Single-cell transcriptomic atlas of different endometriosis indicating that an interaction between endometriosis-associated mesothelial cells (EAMCs) and ectopic stromal cells may influence progesterone resistance

2.1 Study population

In order to explore the relationships and differences among the three main subtypes of endometriosis at the molecular level, we performed scRNA-seq (10X Genomics) on nine surgical biopsy samples from six patients diagnosed with endometriosis, including the different pathologic types: PEM, OEM and DIE (Supporting Information Table 1. Samples’ information). Patients without hormonal therapy before surgery will be included. Patients underwent exploratory laparoscopic surgery for ovarian cysts and had their cysts and peritoneal lesions excised, except for patient NO.6, who had no ovarian cysts detected on preoperative ultrasound or during surgery. Patient NO.6 underwent laparoscopic exploration due to intolerable dysmenorrhea, and lesions were found in Douglas cavity during surgery, which were proved to be PEM and DIE respectively. DIE1 and DIE3 samples were from the sacrouterine ligament at the time of excision of ovarian cyst. DIE2 was resected from the wall of the ovarian cyst. All samples were judged by an experienced surgeon during surgery and by a pathologist on haematoxylin and eosin (H&E)-stained sections.

After quality control as described in Section 3 (Figure S1A–C), a total of 74 643 cells from the three types of lesions were sequenced, of which 65 185 cells entered subsequent analysis (Figures 1A and S1E). DIE2 sample from the ovarian cystic wall was not included in subsequent analysis. Although it was consistent with the pathologic diagnosis of DIE. It differed significantly from the DIE group in transcriptomic similarity analysis and was similar to the OEM (Figure S1D). Perhaps some cyst wall was involved in this sample. Because the sample may be non-classical DIE tissue, it was not included in subsequent analysis.

2.2 scRNA-seq reveals multiple cellular components indifferent types of endometriosis

Unsupervised cluster analysis divided all cells into 25 clusters (Figure S1E). Based on marker genes (Figure 1B,C) in the previous literature,^{14, 18} cells were categorised into five cell types (Figure 1D): epithelial, stromal, endothelial, lymphoid and myeloid. Cells in each category were further to be divided into 44 subpopulations, with one cluster defined as unknown because it lacked of clearly known marker genes, demonstrating the cellular complexity of endometriosis.

When comparing the differences in the analysis of cellular component ratios among the three types at single-cell level, mesothelial cells were excluded, for they are not an original component of the endometriotic lesion but were parts of the microenvironment. In addition, we defined an increase of more than 20% in the comparison of cell proportions as a significant cellular component change. Since both PEM and DIE lesions were located on the peritoneum, we combined them as peritoneal lesions (peritoneal ones) to compare with OEM. In this way, the heterogeneity of the cellular composition of the three pathologic types of endometriosis was evident (Figure S1G–I). Overall, more stromal cells and fewer immune cells are observed in OEM (Figure 1E). In GSVA enrichment analysis (Figure S1F), ‘interleukin-1 receptor binding’ and terms containing ‘chemokine’, ‘fibroblast growth factor receptor’ and ‘type 2 fibroblast growth factor receptor’ GO_MF are enriched in endometriosis, which remind us that endometriosis is an inflammatory disease and closely relate to fibrosis. DIE was enriched with more ‘ferric iron binding’ and ‘type 2 fibroblast growth factor receptor’ GO_MF (Figure S1F), suggesting that DIE might associate with fibrosis and iron overload. Next, let us look at each of the cell classifications.

2.3 Epithelial cells

Histologically, endometriosis is diagnosed by the existence of ectopic stromal and epithelial cells. Firstly, we define epithelial cells by KRT18 and EPCAM (Figure 2A–C). Ciliated and luminal/glandular cells were firstly annotated in single-cell sequencing studies of the endometrium^{14, 17, 18}; however, it was not possible to differentiate between luminal and glandular epithelium in our data, probably because the number of epithelial cells was too small. In addition, two populations worth mentioning are Epi-COL1A1 and Epi-ACTA2, expressing fibroblast-associated genes—COL1A1 and ACTA2 (Figure 2A). GSVA scores of epithelial cells showed that these two populations associate with angiogenesis and undergo EMT process (Figure 2D), and both of them evidently expressed ZEB2 (Figure 2E), a gene that is a marker gene in the EMT process. Above all, the two subpopulations are not classic epithelial cells, which may be experiencing EMT process and acquiring function of angiogenesis. As a result, FN1, one of the mesenchymal marker genes, is highly expressed in Epi-COL1A1, Epi-ACTA2 and Mesothelial cells (Figure 2F). And what confuses us is that Epi-COL1A1 has a significantly high account in OEM lesions (Figure 2G). Thus, they deserve more and deeper researches in the future.

