Single-cell transcriptome profiling of the human endometrium of patients with recurrent implantation failure.

Successful implantation requires synchronized and coordinated crosstalk between the embryo and endometrium 1 . In recent decades, in vitro fertilization-embryo transfer (IVF-ET) has become an effective treatment for infertility with improvements in laboratory procedures and ovarian stimulation. However, it is estimated that approximately 10% of women receiving IVF treatment will experience recurrent implantation failure (RIF), which refers to failure to achieve a clinical pregnancy after transfer of at least four good-quality embryos in a minimum of three fresh or frozen cycles in a woman < 40 years of age 2 . Excluding structural and chromosomal abnormalities (e.g, abnormal uterine cavity, hydrosalpinx and abnormal karyotype), disorders of endometrial receptivity (i.e., progesterone resistance, shifted window of receptivity, decreased mucin 1 (MUC1) and integrin expression, and immunologic disturbances) have been considered important contributors to unexplained RIF 3 , 4 . However, a systematic characterization of endometrial dysfunction in unexplained RIF has not yet been fully revealed in the window of implantation (WOI). Unlike other tissues, the human endometrium undergoes periodic variations in menstruation, menstrual repair, proliferation, and secretory differentiation, which are controlled by a sequential, and sophisticated timed interplay of female sex hormones during the menstrual cycle 5 . Endometrial receptivity is defined as “the period of endometrial maturation during which the trophectoderm of the blastocyst can attach to the endometrial epithelial cells and subsequently invade the endometrial stroma and vasculature” 6 . The establishment of endometrial receptivity is primarily coordinated by estrogen and progesterone, leading to dramatic functional changes in all endometrial cell types, including the stroma, glandular and luminal epithelium, resident immune cells, and endothelium 6 . There is an abundance of molecular mediators involved in regulating endometrial receptivity, including adhesion molecules, cytokines, growth factors, and lipids (e.g., IGFBPs, PRL, HOXA10, WNT, and LIF) 7 . During the last few decades, microarray and RNA sequencing techniques in whole-tissue transcriptomic analysis have been translated into clinical practice to evaluate the endometrial receptivity and determine the WOI timing for IVF-ET 8 , 9 , however, the transcriptome characterization of WOI needs further study 10 , 11 . Recent studies have employed single-cell RNA sequencing (scRNA-seq) technology to investigate the cellular composition and intercellular communication events of the human endometrium during the menstrual cycle 5 , 12 . However, the composition changes (cell types, molecular profiles and cellular dialogue regulatory networks) of the endometrium of RIF patients during the WOI and a detailed understanding of cellular interactions in the endometrium that support endometrial receptivity and differentiation are largely unknown. Here, we profiled the endometrial cells present at the WOI timing in RIF patients and healthy controls using scRNA-seq, and provided a detailed molecular and cellular map of a healthy and RIF endometrium at the WOI.

The protocol for this study was approved by the Human Research Ethics Committees of Obstetrics and Gynecology Hospital of Fudan University (2019-103) and Shanghai Ninth People's Hospital Affiliated Shanghai JiaoTong University School of Medicine (SH9H-2020-TK6-1). Written informed consent was obtained from all participants. Human endometrial tissues were collected from women attending the Department of Assisted Reproduction, a dedicated research clinic at the Shanghai Ninth People's Hospital Affiliated Shanghai JiaoTong University School of Medicine. Surplus tissue from endometrial biopsies obtained for diagnostic purposes at the Department of Assisted Reproduction was used for this study. For scRNA-seq, endometrial biopsy was performed five days after ovulation (ultrasonic observation, equated to LH + 7, the WOI time) in a natural cycle. The endometrium from the RIF patient group (n = 6; age range, 32 - 35 years) defined as unsuccessful implantation following transfer of at least six morphologically good-quality embryos in three or more embryo transfer cycles, was collected. The endometrium of the control (Ctrl) group (n = 3, age range, 29 - 35 years) defined as previous fertility history but with mechanical obstruction of fallopian tube or infertility due to male factors, was also obtained. Patient characteristics are shown in Table 1 . Inclusion and exclusion criteria are listed in Table 2 . For immunofluorescence and flow cytometry analyses, the endometrium from the RIF group (n = 12; age range, 25 - 34 years) or control (n = 58; age range, 25 - 33 years) during proliferative, secretory or menstrual phases was obtained. All donors had regular menstrual cycling (6 - 7 days every 28 - 30 days). Women with the following conditions were excluded from tissue collection: recent contraception (intrauterine device and hormonal contraceptive use in the past three months), endocrine metabolic abnormalities (i.e., polycystic ovary syndrome, diabetes, insulin resistance, and hypothyroidism), genetic abnormalities, severe adenomyosis or endometriosis, severe hydrosalpinx, moderate to severe intrauterine adhesions, uterine malformations, recurrent miscarriage, thrombosis, autoimmune diseases, and body mass index (BMI) > 30. For in vitro trials, normal endometrial samples were collected from the Obstetrics and Gynecology Hospital of Fudan University, and taken from 58 patients (age range; 30 - 45 years) who underwent diagnostic curettage or hysterectomy for benign reasons (e.g., septate uterus) unrelated to endometrial dysfunction as healthy controls. These samples were evaluated by a histopathologist to identify the cyclic phase as the secretory phase and exclude endometrial pathology. The endometrial