Single‐cell transcriptome analysis reveals defective decidua stromal niche attributes to recurrent spontaneous abortion

Abstract Objectives Successful pregnancy involves the homeostasis between maternal decidua and fetoplacental units, whose disruption contributes to compromised pregnancy outcomes, including recurrent spontaneous abortion (RSA). The role of cell heterogeneity of maternal decidua in RSA is yet to be illustrated. Materials and methods A total of 66,078 single cells from decidua samples isolated from patients with RSA and healthy controls were analysed by unbiased single‐cell RNA sequencing (scRNA‐seq). Results Our scRNA‐seq results revealed that stromal cells are the most abundant cell type in decidua during early pregnancy. RSA samples are accompanied by aberrant decidualization and obviously obstructed communication between stromal cells and other cell types, such as abnormal activation of macrophages and NK cells. In addition, the over‐activated TNF superfamily member 12 (TNFSF12, TWEAK) and FASLG in RSA are closely related to stromal cell demise and pregnancy failure. Conclusions Our research reveals that the cell composition and communications in normal and RSA decidua at early pregnancy and provides insightful information for the pathology of RSA and will pave the way for pregnancy loss prevention.


| INTRODUC TI ON
Recurrent spontaneous abortion (RSA), experienced by 5% of fertile couples, is a pregnancy complication with an elusive underlying mechanism due to multiple reasons, including genetic predisposition, inflammation, immunology imbalance, maternal ageing, hormonal imbalance and environmental stresses. [1][2][3] The logistical and ethical difficulties in accessing meaningful samples are the major limitations for RSA studies attributing to intangible pathogenesis.
The availability of sufficient samples from diverse groups with varied backgrounds will greatly advance the mechanistic study of RSA.
Successful pregnancy involves the participation of maternal decidua and fetoplacental units with many other cell types, as revealed by single-cell sequencing. 4,5 While the role of decidua in pregnancy maintenance is largely neglected, there is an increasing interest in the role of decidua in orchestrating the homeostatic balance between the mother and foetus, and studies suggest that the decidua is a critical regulator in the cross-talk between foetus and decidua by secreting cytokines, morphogens and signalling molecular to safeguard foetal development. [6][7][8][9] Defective decidualization can lead to aberrations in placentation and adverse pregnancy outcomes. Failed decidualization is a vital contributor to downregulated cytotrophoblast invasion in severe pre-eclampsia. [10][11][12][13] However, the clear com- In this study, scRNA-seq is leveraged to unravel the cell heterogeneity of maternal decidua in normal and RSA groups in the first trimester. We observe that stromal cells are the most abundant cell type and that disrupted decidualization is also present in RSA

samples. The communication between stromal cells and other cell
types is obstructed in RSA samples. Besides, the aberrant activation of macrophages and NK cells is also observed in RSA samples.
Collectively, our study elucidates the role of decidualized stromal cells, macrophages and NK cells in normal pregnancy as well as its role in RSA. The resulting findings reveal a potential mechanism of RSA pathogenesis and provide insightful information for the development of RSA and pave the way for RSA prevention.

| Sample collection
All tissue samples used for this study were obtained with written informed consent from all participants. The study was approved by Patients with two or more previous spontaneous unexplained abortions, normal karyotype of parents and abortus, and absence of uterine malformation, endocrine, metabolic, autoimmune diseases or infection were enrolled in the RSA group (6 for single-cell RNA sequencing, 19 for quantitative reverse-transcriptase polymerase chain reaction (qRT-PCR)). Women who underwent elective termination of normal pregnancies without a history of miscarriages were eligible as healthy controls (5 for single-cell RNA sequencing, 20 for qRT-PCR). The demographic characteristics of participants in the RSA group and the normal group used for single-cell RNA sequencing are displayed in Table S1. Samples of the RSA group were collected from women with RSA by ultrasound-guided curettage immediately after the diagnosis of missed abortion. Decidua was identified macroscopically, washed in phosphate-buffered saline to remove excess blood, and used for the single-cell isolation as described below.