2.4 Stromal cells

Overall, stromal cells are the most complex and numerous, occupying 12 clusters with a total of 40 469 cells (Figure 3A). Stromal cells can be subdivided into three major groups (Figure 3A): fibro C7, which expresses complement C7; perivascular cells, which express RGS5, MYH11, STEAP4 and RERGL (Figure 3F); and other endometrial fibroblasts, in which the eS (non-decidualised endometrial stromal cells) and dS (decidualised endometrial stromal cells) are high in the expression of the hormone-associated genes of ESR1, PGR and PGRMC1 (Figure 3G–I). dS are characterised by the expression of HOXA10 and HOXA11 (Figure 3B,C). The HOXA gene family is associated with the development of the Müllerian duct, and its different genes correspond to different segments of the Müllerian duct, with HOXA10 and HOXA11 corresponding to the uterine and oviduct.¹⁹ The dS subpopulation indicates that lesions may come from endometrium or endosalpin, supporting Sampson's theory of retrograde menstruation.

And few differences in cell component are observed in stromal cells. Fibro_FN1 was less abundant in PEM than DIE (4% vs. 8%; Figure 3E).

2.5 Endothelial cells

Endothelial cells, which are part of the stroma (Figure 4A,B), are independent because they characteristically express VWF and CDH5 (Figure 1B) and are a continuous vascular system with fluctuated expression of gene markers in endothelial subpopulations (Figure 4A). EndoMT co-expresses endothelial and fibroblastic marker genes, which may suggest that it undergoes endothelial–mesenchymal transition.²⁰ (Figure 4A)

Perivascular cells surround endothelial cells in capillaries and veins throughout the body. Perivascular cells (Prv_MYH11, Prv_STEAP4, VSMC) were more abundant in PEM than DIE (35% vs. 23%; Figure 4C). Perivascular cells, also called pericytes, are most in PEM among three pathological types, which interacted with endothelial cells intimately.²¹ Ligand–receptor analysis performed by Cellphone DB infer pericyte interacting with endothelial cells by ‘VEGFA_KDR’, ‘VEGFA_FLT1’, ‘THBS1_integrin_aVb3_complex’, ‘THBS1_CD36’, ‘THBS1_integrin_a3b1_complex’, ‘ANGPT2_TEK’, influencing the process of angiogenesis (Figure 4D–F). Likewise, there were significant differences in endothelial cells (Figures 1E and 4C). Endothelial cells account more in peritoneal lesions and much less in OEM. OEM contained a higher percentage of ECtip and ECartery, which is the symbol of neovascularisation.²² Of the relatively small number of endothelial cells in the OEM, the neoplastic endothelium nevertheless accounted for the most (Figures 1E and 4C). The reasons of this phenomenon are worth to be explored. Post capillary vessel (PCV), also called high endothelial venules, is pivotal for lymphocyte recirculation and immune surveillance,²³ is significantly more prevalent in peritoneal lesions than in OEM (Figure 4C). PCV is an important channel for lymphocytes entering into local lymph nodes from circulation to exert their immune surveillance.^{23, 24} The low levels of PCV in OEM may related to ovarian immunosuppression, perhaps leading that endometriotic cysts are difficult for the immune system to eradicate and are prone to recurrence.

2.6 Immune cells

Immune cells can be separated into two main groups—the lymphoid lineage and the myeloid lineage (Figure 5A,D). In the lymphoid lineage, we annotated T cells, B cells and proliferative cells (Figure 5A,B). Proliferative cells highly express TOP2A and MKI67, suggesting high proliferative activity (Figure 5B). The unknown population co-expresses markers for T cells and fibroblasts (Figure 5B), which could not be annotated.

