Human endometrial cell-type-specific RNA sequencing provides new insights into the embryo-endometrium interplay.

One of the most elegant and fascinating interactions in human physiology takes place between an embryo and the endometrium, in order to initiate and maintain pregnancy. For successful embryo implantation, the development of an embryo into a blastocyst and the differentiation of the endometrium into the receptive phase need to be synchronized. The first physical contact between an implantation-competent blastocyst and the receptive-phase endometrium involves sequential stages of apposition, attachment and invasion, which together steer the entire process of implantation ( Norwitz et al. , 2001 ; Haller-Kikkatalo et al. , 2014 ; Evans et al. , 2016 ; Ashary et al. , 2018 ). The first step of implantation, apposition, is characterized by unstable adhesion where the surface of the blastocyst touches the endometrial epithelium ( Wang and Dey, 2006 ; Hernández-Vargas et al. , 2020 ). During this phase, the blastocyst rolls on the epithelial lining, anchors itself into the mucus and orients itself so that the polar trophectodermal part of the embryo, located nearest to the inner cell mass, faces the uterine decidua ( Pellicer et al. , 2002 ; Deglincerti et al. , 2016 ). During the attachment phase, the embryo adheres to the endometrial epithelial cells via the polar trophectodermal cells ( Meistermann et al. , 2021 ). This phase is characterized by specific receptor–ligand binding and signalling via integrins and E-cadherins ( Aplin, 1997 ). In the early invasion stage, the embryo dislodges epithelial cells, penetrates the basal lamina, establishes contact with the endometrial stromal cells and finally invades into the uterine decidua ( Norwitz et al. , 2001 ). The events that orchestrate these processes in implantation are coordinated by numerous factors under the influence of the ovarian hormones oestrogen and progesterone ( Abbas et al. , 2020 ). Identification of the molecular events, however, remain hardly accessible for ethical reasons, making human embryo implantation a true ‘black box’ in developmental biology. Thus, the molecular dialogue taking place between the embryo and the endometrium is still poorly understood and the proteins involved in the implantation process are largely unknown. We have previously modelled the human embryo–endometrium molecular crosstalk by studying the transcriptome of the whole embryos and full-thickness endometrial biopsies ( Altmäe et al. , 2012 ). Gene expression analyses of whole tissues neglect the cell-type specific profiles and proportions of the cell populations within the sample, thereby likely confounding the results of differential transcriptome analyses ( Suhorutshenko et al. , 2018 ). With a novel approach, we set out to analyse the cell-type-specific transcriptome of freshly isolated pre-receptive and receptive stage endometrial epithelial and stromal cells and combine the data with transcriptome data from trophectodermal cells of blastocyst stage embryos in order to identify the first molecular events taking place in the embryo–endometrium dialogue.

To identify the cell-type-specific transcriptome changes between the pre-receptive and receptive endometria, we conducted separate differential expression analyses on endometrial epithelial and stromal cells. Out of 12 339 gene transcripts detected in the receptive phase endometrial epithelial cells, 499 genes were up-regulated and 101 genes were down-regulated when compared to the pre-receptive phase samples ( Fig. 2 , Supplementary Table SIII ). Out of the 12 944 genes detected in the stromal cells, 581 genes were significantly up-regulated and 116 genes were down-regulated when compared to the pre-receptive phase samples. We compared the up- and down-regulated genes between two cell types to identify the unique epithelium- and stroma-specific genes. As 126 receptivity phase genes were up-regulated and 36 genes were down-regulated in both cell types, 373 genes (74.7%) remained as uniquely up-regulated and 65 genes (64.4%) remained as uniquely down-regulated in epithelial cells, while 455 genes (78.3%) remained as uniquely up-regulated and 80 genes (69.0%) remained as uniquely down-regulated in stromal cells ( Fig. 2 , Supplementary Table SIII ). In the following analyses, we considered all differentially expressed genes as potential endometrial receptivity-specific genes. Cell-type-specific gene expression. ( A ) Up-regulated genes in epithelial and stromal cells in receptive versus pre-receptive phase endometrium. ( B ) Down-regulated genes in epithelial and stromal cells in receptive versus pre-receptive phase endometrium. Gene lists are provided in Supplementary Table SIII . Validation by qRT-PCR confirmed the RNA-seq expression change for five genes selected for validation: APOC1 , CLU , RGS16 , RORC and TEX40 ( Supplementary Table SII ). The list of endometrial cell-type-specific molecules enabled us to predict potential interactions between epithelial and stromal cells in the receptive phase endometrium. In the in silico analyses of the RNA-seq data, we prioritized the interactions with tested gene–protein correlation in the previously published proteome encyclopaedia ( Nusinow et al. , 