| Isolation of decidual cells
Primary decidual cells were isolated as previously described from freshly collected decidual tissue. 18 Briefly, decidua was rinsed several times with PBS until removal of obvious blood clots; then was cut into small pieces, which were digested with collagenase Type IV (0.5 mg/ml, C5138; Sigma-Aldrich) and DNase I (0.1 mg/ml, DN25; Sigma-Aldrich) for 30 min. The released decidual cells were filtered through 70 and 40 µm mesh sieves, centrifuged, and resuspended in 5 ml of red blood cell lysis buffer (Invitrogen, 00-4300) for 10 min to exclude any remaining red blood cells. Finally, the pelleted decidual cells were resuspended in PBS and then were used for single-cell 3′-cDNA library preparation followed by 10× Genomics Chromium Single-Cell 3′ reagent Kits protocol.

| Single-cell RNA-seq data processing
Single-cell libraries were sequenced on Illumina HiSeq X Ten instruments using 150 nt paired-end sequencing. Reads were processed using the Cell Ranger 4.0.0 pipeline with the default and recommended parameters. FASTQs generated from the Illumina sequencing output were aligned to the human reference genome (GRCh38) using the STAR algorithm. Next, Gene-Barcode matrices were generated for each sample by counting unique molecular identifiers (UMIs) and filtering non-cell associated barcodes. Finally, the genebarcode matrix containing the barcoded cells and gene expression counts were generated.
This output was then imported into the Seurat (v3.0) R toolkit for quality control and downstream analysis of our single-cell RNA-seq data. All functions were run with default parameters unless specified otherwise. Low-quality cells (<500 genes/cell) were excluded for each sample of all six RSA patients and five healthy controls. After filtering, a total of 66,078 cells were left for the following analysis.
Finally, a filtered gene-barcode matrix of all samples was integrated with Seurat v.3 to remove batch effects across different samples using the SCT method.

| Identification of cell types and subtypes by Uniform Manifold Approximation and Projection (UMAP)
The Seurat package implemented in R was applied to identify major cell types. Highly variable genes were generated and used to perform PCA. Significant principal components were determined using JackStraw analysis and visualization of heatmaps focussing on PCs 1 to 30. PCs 1 to 30 were used for graph-based clustering (at res = 0.6) to identify distinct groups of cells. We characterized the identities of cell types of these groups based on annotation results of singleR.

| Cluster marker identification
The cluster-specific marker genes were identified by running the FindAllMarkers function in the Seurat package to the normalized gene expression data. To identify differentially expressed genes between two clusters, we used the 'findmarkers' function. We used the R package clusterProfiler to perform biological process enrichment analysis with the top 20 differentially expressed genes in each cluster or subset.

| Constructing cell trajectories
Trajectory analysis was performed separately for the cluster DS1, DS2 and DS3 cells using Monocle 2 (version 2.6.4). We then conducted differential gene expression analysis of the studied cells using the differentialGeneTest function to identify significant genes (BH-corrected p < 0.05), and cell ordering was performed on these genes in an unsupervised fashion. Trajectory construction was then performed after dimensionality reduction and cell ordering with default parameters.

| Cell-cell communication analysis
To investigate potential interactions across different cell types in

| CellChat
To further analyse and compare the intercellular communication

| SCENIC
Transcripts factor analysis was performed using SCENIC, following an automated SCENIC pipeline using the expression matrices (http:// scenic.aerts lab.org). The key regulators of DS, Macro, NK and Endo subsets in normal samples were identified.