Myeloid lineage can be further continued into three major subpopulations (Figure 5D–G): DCs, referring to the annotation method and nomenclature of DCs by Heger et al.²⁵ and Tan et al.¹⁴; monocytes; and macrophages. cDC1s and LAMP3+ DCs are likely to be immunosuppressive dendritic cells because of their expression of IDO1 (Figure 5F), and protein encoded by this gene plays a key role in inducing DC to immunosuppressive phenotype through L-Tryptophan (Trp) metabolism via the kynurenine pathway (KP).²⁶ Macrophages can be subdivided into five subpopulations (Figure 5B), revealing that macrophage was much more complex than M1/M2 dichotomy. MФ-C1q expresses C1QA, C1QB (Figure 5B) and in either tumours or normal tissues, complement C1q is the marker of immunosuppressive macrophages.²⁷ Moreover, C1Q subpopulations co-express CD163 and MRC1 (CD206; Figure 5G), which are markers of M2 cells. In addition, MФ-CD3, a T-cells receptor-positive macrophage, was again supported by our data (Figure 5B).²⁸

Regarding lymphocytes, NK cells and CD8+ T cells were less in OEM, whereas plasma cells were more in OEM (Figure 5H). The difference among types may indicate an immunosuppressive microenvironment of the ovary.

2.7 Mesothelial cells

Mesothelial cells have also been reported in previous single-cell transcriptome sequencing articles on endometriosis.^{14, 17} Previous study reported that mesothelium in women with endometriosis experienced EMT,^29-33 which means that mesothelial cells lose part of epithelial features and acquires characteristic typical of mesenchymal cells. According to the literatures, CALB2 is a protein expressing in mesothelial cells but not in female reproductive system cells.^{34, 35} Considering that mesothelial cells may experience EMT, we defined CALB2+/CD10− epithelial and stromal cells as mesothelial cells (Figure 6A), whereupon we found mesothelial cells in each pathologic type as well as in each sample (Figure 6B). For peritoneal lesions directly plant on peritoneal mesothelial cells, it is reasonable mesothelial cells exist in PEM and DIE, but we also find them in OEM, which surprises us. So, we prove that with immunohistochemistry. We apply CALB2 antibody in OEM slice, and find mesothelial cells (Figure 6C). As for the different proportion of mesothelial cells in different types of lesions, they may be related to the volume of lesions and sampling, and have less practical significance.

2.8 Mesothelial cells experience different level of EMT process in different pathological types of endometriosis

Since mesothelial cells exist in PEM, DIE and also OEM (Figure 6B), then we commit to exploring their similarities and discrepancies. In GSVA analysis of epithelial cells, we found that the predominant epithelial cell type in which EMT occurs is mesothelial cells, but not ectopic endometrial epithelium (Figure 2D). GSEA analysis of mesothelial cells with different pathologic types revealed that EMT-related hallmarks were significantly upregulated in DIE mesothelial cells compared with those in PEM (Figure 6D). Also, we apply AddModuleScore function to score EMT level, demonstrating that mesothelial cells in DIE lesions underwent a stronger EMT process than the other two types with statistical significance (Figure 6E and Supporting Information Table 3. EMTscore gene. Supporting Information Table 4. AddModuleScore result–EMTscore). Then, we did differential gene analysis on mesothelial cells of DIE and PEM (Figure 6F), which showed that 159 genes were differentially expressed in DIE mesothelial cells, of which 95 were involved in EMT (Figure S3A and Supporting Information Table 2. EMT-associated genes, https://www.genecards.org/), such as, TAGLN2, CHI3L1, FN1, IGF1, TGFBI, ANXA4, are all the highest in the DIE (Figure 6G). At the individual cell level, FN1 expression is significantly higher in mesothelial cells from DIE lesions than that from PEM and OEM (Figure 6H).

2.9 Endometriosis-associated mesothelial cells interacted with ectopic endometrial stromal cells via FN1-integrin complex

Although stromal cells are complex and heterogeneous, in the analysis of interactions between all stromal cell types and epithelial cells, we found strong interactions between each stromal cell type and mesothelial cells regardless of pathological types (Figure 3C–E), suggesting that mesothelial cells in endometriotic microenvironment may have a specific influence on the stromal components of ectopic endometrium.

