Spatial transcriptomic analysis identifies epithelium-macrophage crosstalk in endometriotic lesions

Summary

The mechanisms underlying the pathophysiology of endometriosis, characterized by the presence of endometrium-like tissue outside the uterus, remain poorly understood. This study aimed to identify cell type-specific gene expression changes in superficial peritoneal endometriotic lesions and elucidate the crosstalk among the stroma, epithelium, and macrophages compared to patient-matched eutopic endometrium. Surprisingly, comparison between lesions and eutopic endometrium revealed transcriptional similarities, indicating minimal alterations in the sub-epithelial stroma and epithelium of lesions. Spatial transcriptomics highlighted increased signaling between the lesion epithelium and macrophages, emphasizing the role of the epithelium in driving lesion inflammation. We propose that the superficial endometriotic lesion epithelium orchestrates inflammatory signaling and promotes a pro-repair phenotype in macrophages, providing a new role for complement 3 in lesion pathobiology. This study underscores the significance of considering spatial context and cellular interactions in uncovering mechanisms governing disease in endometriotic lesions.

Graphical abstract

Highlights

• •Minimal transcriptome changes were found in the lesion stroma
• •Immune response genes were increased in the lesion epithelium, including C3
• •Epithelium-to-macrophage signaling was 3.7-fold higher in lesions

Reproductive medicine; Integrative aspects of cell biology; Organizational aspects of cell biology; Transcriptomics

Introduction

Endometriosis affects around 10% of reproductive age women and is defined by the presence of endometrium-like tissue growing outside of the uterus.^1,2,3 The disease is clinically associated with life-altering pain and infertility and up to 50% of infertile women are diagnosed with endometriosis.^4,5 Endometriotic lesions are estrogen-dependent and the only effective non-surgical treatment is suppression of ovarian function to limit estrogen production. Endometriotic lesions in the peritoneal cavity are classified by location with lesions present on the visceral or parietal peritoneum termed peritoneal lesions and those on the ovary termed endometriomas. Peritoneal lesions are further separated into superficial or deep infiltrating according to the invasion depth of underlying tissue with deep infiltrating lesions invading more than 5 mm.^6,7 Superficial peritoneal endometriotic lesions likely develop from endometrial tissue fragments that are refluxed into the peritoneal cavity during menstruation.^8,9 This hypothesis is supported by the baboon model of endometriosis in which menstrual tissue inoculated into the peritoneal cavity results in endometriotic lesions and disease that persists for at least 15 months.^10,11,12 Endometriosis can also be induced in mice with tissue collected from an artificial menstruation cycle.^13,14

Superficial peritoneal lesions generally consist of endometrium-like epithelium and stroma, fibroblasts that express smooth muscle actin (ACTA2), and immune cells.¹⁵ The presence of ACTA2+ fibroblasts is positively correlated with fibrotic lesions where the presence of epithelium is less common, in contrast to lesions of endometrium-like stroma and epithelium.^16,17 During lesion development, activation of pro-inflammatory immune populations, specifically macrophages, provide a hospitable inflammatory environment to protect the ectopic endometrial tissue.¹⁸ Previous bulk and single-cell RNA-sequencing analyses have uncovered thousands of differentially expressed genes (DEG) in endometriotic lesions, notably including inflammation- and immune-related genes.^{19,20,21,22,23,24,25} While bulk RNA-sequencing offers comprehensive data, it lacks the ability to discern information about individual cell types. Conversely, single-cell RNA-sequencing provides insights into cell type-specific expression but relies on post-hoc cell type identification using known markers and lacks spatial context. The emergence of spatial transcriptomics has addressed these limitations by providing a powerful tool to understand the spatial context of gene expression within tissues, particularly when coupled with canonical cell type immunostaining. Two of the most prominent platforms for spatial transcriptomics, NanoString GeoMx and 10X Visium, offer distinct advantages and limitations depending on experimental design. In this study, we chose the GeoMx platform based on the ability to provide cell type specific data from formalin fixed paraffin embedded (FFPE) tissues with high specificity based on antibody-based fluorescent segmentation.^26,27,28 Our aim was to determine the genes and pathways altered in the glandular epithelium, stroma, and macrophages within endometriotic lesions compared to matched endometrium with spatial transcriptomics. We hypothesized that gene expression changes in superficial peritoneal endometriotic lesions would manifest in a cell type-specific manner, with disruption of the crosstalk among the stroma, epithelium, and macrophages being evident compared to patient-matched eutopic endometrium.

Results

Spatial transcriptomics of human superficial peritoneal lesions

We sought to determine the contributions of epithelium, stroma, and myeloid cell compartments of endometriotic lesions to altered genes and pathways in endometriotic lesions by spatial transcriptomic analyses. Superficial peritoneal lesions and eutopic uterine tissue were collected from women in the secretory phase of the menstrual cycle. Matched lesion and eutopic tissues from five women (n = 10) were selected based on the histological presence of endometrial gland-like structures in lesions confirmed by trichrome staining (Figure 1). The average age of the patients was 42 (Table 1, SD = 6.1). Tissues were stained for pan-cytokeratin, smooth muscle actin, and CD68 to identify epithelium (pan-cytokeratin+), stroma (ACTA2-) and myeloid-lineage cells (CD68⁺, monocytes and macrophages). Myeloid cells were present in both eutopic endometrium and endometriotic lesions within stromal and epithelial compartments (Figure 1). ACTA2+ cells were present in the vasculature and stroma, although not in the immediate sub-epithelial stroma. Duplicate regions of interest were selected for each tissue and segmented based on fluorescence intensity for macrophages, epithelium, and stroma, resulting in 60 segments.

