MicroRNA Signature of Epithelial-Mesenchymal Transition in Group B Streptococcal Infection of the Placental Chorioamniotic Membranes

Abstract

Background

Infection-induced preterm birth is a major cause of neonatal mortality and morbidity and leads to preterm premature rupture of placental chorioamniotic membranes. The loss of amniotic epithelial cells and tensile strength preceding membrane rupture is poorly understood. We hypothesized that intrauterine bacterial infection induces changes in microRNA (miRNA) expression, leading to amniotic epithelial cell loss and membrane weakening.

Methods

Ten pregnant pigtail macaques received choriodecidual inoculation of either group B Streptococcus (GBS) or saline (n = 5/group). Placental chorioamniotic membranes were studied using RNA microarray and immunohistochemistry. Chorioamniotic membranes from women with preterm premature rupture of membranes (pPROM) and normal term pregnancies were studied using transmission electron microscopy.

Results

In our model, an experimental GBS infection was associated with changes in the miRNA profile in the chorioamniotic membranes consistent with epithelial to mesenchymal transition (EMT) with loss of epithelial (E-cadherin) and gain of mesenchymal (vimentin) markers. Similarly, loss of desmosomes (intercellular junctions) was seen in placental tissues from women with pPROM.

Conclusions

We describe EMT as a novel mechanism for infection-associated chorioamniotic membrane weakening, which may be a common pathway for many etiologies of pPROM. Therapy based on anti-miRNA targeting of EMT may prevent pPROM due to perinatal infection.

Preterm premature rupture of membranes (pPROM) is a major clinical antecedent to preterm birth and is associated with intraamniotic infection in 30%–50% of cases [1]. Several biological pathways have been identified to act synergistically to weaken membranes [2], including the catabolic degradation of collagen mediated by matrix metalloproteinases [3, 4], apoptosis [5–7], oxidative stress [7], and involvement of other proteases contained within neutrophils [8]. Of these processes, matrix metalloproteinase activation represents the best-studied mechanism. Interleukin-6 (IL-6)–mediated local inflammation at the site of membrane rupture also recruits neutrophil trafficking into the membranes with resultant protease production [8]. Whether master regulators of matrix metalloproteinase activation, inflammation, and other biological events leading to membrane weakening and rupture exist across heterogeneous etiologies of pPROM remains unknown.

MicroRNAs (miRNAs) are a class of small noncoding RNAs, which exert powerful effects on gene expression [9]. miRNAs play critical roles in many physiologic and pathologic processes through both mRNA degradation and the inhibition of translation initiation of target mRNAs [10]. Previous studies of miRNAs in pregnancy have investigated their roles in placental development [11] and trophoblast differentiation [11, 12], preeclampsia [11, 12], events of term and preterm parturition leading to myometrial contractility [13], and cervical remodeling. Differential miRNA expression in the context of advancing gestation [14], chorioamnionitis [14], and labor [15] has also been examined. The effects of choriodecidual infection on chorioamnion miRNA expression profiles and the subsequent downstream transcriptional/translational regulatory effects on the placental chorioamniotic membranes are unknown.

We have previously demonstrated that a group B streptococcal (GBS) infection in the choriodecidua of a nonhuman primate (NHP) model leads to weakening of the fetal membranes through dysfunction of the cytokeratin network along with associated mRNA changes [16]. Therefore, we hypothesized that changes in miRNA expression after infection and resultant inflammation could induce early pathways responsible for these observations leading to loss of amnion integrity and chorioamniotic membrane weakening. The results indicate differential regulation of miRNA clusters associated with epithelial-mesenchymal transition (EMT). During EMT, an epithelial cell undergoes a rapid change in morphology, loses cell-cell adhesions, retracts its cytoskeleton, and becomes motile; this process in the chorioamniotic membranes would reduce tissue integrity. Recently, amniotic epithelial cells have been demonstrated to have the ability to perform EMT [17, 18]. In the fetal membranes, epithelial cell loss would predispose to membrane rupture and preterm birth. Our findings provide new insight into the mechanism of pPROM after GBS infection and demonstrate the relevance of miRNAs to chorioamniotic membrane weakening and preterm birth.