By performing ligand–receptor analysis of all epithelial and stromal cell types by CellPhone DB (Figure 7A), we found that mesothelial cells in the epithelium interacted strongly with individual stromal cells, and so we hypothesised that mesothelial cells underwent various degrees of the EMT process, which then induced various degrees of the endometriotic stromal cell progesterone resistance via AKT pathway.^36-40 Between mesothelial and stromal cells, results of CellPhone DB showed a total of 268 statistically significant receptor–ligand pairs (Supporting Information Table 6. Cellphone DB results). After screening, there were a total of six genes in the above ligand pairs that were both involved in EMT and differentiated genes in mesothelial cells among the three types of lesions (Figures S3F and 7C). Of these, FN1 was the most clearly differentiated among mesothelial cells in the three types of endometriosis, both in terms of average cellular expression (Figure 7C) and in terms of expression analysis at the level of individual cells (Figure 6H). Thus, FN1_integrin_a8b1_complex, FN1_integrin_a5b1_complex, FN1_integrin_aVb5_complex, FN1_integrin_aVb1_complex, FN1_integrin_a11b1_complex and FN1_integrin_a3b1_complex are determined to be the major receptor ligand pairs of mesothelial–stromal cell interactions (Figure 7B), which may plays an important role in intercellular junctions and adhesion,⁴¹ inducing the initial planting of retrograde menstruation.

2.10 Endometriosis-associated mesothelial cells may induce different progesterone resistance of ectopic stromal cells via FN1-PI3K-AKT pathway

Previous studies have shown that PGR and PR levels are significantly lower in ectopic lesions than in normal or eutopic endometrium,⁵ but there is an inconsistency in the effectiveness of oral contraceptives with progestins or oestrogen–progestogen combinations in the treatment of different types of endometriosis (Supporting Information Table 5. Progesterone efficacy in clinical practice).^{5, 42-54} In our data, PGR expression is low in stromal cells of endometriosis lesions with significant individual heterogeneity (Figure 7D). Therefore, we hypothesised that the differences in progesterone resistance were due to abnormalities in progesterone signalling. Previous studies have found that the AKT pathway,^{37, 40} downstream of progesterone signalling was abnormally activated in ectopic stromal cells. Meanwhile, GSEA analysis of stromal cells revealed that stromal cells from peritoneal lesions were enriched with more PI3K-AKT pathway-related genes than those of OEM (Figure 7E). According to a study of platinum resistant ovarian cancer,³⁸ peritoneal mesothelial cells, which experience EMT process, can activated ovarian cancer cells’ AKT pathway by FN1 expression on mesothelial cells, inducing to platinum resistance. Similarly, we surmise that endometriosis-associated mesothelial cells, which experience different EMT process and express different FN1, induced different progesterone resistance of endometrial stromal cells via the FN1-PI3K-AKT pathway. In order to prove this inference, we need further experiments in the future.

3.1 Tissue processing

Surgical specimens were obtained with the informed consent of the patients, and the study was approved by the Ethics Committee of Capital Medical University Affiliated Beijing Chaoyang Hospital (Ethics File No. 2016-11-1-2).

After surgical sampling, non-target tissues were eliminated, blood stains were washed with saline and tissues were placed in brown sample preservation tubes containing MACS® Tissue Storage Solution (Miltenyi Biotec), transported to the laboratory at 4°C, and tissue dissociation experiments were carried out within 1 h. Tissues were removed from the sample storage tubes, washed in cooled phosphate-buffered saline (PBS), cut into 2–3 mm pieces and dissociated for 40 min at 37°C using 3 mL of enzyme dissociation solution, which consisted of 10% fetal bovine serum (FBS) + 1 mg/mL collagenase V (Sigma: C5138)+ .1 mg DNaseI (Roche: 10104159001), and the reaction was terminated by the addition of 20 mL of Roswell Park Memorial Institute (RPMI) medium with 10% FBS after the enzyme dissociation, and then filtered through a 70 um cell sieve, and the cells were recovered by centrifugation at 4°C and 450 × g for 5 min. The cells were collected by centrifugation at 450 × g for 5 min, the supernatant was discarded and the cells were washed once using 10 mL PBS.

The undissociated tissue on the cell sieve was dissociated for 20 min at 7°C using 10 mL of enzyme dissociation solution, which consisted of .25% Trypsin EDTA + .1 mg/mL DNaseI enzyme dissociation solution. After dissociation, the reaction was terminated by the addition of 20 mL of RPMI medium with 10% FBS, filtered through a 70um cell sieve, and the cells were recovered by centrifugation at 450 × g for 5 min at 4°C, and the supernatant was discarded. The cells were recovered by centrifugation at 450 × g for 5 min at 4°C, and the supernatant was discarded.

After dissociation of collagenase and Trypsin, the cells were mixed and lysed with Red Blood Cell Lysis Solution (Miltenyi Biotec), washed once with 1× PBS and finally resuspended with PBS + .04% bovine serum albumin (BSA) + RNA inhibitor (1 U/uL), LUNA-FL™, and LUNA-FL™. Automated Fluorescence Cell Counter machine, AOPI dye stained the cells to detect cell activity, concentration, clumping rate and other parameters.