Figure 1.

Segmentation strategy for spatial transcriptomics

(A) Superficial peritoneal endometriotic lesions and matched endometrium (n = 10) were visualized with trichrome staining to confirm the presence of epithelial and stromal compartments. Representative images of selected samples are shown.

(B) Tissues were immunostained for pan-cytokeratin, CD68, and ACTA2 for segmentation and spatial transcriptome collection on the GeoMx platform. Representative regions of interest (ROI) are shown for each tissue type. Duplicate ROI were collected for each tissue and stroma, epithelium, and macrophages were segmented from each for a total of 20 ROI and 60 segments. Scale bar, 100 μm.

Table 1.

Patient age, indications for surgery, and post-operative diagnosis

Patient	Age	Indication(s)	Diagnosis
2756	47	Uterine Leiomyoma Adenomyosis	Endometriosis L ovary and oviduct
4909	36	Uterine Leiomyoma Pelvic pain	Endometriosis, Leiomyoma
5322	49	Uterine Leiomyoma	Endometriosis L&R ovary, Leiomyoma
9997	36	Pelvic pain	Endometriosis L ovary and R oviduct, Pelvic adhesions
12529	44	Uterine Leiomyoma	Endometriosis L&R oviduct, Leiomyoma

Based on the limit of quantification, 10 segments were removed with a gene detection rate of lower than 10% (Figure 2A). Read counts were normalized for the remaining 50 segments using the quartile 3 (Q3) method. Separation between the geometric mean of the negative control probes, used to calculate the limit of detection, and Q3 of all gene read counts was confirmed at the segment and distribution levels to ensure compatibility with Q3 normalization (Figure 2B). Next, genes detected in less than 10% of segments were removed leaving 7,945 genes (Figure 2C). The normalized read counts for each segment, separated by cell type, were visualized by boxplots and to confirm similar distributions (Figures S1A–S1C). Separation of epithelium, stroma, and macrophages was confirmed with a dot plot of marker genes expression (Figure 2D). The mean z-scores for epithelial markers (n = 9), stromal markers (n = 6), and macrophage markers (n = 6) were plotted and demonstrated enrichment of appropriate marker genes for each cell type (Figures S1D–S1F). The segment areas were similar within cell types, with the order of decreasing area being stroma, epithelium, and macrophage segments (Figure S2).

Figure 2.

Spatial transcriptomics results from eutopic endometrium superficial peritoneal endometriotic lesions

(A) Bar plot of all 60 segments separated by gene detection rate. Segments with a detection rate of less than 10% were removed.

(B) Confirmation of gene and negative probe count separation by segment in a scatterplot and by distribution for each cell type in histograms. Separation after gene and segment filtering is required for effective Q3 normalization.

(C) Bar plot of detectable genes per segment used for filtering genes found in less than 10% of segments leaving 7,945 genes for further analysis.

(D) Dot plot of marker genes for epithelium (n = 9), stroma (n = 6), and macrophages (n = 6) confirming transcriptomic separation between cell types. See also Figures S1 and S2.

We compared the epithelium and stroma segments by tissue type to confirm cell type separation and identify differences in the cell identity in the lesion tissue. The eutopic endometrial epithelium was compared to matching stroma and 1,372 DEG corresponding to cell type were found (Figure S3A) with 521 increased, or epithelial, and 851 decreased, or stromal. Similarly, there were 1,313 DEG in lesion epithelium versus stroma (Figure S3B), with 579 increased and 734 decreased. Notably, 14 of the top 20 genes from each comparison were in common. Gene expression patterns were similar in the endometrium and lesion tissues with a correlation coefficient of 0.81 ($p<2.2\times {10}^{-16}$, Figure S3C). Stromal and epithelial biological process terms were distinct and consistent with cell type (Figure S3D). Vascularization, extracellular matrix, and cell adhesion terms were enriched in the stroma while epithelial cell differentiation and cell-cell junction terms were enriched in the epithelium. Terms were very similar in the stroma with 9 of 10 terms common between the eutopic endometrium and lesions. However, five terms were unique to the lesion epithelium, related to cell-cell adhesion and cell projections or cilia, and two terms to the eutopic endometrial epithelium, cellular transition metal ion homeostasis and circulatory system process.

Endometriotic stroma is minimally altered

The sub-epithelial stroma from lesions was compared to matched eutopic endometrium to determine transcriptome alterations in endometriotic lesions. Separation of eutopic stroma from the lesions was strongest on PC3, accounting for 12.37% of variation in the dataset (Figure 3A). Two downregulated genes were identified in lesions (Figure 3B), DHRS1, dehydrogenase/reductase 1, and SCGB2A1, secretoglobin family 2A member 1, both related to steroid metabolism. Six hallmark gene sets²⁹ were significant in a gene set enrichment analysis (GSEA), four increased and two decreased (Figure 3C). The increased gene sets were inflammation-related, TNFα signaling via NF-κB, interferon gamma response, interferon alpha response, and inflammatory response, and the decreased gene sets were MYC targets V1 and epithelial mesenchymal transition.

