Endometriosis-Associated Mesenchymal Stem Cells Support Ovarian Clear Cell Carcinoma through Iron Regulation Open Access

Ovarian cancer is collectively the most deadly gynecologic malignancy in the United States (1). However, ovarian cancer encompasses a diverse set of histologic subtypes, which demonstrate dramatic differences in molecular profile, proposed site of origin, and clinical outcome. Ovarian clear cell carcinoma (OCCC) is one of the most aggressive and understudied subtypes of ovarian cancer. When diagnosed in advanced stages (stage III or IV), OCCC has an almost 2-fold increased mortality compared with the most common subtype, high-grade serous ovarian cancer (HGSC; ref. 2). OCCC has a distinct molecular profile characterized by ARID1A mutations and commonly co-occurring PI3K and PTEN mutations (3). Although the preponderance of evidence points to a fallopian tube origin of HGSC, OCCC is thought to arise within sites of endometriosis (4). Endometriosis, an inflammatory disease marked by ectopic growth of uterine tissue containing both epithelial and stromal components, is a known risk factor for OCCC. Indeed, women with histologically proven endometriosis involving the ovary have a 2.3-fold increased risk of OCCC (5). The endometriosis microenvironment is marked by inflammatory changes and high levels of iron due to continued menstrual cycling of the epithelial lining (6). Both chronic inflammation and high levels of iron are risk factors for cancer development; however, the mechanisms by which the endometriotic microenvironment supports OCCC formation and growth are largely unknown (7, 8).

Previous work focused on HGSC demonstrated that a key stromal component in the tumor microenvironment (TME), the carcinoma-associated mesenchymal stem cell (CA-MSC), strongly supports ovarian cancer growth (9). These CA-MSCs arise from normal tissue MSCs, which become cancer-educated to support ovarian cancer survival, growth, and spread (10). MSCs are also readily present within both endometrial and endometriosis tissue and may be critical to the formation of endometriosis (11). Here we report the discovery of a subset of endometriosis-derived MSCs (enMSC), which specifically support OCCC growth and metastasis. OCCC-supportive enMSCs are marked by loss of CD10 expression. CD10 negative (CD10⁻) enMSCs increase OCCC proliferation, chemotherapy resistance, sphere formation, and in vivo tumor growth. Mechanistically, CD10⁻ enMSCs increase labile iron within tumor cells (TC) necessary for OCCC growth. However, this creates a unique susceptibility to ferroptosis. Treatment with a ferroptosis inducer, erastin, specifically kills OCCC grown with CD10⁻ enMSCs, thus presenting a potentially powerful therapeutic target.

enMSCs were isolated from primary human benign endometriosis deposits involving the ovary or fallopian tubes obtained from the surgical resection specimens of females (including cisgender, transgender, and nonbinary individuals) undergoing surgery for benign indications in accordance with protocols approved by the University of Pittsburgh's Institutional Review Board (PRO17080326). Written informed consent was obtained from all patients. MSCs were isolated following previously described protocols (9,12). Clinical characteristics provided in Supplementary Table S1. Briefly, dissociated cells were plated in mammary epithelial cell basal medium (MEBM) supplemented with 10% heat-inactivated FBS, 1× B27, 20 ng/mL EGF, 1 ng/mL hydrocortisone, 5 μg/mL insulin, 100 μmol/L β-mercaptoethanol, 10 ng/mL β-fibroblast growth factor (FGF), 1% penicillin/streptomycin, and 20 μg/mL gentamicin. MSCs were selected for plastic adherence and cell surface marker expression CD105 (BD Biosciences, Catalog No. 561443, RRID:AB_10714629) CD90 (BD Biosciences, Catalog No. 555596, RRID:AB_395970), CD73 (BD Biosciences, Catalog No. 560847, RRID:AB_10612019) positive; CD45 (BD Biosciences, Catalog No. 340664, RRID:AB_400074), CD34 (BD Biosciences, Catalog No. 555824, RRID:AB_398614), CD14 (BD Biosciences, Catalog No. 340436, RRID:AB_400509), and CD19 (BD Biosciences, Catalog No. 340364, RRID:AB_400018) negative. Adipocyte, osteocyte, and chondrocyte differentiation capacity was verified (following guidelines presented by the ISCT on the minimal criteria for defining multipotent mesenchymal stem cells; ref. 13) and as described previously (9,12,14). MSCs (CD10_high or CD10_low) were propagated for <8–10 passages and used for functional experiments at passage 5 or below. Each MSC sample was derived from an individual patient and the individual patient MSC samples were maintained separately. All experiments were performed with at least three MSC samples from independent patients to account for intra-patient heterogeneity.