2020 ). Among the up-regulated 499 endometrial epithelial genes, 418 (83.8%) and among the up-regulated 581 endometrial stromal genes, 458 (78.8%) showed positive correlation between the gene and protein expression ( Supplementary Table SIV ). These genes were mapped to respective protein products and the epithelial–stromal protein–protein interaction networks were constructed. This epithelium–stroma network contained 194 predicted interactions, where 36 (18.6%) were unique to the epithelial cells and 67 (34.5%) to the stromal cells ( Supplementary Table SV ). Next, we clustered the constructed PPI networks, which resulted in 21 clusters, where the three biggest clusters represented 53% of all the nodes ( Supplementary Fig. S1 ). Enrichment analyses were performed for the three biggest clusters: Cluster A that integrated 27 proteins and processes involved in wound healing and regulation of cell proliferation; Cluster B that integrated 13 proteins and antigen processing and presentation process; and Cluster C that integrated 11 proteins and regulation of immune response process ( Supplementary Table SV ). The unique list of cell-type-specific molecules of the receptive phase endometrium allowed us to predict potential interactions that could take place between embryonic trophectodermal cells and endometrial epithelial or stromal cells at the late apposition, attachment and initiation of invasion processes. To find interacting proteins, we used cell-type-specific endometrial transcriptomic RNA-seq data and single-cell transcriptomic data of polar trophectodermal cells, and prioritized the interactions where there is also evidence of gene–protein correlation ( Nusinow et al. , 2020 ). Of the up-regulated endometrial epithelial genes, 83.8%, and of the up-regulated endometrial stromal genes, 78.8% showed positive correlation between the gene and protein expression. Similarly, 4812 out of 5578 (86.3%) polar trophectodermal genes showed positive correlation between gene and protein levels ( Supplementary Table SIV ). Those genes were mapped to their respective protein products and used for the construction of embryo–endometrial protein–protein interaction networks. This network contained 558 predicted interactions, encompassing 222 proteins ( Supplementary Table SVI ). Within the PPI network, 125 (56.3%) were unique to the polar trophectodermal cells, 16 proteins (7.2%) were detected in trophectodermal and stromal cells and 13 proteins (5.9%) were detected in trophectodermal and epithelial cells ( Supplementary Table SVI ). Only seven proteins (3.2%) (HLA-B, TIMP1, APOL1, AGR2, SERPING1, VCAN and CALCRL) were expressed in both epithelial and stromal cells but were not detected in trophectodermal cells. There were 16 (7.2%) proteins uniquely expressed in epithelial cells and 38 (17.1%) uniquely expressed in stromal cells and 7 (3.2%) proteins were detected in all three cell types. We further clustered the constructed PPI networks and performed enrichment analyses for each cluster to interpret it and describe the possible role of the predicted interactions. Altogether, 222 interacting proteins were divided into 15 clusters, where the top five clusters included more than 18 proteins ( Supplementary Table SVI , Supplementary Fig. S2 ). The formed clusters were ordered based on the number of interacting proteins and named alphabetically. In the next section, we will explore the five largest clusters (Clusters A, B, C, D and E) that contained more than half of the interacting proteins. The largest cluster (Cluster A, Fig. 3 ) contained more than a fifth of the interacting proteins (46 proteins, 20.7%) and a tenth of the interactions (85 interactions, 15.2%) and was mainly enriched in proteins known to be involved in cell-adhesion (adj. P -value 6.5 × 10 −20 ). Indeed, 24 (52.2%) of the interacting proteins are transmembrane proteins ( Fig. 3 ). The protein with the highest degree of interactions in this cluster was integrin β1 (ITGB1), a transmembrane protein expressed on both epithelial and trophectodermal cells. Epithelial MET, a proto-oncogene that interacts with trophectodermal ITGB1 ( Fig. 3 ) has been detected as an important endometrial receptivity gene product, physically interacting with foetal interface at implantation ( Ntostis et al. , 2021 ). Integrin β1 interaction with basigin (BSG; expressed in stromal and trophectodermal cells ( Fig. 3 )) has been shown to regulate embryo implantation ( Lee et al. , 2013 ) and integrin β1 regulation by all cells expressed galectin-3 (LGALS3) that mediates trophoblast cells invasion ( Bojić-Trbojević et al. , 2019 ). The LGALS3 interaction with trophectodermal CD98 (SLC3A2) is, in turn, important for trophoblast cell fusion ( Dalton et al. , 2007 ). We also identified the interaction between the epithelial transmembrane protein CD44 and trophectodermal syndecan-1 (SDC1), likely regulating trophoblast invasion ( Ibrahim et al. , 2017 ). A previous in vitro study showed that functional blocking of epithelial CD44 led to delayed attachment during early stages of implantation ( Berneau et al. , 2019a ), and transcriptome analysis of trophoblast-epithelium in in vitro model highlighted CD44 involvement in early trophoblast attachment ( Vergaro et al. , 2021 ). Cluster