Single-cell analysis of menstrual endometrial tissues defines phenotypes associated with endometriosis

Menstrual effluent (ME) was collected as previously described [11, 15]. Briefly, women of reproductive age (N = 33, age 20–45 years, average age 33.6 years) living in North America who were not pregnant or breastfeeding, who were menstruating, and who were willing to provide ME samples were recruited mainly via social media and consented to the ROSE study (IRB#13-376A; https://​feinstein.​northwell.​edu/​institutes-researchers/​institute-molecular-medicine/​robert-s-boas-center-for-genomics-and-human-genetics/​rose-research-outsmarts-endometriosis). Women with histologically confirmed endometriosis (determined following excision laparoscopic surgery and documented in a pathology report without revised American Society for Reproductive Medicine (rASRM) staging/classification) were enrolled as “endometriosis” subjects (N = 11). Women who reported chronic symptoms consistent with endometriosis (e.g., recurrent dysmenorrhea, persistent abdominal bloating, dyspareunia, dysuria, and/or dyschezia), but not yet diagnosed with endometriosis (or not) were enrolled as ‘symptomatic’ subjects (N = 13). Control subjects living in North America who self-reported no gynecologic history suggestive of a diagnosis of endometriosis (and the absence of polycystic ovarian syndrome, and pelvic inflammatory disease) were recruited mainly via social media and enrolled as “controls” (N = 9).

Endometriosis, symptomatic, and control subjects collected their ME using an “at home” ME collection kit for 4–8 h on the day of their heaviest menstrual flow (typically day 1 or 2 of the cycle) with a menstrual cup (provided by DIVA International), except for one subject who collected ME using a novel menstrual collection sponge (as previously described [15]). After collection, ME was shipped priority overnight at 4^°C to the laboratory for processing. ME collected from menstrual cups was mixed 1:1 with DMEM for processing. For the saturated menstrual collection sponge, ME tissue was collected after rinsing the sponges with PBS to collect cells and tissue. Demographic and gynecologic/health data (including hormone usage, menstrual cycle information, and pain/pain medications) for controls, endometriosis subjects, and symptomatic subjects (and the total cohort) are shown in Table 1.

ME-derived tissue fragments were obtained from controls, symptomatic subjects, and endometriosis patients (n = 2 each); tissue fragments were collected by pouring ME over a 70μ filter, fixed, and transferred to the clinical pathology lab for paraffin embedding and hematoxylin and eosin (H&E) staining. CD10 was chosen for immunohistochemical analysis because it is a sensitive marker of eutopic endometrial stroma [16] and because adjunctive use of CD10 immunostaining with H&E staining enhances the histologic detection of endometriosis [17]. CD56 was chosen because uNK cells stain brightly with CD56. H&E slides and immunostained slides were examined microscopically and imaged by a pathologist. Representative images are shown in Fig. 1.

Whole (unfractionated) ME (2.5–10 ml) was digested with Collagenase I (1 mg/ml, Worthington Biochemical Corporation, Lakewood, NJ) and DNase I (0.25mg/ml, Worthington Biochemical Corporation) at 37 °C for 10–30 min using the gentleMACS^TM Tissue Octo Dissociator (Miltenyi Biotec, Cambridge, MA) using C tubes and Program 37CMulti_E_01 (31 min). After digestion, the sample was sieved over a 70μ filter and washed with DMEM 10% fetal bovine serum (FBS) to neutralize digestion enzymes; the flow through was sieved over a 40μ filter and washed with DMEM 10%FBS. After collecting the single cells (from the flow through) following centrifugation (350×g for 5 min), neutrophils were removed using the EasySep^TM HLA Chimerism Whole Blood CD66b Positive Selection Kit (STEMCELL, Cambridge, MA), according to the manufacturer’s protocol. The neutrophil pellet was frozen at – 80 °C and used as a source of subject DNA for genotyping (see below). The resultant cells were depleted of red blood cells using the EasySep™ RBC Depletion Reagent (STEMCELL), according to the manufacturer’s protocol, and then washed and subjected to density gradient centrifugation using Ficoll-Paque PLUS (Sigma-Aldrich, St. Louis, MO) to collect mononuclear cells, according to manufacturer’s directions. To collect ME-tissue, whole ME (2.5–10 ml) was sieved over a 70μ filter and washed with DMEM; the ME-tissues trapped on the filter were collected and digested with Collagenase I (1mg/ml, Worthington Biochemical Corporation, Lakewood, NJ) and DNase I (0.25mg/ml, Worthington Biochemical Corporation) at 37 °C for 10 min and processed as described above for whole ME, except without a density gradient centrifugation step. The resultant whole ME cells were enumerated, and viability was assessed using ViaStain™ AOPI Staining Solution and the Nexcelom Cellometer (Lawrence, MA). Preparations with > 80% viability were processed for scRNA-Seq. Cells were immediately fixed in methanol for scRNA-Seq, as described by Chen for peripheral blood mononuclear cells [18]. Briefly, cells were washed and resuspended in a 200 μl Ca⁺⁺ and Mg⁺⁺-free PBS, followed by dropwise addition of chilled 100% methanol (800 μl, final 80% w/v). Fixed cells were stored at – 20 °C for 20 min and then stored at – 80 °C until used for scRNA-Seq. A pilot experiment was performed with a single ME sample, which was processed and either prepared immediately for scRNA-Seq (without methanol fixation and freezing) or was fixed in methanol and frozen, as described above to optimize our scRNA-Seq methods. The data showed almost identical scRNA-Seq results using both methods (see Additional file 1) reassuring us that the methods of Chen et al. [18] can be applied to ME samples. Thus, all ME samples were methanol fixed and frozen. An advantage of this approach is that it allows cost-effective pooling of samples collected at different times and reduces the potential batch effects of running samples separately for scRNA-Seq.