The TOV-21G cell line was purchased from ATCC (RRID: CVCL_3613) and maintained in a 1:1 mixture of MCDB 105 medium and Medium 199 supplemented with 15% FBS. The OVISE cell line was a kind gift of Dr. R Zhang and was maintained in RPMI1640 medium supplemented with 10% FBS and penicillin/streptomycin (penicillin: 100 units/mL, streptomycin: 0.1 mg/mL). The RMG1 cell line was a kind gift of Dr. R. Zhang and maintained in DMEM/F12 medium supplemented with 10% FBS and penicillin/streptomycin (penicillin: 100 units/mL, streptomycin: 0.1 mg/mL). All three OCCC lines were cultured at 37°C in a 5% CO2 humidified incubator.

Human ovarian HGSC cell line OVCAR3 (RRID: CVCL_0465) were purchased through ATCC. The primary patient derived HGSC cell line, PT412, and OVSAHO cell line was a kind gift from Dr. R. Buckanovich. All cells were maintained in DMEM medium supplemented with 10% FBS and penicillin/streptomycin (penicillin: 100 units/mL, streptomycin: 0.1 mg/mL). Cells were cultured at 37°C in a 5% CO₂ humidified incubator.

All cancer cell lines have been validated using STR testing though ATCC and all cells are tested on a bimonthly basis to verify Mycoplasma negativity (last test 8/2022). Cells were used at passage 10 or less from thawing.

enMSCs were stained with APC conjugated anti-CD10 antibody (Thermo Fisher Scientific, Catalog No. 17–0106–42, RRID: AB_11043552) for 20 minutes, followed by washing with PBS. Cells were then resuspended in PBS with 2% FBS and sorted into CD10⁺ and CD10⁻ populations with gating based on APC-IgG control using The BD FACSAria Fusion system. Sorted cells were collected directly into supplemented MEBM.

OCCC lines were stained with CellTrace Violet (CTV - Invitrogen C34571) following the manufacturer's protocol. Briefly, cells were washed once with PBS and resuspended in PBS at 1 × 10⁶ cells/mL, followed by staining with 1 μL (5 mmol/L) of CTV per 1 × 10⁶ cells for 25 minutes. Cells were then centrifuged and resuspended in fresh complete culture medium.

1 × 10⁴ CTV-labeled OCCC cells were cultured alone, with 1 × 10⁴ CD10_low enMSCs, or with 1 × 10⁴ CD10_high enMSCs in a 12-well plate. For the chemoresistance assay, Cisplatin was added to the culture medium at 0.5, 1, and 2 μg/mL concentrations. For ferroptosis induction assay, erastin (MedChemExpress- HY-15763/CS-1675) or dihydroartemisinin (DHA; Sigma-Aldrich D7439) was added to the culture medium at 0.5, 1, 2, 5, and 10 μmol/L. The OCCC cell numbers were followed for 4 days. Cells were collected from 2 wells per condition per day and counted using a hemocytometer. Flow cytometric analysis was used to determine the percentage of the CTV-labeled TC to quantify viable TCs per condition. Total OCCC cells (CTV-labeled) = total cell number × CTV positive cell %.

OCCC lines were stained with CellTrace Yellow (CTY - Invitrogen C34573), following the manufacturer's protocol (CTY falls in the “orange/red” spectrum for microscopy). Briefly, cells were washed once with PBS and resuspended in PBS at 1 × 10⁶ cells/mL, followed by staining with 5 mmol/L of CTY per 1 × 10⁶ cells for 25 minutes. Cells were then centrifuged and resuspended in fresh complete culture medium. Using ultra-low-adherent 96-well plates, labeled OCCC cells were plated alone (1 × 10³ cells/well), with CD10_low enMSCs (0.5 × 10³ OCCC cells + 0.5 × 10³ enMSCs/well), or with CD10_high enMSCs (0.5 × 10³ OCCC cells + 0.5 × 10³ enMSCs/well) in 300 μL MSC media (described above). After 7 days, total spheres (defined as >4 cells spheroid, at least 1 TC per sphere) in the entire well were counted.