tissues were washed with ice-cold PBS to remove any remaining blood. The endometrial tissues were then sectioned into 1 mm 3 pieces on ice and digested with 1 mg/mL collagenase type IV (Sigma-Aldrich, USA) for 15-20 min at 37 °C with constant agitation. After digestion, the samples were sieved through a 70 µm cell strainer (Falcon, USA), and the cell suspensions were centrifuged at 400 × g for 7 min to collect all the cells. To remove the remaining erythrocytes, 15 mL of red blood cell lysis buffer (Beijing Solarbio Science & Technology Co., Ltd., Beijing, China) was added to the pellet for 15 min on ice. After washing with PBS containing 0.04% BSA, the cell pellets were re-suspended in PBS containing 0.04% BSA and re-filtered through a 35 μm cell strainer (Falcon, USA), and the filtrate was collected. Dissociated single cells were then stained with acridine orange/propidium iodide (AO/PI) for viability assessment using a Countstar Fluorescence Cell Analyzer. The proportion of living cells was great than 90%. The single-cell suspension was further enriched with a MACS Dead Cell Removal Kit (Miltenyi Biotec, Germany). The scRNA-Seq libraries were generated using the 10X Genomics Chromium Controller Instrument and Chromium Single Cell 3' V3.1 Reagent Kits (10X Genomics, Pleasanton, CA, USA). Briefly, cells were concentrated to 1000 cells/µL and approximately 8,000 cells were loaded into each channel to generate single-cell gel bead-in-emulsions (GEMs), which resulted in the expected mRNA barcoding of 6000 single-cells for each sample. After the RT step, the GEMs were broken and barcoded-cDNA was purified and amplified. The amplified barcoded cDNA was fragmented, A-tailed, ligated with adaptors and amplified using PCR. The final libraries were quantified using the Qubit High Sensitivity DNA assay (Thermo Fisher Scientific, USA), and the size distribution of the libraries was determined using a High-Sensitivity DNA chip on a Bioanalyzer 2200 (Agilent). All the libraries were sequenced using an Illumina sequencer (Illumina, San Diego, CA, USA) on a 150 bp paired-end run. Single-cell RNA-seq data analysis was performed by NovelBio Bio-Pharm Technology Co., Ltd. with the NovelBrain Cloud Analysis Platform ( https://singlecell.novelbrain.com/login ). We applied fastp 13 with default parameters to filter the adaptor sequence and removed low-quality reads to achieve clean data. Feature-barcode matrices were then obtained by aligning reads to the human genome (GRCh38 Ensemble: version 91) using CellRanger v3.1.0. We applied the down sample analysis among samples sequenced according to the mapped barcoded reads per cell of each sample and finally achieved the aggregated matrix. Cells containing over 200 expressed genes and mitochondrial UMI rates below 40% passed the cell quality filtering and mitochondria genes were removed from the expression table. Seurat package (version: 3.1.4, https://satijalab.org/seurat/ ) was used for cell normalization and regression based on the expression table according to the UMI counts of each sample and percentage of mitochondria rate to obtain scaled data. We used the cellranger aggr pipeline to combine multiple samples. The samples were down sampled according to the mapped barcoded reads per cell of each sample and the aggregated matrix was finally achieved. To remove batch effects among samples, we integrated the samples using canonical correlation analysis, which was implemented in the Seurat workflow. PCA was constructed based on the scaled data with top 2000 high variable genes and top 10 principals were used for tSNE and UMAP constructions. CCA analysis in the Seurat package was used to correct for batch effects among the samples. Utilizing the graph-based cluster method, we acquired the unsupervised cell cluster result based on the PCA top 10 principal components. We calculated the marker genes using the FindAllMarkers function with the Wilcoxon rank sum test algorithm under following criteria: 1) lnFC > 0.25; 2) P value 0.1. To identify the cell types in detail, clusters of similar cell type were selected for re-tSNE analysis, graph-based clustering and marker analysis. Single-cell Trajectory analysis was performed using Monocle2 ( http://cole-trapnell-lab.github.io/monocle-release ) DDR-Tree and default parameters. Before Monocle analysis, we selected marker genes from the Seurat clustering results and raw expression counts of the filtered cells. Based on the pseudo-time analysis, branch expression analysis modeling (BEAM Analysis) was applied for branch fate determined gene analysis. To enable a systematic analysis of cell-cell communication molecules, we applied cell communication analysis based on CellPhoneDB 14 , a public repository of ligands, receptors, and their interactions. Membrane and secreted and peripheral proteins of these clusters were annotated. Significant mean and cell communication significance (p-value < 0.05) were calculated based on the interaction and normalized cell matrix achieved by Seurat normalization. To characterize the relative activation of a given gene set such as pathway activation, “Angiogenesis” and “Fatty Acid Metabolism” as described before, we performed QuSAGE 15 (2.16.1) analysis. To identify differentially expressed genes among samples, the FindMarkers function with the Wilcoxon rank sum test algorithm was used under the following criteria: 1) lnFC > 0.25; 2) p value 0.1. Functional enrichment analysis was performed using GO enrichment analysis ( http://www.geneontology.org ), and each enriched ontology hierarchy (false discovery rate (FDR) < 0.05) was reported with two terms in the hierarchy: 1) the term with the highest significance value; and 2) the term with the highest specificity. Paraffin sections were technically supported by Wuhan Servicebio Technology Co., Ltd. (China). Endometrial tissues fixed in 4% paraformaldehyde were embedded in paraffin and sliced to thickness of 4 μm for immunofluorescence. Endometrial tissue sections were baked at 60 °C for 2 h, deparaffinized