Figure 3.

Endometriotic sub-epithelial stroma is minimally altered compared to eutopic endometrium

(A) Principal component (PC) plot of lesion stroma compared to eutopic endometrium showed group separation most clearly on PC3.

(B) Volcano plot of two down-regulated genes in the endometriotic stroma.

(C) Gene set enrichment analysis of hallmark pathways found four increased inflammatory gene sets and two decreased.

Immune response is increased in the endometriotic lesion epithelium

The endometriotic epithelium was compared to the matched eutopic glandular epithelium and was most separated from the eutopic endometrial epithelium on PC3, representing 9.45% of variation in gene expression (Figure 4A). Differential expression analysis found 76 DEG with 33 increased in the lesion epithelium and 43 decreased (Figure 4B). Notably, multiple immune-related genes were increased including, C3, or complement 3, CD74, CLU, or clusterin, and major histocompatibility complex class II members HLA-DRB1 and HLA-DPB1. There were 10 significant hallmark gene sets from a GSEA with four terms increased and six decreased (Figure 4C). Among the increased gene sets were interferon gamma response and allograft rejection, which matched the increase in immune-related genes in the volcano plot. Genes decreased in response to damage by UV radiation, a source of tissue damage and inflammation, were decreased in lesions. Compared to the eutopic endometrium, glycolysis, angiogenesis, androgen response, and epithelial mesenchymal transition gene sets were decreased in the lesion epithelium. Genes increased in the lesion epithelium were enriched for 33 gene ontology (GO) biological process terms and nine of the top 10 were related to immune response (Figure 4D). In fact, nine of the 33 upregulated genes were associated with positive regulation of immune response: CD74, HLA-DRB1, CLU, C3, HLA-DPB1, RPS19, HLA-DQB1, CFB, and VTCN1.

Figure 4.

Immune response genes are increased in the endometriotic lesion epithelium

(A) Principal component (PC) plot of lesion epithelium compared to eutopic endometrium showed group separation most clearly on PC3.

(B) Volcano plot of 76 differentially expressed genes (DEG) in the endometriotic epithelium with 33 increased and 43 decreased. Of note, complement 3, C3, was one of the top increased DEG and several others were major histocompatibility complex class II related genes, like CD74 and HLA-DRB1.

(C) Gene set enrichment analysis of hallmark pathways found four increased gene sets and six decreased, indicating increasedinflammation and proliferation.

(D) Nine of the top 10 enriched gene ontology biological process terms were immune related.

Neither stroma nor epithelium clustered by tissue

Hierarchical clustering dendrograms were plotted for the stromal and epithelial segments (Figures S4A and S4B) as an unsupervised measure of segment similarity. Neither cell type clustered by eutopic or lesion origin. Segments from two patients were clustered in the stroma and epithelium that is lesion segments were more like eutopic tissue from the same patient than another lesion. Principal component (PC) analyses of the stroma and epithelium (Figures S4C and S4D) segments revealed that PC1 was driven by patient ID, representing 23.28% and 16.68% of variation in gene expression, respectively.

Macrophages are altered in peritoneal lesions

Lesion macrophages were most separated from eutopic endometrium macrophages on PC3, representing 8.87% of variation in gene expression (Figure 5A). Differential expression analysis found 148 DEG with 35 increased and 113 decreased in the lesion macrophages (Figure 5B). Nine hallmark gene sets were significantly enriched from a GSEA with seven terms increased and two decreased (Figure 5C). The seven increased terms were associated with inflammation and included TNFα signaling via NF-κB, inflammatory response, allograft rejection, and IL2 STAT5 signaling. Gene sets that were decreased were myogenesis and epithelial mesenchymal transition. Neither macrophage number nor proximity to the epithelium was altered in lesions compared to eutopic endometrium (Figure S5).

Figure 5.

Macrophages are altered in endometriotic lesions

(A) Principal component (PC) plot of lesion macrophages compared to eutopic endometrium showed group separation most clearly on PC3.

(B) Volcano plot of 148 differentially expressed genes (DEG) with 35 increased and 113 decreased in the endometriotic macrophages.

(C) Gene set enrichment analysis of hallmark pathways found seven increased gene sets and two decreased, indicating increased inflammation likely reflective of the presence of monocyte-derived macrophages in the lesion tissue. See also Figure S5.

To provide context for our findings, we sought to compare our results with bulk sequencing data from human peritoneal lesions.²¹ Human peritoneal lesions were separated from eutopic endometrium on PC one (39.43% of variation) and 3,656 DEG were identified in an analysis of bulk RNA-sequencing data (Figures S6A–S6C). GSEA³⁰ found TNF⍺ signaling via NF-κB, myogenesis, immune response, and adipogenesis hallmark genes²⁹ increased, while estrogen response, glycolysis, and multiple cell cycle gene sets were decreased in endometriotic lesions (Figure S6D).