A also contains secreted Insulin-like growth factor-binding protein-2 (IGFBP2), which binds to integrin β1 and this interaction has been shown to regulate cell migration and invasion ( Han et al. , 2014 ). Cluster A. Molecular interactions of polar trophectodermal cells with endometrial epithelial and stromal cells. Pink: stromal proteins, red: epithelial proteins, light blue: trophectodermal proteins, violet: epithelial and stromal proteins, green: epithelial and trophectodermal proteins, light green: stromal and trophectodermal proteins, grey: epithelial, stromal and trophectodermal proteins. Ellipse: secreted proteins, diamond: extracellular matrix proteins, octagon: membrane-bound proteins, rectangle: transmembrane proteins. Interacting proteins and interactions are provided in Supplementary Table SVI . The second largest cluster was Cluster B ( Fig. 4 ) with 46 interacting proteins (20.7%), including more than quarter of the PPIs (161 interactions, 28.9%). Most of the involved proteins (24 proteins, 52.2%) undergo post-translational protein modifications (adj. P -value 9.1 × 10 −26 ) and could therefore participate in cellular signalling. In fact, ∼60% of the interacting proteins were secreted proteins ( Fig. 4 ). For example, proteolytically cleaved extracellular matrix cytokine osteopontin ( SPP1 ) in the epithelium may interact with trophectodermal transmembrane integrin α5β1 (ITGA5) and therefore support cell adhesion and migration ( Barry et al. , 2000 ). Epithelially secreted osteopontin interacted with trophectodermal extracellular matrix protein galectin-1 ( LGALS1 ), which has been related to blastocyst attachment and trophoblast migration ( Barrientos et al. , 2014 ). Versican (VCAN), an extracellular matrix protein, expressed by both endometrial epithelial and stromal cells, and involved in cell adhesion and migration, with a role in implantation ( San Martin et al. , 2003 ; Altmäe et al. , 2012 ), was detected as one central molecule from the endometrial interface interacting with the trophectodermal proteins APOE, APOA1, APOA2, LAMB1, LAMC1, LGALS1 and MXRA8 that are all involved in cell adhesion and migration processes. Cluster B. Molecular interactions of polar trophectodermal cells with endometrial epithelial and stromal cells. Pink: stromal proteins, red: epithelial proteins, light blue: trophectodermal proteins, violet: epithelial and stromal proteins, green: epithelial and trophectodermal proteins, light green: stromal and trophectodermal proteins, grey: epithelial, stromal and trophectodermal proteins. Ellipse: secreted proteins, diamond: extracellular matrix proteins, octagon: membrane bound proteins, rectangle: transmembrane proteins. Interacting proteins and interactions are provided in Supplementary Table SVI . Proteins in cluster C ( Fig. 5 ) were enriched for proteins regulating the secretion processes (adj. P -value 2.4 × 10 −7 ), key processes in releasing hormones, enzymes and other factors, and intercellular signalling ( Trikha et al. , 2010 ). This cluster contained 31 interacting proteins (14.0%) and 32 interactions (5.7%). Almost half of the proteins (45.2%) were secreted proteins, including epithelial cell-specific interferon (IFN)-stimulated gene 15 (ISG15), stromal cell-specific chemokine (C-X-C motif) ligand 13 (CXCL13), epithelial and trophectodermal-specific proteins stratifin (SFN) and granulin (GRN), trophectodermal NPC intracellular cholesterol transporter 2 (NPC2) and gamma-glutamyl hydrolase (GGH), as well as stromal and trophectodermal-specific proteins cathepsin B (CTSB) and fatty acid-binding protein, epidermal (FABP5). We predicted six of these proteins (NPC2, GGH, ISG15, CTSB, FABP5 and GRN) to interact with the extracellular matrix protein annexin A2 (ANXA2), which was expressed by all cell types. This secretion pathway protein is implicated in endometrial epithelial cell migration and trophoblast outgrowth, being therefore essential for embryo adhesiveness to the human endometrium ( Garrido-Gómez et al. , 2012 ). Indeed, ANXA2 was recently detected as an important endometrial receptivity gene product that physically interacts with trophectodermal cells at implantation ( Ntostis et al. , 2021 ). Cluster C. Molecular interactions of polar trophectodermal cells with endometrial epithelial and stromal cells. Pink: stromal proteins, red: epithelial proteins, light blue: trophectodermal proteins, violet: epithelial and stromal proteins, green: epithelial and trophectodermal proteins, light green: stromal and trophectodermal proteins, grey: epithelial, stromal and trophectodermal proteins. Ellipse: secreted proteins, diamond: extracellular matrix proteins, octagon: membrane bound proteins, rectangle: transmembrane proteins. Interacting proteins and interactions are provided in Supplementary Table SVI . Cluster D is rather small, including 25 (11.2%) of the PPI proteins and 27 (4.6%) of the interactions, and is enriched in cell–cell adhesion molecules (adj. P -value 1.4 × 10 −7 ). Of the interacting proteins in the network 76% were transmembrane proteins and the cluster formed around the transmembrane E-cadherin (CDH1) ( Fig. 6 ), an important molecule in cell–cell adhesion with implications in the implantation process ( Haller-Kikkatalo et al. , 