3.2 Single-cell capture, library preparation, sequencing

Chromium Next GEM Single-cell 3ʹ Reagent Kits v3.1 (Dual Index) kits, each sample according to 16 000 up-sampling to complete water-in-oil generation, RT-PCR amplification, cDNA amplification and library construction.

Download the kit manual at: CG000315_ChromiumNextGEMSingleCell3-_GeneExpression_v3.1_DualIndex__RevE.pdf (ctfassets.net)

Library mass concentration was detected using Qubit 4.0 machine Qubit™ 1X dsDNA Assay Kits, high sensitivity kits, StepOnePlus™ Real-Time PCR System for library molar concentration, LabChip Touch for library insertion fragments, illumina StepOnePlus™ Real-Time PCR System for library molar concentration, LabChip Touch for library inserts, and illumina Novaseq6000 Sequencing Platform PE150 for read length sequencing.

3.3 Single-cell data quality control and analytical processing

Illumina base call files for all libraries were demultiplexed and converted to FASTQ files using bcl2fastq v.2.20.0.422 (Illumina). Raw reads were compared to the hg38 reference genome (GRCh38 10X Genomics reference 3.0.0), matrix files were generated using cellranger 6.0.1 (10X Genomics) for raw data, and unique molecular indentifier (UMI) count matrices were processed using the R package Seurat (v.3.0.0). The resulting count matrices were further processed using the Scanpy package (v.1.7.1) to exclude genes detected in fewer than three cells and the top 2000 highly variable genes, and to exclude cells with parameters such as (1) fewer than 200 genes, (2) fewer than 1000 UMIs, (3) more than 100 000 UMIs and (4) maximum mitochondrial content of 20%. The filtered matrix was library normalised in Seurat to obtain normalised counts. Principal component analysis and neighbourhood graph generation were performed based on highly variable gene sets. Harmony (v.0.1.0) was used to minimise batch effects.

Principal component with batch effects removed was downscaled using UMAP (dimension 30, resolution .6). Differentially expressed genes in each cluster were compared to previously documented sets of cell type marker genes to annotate cell types, lineages and sub-lineages.

Analysis were executed with functions implemented in Scanpy (1.7.1). Similarities among samples were based on hierarchical clustering calculated from Pearson correlation using the Ward linkage algorithm. In Seruat, AddModuleScore function was used to calculate the combined scores of the expression levels of a collection of different genes in each single-cell, which are computed by averaging the expression levels of all genes in each program at the single-cell level and subtracting the aggregated expression levels of a randomly selected set of control features. A supporting information table (Supporting Information Table 3. AddModuleScore–EMTscore gene) lists the genes used for the EMT score. EMT-associated genes refers to https://www.genecards.org/. Gene ontology enrichment analysis and Kyoto Encyclopedia of Genes and Genomes pathway enrichment analyses were applied with the clusterProfiler in the R package for differential genes. GSEA analyses were performed by applying GSEAv.3.0 The GSEA analysis was performed using the GSEAv.3.0 software (http://software.broadinstitute.org/gsea/index.jsp).^{55, 56} We used the R package ‘GSVA’ for GSVA analysis.^{57, 58} To explore the interactions between different cell types in ectopic endometrium, we used CellPhone DB to do cell-cell interaction analysis of all epithelial cell subpopulations and stromal cell subpopulations.^{59, 60}

3.4 Immunohistochemistry

Briefly, 5-µm thick formalin-fixed paraffin embedded (FFPE) endometriotic tissue sections were stained with the Anti-Calretinin (ab92341 abcam, 1:100 dilution) antibody per manufacturer's instructions at 4°C overnight after dewaxing, antigen retrieval and blocking. Antigen retrieval was performed by incubating tissue slices in citric acid buffer (pH 6.0) at 95°C for 20 min. Samples were washed and incubated with a second antibody (goat anti-rabbit IgG H&L [HRP], Abcam: ab205718) at room temperature for 1 h. The immunohistochemistry signal was detected by 3,3′-diaminobenzidine substrate buffer (ZLI-9017, ZSGB-BIO) and counter-stained by haematoxylin (ZLI-9610, ZSGB-BIO). The mouse brain tissue pairs served as positive controls. All the staining results were scanned and determined by the Flexacam (Leica) scanning system in the Medical Research Center of Beijing Chaoyang Hospital.