Ligand-receptor analysis identified epithelium-macrophage signaling axis in lesions

CellChat³¹ ligand-receptor analysis was performed to determine the cell-cell communication landscape in endometriotic lesions and eutopic endometrium. Based on 976 significant ligand-receptor interactions, 29 signaling pathways were identified in total. Twenty-six pathways were present in lesions, 24 in the endometrium, and 21 pathways were common to both. Three pathways were unique to the endometrium, CXCL, SPP1, and TGF-β, and five unique to lesions, CD45, CD46, ITGB2, PTPRM, and SEMA3. A heatmap of signaling pathways consisting of all possible connections, inter- and intra-tissue, ranked by signaling strength (Figure 6A) found collagen signaling to be the strongest followed by MHC-II, laminin, macrophage migration inhibitory factor (MIF) and complement. The three top interacting cell types were lesion stroma, endometrium stroma, and lesion macrophages. Visualization of the cell-cell communication networks in the endometrium and lesions by weight in circle plots showed strong connections from the stroma compartment to the epithelium in both tissue types (Figure 6B). Based on combined network weights, confined to tissue type, the top senders were endometrial (0.172) and lesion stroma (0.170), and the top receivers were lesion macrophages (0.184) and epithelium (0.104). Particularly striking was a 3.70-fold increase in epithelium-to-macrophage signaling in lesions compared to eutopic endometrium. Interaction strength heatmaps (Figure 6C), confirmed the strong connection from the endometrial stroma received by the epithelium. In lesions, the strongest interaction was macrophage-macrophage signaling followed by epithelium-macrophage, surpassing stroma to epithelium signaling. These results point to a strengthened epithelium-macrophage signaling axis in endometriotic lesions.

Figure 6.

CellChat ligand-receptor analysis identified increased communication from the lesion epithelium to macrophages

(A) Signaling pathway heatmap of all 29 significant pathways identified by CellChat sorted by signaling strength, indicated by the gray bar. Interactions, combined incoming and outgoing, totaled for each cell type in the bar plot at the top of the heatmap and scaled per row.

(B) Circle plots of communication networks including 24 pathways in the endometrium and 26 in lesions where line thickness corresponds to strength. Stroma to epithelium communication was strong in the eutopic endometrium and similarly present in the lesions. However, epithelium to macrophage communication was increased 3.7-fold in lesions.

(C) Interaction strength heatmaps highlight the gain of epithelium to macrophage signals in lesions and strong macrophage-macrophage interactions.

Given that multiple inflammatory pathways were increased in lesions, including MHC-II and complement, and since increased signaling was noted between the lesion epithelium and macrophages, we sought to determine the source of inflammation-related ligands in lesions. A ligand-receptor bubble plot (Figure 7A) demonstrated that complement and MHC-II ligands were originating in the lesion epithelium and likely binding to receptors on macrophages. Within the stroma, MIF was the only visualized inflammatory pathway with likely communication probability indicating that inflammatory signals originated largely in the lesion epithelium. Indeed, a chord diagram of complement signaling (Figure 7B) highlighted the increased complement signaling present in lesions compared to the endometrium and the lesion epithelium as the source.

Figure 7.

Inflammatory ligands originate in the endometriotic lesion epithelium

(A) Ligand—receptor dot plot of inflammatory pathways in endometriotic lesions. Pathway names are indicated on the left side of the plot followed by specific ligand and receptor interactions. All signals originating from the lesion epithelium are in the left three columns, macrophages in the center, and stroma on the right. Complement and major histocompatibility complex class II (MHC-II) pathways originated in the lesion epithelium and not the stroma. Macrophage migration inhibitory factor (MIF)

signaling was strongest in the epithelium, but also sent from the stroma and macrophage cell types.

(B) Complement signaling chord diagram of the weighted complement signaling pathway indicating increased signaling from the lesion epithelium compared to the eutopic endometrium epithelium.

Discussion

Endometriosis is a complex reproductive disorder marked by the presence of endometrium-like tissue outside the uterus. Focusing on superficial peritoneal lesions, we hypothesized that endometrium-like tissue in these lesions would display an altered molecular phenotype influenced by the peritoneal microenvironment. To explore this hypothesis, we utilized spatial transcriptomic, receptor-ligand, and bulk RNA-sequencing analyses of endometriotic lesions compared to eutopic endometrium.

Bulk RNA-sequencing identified thousands of DEG and inflammatory pathways in lesions compared to the eutopic endometrium, highlighting the inflammatory and fibrotic nature of these lesions. Elevated cytokines in the peritoneal fluid³² likely influence these changes, alongside altered miRNAs and Notch signaling, aromatase, prostaglandins, and progestin responsiveness.^{33,34,35,36,37,38,39,40,41,42,43,44}

In contrast, spatial transcriptomics revealed striking similarities in gene expression across stromal, epithelial, and macrophage compartments of lesions and matched eutopic endometrium, with macrophages showing the most differences. This supports the hypothesis that superficial lesions originate from endometrium via retrograde menstruation.^8,9 PC analysis further indicated that patient-specific factors, rather than tissue type, primarily drive gene expression differences. The differences in macrophage gene expression are likely attributable to the presence of peritoneal and monocyte-derived macrophages in lesions, as opposed to the tissue-resident macrophages found in the eutopic endometrium.^45,46,47

The divergence between bulk RNA-seq and spatial analyses reflects their distinct methodologies. While bulk RNA-seq captures broad transcriptomic differences across homogenized tissue, spatial transcriptomics targets localized gene expression within specific regions. This specificity revealed few DEG, suggesting that transcriptomic differences in bulk data arise from peripheral tissue rather than endometrium-like areas in lesions. These findings highlight the value of spatially resolved approaches in studying heterogeneous tissues like endometriotic lesions.