2014 ). E-cadherin was expressed in both epithelial and trophectodermal cells, and we predicted it to bind claudins (CLDN3, CLDN4, CLDN7), flotillins (FLOT1, FLOT2), cadherins (CDH3, CDH13), plasminogen activator inhibitor-1 (SERPINE1), alpha-catulin (CTNNAL1) and epithelial cell adhesion molecule (EPCAM). However, it is unclear whether these interactions are heterotypic trans-interactions between the epithelial and the trophectodermal cells or take place on the surface of trophectodermal cells. Interestingly, knock-down of SERPINE1 in the human ECC-1 cell line demonstrated significantly reduced trophectoderm spheroid adhesion ( Evans et al. , 2020 ). Cluster D. Molecular interactions of polar trophectodermal cells with endometrial epithelial and stromal cells. Pink: stromal proteins, red: epithelial proteins, light blue: trophectodermal proteins, violet: epithelial and stromal proteins, green: epithelial and trophectodermal proteins, light green: stromal and trophectodermal proteins, grey: epithelial, stromal and trophectodermal proteins. Ellipse: secreted proteins, diamond: extracellular matrix proteins, octagon: membrane-bound proteins, rectangle: transmembrane proteins. Interacting proteins and interactions are provided in Supplementary Table SVI . Cluster E was enriched in proteins known to be involved in extracellular matrix organization (adj. P -value 1.8 × 10 −6 ), where seven (40%) of the interacting proteins were extracellular matrix proteins and another seven (40%) were transmembrane proteins ( Fig. 7 ). For example, the secreted protein oestrogen-regulated anterior gradient 2 (AGR2), unique to epithelium/stroma in our study, interacting with Ly6/PLAUR domain-containing protein 3 (LYPD3) and dystroglycan 1 (DAG1), has been shown to increase the aggressiveness of cancer cells through the regulation of receptor adhesion and interaction with the extracellular matrix ( Salmans et al. , 2013 ; Arumugam et al. , 2015 ). Trophoblast cell-secreted procollagen-lysine, 2-oxoglutarate 5-dioxygenase 3 (PLOD3) could regulate extracellular matrix protein COL12A1 in the stroma to promote cells invasion and migration ( Li et al. , 2020 ). Cluster E. Molecular interactions of polar trophectodermal cells with endometrial epithelial and stromal cells. Pink: stromal proteins, red: epithelial proteins, light blue: trophectodermal proteins, violet: epithelial and stromal proteins, green: epithelial and trophectodermal proteins, light green: stromal and trophectodermal proteins and grey: epithelial, stromal and trophectodermal proteins. Ellipse: secreted proteins, diamond: extracellular matrix proteins, octagon membrane-bound proteins, rectangle: transmembrane proteins. Interacting proteins and interactions are provided in Supplementary Table SVI .

The study was approved by the Research Ethics Committee at the University of Tartu (Estonia) and Instituto Valenciano de Infertilidad (Spain), and written informed consent was obtained from all participants. All methods were performed in accordance with the relevant guidelines and regulations. Altogether, 16 healthy fertile women from Estonia and Spain were recruited. The average age of the women was 29.7 ± 3.2 years, all had normal BMI (average BMI: 22.8 ± 2.3), a regular menstrual cycle (25–35 days), had not used hormonal medication for 3 months before recruitment, had normal serum levels of progesterone, prolactin and testosterone, negative screening results for sexually transmitted diseases, no uterine pathologies, no endometriosis nor polycystic ovary syndrome and had at least one live-born child. Menstrual cycle dating was confirmed by combining menstrual cycle history and luteinizing hormone (LH) peak estimation (BabyTime, Pharmanova, Beograd, Serbia), vaginal ultrasound and endometrial histology by Noyes’ criteria ( Noyes et al. , 1975 ). Each participating woman provided two endometrial biopsy samples within the same menstrual cycle: from early secretory phase (2 days after the LH peak, LH + 2, defined as pre-receptive phase) and from mid-secretory phase (LH + 7/+8, defined as receptive phase) ( Fig. 1 ). The receptivity status of all analysed samples was confirmed using the beREADY endometrial receptivity test ( www.beready.ee ) that is based on the transcriptomic profiling of receptivity biomarkers ( Altmäe et al. , 2017 ). The biopsies were obtained using a Pipelle catheter (Laboratoire CCD, Paris, France) and the samples were frozen at –80°C using cryopreservation media to keep cells alive, as described in our previous study ( Krjutškov et al. , 2016 ). The preparation of the endometrial biopsies after thawing for sorting of CD9-positive epithelial and CD13-positive stromal cells is summarized in Fig. 1 ; a detailed protocol has been described previously ( Kato et al. , 2007 ; Krjutškov et al. , 2016 ; Masuda et al. , 2016 ). Total RNA was isolated immediately from the sorted cells using the RNeasy Micro kit (Qiagen, Hilden, Germany). Overview of endometrial sample collection and study design. The preparation of the endometrial biopsies for the cell sorting has been described in detail ( Krjutškov et al. , 2016 ). Briefly, the biopsied tissue samples were thawed and dissociated; endometrial stromal cells were stained with fluorochrome-conjugated mouse