddy TE, Pauli-Behn F, et al.Dynamic DNA methylation across diverse human cell lines and tissues.Genome Res. 2013;23(3):555–67.•  We accept pre-submission inquiries •  Our selector tool helps you to find the most relevant journal•  We provide round the clock customer support •  Convenient online submission•  Thorough peer review•  Inclusion in PubMed and all major indexing services •  Maximum visibility for your researchSubmit your manuscript atwww.biomedcentral.com/submitSubmit your next manuscript to BioMed Central and we will help you at every step:Leavey et al. Clinical Epigenetics  (2018) 10:28 Page 13 of 13

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.52383.1-0364135/manifest

Comment

Related Items