METHODS

Ethics Statement

This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Research Council and the Weatherall report [19]. The protocol was approved by the Institutional Animal Care Use Committee of the University of Washington (permit number, 4165-01). All surgery was performed under general anesthesia and all efforts were made to minimize discomfort.

Human chorioamniotic membrane samples were obtained under the approval of the University of Washington Institutional Review Board (IRB) 34004 and through the Global Alliance for the Prevention of Prematurity and Stillbirth (GAPPS, Seattle, WA) tissue repository under the approval of Seattle Children’s Hospital IRB 12879 and 13975. For both the University of Washington and GAPPS samples, subjects provided written informed consent.

Study Design

Ten pregnant NHPs (Macaca nemestrina) were chronically catheterized at 118–125 days gestation. Following a recovery period, they received 1 of 2 experimental treatments: (1) choriodecidual and intraamniotic saline infusions (n = 5), or (2) choriodecidual inoculation of GBS (COH-1) (n = 5). Experiments were ended at 3 days after GBS infection or earlier if preterm labor occurred; in the saline control group, experiments were ended 7 days after infusion. The chorioamniotic membranes were harvested following cesarean delivery and the inoculation site was identified. Samples from the inoculation site were then stored in RNALater; the chorioamniotic membranes were rolled on a Dacron swab to create tissue “rolls,” which were then frozen in optimal cutting temperature (OCT) compound as well as fixed in formalin and paraffin embedded. Note that term gestation in our colony is 172 days; the timing of catheterization corresponds approximately to 27–29 weeks human gestation (late 2nd/early 3rd trimester). At the experimental endpoint, a cesarean delivery was performed and the fetus was euthanized. Next, a fetal necropsy was performed to recover placental and fetal tissues for analysis. Additional details regarding the experimental methods can be found in the Supplementary Material and previous publications [20, 21]. Following a recovery period, the dam was returned to the colony.

RNA Extraction, Microarray Processing, Data Analysis, and Validation

RNA extraction of the chorioamniotic membranes near the GBS or saline inoculation site was performed using the Qiagen miRNeasy Kit. Total RNA from each sample (500 ng) was processed for miRNA array analysis using the Affymetrix GeneChip miRNA 2.0 Array. Hybridized Affymetrix arrays were scanned with an Affymetrix GeneChip 3000 fluorescent scanner. Image generation and feature extraction were performed using Affymetrix GeneChip Command Console software. Raw microarray data were preprocessed and analyzed with Bioconductor (http://www.bioconductor.org/) [22]. Additional quality control steps can be found in the Supplementary Material. The data were quantile normalized and summarized using the Bioconductor oligo package. From the normalized data, miRNA transcripts with significant evidence for differential expression were identified using the limma package in Bioconductor. P values were calculated with a modified t test in conjunction with an empirical Bayes method to moderate the standard errors of the estimated log-fold changes. Differentially expressed miRNA transcripts were selected by an unadjusted P value < .05, an absolute fold change ≥ 2.0, and log₂ signal intensity > 5. The miRNA and mRNA data have been deposited in the NCBI Gene Expression Omnibus database through GEO Series accession number GSE77152 (miRNA) and GSE46290 (mRNA).

Next, the data were analyzed using Gene Set Analysis in R code (http://www-stat.stanford.edu/~tibs/GSA/) [23, 24]. In Gene Set Analysis, the P values that are calculated to test the null hypothesis are based on permutations of the sample labels using the Bioconductor romer function from the limma package. We used gene set databases from the Broad Institute and Gene Ontology [25]. We also used the microRNA target filter feature of the Ingenuity Pathway Analysis (IPA) software (Qiagen) to determine the predicted mRNA targets of the differentially expressed miRNAs, using the default filter settings.

Reverse transcription was performed using the TaqMan MicroRNA RT Kit (Life Technologies) following the manufacturer’s protocol to validate the results. Quantitative reverse transcription polymerase chain reaction (qRT-PCR) was carried out on an ABI Prism 7900 Sequence Detection System (Life Technologies). U6 snRNA was used as an endogenous control.