anti-human CD13 monoclonal antibody (clone TUK1, R-Phycoerythrin, Invitrogen, CA, USA); epithelial cells were stained simultaneously with fluorochrome-conjugated mouse anti-human CD9 monoclonal antibody (clone MEM-61, FITC, Novus Biologicals, Littleton, USA); and all dead cells were stained with DAPI (final concentration of 0.5 µg/ml) (Invitrogen). CD9 or CD13 positive and DAPI low signal or negative cells were sorted by fluorescence-activated cell sorting (FACS) and collected separately into QIAzol Lysis Reagent (Qiagen). The percentages of both epithelial (CD9 positive) and stromal (CD13 positive) cells varied considerably from patient to patient: in the early secretory phase CD9 positive (epithelial) cells composed 26.4–87.7% of the cells, while the number in mid-secretory samples was 8.2–61.8%. Divergence of CD13-positive (stromal) cells was 3.4–50.2% in early secretory biopsy samples, and 6.6–59.0% in mid-secretory phase samples. To verify the purity of our isolated cell populations, we performed additional FACS analysis using a leukocyte-specific CD45 antibody (BD Pharmingen, cat: 555485). CD45 is present on all human leukocytes including lymphocytes, monocytes, granulocytes, eosinophils and thymocytes. We detected that only 0.2% of isolated CD13+ and only 1.1% of CD9+ cells were CD45 positive. As the fractions of CD45+ cells were small and all our isolated cells from early and mid-secretory endometria were manipulated in the exact same way, we believe that the purity of the cell populations between different samples is comparable and does not affect the overall expression profile. Full transcriptome analysis of the sorted endometrial cells was performed using the single-cell tagged reverse transcription (STRT) protocol ( Krjutškov et al. , 2016 ) with modifications that are required for bulk RNA analysis (RNA-seq data are available at www.ncbi.nlm.nih.gov/geo under accession number GSE97929 ). Samples of 10 ng of high-quality RNA extracted from enriched epithelial and stromal cell populations was converted into cDNA and amplified using 15 plus 10 PCR cycles to form an Illumina-compatible sequencing library with single 8 bp indexes. The STRTprep pipeline, available at https://github.com/shka/STRTprep/tree/v3dev , was used to process the raw RNA-seq reads, aligning to the hg19 genome, quantitating the expression levels and running the differential expression tests. Significance of fluctuation on the gene expression was tested by comparison with fluctuation of spike-in levels as described before ( Krjutškov et al. , 2016 ). Differential expression between sample types was tested by SAMstrt ( https://github.com/shka/R-SAMstrt ) ( Katayama et al. , 2013 ), which is based on Wilcoxon statistics with multiple Poisson resampling to equalize the spike-in depths, with the diffexp score representing the average of the Wilcoxon statistics among the resampling. Significantly differentially regulated genes were defined by Benjamini–Hochberg adjusted P -value <0.05 for the fluctuation and Q -value <0.05 for the differential expression. The gene expression was compared between the pre-receptive and receptive phase endometria for the enriched fraction of CD9-positive epithelial cells and CD13-positive stromal cells. The endometrial genes that were identified as significantly up-regulated in receptive versus pre-receptive phase, but were detectable (expression value > 0) in <75% of the receptive phase samples were filtered out. The transcripts expressed at the polar trophectoderm of the blastocyst stage embryo were obtained from our previously published human embryo transcriptional mapping, which was based on RNA-seq of 466 single embryonal cells from 17 human blastocyst-stage embryos from days 6 and 7 ( Petropoulos et al. , 2016 ). Briefly, cDNA libraries were generated using the Smart-seq2 protocol ( Picelli et al. , 2014 ). Gene expression levels were estimated in terms of reads per million mapped reads (RPKM). Based on the fact that human embryo attaches to the endometrial surface with the polar side ( Johnson, 2012 ), we used the subset of genes that were expressed on the polar trophectodermal cells and filtered in the transcripts with mean expression of >10 RPKM. As confirmation, the recently identified maturation marker of polar trophectoderm, NR2F2 ( Meistermann et al. , 2021 ), was detected in our dataset of polar trophectodermal cells. Human protein–protein interactions (PPI) were downloaded from the UniHI version 7.1 database ( http://www.unihi.org ) ( Kalathur et al. , 2014 ) and STRING 11.0 database ( http://version11.string-db.org/ ) ( Szklarczyk et al. , 2015 ). These two databases involve the largest number of experimentally verified interactions to date ( Bajpai et al. , 2020 ). The most relevant interactions from the STRING database were attained by retaining only interactions with confidence score of ≥0.7. PPI between (i) endometrial epithelial and stromal cells and (ii) the embryo and the endometrium were defined between the genes that were significantly up-regulated in the receptive phase endometrial epithelial and stromal cells and all identified genes expressed in the polar trophectodermal cells. The detected transcript identifiers were converted into corresponding proteins and only these