| RNA extraction and quantitative reversetranscriptase polymerase chain reaction analyses
Total RNA was extracted from decidua tissues obtained from RSA (n = 19) and normal (n = 20) groups using TRIzol (Takara, Japan), according to the manufacturer's protocol. Extracted RNA was diluted with DEPC-treated water and quantified using a ratio of measurements at 260 and 280 nm (Nanodrop; Thermofisher), then were reversed to cDNA using a PrimeScript™ RT reagent Kit with gDNA Eraser (Takara, China) according to the manufacturer's instructions.
The real-time PCR system used RR420A TB Green™ Premix Ex Taq™ (Tli RNaseH Plus) (Takara, Japan), with primer sequences provided in Table S4. All real-time PCR (qPCR) reactions were carried out using the Q3 Real-Time System (Applied BioSystems). The results were normalized to the expression of SDHA. Relative fold change in gene expression were calculated using the 2 −ΔΔCt method, normalized with respective controls.

| IHC analysis and Western blot analysis
For immunohistochemistry (IHC), tissue samples were fixed with 10% buffered formalin at room temperature for 6 h. Samples were embedded in paraffin. After routine rehydration and antigen retrieval, 5μm paraffin sections were stained for HE or with the primary antibodies (Table S5). Macrophages and dNKs were identified by anti-CD68 and anti-CD56 immunostaining, respectively. The sections were further incubated with HRP-conjugated secondary antibodies and visualized with a DAB solution containing 0.03% H 2 O 2 .
Negative/normal controls were performed by replacing the specific antibody with rabbit or mouse IgG. Whole lysates from tissues were extracted with RIPA buffer containing protease inhibitor cocktail (Sigma-Aldrich). Protein concentrations were determined using the BCA™ Protein Assay Kit (Pierce). Western blotting was performed as described previously. Antibodies used for western blotting include Fn14 and FAS (Table S6).

| In-situ hybridization and Immunofluorescence staining
Frozen sections from normal decidua were used for In-situ hybridization and immunofluorescence (IF) staining. Digoxigenin (DIG)-labelled IGFBP1 probes were generated according to the manufacturer's protocol (Roche). In-situ hybridization with DIG-labelled probes was performed as described. 19 Antibodies used for IF staining include PR and HAND2 (Table S5).

| Single-cell atlas in normal and RSA decidua
Cumulative evidence suggests that the decidua plays a critical role in pregnancy haemostasis by regulating programmed decidual senescence and local inflammation. 20 Furthermore, the endothelial cells in the decidua act as sentinels and are the first line of defence to combat infection. 21 Numerous immune cells are indispensable for immune balance in the maternal-foetus interface during pregnancy. 22 However, the detailed mechanism of each cell type in the decidua during pregnancy is largely unknown. To decipher the heterogeneity of the decidua, single-cell RNA-seq was performed in first-trimester (5-8 weeks of gestation, 6.20 ± 1.09 weeks) decidua from five normal samples using 10× Genomics, as well as six RSA (5-8 weeks of gestation, 6.83 ± 0.75 weeks) samples who experienced more than two unexplained miscarriages without foetal chromosomal abnormality as confirmed by karyotype or array comparative genomic hybridization (CGH) analysis ( Figure 1A and Figure S1A, Table S1).
After quality filtering and batch correction, graph-based clustering was performed to group a total of 66,078 cells composed of 31,140 cells from normal decidua and 34,938 cells from RSA decidua, respectively, according to their gene expression profile (Table S2) Table S3). Among them, decidual stromal cells were the most abundant cells, followed by NK cells, macrophages, T cells and others ( Figure 1B). Similar cell composition was observed in normal and RSA decidua ( Figure S1B).