Spatial transcriptomic analysis resulted in an overall reduced transcriptome size compared to the bulk data, including 7,945 compared to 17,405 genes. Consequently, many of the lowest expression genes from the bulk RNA-sequencing data were filtered from the spatial transcriptome before DEG analysis. The advantage of spatial transcriptomics, however, is gaining information from genes expressed only by a subset of cells or a single cell type that would be lost in the bulk tissue. This is evidenced by the inclusion of 551 genes in the spatial transcriptome that were considered as not expressed in the bulk RNA-sequencing analysis. Further, spatial transcriptomics indicated increased proliferation of the lesion epithelium while bulk RNA-sequencing found an overall reduction in cell cycle gene sets. The epithelium of lesions is likely to be a minor contributor to gene expression from homogenized tissue since lesions often have little or even no remaining epithelial component. Interestingly, these results suggest the lesion epithelium may proliferate in response to inflammatory peritoneal environment while the stroma reduces proliferation and may even become senescent.⁴⁸ A growing body of evidence has found that senescence plays a role in multiple pathologies and may contribute to endometriotic lesions.^49,50,51

Receptor-ligand analysis identified an almost 4-fold increase in signaling between epithelium and macrophages in lesions compared to the endometrium, including multiple inflammatory pathways. MHC-II ligands expressed in the lesion epithelium were predicted to communicate with macrophages through the CD4 receptor. The canonical macrophage marker from this study, CD68, is expressed by monocytes and macrophages.⁵² Notably, human monocytes and macrophages express CD4, unlike the mouse, serving as an important target for HIV-1 infection.^53,54,55 Although monocytes rapidly differentiate into macrophages within a tissue, some monocytes were likely included in the spatial transcriptome analysis. Canonically, MHC-II binding of the CD4 receptor on monocytes promotes differentiation into macrophages,⁵⁴ thus increased MHC-II in the lesion epithelium would drive macrophage differentiation from circulating monocytes homing to areas of hypoxia and inflammation. Increased expression of MHC-II in the epithelium aligns with recent single cell RNA-sequencing analysis of ovarian endometriosis, which found increased express of MHC class II molecules in the epithelium.⁵⁶

Complement 3 expression was increased in bulk RNA-sequencing results and found in the lesion epithelium from spatial transcriptomic analysis. Endometriotic lesions were reported to express C3 by Weed and Arquembourg in 1980⁵⁷ and later found to be secreted by the lesion epithelium.^58,59 In fact, serum levels of C3 are elevated in patients with endometriosis and C3 was proposed as a potential diagnostic biomarker.⁶⁰ Although the presence of C3 in endometriotic lesions has been known for 40 years, its potential contribution to the disease is unclear.⁶¹

The complement system is a signaling cascade that plays a pleiotropic role in the immune response, including production of proinflammatory mediators and recognition and clearance of pathogens by lysis and enhanced phagocytic uptake by immune cells.⁶² Multiple stimuli can activate the complement system, but all pathways converge on activation of C3 by a convertase. C3a and C3b are produced following activation of C3.⁶² C3a can activate a pro-repair or tumor-associated macrophage (TAM) phenotype by binding to C3AR1.⁶³ This provides a mechanism by which C3 activation from the superficial peritoneal endometriotic lesion epithelium promotes the pro-lesion macrophage phenotype described as pro-repair, tumor-associated, or M2 which contributes to tissue remodeling including neovascularization and fibrosis.^18,64,65 C3b induces phagocytic activity of macrophages, but lesion tissue is refractory to complement targeting through multiple mechanisms.^62,66,67,68

Macrophage migration inhibitory factor (MIF) is released by monocytes, macrophages, and other immune cells upon exposure to proinflammatory mediators, such as C3, to reinforce a local proinflammatory environment.^61,69,70,71 MIF is constitutively expressed by many cell types, including epithelium, and has pleiotropic autocrine and paracrine effects on tissues.⁷² Receptor-ligand analysis identified MIF in the top five signaling pathways and that the lesion epithelium was predicted to activate MIF signaling in lesion macrophages. Activation of MIF signaling in macrophages sustains proinflammatory function by inhibiting apoptosis, counteracting the immunosuppressive effects of glucocorticoids, and promoting secretion of prostaglandin E2.^72,73 Increased prostaglandins can activate steroidogenic enzymes and, potentially the production of estradiol and estrone by the lesion.^33,35,37,74 These estrogens, in turn, activate complement,⁷⁵ creating a feedback loop that sustains local inflammation and pro-repair macrophages.