genes whose expression was positively correlating (Spearman R > 0) with the protein expression profile in the previously published proteome encyclopaedia ( Nusinow et al. , 2020 ) were added to the subsequent analyses. To explore the possible epithelium–stroma interactions within receptive endometrium and embryo–endometrium interaction partners, we focussed only on the genes that encode proteins localized at the cell surface, extracellular matrix or secreted by the cells, according to the UniProt database (Accessed date: 10 April 2020). Only the PPI that took place between different cell types were described. To delineate the most relevant PPI for implantation processes, predicted interactions were clustered using the Cytoscape application AutoAnnotate ( http://apps.cytoscape.org/apps/autoannotate ) using the default settings ( Kucera et al. , 2016 ). Enrichment analyses for the interacting molecules within endometrium were performed for the three largest clusters (n > 11) and analyses for the embryo–endometrium interaction were performed for the five largest clusters (n > 18) by g:Profiler ( Raudvere et al. , 2019 ). Enrichment analyses were conducted by hypergeometric test and the resulting enrichment P -values were further adjusted for multiple testing using the g: SCS (Set Counts and Sizes) method, developed to address the non-independent structure of tested GO terms ( Raudvere et al. , 2019 ). Endometrial cell-type-specific RNA-seq was validated by quantitative real-time PCR (qRT-PCR) on a selected set of genes ( APOC1 , CLU , RGS16 , RORC and TEX40 ), using flow cytometer isolated endometrial LH + 2 epithelial (n = 10) and stromal (n = 8) cells, and LH + 8 epithelial (n = 15) and stromal (n = 12) cells obtained from the same individuals used in the RNA-seq analyses. qRT-PCR primers are listed in Supplementary Table SI . Samples of 10 ng of DNase-treated (TURBO DNA-free™ kit, Ambion Inc., Austin, TX, USA) RNA was converted into cDNA using RevertAid First Strand cDNA Synthesis Kit (Thermo-Fisher Scientific Inc. MA, USA). qRT-PCR was performed using 1× HOT FIREPol EvaGreen qPCR Mix Plus (Solis BioDyne, Estonia) according to the conditions specified by the manufacturer. The list of primers is provided in Supplementary Table SI . SDHA was used as endogenous control and the expression differences between LH + 2 and LH + 8 cell populations were calculated using Student’s t-test, using a P -value cut-off of <0.05. The 2-ΔΔCt method was used for calculating the relative gene expression and to determine the fold change in gene expression between LH + 2 and LH + 8 samples. The comparisons with RNA-seq fold change values are provided in Supplementary Table SII .

Implantation of a healthy embryo into a receptive endometrium is a critical step in the establishment of pregnancy, but the underlying molecular mechanisms of the first steps of implantation are still not clearly established. Our study reveals in silico molecular networks between the polar trophectodermal cells of implantation-competent embryo and the epithelial and stromal cells of receptive endometrium, extending the knowledge of the first steps of embryo implantation processes in humans. We found that the gene expression profiles are significantly different between the two main cell subtypes in the human endometrium, epithelial and stromal cells, which is in line with the previous single-cell endometrial transcriptome study ( Wang et al. , 2020 ; Garcia-Alonso et al. , 2021 ). When analysing the possible epithelial–stromal cellular interplay within the receptive-phase endometrium, we detected a big part of the interactions highlighting the importance of immune responses, wound healing and regulation of cell proliferation, which are important processes in endometrial receptivity ( Evans et al. , 2016 ; Altmäe et al. , 2017 ). In fact, a number of common endometrial receptivity biomarkers ( Díaz-Gimeno et al. , 2011 ; Altmäe et al. , 2017 ; Enciso et al. , 2018 ; Giacomini et al. , 2021 ) were detected in the endometrial cell subpopulations including ANXA4 , ARG2 , C4BPA , CFD , CLDN4 , CP , DKK1 , GPX3 , IL15 , MAOA , MT1H , NNMT , PAEP , S100P , SERPING , SPP1 , SOD2 and TCN1 among other genes. Considering the feasibility of the whole-tissue biopsies versus isolation of cell subpopulations within the sample, we propose that the differentially regulated genes C4BPA , GPX3 , MT1H and S100P that were identified in both endometrial cell types, and highlighted by previous studies, could serve as high-confidence candidates for endometrial receptivity biomarkers. A few studies have investigated the possible human embryo–endometrium interactions where whole tissue samples have been analysed ( Altmäe et al. , 2012 ; Haouzi et al. , 2012 ; Vilella et al. , 2015 ). Here, we focussed on the specific cell subpopulation transcriptome profiles of freshly isolated non-cultured endometrial epithelial and stromal cells, and model the initial molecular embryo–endometrium dialogue together with the potential paracrine interactions ( Fig. 8 ). While we confirmed the presence of several previously detected proteins in the crosstalk (FBLN1—MFAP5, NID1—FBLN2, LAMA1—FBLN2, DAG1—LAMA2, VEGFA—ADAMTS, SPP1—ITGAV, PDGFA—PDGFRA) ( Altmäe et al. , 2012 ), our focus on