with dimethylbenzene, and rehydrated using ethanol series. Antigen retrieval was performed by boiling the tissue sections in 10 mM Tris-EDTA buffer (pH 9.0) (Beijing Solarbio Science & Technology Co., Ltd., China) for 20 min, followed by immediate cooling in cold water for 30 min. Tissue permeabilization was performed using 0.25% Triton X-100 in PBS for 5 min, followed by washing twice with 0.05% Triton X-100 in PBS for 5 min. Nonspecific binding was blocked with 5% BSA/0.05% Triton X-100/4% goat serum in PBS for 1 h at room temperature. Tissue sections were then incubated with primary antibodies overnight at 4 °C and secondary antibodies for 1 h at room temperature. The primary antibodies and dilution ratios were as follows: CD63 (1:100; no. ab1318, Abcam, USA), CCNL2 (1:200; no. PA5-62738; Thermo Fisher Scientific), KI67 (1:200; no. ab16667, Abcam), MMP14 (1:100; no. ab3644, Abcam), PGR (1:100; no. ab63605, Abcam), and RPL10 (1:100; no. PA5-101098, Thermo Fisher, USA). The secondary antibodies used and dilution ratios were as follows: donkey anti-rabbit antibody (1:500, no. ab150075, Abcam). All the sections were counterstained with DAPI (Thermo Fisher Scientific) and mounted with buffered glycerol. Images were visualized using fluorescent signals from different lasers and captured using an optical and epifluorescence microscope (BX53 Microscope, Olympus Corporation, Japan). Human antibodies for flow cytometry assays (all antibodies were purchased from BioLegend, CA, USA) were used for the measurement of cell markers, as listed in Table S1 . Isotype IgG antibody (5 μL separately) was used as control. Human Trustain FcX (BioLegend) was used to block Fc receptors prior to flow cytometry. Subsequently, the cells were washed twice and resuspended in PBS for flow cytometry. Samples were analyzed using a CytoFLEX flow cytometer (Beckman Coulter, Inc., USA) and data were analyzed using FlowJo (version 10.07 (FlowJo LLC, USA). The STRING database (available online: http://string-db.org ) was used for PPI network prediction. The endometrial tissues were digested and isolated as previous procedures 16 , 17 . After centrifugation, the supernatant of single cells was discarded, and the cells were resuspended in DMEM/F-12 containing 10% FBS (Gibco, Germany), plated in culture flasks, and incubated in a humidified incubator with 5% CO 2 at 37 °C. Primary fibroblast-like endometrial stromal cells (ESCs) were allowed to adhere for 20 min. The culture medium was replaced every 2-3 days. The human endometrial epithelial cell line (hEEC, WHELAB C1225) was provided by SHANGHAI WHELAB BIOSCIENCE LIMITED, and was cultured and resuspended in MEM containing 10% FBS and 1% penicillin/streptomycin. The hEECs and primary ESCs were seeded in a 24 well plate, adhered for 12 h, washed with PBS, and fixed with 4% paraformaldehyde. For characterization of hEECs, cells were then immunostained with the anti-PAX8 (1:200; no. 10336-1-AP, Proteintech, USA), anti-ER-α (1:200; no. ab32063, Abcam, USA), anti-EpCAM (1:800; no. 2929S, CST, USA) and anti-CK7 (1:100; no. ab185048, Abcam, USA) antibodies overnight at 4 °C and secondary antibodies (donkey anti-rabbit antibody (1:500, no. ab150075, Abcam) or donkey anti-mouse antibody (1:500, no. ab150105, Abcam) for 1 h at room temperature. Cells were then counter stained with 1 µg/mL DAPI for 10 min for nuclear labling, and visualized by a fluorescence microscope (BX53 Microscope, Olympus Corporation, Japan). Images were processed using ImageJ (National Institutes of Health, USA). The hEECs were treated with the vehicle (0.1% DMSO, Sigma, USA) or medroxyprogesterone acetate (MPA) (1 μM, Sigma USA) for 24 h, in vitro . The cell culture supernatant was collected to extract exosomes. The hEECs were treated with rh-IGF1 (2 ng/mL, Abcam, USA) for 48 h, and collected to determine the mRNA expression levels of ATG5 , ATG7 , BECN2 , MAP1LC3B , mTOR , and MUC1 using quantitative real-time polymerase chain reaction (qRT-PCR). Additionally, primary ESCs were treated with vehicle or palmitic acid (PA) (10 μM, Xi'an Kunchuang Co., Ltd., China) for 48h, in vitro . The hEECs were also treated with the vehicle or PA (10 μM) for 48h, or GW4869 (1 μM, MedChem Express, USA) for 24h, in vitro . These cells were then collected, and the expression of APOD, APOE, IL15 and CXCL12 was detected by qRT-PCR. The endometrium tissues were digested and isolated as a previous procedure 16 , 17 . Single cells were collected to isolate endometrial NK cells by MASC, a human NK cell isolation kit (130-092-657, Miltenyi Biotec, Germany) for in vitro experiments. NK cells were co-cultured with EECs pre-treated with GW4869 or the vehicle (0.1% DMSO, USA) for 24h, and then NK cells were collected and further analyzed by flow cytometry assays. Details can be found in Supplementary information . Total RNA from primary ESCs and hEECs was extracted by TRIzol regent (Invitrogen, Carlsbad, CA, USA). Subsequently, a NanoDrop spectrophotometer (NanoDrop Technologies; Thermo Fisher Scientific, MA, USA) was used to quantify the concentration and purity of RNA. The PrimeScript RT Reagent Kit (TaKaRa Biotechnology, Co., Ltd., Dalian, China) was used to hereversely transcribe total RNA to cDNA. Next, qRT-qPCR was performed using the SYBR Green PCR Master Mix (TaKaRa Biotechnology). The qRT-PCR primers used are listed in Table S2 . The target mRNA expressions were normalized to ACTB expression. All reactions were performed using an Applied Biosystems 7500 Real-Time PCR System (Thermo Fisher Scientific). The test results were analyzed using the 2 -ΔΔ Ct method. The continuous variable is shown as mean ± SEM for normally distributed data, and as median ± inter-quartile range (IQR) for non-normally distributed data. Continuous variables were analyzed using Student's t -test for normally distributed data or two-tailed Mann Whitney test for non-normally distributed data. All analyses were performed using SPSS 21.0 Statistical Package. P < 0.05 was considered to indicate a statistically significance.