Spatial transcriptomics has provided novel insights into the intricate cellular crosstalk within lesions. Our data indicate signaling between the stroma and epithelium, as well as between the lesion epithelium and macrophages. Notably, the epithelium emerged as a central player in driving inflammation within the lesion, promoting inflammatory signaling independent of the stromal compartment. Drawing from these observations, we present a model wherein the superficial lesion epithelium orchestrates inflammatory signaling within the lesion while promoting a pro-repair phenotype in infiltrating macrophages. This novel insight sheds light on the pathological consequence of C3 expressed by the epithelium in modulating macrophage phenotype to support lesion development. In summary, our study underscores the critical role of considering spatial context and cellular interactions within lesions to unravel the physiology of superficial peritoneal endometriotic lesions. Further investigation is imperative to elucidate the mechanisms governing epithelium-driven inflammation and its implications for disease progression and therapeutic intervention.

Limitations of the study

This study’s paired design enhanced statistical power despite a limited sample size, yet several factors may influence the generalizability of the findings. The absence of healthy control endometrium and the unknown duration of symptom onset are notable considerations, particularly for understanding early disease initiation. While uterine fibroids affect up to 80% of women and adenomyosis incidence ranges from 1% to over 20%, this study specifically avoided selecting endometrium adjacent to adenomyosis or leiomyomas to minimize confounding influences. However, the use of eutopic tissue from patients with other pathologies may still reflect underlying endometrial dysfunctions, potentially influenced by the older age of the cohort. Limited clinical information, including hormonal therapy details, may further influence the interpretation of gene expression profiles. The sample size, while adequate for identifying intra-patient similarities and differences, restricts the ability to capture inter-patient variability.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Asgerally T. Fazleabas (fazleaba@msu.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •Spatial transcriptomics data have been deposited at the NCBI Gene Expression Omnibus and are publicly available as of the date of publication. This paper analyzes existing, publicly available RNA-sequencing data. Accession numbers are listed in the key resources table.
• •This paper does not report original code.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

The authors would like to sincerely thank all members of the Fazleabas laboratory and the Van Andel Genomics core, particularly Becca Siwicki, for their contributions to this project. This work was supported by NanoString Technologies, Incorporated (G.W.B.) and National Institutes of Health grants F32HD104478 (G.W.B.), R01HD099090 (A.T.F.), R01HD050559 (I.F.), U56CA126379 (I.F.), and T32HD087166 (G.W.B. and A.T.F.). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Author contributions

Conceptualization, G.W.B., E.G., and A.T.F.; data curation, G.W.B., Z.F., Z.B.M., and I.F.; formal analysis, G.W.B., Z.F., E.L.V., and Z.B.M.; funding acquisition, G.W.B., I.F., and A.T.F.; investigation, G.W.B. and E.L.V.; methodology, G.W.B., Z.F., E.L.V., and Z.B.M.; resources, I.F. and A.T.F.; software, G.W.B., Z.F., and Z.B.M.; supervision, E.G. and A.T.F.; visualization, G.W.B.; writing—original draft, G.W.B and A.T.F.; writing—review and editing, G.W.B., Z.F., E.G., I.F., and A.T.F.

Declaration of interests

The authors declare no competing interests.

Declaration of generative AI and AI-assisted technologies in the writing process

During the preparation of this work, the authors used ChatGPT (v3.5) for proofreading and to improve readability. After using this service, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
ACTA2 Alexa Fluor® 594	Abcam	Cat# ab202368; RRID: AB_2924381
CD68 Alexa Fluor® 647	Santa Cruz Biotechnology	Cat# sc-20060 AF647; RRID: AB_3073741
Cytokeratin, Pan Alexa Fluor® 488	Novus Biologicals	Cat# NBP2-33200AF488; RRID: AB_3284601
Chemicals, peptides, and recombinant proteins
Antigen Retrieval Reagent	Invitrogen	Cat# 00-4956-58
SSC Buffer	Sigma Aldrich	Cat# S6639
Syto13 Nucleic Acid Stain	Invitrogen	Cat# S7575
Trichrome Histology Stain	StatLab	Cat# KTTRBPT
Critical commercial assays
GeoMx Human Whole Transcriptome Atlas FFPE-PCLN	NanoString Technologies	Cat# 21300313
High Sensitivity DNA Kit	Agilent Technologies, Inc	Cat# 5067-4626
NovaSeq 6000 S2 Reagent Kit v1.5	Illumina, Inc.	Cat# 20028316
QuantiFluor dsDNA System	Promega Corporation	Cat# E2670
Deposited data
Spatial Transcriptomics	This paper	GEO GSE263897
Bulk RNA-sequence	Tan et al.21	GEO GSE179640
Software and algorithms
bcl2fastq Conversion Software v1.9.0	Illumina, Inc.	RRID: SCR_015058
CellChat v1.6.10	https://github.com/sqjin/CellChat	RRID: SCR_021946
clusterProfiler v4.6.0	Bioconductor	RRID: SCR_016884
dendextend v1.17.1	https://CRAN.R-project.org/package=dendextend	N/A
edgeR v4.0.3	Bioconductor	RRID: SCR_012802
EnhancedVolcano v1.16	Bioconductor	RRID: SCR_018931
GeoMx NGS Pipeline v2.3.3.10	NanoString Technologies, Inc.	N/A
GeomxTools v3.2.0	Bioconductor	RRID: SCR_023424
GraphPad Prism v10.2.2	GraphPad Software	RRID: SCR_002798
ImageJ v2.14.0	https://imagej.net/ij/	RRID: SCR_003070
Limma v3.54.2	Bioconductor	RRID: SCR_010943
PCAtools v2.10	Bioconductor	RRID: SCR_025593
R v4.2.2	https://www.r-project.org	RRID: SCR_001905
RStudio Desktop v2024.04.2	Posit Software	RRID: SCR_000432

Experimental model and study participant details

All protocols involving tissue collection were approved by the institutional review board (IRB) Committee of Ponce Health Sciences University, School of Medicine (PHSU). Endometriotic lesions and matched eutopic uterine tissues from five female patients who received surgery for lesion excision and hysterectomy were obtained as archived deidentified formalin-fixed paraffin-embedded (FFPE) tissue blocks.^76,77,78 All tissues were previously evaluated by a pathologist to confirm the diagnosis of endometriosis and menstrual cycle stage of eutopic endometrium using Noyes criteria.⁷⁹ All available patient data are included in Table 1.