embryo–endometrium interactions using confirmed PPI based on cell-type RNAseq data provided a model with increased specificity. Indeed, the top molecules detected in the endometrium ( Fig. 8 ) have been identified on the single-cell RNA-seq resolution in the luminal epithelium and stroma, respectively ( www.reproductivecellatlas.org , Garcia-Alonso et al. , 2021 ). With the highly curated protein–protein network, we highlight the involvement of cell adhesion, post-translational protein modifications, regulation of secretion processes and extracellular matrix organization processes in the early implantation events such as apposition, attachment and initiation of invasion. A previous human proteome study on cultured cells detected cellular protein network changes between monolayer and spheroid trophectoderm models with regard to their adhesive protein landscape ( Evans et al. , 2020 ). Furthermore, a previously published modelling of embryos breaching the endometrial epithelium using human blastocysts or trophoblast stem cell spheroids cultured with endometrial epithelial cells predicted a set of genes in embryo–endometrium networks providing new knowledge to understand the embryo implantation process ( Ruane et al. , 2020 ). A number of the interacting proteins were also detected in the current study of freshly separated non-cultured endometrial epithelial and stromal subpopulations, such as CP, VCAN, SERPING1, ANXA2, DCN and APOD, and on the trophectodermal side, HSP90AB1, FURIN, HAPLIN1, CD55 and CTSB. Molecular crosstalk between the trophectodermal cells and the endometrial epithelial and stromal cells in implantation. The initial stages of the process: apposition-attachment ( A ) and the initiation of invasion ( B ). The molecules involved in the embryo–endometrium dialogue are highlighted; in green: trophectodermal proteins, pink: endometrial epithelial proteins, blue: stromal proteins. All proteins involved in the interactions are provided in Supplementary Table SVI . Embryo implantation is a complex process, likely to require a cascade of interacting proteins. A previous extensive endometrial proteome study identified 188 proteins crucial in the acquisition of endometrial receptivity ( Pérez-Debén et al. , 2019 ), out of which we identified 41 (22%) proteins among the endometrial cell subpopulations, where 8 were interacting as stroma-specific secreted proteins (C3, A2M, NAMPT, CFB, VTN, CST3, APOD, CST3) and CP was interacting as an epithelial-cell-specific secreted protein in the modelled embryo–endometrium crosstalk. The NAMPT gene product in the receptive phase endometrium has also been detected as physically interacting with trophectodermal cells ( Ntostis et al. , 2021 ). One of the hub molecules in our PPI networks was the transmembrane protein integrin β1, an important component of the surface of the trophectodermal and receptive-phase endometrial epithelial cells. A previous clinical trial showed that integrin β1 is a promising biomarker for evaluating uterine receptivity and determining the optimal time for embryo transfer ( Chen et al. , 2016 ). Interestingly, mouse blastocysts that lack the integrin β1 subunit fail to implant, which is due to the inability to adhere to or invade the subepithelial stroma ( Brakebusch et al. , 1997 ). Although there are differences between the initial implantation processes between human and mice (i.e. in mice, epithelial cells are lost by entosis and the trophectoderm interacts directly with the stroma ( Ye, 2020 )), animal models provide additional information of the process. In our study setting, ITGB1 seems to mediate a variety of interactions, which likely regulate embryo implantation. Integrin β1 levels on the surface of a trophectodermal cell are regulated by stromal/epithelial cell-secreted galectin-3, which therefore controls trophoblast cell migration ( Bojić-Trbojević et al. , 2019 ). Galectin-3 expression on the uterine epithelia of pregnant mice was found immediately after implantation, but not during the pre-implantation stage ( Phillips et al. , 1996 ). We also predicted integrin β1 to interact with the transmembrane glycoprotein BSG on the surface of trophectodermal and stromal cells. It is known that female mice lacking the BSG gene are infertile due to implantation failure ( Kuno et al. , 1998 ). Another galectin family member detected in our PPI network is galectin-1 (LGALS1), expressed by trophectoderm cells. Although mice embryos lacking galectin-1 implant normally ( Colnot et al. , 1998 ), galectin-1 expression in the trophectoderm of blastocysts has been suggested to play an important role in embryo implantation, by improving embryo adhesion or regulating maternal immunity ( Tirado-Gonzalez et al. , 2013 ). One of the interacting molecules with trophectodermal galectin-1 is epithelial cell secreted osteopontin. Osteopontin (SPP1) is one of the few genes repeatedly identified in transcriptome studies as endometrial receptivity-specific genes ( Altmäe et al. , 2017 ; Rekker et al. , 2018 ; Wang and Yu, 2018 ), and it has been shown to function in cell-to-cell adhesion during the early stages of implantation in different in vitro models ( Kang et al. , 2014 ; Berneau et al. , 2019b ; Vergaro et al. , 2021 ). Furthermore, like integrin