Immunohistochemistry

E-cadherin, N-cadherin, and vimentin immunohistochemical staining was performed using the following primary antibodies: (1) monoclonal mouse anti-human IgG1 (HECD1/E-cadherin, 1:400 dilution, made from a hybridoma culture supernatant; gift from W. G. Carter); (2) monoclonal mouse anti-human IgG1 (N-cadherin, 1:2000, ab98952; Abcam); (3) monoclonal mouse anti-human IgG1 (vimentin, 1:3000, M0725; Dako); (4) a mouse IgG isotype control (1:200, I-2000; Vector Laboratories), and (5) a rabbit IgG isotype control (1:1000, AB-105-C; R&D Systems). We used OCT-embedded cryosections for E-cadherin staining and formalin-fixed paraffin-embedded sections for N-cadherin and vimentin staining. Details on staining protocols and immunohistochemistry analysis are provided in the Supplementary Material.

Transmission Electron Microscopy

Formalin-fixed and paraffin-embedded tissues from human chorioamnion were used for transmission electron microscopy (see Supplementary Material). The number of desmosomes between 2 neighboring amniotic epithelial cells in human tissues was counted by using overlapping images (20 000 × and 40 000 ×) with visual reconstruction of the cell-cell junction from apex to base. Typically, 20–40 images were reviewed for each case or control to determine desmosome counts along at least 6 cell-cell junctions. Two observers blinded to the case-control status performed the desmosome counts (S.W. and K. M. A. W.).

EMT in Human Amniotic Epithelial Cells

Details for isolation, transfection, and evaluation of EMT in human amniotic epithelial cells can be found in the Supplementary Material. Briefly, human amniotic epithelial cells (n = 6/condition) were transfected with hsa-miR-200a-3p (Millipore Sigma), or SilencerSelect Negative Control No 1 (Abcam), or media only. Then, the amniotic cells were treated with GBS COH-1 and uninfected cells were also included as controls (n = 3/condition). Three independent cultures of GBS were used. After 24 hours, the cells were harvested, stained with phycoerythrin-conjugated E-cadherin (1:30; clone 36; BD Biosciences), washed, and then analyzed for E-cadherin expression on an LSR II flow cytometer (BD Biosciences). Cells in 3 separate wells were used for each condition. mData were analyzed using FlowJo software version 10.

Statistical Analysis

A Wilcoxon rank sum was used for analysis of qRT-PCR and immunohistochemistry. A linear regression correlated log-transformed qRT-PCR results with log-transformed peak amniotic fluid cytokine or prostaglandin values. For the analysis of desmosome number, linear regression with robust standard errors was used with clustering by person. An unpaired t test compared E-cadherin expression in GBS and untreated human amniotic epithelial cells.

RESULTS

miRNA and mRNA Expression Profile Characteristic of EMT Is Associated With Choriodecidual GBS Infection

We observed differential expression of 71 out of 1105 probe sets in the GBS group compared to saline-treated controls. Mapping differentially expressed probe sets using IPA yielded 67 miRNA gene sequences, 6 of which were duplicated (select miRNAs in Table 1 and complete set in Supplementary Table 1). We also fit miRNA expression data using linear models to amniotic fluid inflammatory cytokines (IL-1β and tumor necrosis factor-α [TNF-α];Supplementary Figure 1). miRNA expression was significantly correlated with peak amniotic fluid IL-1β and TNF-α(P = 3.7 × 10^–21 and 1.0 × 10^–13, respectively).

To validate differential expression of miRNAs on the array, we performed qRT-PCR. There was a significant downregulation of miR-149-5p, miR-182-5p, miR-200b, miR-203-5p, miR-205-5p, miR-495-5p, miR-183-5p, and miR-506-5p (P < .05; Figure 1). Cycle threshold values were greater than 35 for miR-183-5p and miR-506-5p, indicating that these miRNA were expressed at extremely low levels. All of these validated miRNAs have been implicated in EMT or cancer [26, 27].

Next, we performed a gene set analysis to evaluate gene expression among groups of genes. Multiple gene sets involved with EMT were significantly differentially expressed (eg, Broad Institute Hallmark EMT Gene Set, P = .007, Figure 2; Epithelial to Mesenchymal Transition GO:0001837, P = .003; Gotzmann Epithelial to Mesenchymal Transition Down, P = .027; Gotzmann Epithelial to Mesenchymal Transition Up, P = .007).