To determine the full repertoire of cell types and gene expression programs present in the endometrium, we isolated cells from the endometrium of patients with RIF (n = 6) and healthy controls (n = 3) at the time of WOI (see the Materials and Methods section), and generated single-cell transcriptome libraries on the droplet-based 10X Genomics Chromium System ( Figure 1 A ). After computational quality control and integration of transcriptomes, we obtained a total of 60222 endometrial single-cell transcriptomes, preformed graph-based clustering of t -disturbed stochastic neighbor embedding ( t -SNE) and used cluster-specific marker genes to annotate the clusters ( Figure 1 A and Figure S1 ). Overall, all sequenced endometrial cells were assigned to four main classes: fibroblast-like cells (FIB, expressing HOXA10 , MME and DCN ; 52825 cells), epithelial cells (EC, expressing KRT18 , KRT8 and EPCAM ; 1648 cells), immune cells (IC, expressing PTPRC ; 4790 cells) and vascular cells (VASC, expressing CLDN5 , PECAM1 , and VWF ; 959 cells) ( Figure 1 A - B ). Although the exact mechanism is largely unknown, there is evidence of a large amount of immune cell infiltration in the human endometrium at the WOI, including NK cells, macrophages and T cells 18 . Here, we also observed a very rich population of immune cells, such as NK cells ( PTPRC + CD3E - NACM1 + NKG7 + ), monocyte/macrophage (Mo/Mφ, PTPRC + CD14 + TYROBP + ), CD4 + T cells ( PTPRC + CD3E - CD4 + CD8A - ), CD8 + T cells ( PTPRC + CD3E - CD4 - CD8A + ), B cells ( PTPRC + CD3E - CD19 + CD79A + ), mast cells ( PTPRC + HPGD + ), ILC ( PTPRC + IL4I1 + ), and T/NKp cells ( PTPRC + CD3E + NACM1 + NKG7 + MKI67 + ) ( Figure S2 A - C , Figure S3 ). Here, digestive conditions and filtration operations may limit EC cell acquisition. Notably, we observed no difference in the total cell number and proportion of FIB, EC, VASC, IC and the subpopulation of IC (NK, T, Mo /Mφ, B, Mast, ILC and T/NKp cells) in the endometrium between the Ctrl and RIF patients ( Figure 1 C ). The three main tissue compartments of the uterus support and regulate pregnancy, including the stroma, endometrial epithelium, and myometrium. As FIB is the most abundant cell type in the endometrium, we initially explored endometrial FIBs, and identified seven subset clusters of endometrial FIBs: TOP2A + MKI67 + CDC20 + endometrial fibroblast-like cells (FIBp), RPL10 + RPS10 + PTN + IGFBP2 + FIB with high levels of protein synthesis and secretion-related genes (FIB1), MMP14 + IGF2 + COL6A1 + IGFBP2 + FIB with high levels of tissue remodeling-related genes (FIB2), ACTA2 + RGS5 + myofibroblasts (MFC), and groups of RPL10 low CD74 + IL32 high FIB5, CCNL2 + MMP14 + TIMP2 low FIB3 and MT1G + TM4SF1 + FIB4 ( Figure 2 A , Figure S4 A - B ). However, further analysis showed that FIB4 and FIB5 existed only in a single tissue sample as small subsets ( Figure S4 C ). Considering the key role of FIB in embryo implantation 19 , our subsequent analysis focused primarily on four groups of FIBs (FIBp, FIB1, FIB2 and FIB3). Among these, FIBp had high levels of cell cycle and proliferation-related genes (e.g., CDC20 , PTTG1 , PCNA and MKI67 ), suggesting a high proliferative ability ( Figure 2 B and Figure S5 ). Additionally, this subset of FIBs expressed a certain level of ribosomal protein-related genes (e.g., RPL10 , and RPS11 ) and HLA class I histocompatibility antigens (e.g., HLA-A , HLA-B , HLA-C ). Maternal HLA-C has been reported to inhibit the cytotoxicity of maternal NK cells, thereby establishing maternal-fetal immune tolerance 20 , 21 . Notably, FIB1 had strong protein synthesis and secretion capacities, and high levels of endometrial receptivity-related molecules ( Figure 2 B - C , Figure S5 , and Figure S6 A ), which were characterized by high levels of ribosome protein (e.g., RPL10 , RPS11 ), membrane glycoprotein (e.g., CD81 ), proteoglycan ( DCN ), calcium-binding proteins ( S100A4 , S100A6 , and ANAX2 ), HLAs, apolipoprotein ( APOD , and APOE ), cytokine ( PTN , and IL15 ) and insulin like growth factor binding protein (IGFBP)-coded genes. Additionally, FIB2 highly expressed cell adhesion (e.g., ICAM , GJA1 , and CD44 ) and extracellular matrix (ECM) remodeling (e.g., MMP2 , and MMP14 )-related genes, as well as IGF2 and IGFBP . Several adhesion-promoting molecules, such as CD44 22 , support the presence of cross-talk between blastocysts and the endometrial epithelium/stroma during human embryonic implantation. The ECM remodeling is essential for successful implantation and placentation and multiple MMPs (MMP14 and ADAM10) and their substrates are involved in this process 23 . Therefore, FIB2 had strong adhesion and tissue remodeling capacities, and high endometrial receptivity ( Figure 2 B - C , Figure S5 , and Figure S6 A ). In contrast to decidualized stromal cells 14 , some classical marker genes (e.g., PRL ) of endometrial receptivity were expressed in all endometrial FIB at low levels ( Figure 2 C ), which partly echoed the previous report 5 . Compared with FIB2, the genes for cell adhesion, MMPs and endometrial receptivity in FIB3 were decreased ( Figure 2 B - C , Figure S5 , and Figure S6 A ). Contrastingly, FIB3 highly expressed ECM organization-related genes, as well as a certain level of cell cycle and proliferation molecules. Further analysis showed decreased percentages of FIB1 and ratio of FIB2 to FIB3, and increased FIBp in RIF patients, based on the average of samples ( Figure 2 D, and Figure S6 B ). Based on receptor-ligand pairs, potential interactions between these four clusters of FIBs and other cells (EC, VASC, NK, macrophage and T cells) in the endometrium were predicted ( Figure 2 E ). Particularly, IL15 and HLA-E derived FIB1 were predicted to promote the proliferation and decrease the cytotoxicity of endometrial NK cells; the interaction of ICAM1/ITGB1, LAMC1/a2b1 complex, FN1/aVb1 complex, and COL3A1/COL6A3/a1b1 complex possibly contributed to the cell adhesion between FIB2 and other FIBs or NK cells. The IGF2 and BMP1 expressed by FIB2 and FIB3 were involved in the differentiation regulation of EC and FIB, which was required for endometrial receptivity and implantation 24 ; notably, VEGFA and VEGFB produced by FIB3 and FIBp resulted in angiogenesis of the endometrium by binding to receptors ( Figure 2 F ). Previous reports have shown that the gene expression profile of total endometrial tissues displays high cellular proliferation, DNA synthesis, angiogenesis and vasculogenesis during the proliferative phase of the menstrual cycle 25 . Subsequently, cell proliferation is inhibited, but the transformation and differentiation of the endometrium begins to occur, and the gene expression of metabolism, cell differentiation and communication, innate immune response, adhesion, and ECM degradation is up-regulated during the secretory phase, as well as in the glandular section 26 . To further explore the relationship between the cell transformation and differentiation of these four FIBs, a standard pseudotime analysis was performed, and a new trajectory for FIBp, FIB1, FIB2 and FIB3 was constructed ( Figure 3 A ). Notably, we observed a notable discontinuity among the four groups of FIBs. As shown, FIBp was the starting point, which went through FIB1 and FIB2, and the final endpoint was FIB3 ( Figure 3 A - B ). Additionally, t -SNE with RNA velocity also demonstrated the evolution from FIBp to FIB3 ( Figure 3 C ). The expression of cell cycle, division and DNA replication genes (e.g., NCAPD2 , MKI67 , and CDC20 ) was markedly reduced from the starting point (FIBp) to the other FIBs branches in the trajectory ( Figure 3 D ). Gene expression (e.g, RPL10 , RPS13 , and SLC25A6 ) in ribosome biogenesis, protein export, and mRNA metabolic process was up-regulated early in FIB1 differentiation and down-regulated in cell differentiating into both FIB2 and FIB3 ( Figure 3 D ). Importantly, gene expression (e.g, COL6A1 , FBLN1 , and FN1 ) in focal adhesion, embryo implantation, ECM remodeling, blood vessel remodeling peaked during FIB2 differentiation, and a decreasing trend was then observed in the FIB3 differentiation, which was characterized by high expression of histone methylation, positive regulation of cell killing, necrotic cell death, chromatin remodeling and lysine degradation-related genes (e.g., JMJD1C , KMT2C , ELN , and REV3L ) ( Figure 3 D , Figure S7 ). To confirm the single-cell trajectory, the expression of KI67, PRL10, MMP14 and CCNL12 was detected by immunofluorescence staining. As expected, MKI67 + FIBp was mainly localized in the endometrium during the proliferative phase of menstruation. During the secretory phase, PRL10 + FIB1 and MMP14 + FIB2 were enriched in the endometrium, while CCNL2 + MMP14 + FIB3 rapidly accumulated during the menstrual period ( Figure 3 E ). These data suggest that enrichment of FIB1 and FIB2 with high protein biogenesis, tissue remodeling and good endometrial receptivity properties at the WOI is very important for embryo implantation. Further analysis of the volcano plot showed that the gene expression of endometrial receptivity (e.g, IGFBP3 , S110A3 , APOD , and DCN ), immunoregulation (e.g, CXCL12 , IL15 , and HLA-C ) and protein biosynthesis (e.g., RPL22 , and RPS27 ) was markedly decreased in these four FIBs from RIF patients ( Figure 3 F ). Particularly, the rose diagram showed the genes in the pathways of cell proliferation, embryo implantation, ECM remodeling, senescence, sex hormone signaling and immune pathway was significantly down-regulated in these four FIBs from RIF patients ( Figure 3 G , Table S1 ). Therefore, the results suggest that the decreased levels of endometrial receptivity and immune regulation-related genes mainly expressed by FIB1 and FIB2 may lead to poor endometrial receptivity in RIF patients. As the site of blastocyst adhesion, the epithelium is perceived as a crucial site for uterine receptivity, which transmits signals to other compartments 25 . Markers that distinguished the different endometrial epithelial cell populations identified four clusters (referred to as Ciliated EC, EC1, EC2 and EC3): FOXJ1 + IGFBP7 + Ciliated endometrial epithelial cell (Ciliated EC, as reported recently 5 ), ALCAM + CD63 + ESR1 + PRDX6 + EC1, ESR1 + DUOXA1 + PRDX6 - LDHB - EC2, and ALCAM - CD63 - IGF1 + PRDX6 + LDHB + EC3 ( Figure 4 A , Figure S8 A ). Functional enrichment analysis, FOXJ1 + Ciliated EC highly expressed genes involved in epithelial and ciliated cell development, cilium movement and beat frequency ( Figure 4 B ). More importantly, genes involved in multiple biological pathways (e.g., cell adhesion, immune response and leukocyte migration, ECM organization, cytokine-mediated signaling pathway, decidualization, autophagy and embryo implantation) were enriched in EC1, possibly regulating endometrial receptivity during implantation ( Figure 4 B , Figure S8 B ). Genes associated with angiogenesis, blastocyst and embryo development, and cellular adhesion, were enriched in EC2. Contrastingly, the most pronounced functional feature of EC3 was a high proportion of cell senescence, replicative senescence and response to oxidative stress ( Figure 4 B ). Interestingly, EC1 displayed an activated