Methanol-fixed cells were removed from – 80 °C and placed on ice for 5 min before centrifugation (1000×g for 5 min). Methanol-PBS supernatant was completely removed and cells were rehydrated in 0.04% bovine serum albumin (BSA) + 1 mM dithiothreitol (DTT) + 0.2 U/μl RNase Inhibitor in 3X SSC (saline sodium citrate buffer solution) Buffer (Sigma). An aliquot of fixed cells was stained with Trypan Blue and visualized under the microscope. The cells were counted and pooled from different donors at equal ratios, filtered using 40 µ strainer (Falcon), recounted and brought up to a final conc. of 2000 cells/μl, and proceeded immediately for GEM generation and barcoding on a 10X Chromium using Next GEM 3’ v3.1 reagents (10X Genomics). Libraries were constructed following 10X Genomics’ recommendations and quality was assessed on a High Sensitivity DNA chip on a BioAnalyzer 2100 (Agilent) before loading (1.8 pM) and sequencing on an Illumina Nextseq 500 using a High Output kit v2.5 (150 cycles).

Five subjects were pooled together into a single 10X lane with at least one of each phenotype per run with a total of 8 runs for ME-tissue and 3 runs of whole ME. The ME-tissue runs had 44,135 total cells, of which 5147 had ambiguous calls in Demuxlet, 2632 were doublets, and 36,356 were singlets; only the singlets were analyzed. The whole-ME runs had 30,090 total cells, of which 5556 had ambiguous calls, 2776 were doublets, and 21,758 singlets (and hence analyzed). A total of 43,054 cells were analyzed in this study following filtering and QC (thresholds of > 10% mitochondrial reads < 500 nUMI (number of unique molecular identifiers) or > 50000 nUMI or > 6000 unique features per cell).

Samples were converted from raw bcl files to gene by cell matrices using CellRanger 6.0 aligned to 10x Genomics’ GRCh38-3.0.0 reference. Individuals were demultiplexed via Demuxlet [19] using genotypes taken from SNPs on the Illumina GSAv3 genotyping array, run on DNA prepared from neutrophils isolated from ME. The thresholds in Demuxlet were adjusted to the expected doublet rate, and those marked as doublets were removed. Downstream analysis and visualization were done using Seurat 4.0 [20]. Briefly, there were at least 25,000 reads per cell on average per 10X run, and the mean number of genes captured was 1388 (± 896) (mean ± standard deviation [SD]). There was no significant difference between the various clinical groups (controls, cases, symptomatics) in these values. Genes were filtered out if they were expressed in less than 3 cells while cells were filtered out if they had > 10% mitochondrial reads, 500 < nUMI < 50000 and > 6000 unique features. For the analysis of ME-tissue samples, only subjects with information on at least 500 cells per subject were retained. After filtering, the cell yields were comparable in each group (mean ± SD: 1256 ±732 and 1319 ±767 in ME-tissue and whole ME, respectively). Gene expression normalization and cell clustering was done using the SCTransform pipeline [21] with percent mitochondrial reads regressed out and person specific batch effects corrected using Harmony [22]. Identification of cell clusters was done using known marker genes (Additional file 2) [23‐30] with differential gene expression calculated using a Wilcoxon rank sum test. Enrichment of cell clusters of specific phenotypes was done using MASC (mixed-effects modeling of associations of single cells) (https://​github.​com/​immunogenomics/​masc), which essentially uses a percentage of cells per cluster while also taking into account technical covariates; 10X library batch, preparation (whole ME or ME-tissue), nUMI per cell, percent mitochondrial reads, and phase are accounted for. All datasets are deposited in the National Center for Biotechnology Information/Gene Expression Omnibus (GEO) accession number GSE203191.