Method details

Histology

Tissue blocks were sectioned and trichrome stained (StatLab #KTTRBPT) to confirm the presence of defined epithelial and stromal compartments in lesions. Blocks were serial sectioned at 6 μm thickness and floated onto positively charged microscope slides. Tissues were attached to the slides prior to processing by heating on a slide warmer for 15 minutes. Tissues were then deparaffinized and rehydrated through three changes each of xylenes for five minutes and 100% ethanol for one minute before a rinsing in running tap water for one minute. Slides were incubated in Bouin’s fluid at room temperature overnight. Slides were rinsed in running tap water for three minutes and immersed in modified Mayer’s hematoxylin stain for four minutes at room temperature. After a three-minute rinse in running tap water, slides were immersed in one step trichrome stain for five minutes and rinsed in running tap water for five seconds. Slides were then dehydrated through three changes of 100% ethanol for one minute, cleared in three changes of xylenes for one minute, and a coverslip was applied with Permount mounting medium (Fisher Scientific #SP15).

Spatial transcriptomics

Human tissue slide preparation, library generation, and sequencing for digital spatial profiling (DSP) on the NanoString GeoMx DSP instrument (RRID: SCR_021660) were performed by the Van Andel Institute Histology and Genomics Cores. Sections were cut at 5 μm thickness and mounted on plus-charged slides (Epredia Colormark Plus #CM-4951WPLUS-001, Erie Scientific). Slides were baked at 60°C for one hour and stored at 4°C in a vacuum-sealed container containing desiccant for up to two weeks. All subsequent steps were performed under RNase-free conditions with DEPC-treated water. Slides were deparaffinized with three sequential five-minute washes in xylenes, followed by two washes in 100% ethanol for five minutes, one wash in 95% ethanol, and one wash in 1X PBS. Antigen retrieval was performed in target retrieval reagent (EDTA, pH 9.0; Invitrogen #00-4956-58) diluted to 1X in the BioGenex EZ-Retriever System for 10 minutes at 95°C. Slides were then washed with 1X PBS for five minutes. Slides were then incubated in 1 μg/mL proteinase K 10 minutes at 37°C and washed in 1X PBS for five minutes at room temperature. Slides were fixed for five minutes in 10% neutral buffered formalin followed by two washes in NBF stop buffer, for five minutes each, and one wash in 1X PBS for five minutes. Slides were then incubated with UV-photocleavable Human Whole Transcriptome Atlas hybridization probes diluted in buffer R (GeoMx RNA Slide Prep FFPE-PCLN kit, #121300313) in a hybridization oven at 37°C for 16-20 hours. Following probe incubation, slides were washed with stringent buffer (1:1, formamide:4X SSC buffer; Thermo Fisher #AM9342; Sigma-Aldrich #S6639) at 37°C twice for 25 minutes each. Then slides were washed twice in 2X SSC buffer. Slides were blocked in 200 μL buffer W (GeoMx RNA Slide Prep FFPE-PCLN kit) for 30 minutes and incubated at 4°C overnight with SYTO 13 nucleic acid stain (500 nM; Invitrogen #S7575) and antibodies for pan-cytokeratin (panCK; 1 μg/mL; Novus Biologicals #NBP2-33200AF488), smooth muscle actin (ACTA2; 1.25 μg/mL; Abcam #ab202368), and CD68 (0.5 μg/mL; Santa Cruz Biotechnology #sc-20060AF647) diluted in buffer W. Slides were washed four times in 2X SSC buffer for 3 minutes each and placed in the NanoString GeoMx DSP instrument.

Duplicate 500 μm x 500 μm regions of interest (ROI) were selected for each based on fluorescent imaging on the GeoMx DSP instrument (NanoString Technologies). ROI were drawn in the central portion of the endometrium stratum functionalis, to include glandular and not luminal epithelium, for eutopic uterine tissue and in lesions areas with defined epithelium. ROIs were then segmented in order of macrophages (CD68⁺), epithelium (panCK+), and stroma (ACTA2-) areas of illumination (AOI) for individual collection. AOI were processed sequentially on the instrument using UV light focused through each AOI that released the photocleavable oligos into wells of a 96 well plate. After AOI collection, indexing primers were hybridized during NanoString library preparation. Quality and quantity of the finished library pools were assessed using a combination of Agilent DNA High Sensitivity chips (Agilent Technologies, Inc.) and the QuantiFluor dsDNA System (Promega Corp.). Paired-end, 50 bp sequencing was performed on an Illumina Novaseq 6000 sequencer (RRID: SCR_016387) using an S2, 100 cycle sequencing kit v1.5 (Illumina Inc.). Base calling was done by Illumina RTA3 and output was demultiplexed and converted to fastq format with bcl2fastq v1.9.0.