| The characterization of different stromal cells in decidua
After embryo implantation, the stromal cells undergo extensive decidualization to support the growth of the developing foetus.  Figure S3A,B). MME, known as CD10, is a wellrecognized human endometrium stromal cell marker. 23,24 The expression of CD10 was only observed in non-pregnant endometrium but dramatically decreased in the decidua of early gestation, 25 which is consistent with our results that CD10 is mainly reduced in DS3.
PLA2G2A, a member of the phospholipase A2 family (PLA2) producing biologically active lipid mediators such as lysophosphatidic acid F I G U R E 1 Overview of the 66,078 single cells from normal or RSA decidua. (A) Summary of the sample origins and analysis workflow. The decidua tissue was collected and processed into a single-cell suspension and single-cell RNA sequencing was performed using the 10× Genomics platform followed by bioinformatics analysis. (B) UMAP of the 66,078 cells with its sample type of origin (RSA or normal), the associated cell type and the proportion of each cell type in decidua samples. (C) Dotplot map showing the expression of classical cell typespecific marker genes in each cluster (LPA) and arachidonic acid (AA), is a critical enzyme for inflammatory processes and the regulation of the phospholipid metabolism. 26,27 Intracellular PLA2 was reported to restrict Wnt signalling during haemostasis in the small intestine. 28 This speculation was also supported by the expression of SFRP1 and SFRP4, critical inhibitors of Wnt signalling directed differentiation. 29,30 These reports suggested that DS1 stromal cells were less decidualized, but DS3 underwent extensive decidualization.
In decidualization, 3,5,3′-triiodothyronine (T3) is inactivated in decidualized stromal cells to facilitate decidualization. DIO2, a critical enzyme for producing active T3 from 3,5,3′,5′-tetraiodothyronine (T4), is highly expressed in sub-epithelial stromal cells before embryo implantation in mice; while its expression is dramatically decreased after the initiation of decidualization. 31 The downregulation of DIO2 in in-vitro decidualized endometrial stromal cells as examined by real-time PCR further corroborated this discovery ( Figure 2C and  Figure 2N). Most autophagy-associated genes were specifically expressed in FBs; whereas most oxidative phosphorylation genes were expressed in DSs. We also noticed that some histone modification enzymes were significantly enriched in FBs ( Figure 2N). Moreover, a few DS4 (246) cells remained in the normal decidua. Compared to DS1, DS2 and DS3 cells, DS4 cells expressed higher epithelial markers, such as PAEP and KRT18, and lower mesenchymal markers, such as ZEB1 and TWIST2 ( Figure S3G).

| Defective stromal decidualization contributes to RSA
As the normal decidual stromal cells were delineated, we explored the difference between normal and RSA decidua. Notably, the number of DS cells was significantly decreased in RSA decidua ( Figure 3A Figure 4A,B). In RSA patients, the trajectory of decidualization was largely compromised, and one trajectory branch had almost disappeared ( Figure 4A,B).
In the pseudotime heatmap, activating transcription factor 3 (ATF3) and MME were highly enriched in the middle; WNT6, WNT4, semaphorin 3B (SEMA3B) and IL1B were enriched in one end and cytochrome P450 family 1 subfamily B member 1 (CYP1B1) was enriched in the other end ( Figure 4C). The expression profile of ATF3 and MME showed that they were highly expressed in undifferentiated stromal cells ( Figure 4D,E), SEMA3B and WNT4 were expressed in the decidualized DS3 (Figure 4F,G). The branch analysis also showed that WNT4 and PLA2G2A were critical branch-relevant genes ( Figure 4H). Some cells were decidualized along with high expression of WNT4, while some cells highly expressed PLA2G2A (Figure 4I,J). interesting that the overall cell-cell interactions in RSA decidua were significantly increased ( Figure 5A). For each cell type, the signallings sent or received were grossly similar between normal and RSA ( Figure 5B,C). As the connection abundance between DSs and the endothelium was listed at the top, the detailed connections between these cell types were enumerated ( Figure 5D). The WNTs communications were mainly presented between DSs and the endothelium.

| Cell connection in the decidual niche
A novel mutual connection of SEMA3A (Semaphorin 3A)-NRP1 (Neuropilin 1) was identified between the endothelium and DS cells ( Figure 5D). The major function of SEMA3A was reported to inhibit angiogenesis of endothelial cells via neuropilins and plexins. 34 As the expression of SEMA3A was observed in the proliferative phase of the endometrium, 35 it was possible that the developing decidua resisted new blood vessel formation by SEMA3A secretion.