progesterone receptor signaling (e.g, high levels of PGR , PGRMC1 / PGRMC2 , FKBP4 / FKBP5 , HSPA1A / HSPA1B , and NCOR1 / NCOR2 ), contributing to the strong exosome production, transport and secretion capacities ( CD63 + CD9 + CD81 + TSG101 + MUC1 + VPS28 + TM4SF1 + STX18 + ) possibly by the up-regulation of cell autophagy ( MAP1LC3B + ATG5 + ATG12 + ) ( Figure 4 C - D ). In contrast, other epithelial cells were less capable of secretion properties, especially EC2 and EC3. Notably, a close interactive dialogue between endometrial stromal cells and epithelial cells was observed, particularly in the four populations of FIBs and EC1 ( Figure 5 A ). To explore the potential functions of EC cells, we employed the hEEC cell line. We have verified by in vitro experiments that hEECs express markers characteristic of epithelial cells 27 ( Figure S9 ). Further analysis showed that activated autophagy of EC1 may be induced by IGF-1, which is produced by FIB2 and FIB3, in a mTOR-independent manner ( Figure 2 F , Figure 5 B , and Figure S10 A ). Further, MUC 1 (a highly glycosylated polymorphic mucin-like protein) is more abundant in fertile women than in infertile women, and has also been shown to be progesterone- rather than estrogen-dependent in baboons, serving as a marker of the pre-implantation phase 28 . Here, we observed that MUC1 + EC1 with many exosomes had better endometrial receptivity (expressed high levels of IGFBP2 , IL-1 , and IL6 ; Figure S8 ), possibly by interacting with IGF1 signaling ( Figure 5 C ). Stimulation with MPA led to increased exosome production, in vitro ( Figure S10 B - C ). Extracellular vesicles (EVs) were recently shown to play a role in embryo-mother cross communication, even at preconception from gamete maturation to implantation and throughout pregnancy 29 , 30 . However, further analysis and verification showed that the percentage of CD63 + EC1 with high response to progesterone was decreased significantly in RIF patients ( Figure 5 D - E ). Moreover, many candidate genes of endometrial receptivity (such as LIF , IL6ST , and ITGA3 ), senescence (such as COMP , SOD2 , and EDNRB ), exosome (such as CD9 , and VSP28 ) and autophagy (such as ATG9B , and APOL1 ) were downregulated in ECs of RIF patients ( Figure 5 F ). Endometrial NK cells play an important role in the decidualization, angiogenesis and embryo implantation 31 , 32 . We then identified four main NK cellular subsets (NK1, NK2, NK3 and NK4) ( Figure 6 A ). The NK1 cells expressed ITGA1 (also known as CD49a , a tissue-resident marker), CD103 and ITGB2 , but not CXCR4 (an important chemokine receptor for NK cell migration and recruitment), CD39 , KIR2DL1 , CD160 , or FCGR3A (also known as CD16 ) ( Figure 6 B , and Figure 6 D ). This subset of resident NK cells should be involved in tissue remolding by ECM-receptor interaction ( Figure S11 A ), and may be further differentiated to CD103 - ITGB2 + CD39 - KIR2DL1 - decidual NK cells (dNK2), as reported in a previous study 14 . The NK2 cells expressed CD49a , CXCR4 , CD103 , ITGB2 , CD160 and TIGIT , but not CD39 , KIR2DL1 and CD16 ( Figure 6 B , and Figure 6 D ), which might be the precursor cells of CD49a + CD103 + ITGB1 + KIR2DL1 - dNK3 in the decidua during early pregnancy 14 . Contrastingly, NK3 cells highly expressed LGALS1 and PGK1 , and negatively expressed CD49a , CXCR4 , CD39 , ITGB2 , KIR2DL1 , PRF1 and CD16 ( Figure 6 B , and Figure 6 D ). CD49a - CXCR4 + CD160 + TIGIT + PRF1 + CD16 + NK4 cells displayed an activated cytotoxicity. Unlike peripheral blood NK cells, the NK4 cells exhibit high levels of ITGB2 . Interestingly, we found that the percentage of the four subsets of human endometrial NK cells also showed dynamic and periodic changes throughout the menstrual cycle ( Figure 6 C ). Among these, NK1 was the predominant population in the endometrium during the proliferative and secretory phases, while the percentage of NK2 cells with high expression of chemokines peaked during the secretory phase and then declined, and the main NK cells of the menstrual endometrium were CD16 high IFNG high GZMB high PRF1 high NK4 cells, which are involved in immune defenses during menstruation ( Figure 6 C - D , and Figure S11 A ). In tumor microenvironment, a heightened capacity for glucose metabolism through glycolysis supports NK cells with the greatest cytotoxic capacity 33 . However, the cytotoxicity of endometrial LGALS1 + LDHA + NK3 cells with an activated glycolysis was not strong ( Figure 6 D , Figure S11 A - B ), which should be the progenitor cells of CD49a + CD103 - CD39 + ITGB1 - dNK1 cells with high levels of granule proteins, activated glycolytic metabolism and high expression of KIR genes 14 . LGALS1, also known as galectin 1, is an important lectin with major functions in embryo implantation, modulation of maternal immune responses and placentation 34 , suggesting that NK3 cells particularly interact with embryonic trophoblasts. More importantly, CD49a + CXCR4 + CCR9 + NK2 cells should be recruited by four subsets of CCL25 + CXCL12 + FIBs and two subsets of EC (Ciliated EC and EC1), and further contribute to the infiltration of other immune cells ( CCR1 + CCR5 + Mo/Mφ, CCR1 + CCR5 + T cells, XCR1 + B cells and XCR1 + mast cells) into the endometrium by high levels of various chemokines (e.g., CCL3, XCL1, and XCL2) ( Figure 6 E - F ). Additionally, Ciliated EC, EC1, EC2 and EC3 are involved in the recruitment of CX3CR1 + CXCR1 + NK4 and mast cells ( Figure 6 E ). These results indicate that CD49a + CXCR4 + NK2 cells should be a core node for the regulation of FIB/EC infiltration in immune cells. Further