Quantification and statistical analysis

Spatial data analysis

Fastq files were converted to digital count conversion (DCC) files using GeoMx NGS Pipeline v2.3.3.10. Sample annotation, PKC configuration, and DCC files were imported with GeomxTools (v3.2.0) in R to generate a NanoStringGeoMxSet object. Read counts with a value of zero were shifted to one for downstream transformations and quality control (QC) analyses. Each AOI, or segment, was assessed for sequencing and tissue quality using the following parameters: minSegmentReads = 1000, percentTrimmed = 80, percentStitched = 80, percentAligned = 75, percentSaturation = 40, minNegativeCount = 1, maxNTCCount = 9000, minNuclei = 20, minArea = 1000. Outlier probes were then excluded before counts were collapsed to gene-level data (18,677 genes) by geometric mean. The limit of quantification (LOQ) was determined per segment based on the distribution of negative control probes and set at two geometric standard deviations above the geometric mean with a minimum value of two. Segments with a gene detection rate less than 10% were filtered before removing genes detected in less than 10% of segments. Separation between the upper quartile 3 (Q3) of gene counts and the geometric mean of the negative probes was confirmed before count normalization using the quartile 3 (Q3) method included with GeomxTools. Briefly, Q3 normalization uses the top 25% of expressed genes to normalize across segments. Principal component analysis of log₂-transformed normalized read counts was conducted with the PCAtools package (v2.10). The biplot and pairsplot functions were used to visualize sample separation across principal components one through five.

Differential expression analyses were completed using log₂-transformed normalized read counts with a linear model approach in the limma package (v3.54.2).⁸⁰ A linear model was fit for each gene with lmFit before an empirical Bayes test was completed using patient as a covariate. P-values were corrected for multiple testing with the Benjamini Hochberg method included in the decideTests function of limma. Genes were considered differentially expressed when the FDR adjusted p-value was below 0.05. Data for all comparisons are included in Tables S1, S2, S3, S4, S5, and S6. A volcano plot was generated with EnhancedVolcano (v1.16) and the top 20 DEG, based on FDR p-value, were labeled. Gene set enrichment and over-representation analyses were completed using the clusterProfiler package (v4.6.0).⁸¹ Genes, converted to entrez identifiers, and log₂-transformed fold changes were used as input for the GSEA function with the mSigDB²⁹ reference. Entrez gene identifiers for upregulated or downregulated DEG were analyzed using the enrichGO function, with all expressed genes set as the background universe, based on the Gene Ontology database.^82,83

Macrophage quantification

The macrophage count distance between from epithelium was manually measured for each segment using the line tool in ImageJ software.⁸⁴ The global scale was set by measuring the scale bar on the image (1 pixel = 2.51 μm). For each image, the green epithelium and red macrophage channels were merged for analysis. A line was drawn from the center of each macrophage to the center of the closest epithelial cell and the measurements. Based on these measurements the macrophages count and the average distance for each segment was calculated. Data were analyzed in Prism (v10.2.2, GraphPad Software) following confirmation of normality (Shapiro-Wilk, p > 0.1). To determine if there was a relative increase in lesions compared to the matched endometrium, macrophage count was tested using a ratio paired t-test. Macrophage distance to the nearest epithelium was tested using a paired t-test.

Bulk RNA-sequencing analysis

Read counts per gene were downloaded from eutopic endometrium (n = 7) and peritoneal endometriotic lesion samples (n = 6) for series GSE179640 ²¹ from the NCBI Gene Expression Omnibus repository. Files were parsed with R (version 4.2.0) and all subsequent analyses, unless otherwise noted, were completed in R using RStudio.⁸⁵ Differential expression analysis was conducted using the edgeR-robust method (v4.0.3).^86,87 Genes with low counts per million (CPM) were removed using the filterByExpr function from edgeR. Multidimensional scaling plots, generated with the plotMDS function of edgeR, were used to investigate group separation prior to statistical analysis. Principal component analysis of normalized log₂-transformed CPM was conducted with the PCAtools package (v2.10). The biplot and pairsplot functions were used to visualize sample separation across principal components one through five. Differentially expressed genes (DEG) were identified based on Benjamini-Hochberg false discovery rate (FDR) p-values less than 0.05. The hclust function was used for hierarchical clustering (ward.D2) of Euclidean distances and plotted with the dendextend package (v1.17.1). Volcano plots, gene set enrichment analysis, and over-representation tests were completed as described for spatial transcriptomic analysis.

Receptor-ligand analysis

Quantile-normalized read counts were used for receptor-ligand analysis with the CellChat package (1.6.10) in R.³¹ Significant genes and interactions were identified using the functions identifyOverExpressedGenes and identifyOverExpressedInteractions. Communication probabilities were calculated between cell types with raw input counts option enabled, as recommended for bulk-sequencing data. The resulting interactions were visualized using the included plotting functions. Weighted communication probabilities were subset for eutopic endometrium or lesions before calculating tissue-specific communication interaction strengths.

Published: January 10, 2025