Expression of several transcription factor genes has been associated with the induction of EMT, including zinc finger E-box binding homeobox 2 (ZEB2) and E-cadherin (CDH1) in our previously published microarray data [16]. We performed qRT-PCR to validate the upregulation of ZEB2 and downregulation of CDH1. There was a significant increase in ZEB2 mRNA in GBS versus controls (P = .03). There was no difference in CDH1 mRNA abundance between cases and controls, but mRNA levels were significantly negatively correlated with peak amniotic fluid IL-6 (P = .03) and IL-8 (P = .009). In summary, ZEB2, a transcription factor known to induce EMT, was significantly upregulated in the chorioamniontic membranes after GBS infection. Lower E-cadherin transcription correlated with higher levels of inflammatory cytokines in the amniotic fluid.

Putative and Validated mRNA Targets by Ingenuity Pathway Analysis

To identify miRNA targets in an unbiased manner, canonical pathways were analyzed using IPA. The regulation of EMT pathway was significantly perturbed in the setting of GBS infection when compared to saline controls (P = 2 × 10^–5). Next, we used IPA to investigate putative and validated mRNA targets of our differentially regulated miRNAs to gain insight into the pathogenesis of chorioamnion injury or repair (Figure 3). Of the 61 differentially regulated, nonduplicated miRNAs, 56 miRNAs had predicted mRNA targets. Using the IPA microRNA target filter feature, we found that these 56 differentially regulated miRNAs had 14 080 predicted mRNA targets. When overlapped with the mRNA microarray data from the same chorioamnion samples previously published [16], 593 mRNA targets were identified with inverse pairing. Twelve of these mRNA targets were previously experimentally validated. Eight of the 12 mRNA targets are known to be upregulated in EMT or cancer: ZEB2 [26], TGFB2 (transforming growth factor, beta 2) [28], BCL6 (B-cell CLL/lymphoma 6) [29], KIF23 (kinesin family member 23) [30], CENPF (centromere protein H) [31], LMNB1 (lamin B1) [32], ACVRB1B (activin A receptor type IB) [33], and FHOD1 (formin homology 2 domain containing 1) [34]. Several miRNAs differentially regulated in our study were predicted to target mRNAs that code for proteins providing cytostructural support to the amnion (ie, collagens, integrins, and cytokeratins) and within cell-cell adhesions (claudins, desmocollin, and desmogleins; Table 2).

Immunohistochemistry of EMT-Associated Proteins

EMT is classically associated with a loss of epithelial markers and gain of a mesenchymal phenotype [35]. To determine if proteins typically associated with EMT changed after GBS infection, we performed immunohistochemistry for E-cadherin (epithelial marker) and N-cadherin and vimentin (mesenchymal markers; Figure 4). In saline controls, there was diffuse and often dark staining of E-cadherin throughout the amniotic epithelium (Figure 4A). In contrast, GBS infection was associated with scattered regions of less-intense E-cadherin staining, as well as patchy loss of the epithelium (Figure 4B), which was expected with EMT. In contrast, vimentin and N-cadherin staining on amniotic epithelial cells appeared to be increased in association with an influx of neutrophils in GBS cases. Overall, there was no significant difference in immunohistochemical staining between cases and controls, but analysis was compromised by amniotic epithelial cell loss or injury in GBS cases (see Supplementary Material). Immunohistochemical staining for several markers was significantly correlated with peak levels of inflammatory cytokines and prostaglandins (E-cadherin negatively correlated with prostaglandin F2-α, P = .04; N-cadherin positively correlated with prostaglandin E2, P = .04; and vimentin positively correlated with IL-6, P = .02).

Loss of Desmosomes in the Amnion of Women With pPROM

To determine if ultrastructural changes associated with EMT were evident in the chorioamniotic membranes from women with pPROM, we counted the number of desmosomes between 2 neighboring amniotic epithelial cells using transmission electron microscopy. Desmosomes are intercellular junctions that provide mechanical resilience and medical cell-cell adhesions, which can also be transcriptionally repressed by factors that promote EMT [36]. We found a significant decrease in desmosomes per cell-cell junction comparing term pregnancies with pPROM (term cesarean delivery, mean 15.5 ± 1.1 [standard error of the mean (SEM)]; pPROM, mean 5.5 ± 0.7 (SEM); P < .0001; Figure 5). In many pPROM cases, desmosomes were smaller and lighter in intensity suggesting less protein within the desmosome.