FCM analysis showed that there was an aberrant percentage of NK2 and NK4 cells in RIF patients ( Figure 6 G ), which contributes to an imbalance in the endometrial immune microenvironment. To systematically study the interactions of endometrial cells at the WOI, we developed a repository of ligand-receptor interacting pairs, representing a complex regulatory network by intercellular communication analysis ( Figure S12 A - B ). Notably, the most intense crosstalk between EC1/Ciliated-EC and NK1/NK2 ( Figure S13 A - B ). In addition to acting as messengers between the uterus and embryo, EVs have also been reported in recent years to be involved in immune regulation in tumor environments 35 , 36 . To further evaluate the potential role of EVs derived from epithelial cells on endometrial NK cell, GW4869, an inhibitor of exosome biogenesis/release, was used to treat hEECs for 24 h, and then co-cultured with primary isolated endometrial NK cells, in vitro ( Figure 7 A ). The percentage of NK1 (nearly 1%) and NK2 (nearly 20%) after 48 h in vitro culture significantly decreased, as well as an obvious increase in NK4 (above 60%), suggesting that the local endometrial environment contributes to the survival, proliferation and/or differentiation of NK1 and NK2 ( Figure S13 C - D ). Further observation showed that GW4869-pretreated hEEC led to a marked decrease in NK2 cells and an increase in NK3 cells in the co-culture system compared with control hEEC ( Figure 7 A ), suggesting that exosomes released by EEC contributes to the survival and proliferation of resident NK1 and NK2 cells in the endometrium. Interestingly, the CellPhoneDB results predicted that CXCL12 / CXCR4 and IL15 / IL15 receptor are involved in the regulation of EC/FIB in the NK cells, especially the NK2 cells ( Figure 7 B ), and these ligand-receptors are important factors for the recruitment, residence and proliferation of endometrial NK cells 37 - 39 . It has been reported that progesterone regulates NK cells by IL-15 40 , 41 . EVs are enriched in certain types of lipids compared with their parent cells. For example, vesicles are enriched in cholesterol and saturated fatty acids 29 , 42 . Notably, both marker genes of exosomes and APOD/APOE were predicted to be associated with CXCL12/CXCR4 and IL15 ( Figure 7 C ). We found that GW4869 down-regulated the expression of CXCL12 and IL15 in hEEC. Contrastingly, exposure to palmitic acid (PA) up-regulated the expression of CXCL12 and IL15 in hEEC and primary ESC in vitro , as well as APOD/APOE in ESC ( Figure 7 D ). These findings indicate that EC1 with rich exosome and FIB1 with high levels of APOD/APOE are involved in the recruitment and survival of NK2 by secreting CXCL12 and IL15 ( Figure 7 E ). The absence of EC1, and decreased levels of CXCL12 and IL15 in ESC contribute to the abnormal decrease of NK2 in the endometrium of RIF patients. In the endometrium, NK1 and NK2 have high levels of NCAM1 (also known as CD56, which is highly expressed in decidual NK cells during early pregnancy), CD49a and EOMES ( Figure 8 A ) and relatively low levels of CD16 , INFG , and GZMB , suggesting that NK1 and NK2 are the progenitor cells of decidual NK cells 14 , 43 . Our previous work showed that TNFSF14 (also known as LIGHT) promotes cell adhesion and tissue remodeling of decidual stromal cells by interacting with TNF4SF14 (also known as HVEM) and further activating MMP9 signaling 32 . Therefore, TNFSF14 + NK2 cells are beneficial to TNF4SF14 + FIB/EC adhesion and NK2 cell residence in the endometrium at the WOI ( Figure 8 A and 8C ). In addition to promoting the infiltration and adhesion of NK1 and NK2 in the endometrium, FIBp, FIB1 and FIB2 were predicted to promote the differentiation of NK4 cells into decidual-like NK cells by TGFB ( Figure 7 B and 7E ) 44 . Furthermore, NK4 and Mo/Mφ-derived TNF were associated with moderate inflammation required for implantation ( Figure 8 C ) 45 . Four subsets of FIBs were also involved in the adhesion and residence of Mo/Mφ via a variety of integrins and LGALS9 / CD44 ( Figure 8 A ). Macrophage migration inhibitory factor (MIF) and IGF2 play critical roles in regulating inflammatory and anti-inflammatory properties of macrophage, respectively 46 , 47 . Here, we predicted that MIF and IGF-2 derived from FIB2 and FIB1 regulated the homeostasis of endometrial Mo/Mφ ( Figure 7 E and 8A ), which would prepare for moderate inflammation during embryo implantation and rapid transformation of maternal-fetal immune tolerance during early pregnancy. In turn, TGFB , PDGFB and HBEGF derived from endometrial Mo/Mφ and or NK cells (i.e., NK2 cell) accelerated the decidualization of FIB and differentiation of Ciliated EC, EC1 and EC2, and trophoblast invasion ( Figure 8 A and 8C ) 48 - 51 . Notably, endometrial Mo/Mφ and NK2 cells are considered to participate in angiogenesis and vascularization through the interaction of VEGF and its receptors ( Figure 8 A and 8C ) 52 - 54 . Additionally, close crosstalk between NK cells and other immune cells (e.g., Mo/Mφ) was observed, including the regulation of focal adhesion, ECM-receptor interaction, cytokine-cytokine interaction, NK cell-mediated cytotoxicity, and primary immunodeficiency ( Figure 8 B ). Therefore, the proper proportion and function of the four major cellular subsets (FIB, EC, IC and VASC) contribute to the homeostasis and endometrial receptivity at the WOI through precise and orderly intercellular interactions. The subset imbalance of these cells (e.g, EC1, and NK2) was predicted to result in the pathogenesis of RIF with an aberrant cell communication network.