RNA was isolated using RNeasy Mini Kit and on-column DNase treatment (Qiagen, 28106). After determining the RNA concentration using the Nano-Drop ND-1000 Spectrophotometer, cDNA was synthesized using SuperScript III First Strand Synthesis System (random hexamer; Invitrogen, 18080–051). SYBR-green based RT-qPCR (Applied Biosystem, 4472908) was performed using CFX96 Real-Time System, samples were run for 40 cycles with GAPDH as the reference gene. Primer sequences are listed in Supplementary Table S2.

Total RNA was extracted using a NucleoSpin RNA Kit (Machery-Nagel) and eluted in nuclease-free water. RNA concentration and quality was assessed via a NanoDrop spectrophotometer (Thermo Fisher Scientific). Samples were then packaged and shipped to Novogene (Novogene Corporation) for further processing and sequencing.

mRNA was purified from total RNA using poly-T oligo-attached magnetic beads. After fragmentation, first-strand cDNA synthesis was carried out using random hexamer primers followed by second strand synthesis using dUTP. Qubit (Thermo Fisher Scientific) and real-time PCR was used for quantification and bioanalyzer for size distribution detection. Quantified libraries were pooled and sequenced on an Illumina platform (Illumina Inc.) and paired-end reads were generated.

Raw reads in fastq format were processed through a Novogene in-house perl script for QC. Briefly, reads containing adapters, ploy-N, and low-quality reads were removed and Q20, Q30, and GC content was calculated. The index of the reference genome hg38 was built and paired-end clean reads were aligned to the reference genome using hisat2 (v2.0.5; ref. 15). Quantification of the gene expression level was done via featureCounts (v1.5.0-p3; ref. 16). The FPKM of each gene was calculated based on the length of the gene and read counts mapped to the gene. Differential gene expression analysis was performed using the DESeq2 R package (v1.20.0; ref. 17). Resulting P values were adjusted using the Benjamini–Hochberg method to control for FDR. Genes with a P_adj ⇐ 0.05 and |log2(FoldChange)| ≥1 were assigned as differentially expressed.

Over-representation analysis (ORA) using gene ontology (GO) terms was done on the differentially expressed gene (DEG) lists obtained from DESeq2 using the clusterProfiler package in R version 4.1.2 (18). Briefly, genes with a log2 fold change greater or lesser than 1 were matched to GO terms and tested for over-representation against the “org.Hs.eg.db” gene universe using the enrichGO function. GO terms with a Benjamini–Hochberg corrected P values with a cutoff of 0.1 were obtained. Redundant GO terms were removed from the ORA result using Revigo, which finds a representative subset of GO terms using a clustering algorithm based on semantic similarity measures (19). Results were visualized via the dotplot function in clusterProfiler (18).

All experimental procedures were performed in accordance with the approved protocol by the Institutional Animal Care and Use Committee (IACUC) at the University of Pittsburgh (PRO21018732). 6- to 8-week-old female NSG (NOD SCID gamma; The Jackson Laboratory, Catalog No. 005557; RRID:IMSR_JAX:005557) mice were used for mouse models.

A total of 2 × 10⁵ TOV-21G cells were injected alone, with CD10_low enMSCs (2 × 10⁵), or with CD10_high enMSCs (2 × 10⁵) subcutaneously into the mammary fat pad of NSG mice (total injection volume was 100 μL), with 5 mice used for each of the three conditions. Initial engraftment was assessed at 2 weeks post-injection and only mice with established tumors at this time point were followed for tumor volume/growth assessment. Mouse weight and health were monitored throughout the experiment, and the end point was determined when the first group of mice met the endpoint criteria (>10% weight loss, >20% weight gain, tumor size >1,000 mm³, ulceration or distress). Mice were then sacrificed, and a necropsy was performed on each mouse to assess for metastasis.