We carried out histological assessment of fresh menstrual effluent (ME)-associated tissues isolated from ME. Representative H&E sections of ME-derived tissue fragments from four subjects (1 control, 2 laparoscopically/histologically confirmed endometriosis subjects, and 1 symptomatic subject) show the presence of endometrial tissues with mucosal and glandular epithelium and areas of stroma. The endometrium had typical late secretory/menstrual morphology with expanded stroma containing scattered inflammatory cells and secretory and inactive-type glands (Fig. 1A–D, upper panels). Immunostaining of ME-derived tissue sections reveals a range of stromal cells stained with antibodies to CD10, a clinically used marker of endometrial stroma [16, 17], and an abundance of uNK cells (stained with antibodies to CD56 (NCAM), an archetypical marker of NK cells [Fig. 1A–D, lower panels]).

We have analyzed ME samples from 33 subjects, including age-matched healthy controls (N = 9), endometriosis cases (N = 11), and subjects with chronic symptoms suggestive of endometriosis but not yet diagnosed (N = 13) (see Table 1). ME samples from either whole ME (unfractionated) or ME samples enriched for tissues (“ME-tissue”) were digested with collagenase I and DNase I, depleted of neutrophils, and processed for scRNA-Seq, as described in the methods. As shown in Fig. 2, a graph-based clustering approach using Seurat distinguishes multiple cell clusters shown on the UMAP (uniform manifold approximation and projection) plot. There is striking diversity of the cell types defined by the cluster analysis. A major group of uterine NK cells is designated cluster uNK1, with a small associated cluster designated uNK2. Sets of clusters related to CD8+ and CD4+ T cells are shown in the central portion of the plot. Endometrial stromal cells and epithelial cells are identified in major clusters in the right side of the UMAP plot. Subclusters of endometrial stromal cells are described below in detail. Based on [31], Epithelial1 appears to be a mix of lumenal and glandular epithelial cells, Epithelial2 is comprised of ciliated epithelial cells, and Epithelial3 is a separate set of CD326-expressing cells that do not overlap with Epithelial1 or Epithelial2. Distinct clusters of B cells and myeloid cells can also be delineated, along with a small cluster of plasmacytoid dendritic cells (pDC). The positive gene markers used to generate the cell clusters shown in Fig. 2 are included in Additional file 2. Overall, the various cell clusters are well represented whether unfractionated whole ME or tissue-enriched ME is processed for scRNA-Seq. Some differences in cell subset frequencies can be observed; in particular, epithelial cells were enhanced when tissue-enriched ME was utilized for sample processing (see Additional file 3).

We compared the relative frequency of the various cell clusters in the freshly processed ME obtained from the diagnosed endometriosis cases (N = 11) compared with controls (N = 9), as shown in Fig. 3. By inspection of Fig. 3, it is apparent that both clusters of uNK cells (uNK1 and uNK2) are markedly depleted in the cases vs. controls (average percentage of uNK approximately 8% in cases, 28% in controls), as well as an increase in the proportion of B cells in cases (~9%) vs. controls (~3%). The odds ratios and confidence intervals for these two cell enrichment patterns are shown in Fig. 4, along with the patterns of enrichment of all the other major cell clusters. While there is some variation among many of the different cell clusters, a formal analysis shows the most striking differences are observed for uNK cells, which are enriched in controls (and depleted in cases; uNK1, P < 10E−16; uNK2, P < 10E−16), along with a relative enrichment in the proportion of B cells in the cases diagnosed with endometriosis (and relatively depleted in controls; P < 10E−16). Note that the stromal cell cluster is not significantly different between cases and controls (P > 0.05).

We also explored whether the various proportions of cell clusters of the ME preparations from the “symptomatic” but undiagnosed group of subjects (N = 13) are different from ME preparations from controls. This is clearly the case, as shown in Additional file 4. Here, we show the relative enrichment of uNK cells is maintained in controls in comparison to the symptomatic group (uNK1, P < 10E−16; uNK2, P = 0.0025), similar to that observed with ME from cases. B cells also show a significant relative enrichment in symptomatic as well as diagnosed cases, compared with controls (symptomatic vs. control, P = 5.8 × 10⁻⁶), similar to that observed with ME from cases (Additional file 4). Perhaps not surprisingly, these significant differences in symptomatic cases vs. controls are less striking than the differences in endometriosis cases vs. controls, given the likely heterogeneity of the symptomatic group.