| Cell communications in the decidual niche
To globally distinguish the sending signalling and receiving signalling in different decidual cell types, CellChat was applied to investigate the cell-cell cross-talk. 36 Figure S6A). WNT4, WNT5A and WNT5B, which were the major players of ncWNT and WNT, were highly expressed in DSs. Their corresponding receptors LDL receptor-related protein 6 (LRP6), frizzled class receptor 4 (FZD4) and FZD6 were primarily expressed in different endothelium, with some LRP6 also expressed in DSs ( Figure 6D). We also provided clear evidence that the SEMA3 generated by DSs, especially DS3, targeted endothelium and DSs ( Figure 6E,F). This data suggested that comprehensive cross-talks between DSs and the endothelium ensure normal foetal development.
For the receiving signalling, DS1, DS2, DS3, FB1, FB2 and PV shared a very similar signalling pattern, including EDN (Endothelin), EGF (Epidermal growth factor), PDGF and others, partially ascribing to a similar origin of these cell types ( Figure 6G). We showed that Endo3 produced the most EDN targeting DS1, DS2, DS3 and FBs ( Figure 6H and Figure S6B). This data further confirmed the closed interaction between stromal cells and endothelial cells. At the early stage of pregnancy, the stromal cells still underwent proliferation and growth, but the source of the growth signalling remains unclear. Here we revealed that multiple growth factors released by macrophages, including HB-EGF, AREG (Amphiregulin) and EREG (Epiregulin) target EGFRs on DSs ( Figure 6I and Figure S6C,D).
Interestingly, PV cells expressed most stromal cell markers, such as PGR, HADN2 and WT1. Our data also revealed that these perivascular cells intimately interacted with stromal cells by PDGF-PDGFRs signalling pathways ( Figure S6E,F). Overall, these data suggested decidual stromal cells functioned as a central hub with extensive communications with other cell types to guide foetal growth.  Figure S8A,B). Among the three non-proliferating macrophages, Mac1 was equipped with higher phagosome and proteoglycans activity, and Mac2 and Mac3 were featured with more robust antigen processing and presentation capability ( Figure S8C,D). The population of Mac1 was increased in RSA compared with normal decidua with a similar number of other macrophages ( Figure S8E). The molecular trajectory of macrophages in these two groups was also comparable ( Figure S8F). Functional enrichment based on the differentially expressed genes uncovered the TNFα and NFkB signalling pathways were significantly upregulated in RSA Macro1 ( Figure S8G,I).

| Derailed communications contributed to RSA
Moreover, phagosome and antigen processing and presentation pathways were enriched in RSA Macro3 upregulated DEGs ( Figure S8H,I).
Another signalling pathway that was varied was the FASLG (Fas ligand) signalling pathway. There was a weak interaction between NK cells and DS2, DS3, DS5 decidual in the normal decidua, which indicated these stromal cells would be eliminated by NK-derived cell death signal FASLG-FAS signalling pathway ( Figure 7E whether RXRs were also independent for decidual macrophage maintenance requires further investigation.