2 × 10⁵ TOV-21G luciferase labeled cells were injected alone, with CD10_low enMSCs (2 × 10⁵), or with CD10_high enMSCs (2 ×10⁵; total injection volume was in <15 μL) into the ovarian bursa of NSG mice, with 10 mice used for each of the three conditions. Tumor growth and metastatic progression were monitored bi-weekly using In Vivo Imaging System (IVIS) as described previously [9,12]. Mouse weight and health were monitored over time. Mice were followed for survival and individual mice were sacrificed as they met the end-point criteria as above. A necropsy was performed on each mouse to assess the metastatic burden.

Using RIPA buffer (Pierce, 89900), cell pellets were homogenized with a complete protease inhibitor cocktail. Samples were then centrifuged at 16,000 × g at 4°C for 15 minutes to remove insoluble material. Protein concentrations were determined using the Pierce BCA Protein Assay Kit (Thermo Fisher Scientific, 23227). Using 4% to 12% NuPAGE SDS gel (Invitrogen), equal amounts of protein were separated and transferred onto a PVDF membrane. Anti-ferritin L (Abcam, Catalog No. ab69090, RRID:AB_1523609), anti-ferritin H (Abcam, Catalog No. ab65080, RRID:AB_10564857), anti-Ferroportin (Abcam, Catalog No. ab58695, RRID:AB_2302072), and anti-transferrin receptor (Abcam, Catalog No. ab84036, RRID:AB_10673794) at 1:1,000 dilution, and anti-HEPH at 1:500 dilution (Abcam, Catalog No. ab108003, RRID:AB_10863278) were used. Bands were visualized using the SuperSignal West Pico PLUS Chemiluminescent Substrate (Pierce, 34577).

Intracellular (labile) iron pool was quantified as described previously (20). Briefly, 1 × 10⁵ CTV-labeled OCCC cells were cultured alone, with 2 × 10⁵ CD10_low enMSCs, or with 2 × 10⁵ CD10_high enMSCs. At day 3, cells were collected and washed twice with PBS. Cells were then incubated with 0.125 μmol/L calcein AM (CA-AM; Invitrogen-C1430) for 15 minutes at 37ºC. After washing with PBS, cells were incubated for 15 minutes with DFO at 100 mmol/L. Cells were then washed with PBS and analyzed by flow cytometry at a rate of 1,000 events/second at 448 nm excitation. Mean fluorescence intensity (MFI) was calculated using FlowJo. Given that CA-AM fluorescence intensity increases as free iron content decreases, the change of MFI represents the free iron pool (labile iron pool, LIP).

ΔMFI = MFICA - AM/DFO - MFICA - AM alone

A total of 2 × 10⁴ CTV-labeled OCCC cells were cultured alone, with 2 × 10⁴ CD10⁻ enMSCs, or with 2 × 10⁴ CD10⁺ enMSCs on a coverslip in a 6-well plate. Erastin was added to the culture medium at 1 μmol/L. Day 3 after the coculture, cells were fixed and incubated with anti-4-hydroxynonenal (4-HNE; Millipore, Catalog No. AB5605, RRID: AB_569332) overnight at 4°C. After washing with PBB (0.5% BSA in PBS), cells were incubated with Alexa Fluor 546 conjugated secondary antibody (Thermo Fisher Scientific, Catalog No. A-11056, RRID: AB_2534103) for 1 hour at room temperature.

BODIPY 581/591 C11 was used to detect ferroptosis as described previously (21). Briefly, 4 × 10⁴ CTV-labeled OCCC cells were cultured alone, with 4 × 10⁴ CD10⁻ enMSCs, or with 4×10⁴ CD10⁺ enMSCs in a 6-well plate. Erastin was added to the culture medium at 1 μmol/L. After 24 hours of erastin treatment, 2 μL of 1.5 mmol/L BODIPY-C11 stock solution was added to erastin-treated cells (no erastin treatment cells were used as a control). Cells were then returned to the tissue culture incubator and allowed to stain with BODIPY for 20 minutes. After washing with HBSS buffer, cells were then harvested for flow analysis using The BD FACSAriaTM Fusion system.

A total of 1 × 10⁴ TOV-21G cells were cultured per well in 12-well plates. Ferric ammonium citrate (FAC; MP Biomedicals, 158040) was added to the culture media at 0, 0.05, 0.1, and 0.2 μmol/L. Cells were collected from 2 wells per condition per day and counted using a hemocytometer with trypan blue exclusion.