Loss of E-cadherin in Human Amniotic Epithelial Cells due to GBS Was Abrogated by an miRNA-200a-3p Mimic

To further validate our findings, we hypothesized that increased levels of miRNA-200a-3p may prevent loss of E-cadherin associated with GBS infection of human amniotic epithelial cells. We used a mimic for miRNA-200a-3p, as this miRNA was significantly downregulated in our array (P < .04; Table 1). First, we isolated human amniotic epithelial cells from a term, nonlabored placenta as previously described [37]. Then, amniotic epithelial cells were transfected with either an miRNA-200a-3p mimic or a scrambled control. The transfected cells were then exposed to the same GBS strain that was used in the NHP experiments (COH-1). To analyze the effect of the miRNA-200a-3p mimic on EMT, we quantified E-cadherin levels on the amniotic epithelial cells using flow cytometry. GBS significantly reduced E-cadherin expression on amniotic epithelial cells, as expected during the course of EMT (P < .05; Supplementary Figure 2); Further, we observed that loss of E-cadherin was abrogated in cells transfected with an miRNA-200a-3p mimic and not in the controls. These data demonstrate the importance of the miRNA-200a-3p in the pathway to EMT in human amniotic epithelial cells during GBS infection.

Discussion

After GBS infection, the NHP chorioamniotic membranes underwent numerous changes characteristic of EMT including miRNA dysregulation, an increase in ZEB2 transcription, and a loss of epithelial and gain of mesenchymal markers. This was accompanied by changes in collagen mRNA expression that can be expected to weaken the placental chorioamniotic membranes [16]. Additionally, in women with pPROM, we also quantitated the loss of desmosomes, which are ultrastructural changes linked to EMT. Finally, we provided a proof of principle that miRNA-200a-3p can prevent EMT-associated changes in human amniotic epithelial cells.

Mechanisms attributed to membrane weakening in pPROM have focused on the catabolic degradation of collagen mediated by matrix metalloproteinases, apoptosis, and oxidative stress [2–7]. Our findings implicate a cluster of miRNAs involved in chorioamniotic membrane weakening through EMT (conceptual model, Supplementary Figure 3). In EMT, epithelial cells rearrange their cytoskeletal architecture to lose their cell polarity and cell-cell adhesions, while gaining migratory and invasive properties associated with metastasis, embryogenesis, and wound healing [35]. Recently, amnion cells have been demonstrated to have the ability to undergo EMT, but our results highlight this pathway in vivo in a well-controlled NHP model of infection [17, 18].

Downregulation of members of the miR-200 family after choriodecidual infection is a key finding in this study; miR-200 expression determines the epithelial phenotype and prevents transition to the mesenchymal state [26, 38]. The miR-200 family (eg, miR-200a and miR-200b) and miR-205 suppress EMT by directly targeting and downregulating ZEB1 and ZEB2 via binding sites located within their 3′ untranslated regions [39]. Downregulation of the miR-200 family and miR-205 releases the repression of ZEB2 and leads to induction of EMT [40]. ZEB2 directly induces EMT through transcriptional silencing of E-cadherin, which is a glycoprotein important for cell-cell adhesions [36]. Although overall quantity of CDH1 mRNA was not significantly different, there was a significant correlation between lower levels of CDH1 mRNA and higher amniotic fluid IL-6 and IL-8 levels. Further analysis of common protein markers of EMT in the chorioamnion was confounded by loss of amniotic epithelial cells in the infection cases, which biased our results towards the null hypothesis. Despite this limitation, we identified significant correlations between inflammation and an immunostaining profile of decreased E-cadherin and increased N-cadherin/vimentin.