Implantation of the embryo into the uterine endometrium is one of the most finely-regulated processes that leads to a successful pregnancy. Disorders of endometrial receptivity and shifted window of receptivity are considered important etiologies for unexplained RIF; therefore, exploring the normal and pathological cell subsets and cell interaction mechanisms of the endometrium at the WOI is particularly critical for the diagnosis and treatment of explained RIF 4 . Recent studies on the human endometrium have been limited to physiological states 5 , 12 , a single cell type 55 , or a single patient with uterine leiomyoma 56 . Herein, we focused on the human endometrium at the time of embryo transfer, and further identified cell subset changes and cell communication networks in the human endometrium, which were supported by multiple healthy and RIF biological replicates at the WOI and across the menstrual cycle. Herein, we defined four major FIB populations (FIBp, FIB1, FIB2, and FIB3). Pseudotime analysis and further verification showed that FIBp with high proliferative activity, and FIB3 cells located in the end of single-cell trajectory mainly existed in the endometrium during proliferative and premenstrual phases, respectively. As the largest proportion of FIBs at the WOI, FIB1 and FIB2 displayed better endometrial receptivity for the opening of WOI, and promoted the recruitment, residence and proliferation of NK cells by the collagens/integrins, CXCL12/CXCR4, CCL25/CCR9 and IL15/IL15R interactions, contributing to the dominance of NK1 and NK2 cells in the endometrium at the WOI. FIB1 and/or FIB2-derived TGFB, and IGF2 signaling are also involved in the differentiation and cytotoxicity regulation of NK4 cells 57 . Additionally, FIBs are involved in the recruitment and infiltration of other immune cells, including mast cells, ILC and T/NKp cells 57 , 58 to build an immune microenvironment conducive to embryo implantation. The Ciliated epithelium, the foreword place of embryo implantation, has been identified, but its characteristics and function remain largely undefined. Here, we observed that Ciliated-ELE exhibited cilium movement, organelle organization, and smoothened signaling pathways. Moreover, a novel group of EC, EC1, with powerful exosomes secretion was identified. The release of exosomes by EC1 should be dependent on progesterone/autophagy signaling and IGF-1 produced by FIB2 and FIB3. More importantly, Ciliated-EC, EC1, FIB1 and FIB2 work together to open the WOI, promote embryo adhesion and implantation by good endometrial receptivity, and trigger the rapid transformation of the immune microenvironment from moderate inflammation during embryo implantation to maternal-fetal immune tolerance during early pregnancy. These processes are dependent on the regulation of NK and Mo/Mφ activation by HLAs/KIRs and BAG6/NCR3 59 interactions, and MIF/CD74 and IGF2/IGF2R, respectively. Notably, NK cells also exhibit dynamic and periodic changes throughout the menstrual cycle. As the potential progenitor cells of decidual NK cells, tissue resident NK1 and NK2 cells expressing CD49a and EOMES were dominant in the endometrium at the WOI, especially the NK2 cells. Decidual CD49a + EOMES + NK cells have been reported to promotes fetal growth during early pregnancy, and CD49a + EOMES + NK cells in menstrual blood and decidua are associated with unexplained recurrent spontaneous abortion 60 . The recruitment and survival of NK2 cells are dependent on exosomes secreted by EC1, and CXCL12/IL15 produced by FIB1. Interestingly, the NK2 cells are predicted to promote cellular adhesion and tissue remodeling in FIB/EC by TNFSF14/TNFRSF14 and MMP9 signaling. More importantly, the NK2 cells may act as an intermediate link between FIB/EC and other immune cell (e.g., Mo/Mφ) infiltration, which deserves further study. NK2 cells, together with Mo/Mφ, are predicted to accelerate the decidualization and differentiation of FIB/EC to open the WOI, angiogenesis and vascularization, and trophoblast invasion by the production of TGFB, PDGFB, HBEGF and VEGF. However, the specific mechanism needs to be studied further. Additionally, this study has several limitations. On the one hand, it is challenging to digest endometrial glands into single cells. The number of epithelial cells obtained in this study was relatively limited and needs to be improved. However, the sample size of this study is limited, and the etiology of RIF was complex. Additionally, there are marked differences in individuals of RIF patients, which need to be further verified by large sample studies. Notably, several classic marker genes (e.g, PRL and LIF) for endometrial receptivity and decidualization were detected at low levels, which was also reported by Wang et al. , 5 . This suggests that embryo implantation plays an important role in rapidly accelerating decidualization. Benign and effective interactions between the mother and fetus during endometrial differentiation for embryo implantation should be emphasized. More interestingly, there was significantly decreased expression of autophagy, exosomes and senescence-related genes in both FIB and EC from RIF patients, suggesting that appropriate autophagy, exosomes and senescence are important characteristics of the endometrium for embryo implantation during the WOI time 61 - 65 . Therefore, abnormally low levels of these functional genes should contribute to the pathogenesis of RIF, and further research needs to focus on their value in the prediction and therapy of RIF.

Collectively, the proper proportion, precise and orderly intercellular communication of constitutive cells (FIB, EC, IC and VASC) are beneficial for endometrial receptivity and the opening of WOI. The abnormal expression of endometrial receptivity and immune regulation-related molecules in FIBs, and the percentages of EC1 and NK2 cells are associated with RIF. The potential mechanisms for these disorders include progesterone insensitivity-induced autophagy and secretion limitation of EC1, deficiency of the APOD/APOE/CXCL12/IL15 axis-mediated poor decidualization, NK2 infiltration and survival, and incorrect selection for embryo transfer time of IVF. Therefore, a comprehensive and detailed analysis of the subpopulations of cells in the endometrium will help us identify possible causes of RIF and pave the way toward new diagnostic and therapeutic strategies for RIF, for example, choosing a more suitable time for embryo transfer to avoid a shifted window of receptivity or poor endometrial receptivity.

Supplementary figures and tables. Click here for additional data file.