Data are presented as mean ± SEM. All experiments were performed on three ovarian clear cell cancer lines and on at least three patient-derived enMSCs. To determine the statistical significance, Student unpaired t test, ANOVA or log-rank (Mantel–Cox) test were used.

The data generated in this study are available upon request from the corresponding author.

As OCCC is thought to arise within endometriosis and MSCs within the TME support other ovarian cancer histologic subtypes, we first assessed the prevalence of MSCs within endometriosis derived from fresh benign endometriosis samples from women undergoing surgery for noncancer-related indications (Supplementary Table S1; refs. 9, 10, 14). We found that endometriosis contains a clearly discernible population of MSCs (as defined by the international society of cellular therapy; ref. 13). We quantified the prevalence of MSCs as a percentage of total viable cells by dissociating the tissue into single cell suspension and analyzing cells using flow cytometry to define the MSCs as CD73/CD90/CD105 positive and CD45/CD34/CD14/CD11b/CD19 negative. MSCs were isolated in parallel as described previously, verifying plastic adherence and trilineage differentiation into adipocytes, osteocytes, and chondrocytes (Supplementary Fig. S1; refs. 9, 12). Given the variability in MSC prevalence within specific organs, we specifically analyzed endometriosis involving the ovary or fallopian tube. We identified MSCs present within all 15 patient samples of endometriosis. We compared the prevalence of MSCs in endometriosis deposits within ovary or fallopian tube tissue to normal ovary or fallopian tubes without endometriosis (or other pathology). At both organ sites, tissues with endometriosis had a significantly higher number of MSCs (Fig. 1A). We also assessed OCCC involving the ovary or fallopian tube and demonstrated that OCCC contained high levels of MSCs consistent with our previous work describing the increased prevalence of MSCs in cancer versus normal tissue (10). Interestingly, the proportion of MSCs within endometriosis tissue was equivalent to the proportion of MSCs within OCCC at both organ sites (Fig. 1A).

We next tested the functional impact of enMSCs on OCCC growth. After isolating and verifying MSCs from 15 endometriosis samples, we cocultured these MSCs with TOV-21G OCCC cells. Interestingly, the impact on growth rate was mixed with a subset of MSCs supporting OCCC growth and a subset of MSCs with no effect or even slight suppression of OCCC growth (Fig. 1B). Intrigued by this result, we assessed known MSC and endometrial stromal surface markers to determine if there was heterogeneity among the populations. MSC markers were identical across the cell lines consistent with all cells meeting the identity of a MSC (Supplementary Fig. S1B). However, we found significant alterations in the endometrial stromal marker, CD10 (Fig. 1C). CD10 is generally ubiquitously expressed on endometrial stromal cells (22). However, MSCs that promoted OCCC growth demonstrated a significantly lower proportion of cells with CD10 surface expression. This was further confirmed at the transcript level as MSCs with low surface CD10 expression similarly demonstrated decreased levels of CD10 mRNA (Fig. 1D and E). We also directly evaluated the proportion of enMSCs with and without CD10 surface expression from five additional endometriosis tissues (involving the ovary) demonstrating CD10⁻ enMSCs made up 1% to 5% of the total viable cells isolated from the patient samples (Supplementary Fig. S1C). We next assessed the CD10 surface expression of MSCs derived from ovarian clear cell cancer samples. The majority of MSCs derived from primary OCCC were negative for CD10 surface expression (Fig. 1D).

On the basis of the above findings, we grouped the enMSCs into CD10_high versus CD10_low populations (using a threshold of <40% of the enMSC population expressing CD10 for the “low” population and >60% for the “high” population). We confirmed the dichotomous impact of enMSCs on OCCC growth with CD10_low populations, supporting OCCC growth in all three lines tested and CD10_high populations restricting OCCC growth in TOV-21G and OVISE and having no impact on RMG1 growth. This was confirmed with eight independent patient-derived enMSCs with three independent OCCC cell lines (TOV-21G, RMG1, OVISE; Fig. 1F and G). MSCs containing between 40% and 60% CD10⁺ cells did not have a significant impact on OCCC growth. Importantly, presence or absence of CD10 expression, or intensity of CD10 staining did not impact the fundamental characteristics of MSCs with all lines meeting the ISCT criteria for MSCs, including marker expression, differentiation capacity, and baseline proliferation rates (Supplementary Fig. S1D).