Previous studies have reported reduced decidualization capacity in endometrial stromal cells grown from biopsies of patients with endometriosis [32]. We have also observed impaired decidualization using stromal cells grown directly from ME [11, 15]. Therefore, we examined whether this trend could be observed in fresh stromal cells analyzed by scRNA-Seq. The stromal cell numbers or percentages did not significantly differ between the control and endometriosis groups, as shown in Figs. 3 and 4. However, subclustering of the stromal cell cluster clearly identified 5 subclusters of interest within the stromal cell population (Fig. 5A). We have designated these subclusters based on the dominant transcripts expressed in each of these subclusters, as shown in the violin plots in Fig. 5B. Two of the five subclusters (2 and 4) are not different between cases and controls (the top genes of subclusters 2 and 4 are described in Additional file 5). The subclusters showing significant enrichment in either cases or controls (subclusters 1, 3, and 5) are indicated by the Log₂ (odds ratios, [OR]) below the UMAP plot (Fig. 5C).

It is striking that an apparently decidualized stromal cell subcluster (expressing IGFBP1 mRNA) is significantly enriched in controls compared with endometriosis cases (Fig. 5A, B). In addition to IGFBP1, the top differentially expressed genes in this subcluster (compared to other stromal cell subclusters) include LEFTY2, DCN, LUM, MDK, C1QTNF6, APOE/D, DCN, and other progesterone sensitive and decidualization/fertility gene markers (see left panel (subcluster 3) in Fig. 6 and Additional file 6 [33‐112]. This suggests that a phenotype of “decidualization” can be measured directly in stromal cells derived from fresh ME and is associated with control vs. disease phenotype. A modest enrichment of a subcluster expressing IL11 was observed in cases, as indicated in Fig. 5A–C. In addition to IL11, this subcluster is associated with transcripts for MMP3, MMP1, MMP9, SERPINB2, S100A6, and CXCL8, among other genes associated with inflammation, fibrosis, and senescence, as well as endometriosis, as shown in the middle panel (subcluster 1) of Fig. 6 (and Additional file 6). A third subcluster, designated by high expression of the gene encoding matrix Gla protein (MGP), is also enriched in the stromal cells of cases (Fig. 5A–C). This subset expresses numerous extracellular matrix genes that have been associated with presence of perivascular stromal cells, senescence, and cell adhesion/cell spreading, including FN1 (which encodes fibronectin-1), a known risk locus for endometriosis [113]. Figure 6 (right panel (subcluster 5) and Additional file 6) also shows the list of top genes expressed in this subset. Additional file 7 demonstrates that the IGFBP1+ and MGP+ subclusters map to stromal cells subsets defined in the decidua found in the first trimester of pregnancy by Vento-Tormo et al. [26].

We also examined the differences between cases and controls in the two uNK subclusters present in digested endometrial tissues in ME (uNK1 and uNK2, see Fig. 2). We noted a distinct subcluster of uNK cells (uNK2) that is characterized by the expression of genes associated with cell proliferation such as MKI67 which encodes Ki67 and TOP2A which encodes topoisomerase 2A (see Additional file 8 for a full uNK subcluster analysis). As discussed below, this cluster also mapped nearly exactly (97%) with a proliferative subset of uNK cells that has been defined by scRNA-Seq in decidua obtained during the first trimester of pregnancy [26]. This is consistent with the proliferation of uNK cells and overall accumulation of uNK cells in the course of decidualization in control subjects vs. cases, as shown in Figs. 3 and 4.

Finally, in order to address the possibility that the use of hormone treatment by one endometriosis subject (see Table 1) might have affected our results, we performed a reanalysis of cases and controls after eliminating this subject. The results are not significantly changed.

Specifically, the overall spatial distribution of the UMAP is not different when the cells from the endometriosis subject on hormones were removed from the data (Additional file 9), and the log₂ odd ratios (OR) for the cell subsets enriched in controls and cell subsets enriched in cases are not different (Additional file 10). Additionally, the clustering of the endometrial stromal cells is not different (Additional file 11) from the original set (with all cases). Comparing all differentially expressed genes with/without the hormone sample in stromal cell subsets and correlating log fold changes of significant genes within each subcluster shows they are highly correlated, R> 0.997 for all samples (data not shown).