| DISCUSS ION
The role of decidua during early human pregnancy has been recognized recently. Still, the cellular and molecular mechanism and the detailed interaction between different cell types in the decidua remain elusive, ascribing to the ethical issue, the difficulty of accessibility of suitable tissue, and the lack of appropriate animal models. Our current study finds that there are different decidualized stromal cells in the human decidua, and each stromal cell poses its unique feature. The matured stroma cells are characterized by a profound secretory nature, including well-established IGFBP1 and PRL, which is encapsulated by many other studies. 38,39 Our investigation also found these matured decidualized stromal cells also communicate with other stromal cells, as well as endothelial cells and macrophages. PDGFA is a well-known stromal cell-specific secreting protein, which interacts with other cell types to maintain the stability of the microenvironment. 40 This secretory nature of these decidualized cells comes with increased energy consumption and an adaptive metabolism change.
Our study also unravels that the FBs, undecidualized stromal cells in the endometrium, expressed DIO2 and some autophagyassociated genes. These decidualization-resistant cells in the endometrium are probably utilized to prevent uncontrolled invasion of EVT and are also responsible for rebuilding the endometrium after parturition. Our observations suggest the stromal cells should undergo sequential and sufficient decidualization to support foetal growth. The number of decidualized stromal cells, especially DS1 and DS2, in RSA decidua is remarkably diminished, indicating insufficient support of DSs in foetal development. Meanwhile, we also noticed that the number of DS5 is dramatically increased in RSA decidua. It is possible that at the early stage of decidualization development, some stromal cells do not differentiate into DS1 and DS2 appropriately but to DS5. In addition, a recent study reported highly proliferative mesenchymal cells (hPMC) were found in mid-luteal human endometrium and hPMC may differentiate into dS3 cells in early pregnancy. 41 It is unclear at present whether defective DS3 cells in RSA decidua in early pregnancy are associated with the loss of hPMC in the mid-luteal endometrium, which is interesting and needs to be further studied. In summary, derailed stromal cell differentiation, contributing to abrupt decidual cell composition, creates unhealthy circumstances and ultimately leads to pregnancy loss.
The immune cells are dynamically changed during pregnancy.
Decidual NK cells, which was the largest population of decidual leukocytes at the maternal-foetal interface during early pregnancy, have abnormal properties, including subset composition, gene expression and function in patients with pregnancy loss. 42,43 Similar to Wang's study, we also found peri_NK (CD16+) and pNK (MKI67+) subsets in decidua tissues. Although the proportion of peri_NK and pNK was not increased in RSA samples according to our data, the more active signalling pathways of FASLG and TRAIL were exhibited in NK subsets of RSA groups. NK cell expression of FASL and TRAIL can contribute to apoptosis in their targets and NK killing. 44 Figure S4B). This intra-endothelial transition towards lymphatic fate is reported to be essential for spiral artery remodelling in both mice and rats mediated by the communication between NK cells and the endothelium. 48 Our results also present compelling evidence that this intra-endothelial transition is also conserved in humans.
The presence of pericytes is noticed for a long time with its function to regulate vasoconstriction. Here, we provide new evidence that pericyte-derived PDGF would also interact with stromal cells through its receptors. PDGFRα is reported to be broadly expressed in many mesenchymal cells and we prove that PDGFRα is highly expressed in human decidual stromal cells. The function of PDGF in adult tissue is not a major mitogenic factor but more prone to differentiation/migration. 49 Uterine stromal cells are derived from reproductive mesenchymal cells at the neonatal stage and differentiate to uterine stroma cells characterized with PR expression. 50,51 It is possible that the PDGF-PDGFR in pericytes (endothelium) -stroma is very important for the maintenance of haemostasis of stromal cells.

| CON CLUS ION
In summary, we present the decidual cell composition of the human maternal-foetal interface at single-cell resolution in both normal and RSA decidua, describing a detailed characterization of varied decidual cells, their developmental trajectories, and the communication between stromal cells and other cell populations. This study identifies novel molecular and cellular mechanisms involved with RSA development. The resulting findings will pave the way to improve the diagnosis and prevention of RSA.

CO N FLI C T O F I NTE R E S T
There are no competing financial or non-financial interests regarding this work.

DATA AVA I L A B I L I T Y S TAT E M E N T
The datasets generated and analysed in the study are available in the NCBI Gene Expression Omnibus (GEO) and Sequence Read Archive (SRA) (PRJNA672658) and can be accessed upon request. All custom scripts can be accessed upon request to the corresponding author.