To determine if the isolated enMSCs were derived from the ectopic endometrial stroma, which generated the endometriotic lesion or if they were derived from the underlying ovarian or fallopian tube tissue, we assessed the presence of markers reported to be expressed on endometrium-derived MSCs: CD146 and CD140b (23–25). Supplementary Fig. S2 demonstrates both the CD10_low and CD10_high enMSCs express surface levels of CD146 and CD140b, resembling the cell surface expression of primary endometrium-derived MSCs (obtained from human endometrium without pathology or endometriosis). In contrast, ovary and fallopian tube–derived MSCs and HGSC ovarian cancer derived CA-MSCs do not express cell surface CD146 and have substantially lower expression of CD140b. In addition, we assessed the impact of normal ovary and normal FT-derived MSCs on OCCC growth and found that neither MSC source altered OCCC growth (Supplementary Fig. S2B). Collectively, this supports an endometrial origin of the MSCs isolated from endometriotic tissue.

Given the unique microenvironment in which OCCC develops, we next sought to determine if CD10_low enMSCs specifically enhance OCCC growth or if their protumorigenic impact is generalizable to other ovarian cancer histologic subtypes. Using a direct coculture system, we grew CD10_low enMSCS with OCCC or HGSC ovarian cancer cells (OVSAHO and OVCAR3) and HGSC primary patient line pt412. As a comparator, we used CA-MSCs derived from primary patient HGSC ovarian cancer involving the ovary, which we previously demonstrated supports HGSC cell growth (9, 14, 26). CD10_low enMSCs specifically enhanced the growth of OCCC while having no impact on HGSC cell growth (Fig. 2A). HGSC-derived CA-MSCs increased HGSC cell growth while not impacting OCCC growth (Fig. 2A). This was repeated in additional OCCC lines, RMG1 and OVISE with similar results (Supplementary Figs. S3A and S3B). These provide further evidence that enMSCs from the endometriotic microenvironment are uniquely supportive of OCCC growth.

We expanded our investigation of the impact of enMSCs on OCCC growth by testing their effect on chemotherapy resistance and cancer cell sphere growth. CD10_low enMSCs significantly enhanced the resistance of OCCC cells to cisplatin therapy. Figure 2B demonstrates the survival of OCCC TOV-21G cells cocultured with CD10_low enMSCs or CD10_high enMSCs when treated with increasing doses of cisplatin. The (fluorescently labeled) TC component of the coculture was quantified. Coculture with CD10_low enMSCs significantly increased the survival of OCCC cells to cisplatin therapy. In contrast, the CD10_high enMSCs decreased the survival of OCCCs cells to cisplatin therapy. We confirmed these results using three independent enMSC lines with two additional OCCC cell lines (OVISE and RMG1; Fig. 2C; Supplementary Figs. S3C and S3D). The same coculture experiments performed under nonadherent, spheroid-forming conditions also resulted in significantly more TC containing spheres with CD10_low enMSCs. In contrast, CD10_high enMSCs significantly decreased TC sphere formation (Fig. 2D).

We next tested the impact of CD10_low versus CD10_high enMSCs on OCCC growth in vivo. We first used a subcutaneous model where TOV-21G cells with and without CD10_low enMSCs or CD10_high enMSCs were injected subcutaneously into mice. After 2 weeks, CD10_low enMSC containing tumors demonstrated increased tumor initiation with 90% engraftment versus 60% engraftment in the TOV-21G alone group and 40% engraftment in the CD10_high enMSC containing tumors (Fig. 3A). In addition, the CD10_low enMSC containing tumors also grew faster than the other two groups (Fig. 3B). Surprisingly, we also noted development of metastasis from the primary subcutaneous tumor with 60% of mice with CD10_low enMSC containing tumors developing intra-abdominal metastasis and 80% developing lung metastasis. In comparison, 20% of mice with TOV-21G alone tumors developed abdominal metastasis related to direct extension of the tumor into the peritoneal cavity. In the CD10_high enMSC containing tumor group, 20% of these mice developed lung metastasis (Fig. 3C).