β1, SPP1 has previously been implicated in the embryo implantation process in other species. For example, experiments on ovine luminal epithelial and trophectodermal cells demonstrated that SPP1 binding into the integrin receptor leads to cytoplasmic reorganization and induction of focal adhesions ( Johnson et al. , 2001 ), possibly triggering embryo attachment. Epithelial cells also uniquely secrete the IFN-stimulated gene 15 (ISG15), shown to be expressed as an early endometrial response to pregnancy in human as well as bovine, ovine, mice and other mammals, strongly suggesting that the ISG15 might be critical for pregnancy maintenance ( Henkes et al. , 2015 ). The penetration of the embryo between the endometrial stromal cells is mainly regulated by the cellular signalling and extracellular matrix restructuring. One of the matrix proteins, expressed by the stromal cells, is collagen Type XII alpha 1 chain (COL12A1), whose remodelling is induced by the trophectodermal protein PLOD3. Stromal cells also secrete several proteins, being involved in trophoblast cell invasion and migration, such as the epithelial and stromal cells secreted Anterior gradient protein 2 homolog (AGR2) interaction with trophectodermal membrane proteins Dystroglycan 1 (DAG1) and Ly6/PLAUR domain-containing protein 3 (LYPD3) ( Salmans et al. , 2013 ; Arumugam et al. , 2015 ; Li et al. , 2020 ) or stromal cell-secreted Insulin Like Growth Factor Binding Protein 2 (IGFBP2) regulation through integrin β1 ( Han et al. , 2014 ). The stromal secreted protein cathepsin B (CTSB) in mouse has been shown to be necessary for normal embryo development and uterine decidualization ( Afonso et al. , 1997 ) and the secreted protein CXCL13 has been related to successfully implanted blastocysts ( Dominguez et al. , 2008 ). Our study has its limitations, such as providing a static snap-shot analysis of a dynamic process and the nature of prediction analysis, which is limited to known interactions available in databases. Our approach also does not account for in vivo cell-to-cell influences, e.g. seminal plasma effects on endometrial gene expression ( Chen et al. , 2014 ) or interactions between different cell types within endometrium ( Hantak et al. , 2014 ). Due to the ethical and technical considerations, we could only study embryo–endometrial dialogue using biopsies taken from non-pregnant women, meaning that stromal cells were not decidualized as would be likely by the time the trophectoderm has penetrated the luminal epithelium in parallel with trophectodermal changes, and we could only use spare IVF embryos, likely missing some of the critical interactions taking place in case of natural conception. Furthermore, the effect of individual cells such as uterine natural killer cells and macrophages together with factors released from extracellular vesicles known to have role in the embryo–endometrium interplay ( Altmäe et al. , 2017 ; Ye, 2020 ; Díaz-Hernández et al. , 2021 ) could not be assessed in the current study setting. Moreover, the cell sorting technique used separated enriched epithelial cells from stromal cells but did not separate luminal from glandular epithelium, which is shown to differ ( Evans et al. , 2012 , 2014 ). Nonetheless, our study has several strengths to be highlighted, such as the analysis of specific types of cells (avoiding thus the cellular heterogeneity associated with whole-tissue biopsies ( Suhorutshenko et al. , 2018 )), the use of a comprehensive RNA sequencing platform, and most importantly, the study of ex vivo rather than cultured endometrial cells. The latter seems to be critical for getting more interpretable results from gene expression profiling, as we have seen in our previous study that culturing of endometrial cells rapidly alters their transcriptome ( Krjutškov et al. , 2016 ). In conclusion, our results provide a unique insight into the molecular mechanisms of endometrial receptivity by highlighting the relevant molecules involved in embryo apposition, attachment and initiation of invasion events in the initial steps of implantation. Our study identifies putative molecular networks between the polar trophectodermal cells of implantation-competent blastocyst and the epithelial and stromal cells of the receptive endometrium. This new knowledge of the endometrial cell-type-specific molecules that coordinate successful implantation is vital for understanding human reproduction and the underlying causes of implantation failure and infertility. Our study results provide a useful resource for future reproductive research, allowing the exploration of unknown mechanisms of implantation. We envision that those studies will help to improve the understanding of the complex embryo implantation process in humans, and hopefully generate new prognostic and diagnostic biomarkers and therapeutic approaches to target both infertility and fertility, in the form of new contraceptives.

Click here for additional data file. Click here for additional data file. Click here for additional data file. Click here for additional data file. Click here for additional data file. Click here for additional data file. Click here for additional data file. Click here for additional data file. Click here for additional data file.