Induction of early regulators of EMT have been demonstrated in many experimental models of development or cancer without the hallmarks of complete transition, also called partial EMT [41]. Variable expression of protein markers associated with EMT in the amniotic epithelium may have been related to several factors: (1) tissue injury and loss of amniotic epithelial cells completing the transition, or (2) the temporal nature of EMT and ending our experiments too early in the process to observe cadherin switching. If loss of amniotic epithelial cells was due to completion of the transition, our analyses would be biased towards the null hypothesis because we could only measure markers in the remaining cells yet to complete EMT. Morphological changes associated with EMT have also been reported to precede cadherin switching, suggesting that increased motility and loss of amniotic epithelial cells may occur before a loss of E-cadherin, which is consistent with our ultrastructure findings [42].

Previously, a qualitative description of desmosome loss in the amniotic epithelium of women with pPROM was observed in the context of an investigation of apoptotic pathways [43]. In this study, we quantified and statistically analyzed the number of desmosomes in women with pPROM, finding a highly significant loss. Desmosome loss is a classic finding of EMT and necessary to develop a migratory phenotype. A loss of desmosomes suggests that the epithelial plasticity in the amniotic epithelium is more likely to be driven by EMT than a process of wound repair. We previously found a significant reduction in gene expression of desmosomal proteins (desmoplakin and desmoglein) [16], but these experiments are the first to demonstrate a statistically significant loss in human tissues. In wound healing, epithelial cells maintain their cell-cell junctions [44], in contrast to the loss of desmosomes in our study, which is more consistent with EMT.

Our experimental model of choriodecidual infection allows us to examine the early pathways contributing to pPROM. EMT may represent a unifying mechanism for several alternative pathways to pPROM beyond infectious etiologies. Oxidative stress is a pathway associated with pPROM [45] in which reactive oxygen species are generated that can damage tissues and the collagen network; reactive oxygen species are implicated in promoting EMT and particularly the first steps of cell-cell dissociation [46]. Placental abruption is another known risk factor for pPROM; thrombin production during the coagulation cascade is known to weaken the amnion, but the mechanism is unknown [47]. Interestingly, thrombin induces EMT in multiple cells types [48–50]. Prior identification of a possible unifying mechanism for pPROM was hindered because the human chorioamniotic membranes are often too damaged at the time of delivery to allow for scientific investigation of early precursors. Our ability to control key experimental variables and obtain tissues before extensive inflammatory injury was a major study strength. Another study strength is our unique NHP model that shares many key features with human pregnancy, which differ in other animal models of preterm birth (eg, murine and sheep). There are several study limitations. The main study limitation is the modest sample size, which was necessary for ethical reasons, conservation of NHP, and expense of the studies. Also, while there was no predefined criteria on animal assignment to either the GBS or saline groups, animals were not randomly allocated in part due to availability of personnel for processing bacterial (GBS) samples at the time/gestational age when the experiment needed to be performed. We also acknowledge that our results with GBS may not be broadly applicable to other pathogens.

In summary, our study provides the first in vivo evidence of miRNA induction of EMT regulators in the chorioamniotic membranes following GBS infection. The study of miRNAs in pregnancy holds great promise for the development of novel biomarkers for placental growth and adverse perinatal outcomes, as well as discovery of the earliest triggers of changes in placental gene and protein expression that can impact pregnancy outcomes [11]. Our study identifies EMT as a new mechanism of infection-associated placental chorioamniotic membrane weakening.

Supplementary Data

Supplementary materials are available at The Journal of Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.

Notes

Acknowledgments. We gratefully acknowledge the technical assistance of Jan Hamanishi and Gina Heidel in preparing the figures and Dr Raj Kapur for reviewing membrane histology; Dr Craig Rubens and Dr Michael Gravett who helped to obtain initial grant funding and contributed to study design of the original NHP experiments; and Dr William G. Carter for antibodies.

Disclaimer. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health (NIH) or other funders. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Financial support. This work was supported by the Eunice Kennedy Shriver National Institute of Child Health and Human Development, the National Institute of Allergy and Infectious Diseases and the National Center for Research Resources, NIH (grant numbers AI133976 and AI145890 to L. R. and K. A. W., U54HD083091, and P30EY01730); and the Office of Research Infrastructure Programs, NIH (grant number P51OD010425 through the Washington National Primate Research Center).

Potential conflicts of interest. All authors: No reported conflicts of interest. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.

References

Author notes