To further evaluate the impact of enMSCs on OCCC growth, we used an orthotopic model where TOV-21G TCs with or without CD10_low enMSCs or CD10_high enMSCs were injected into the unilateral ovarian bursa as described previously (10, 12). TCs stably expressed luciferase and tumor engraftment, growth, and metastasis were monitored with IVIS bioluminescence imaging. Similar to the subcutaneous model, CD10_low enMSC containing tumors initiated significantly faster than CD10_high enMSC containing tumors or TCs alone. In addition, the CD10_low enMSC containing tumors demonstrated faster growth and mice developed significantly more metastases compared with the other two groups (Fig. 3D). Ultimately, mice with CD10_low enMSC containing tumors demonstrated decreased survival compared with mice with CD10_high enMSC containing tumors (Fig. 3E). At the time of necropsy, we isolated both primary and metastatic tumors and assessed the presence and proportion of enMSCs at each site using flow cytometry. Both primary and metastatic tumors contained human enMSCs ranging from 0.2% to 5% of viable cells within the tumor. This is consistent with our previous work in HGSC demonstrating MSCs travel with cancer cells to metastatic sites (10). We quantified the frequency of CD10⁺ cells among all tumor-isolated enMSCs from the initial CD10_low enMSC and CD10_high enMSC tumor groups. Interestingly, although the initial CD10_low enMSC group (28% with CD10 surface expression) continued to demonstrate a low proportion of cells with surface CD10 expression at the end of the in vivo experiment (20.6% with CD10 surface expression, P = 0.2), the initial CD10_high enMSC group (86% with CD10 surface expression) demonstrated a significant reduction in cells with CD10 surface expression at the end of the in vivo experiment (32% with CD10 surface expression, P < 0.001; Fig. 3F). These indicate there was either a selection for MSCs without CD10 expression in the tumors, which grew and metastasized, or TCs induced the reduction of CD10 expression in enMSCs in vivo.

We next sought to determine the mechanism of CD10 surface expression alterations in enMSCs. We first validated that enMSC CD10 surface expression also changes in vitro with cancer stimulation. We grew CD10_high enMSCs with TOV-21G cells for 5 days. Consistent with our in vivo data, over time, the number of MSCs with CD10 surface expression decreased, changing from 69% positive to 14% positive by day 5 (Fig. 4A). We also tested which coculture conditions induced the greatest change in CD10 prevalence. We grew CD10_high enMSCs with TOV-21G conditioned media versus indirect coculture with TOV-21G cells in a transwell system versus direct coculture with TOV-21G cells. After 5 days, the prevalence of CD10 surface expression on enMSCs was quantified with flow cytometry. Direct coculture was the most effective means of altering CD10 prevalence changing from 90% CD10⁺ enMSCs at baseline to 29% CD10⁺ after coculture. The other experimental conditions had minimal effect on CD10 expression with 87.6% CD10⁺ after 5 days of conditioned media and 84.3% CD10⁺ after indirect coculture (Supplementary Fig. S4).

We next tested if differential growth rates between CD10⁻ and CD10⁺ enMSC lead to the altered proportion of CD10 expressing cells. We sorted enMSCs based on CD10 surface expression into CD10⁺ and CD10⁻ groups. We fluorescently labeled the CD10⁺ cells with cell trace blue and the CD10⁻ cells with cell trace yellow to enable tracking and isolation of the original populations. We mixed the labeled CD10⁺ and CD10⁻ enMSCs in a 1:1 ratio. We quantified the proliferation of each population over time. There was no significant difference in the growth rate of the CD10⁺ or CD10⁻ enMSCs grown without cancer cells, demonstrating no inherent proliferation differences (Fig. 4B). We next directly cocultured the labeled CD10⁺ and CD10⁻ enMSCs with TOV-21G cells for 5 days (a time frame previously sufficient for HGSC cancer cells to educate MSCs; ref. 14). Surprisingly, when grown with cancer cells, the proliferation of the CD10⁻ enMSCs increased ultimately growing faster than the CD10⁺ population (Fig. 4B). Isolation of the initial CD10⁺ enMSCs (fluorescently labeled to separate the initial CD10⁺ population regardless of CD10 expression after coculture) also demonstrated decreases in CD10 mRNA after TC coculture (Fig. 4C). This is consistent with both decreased CD10 transcriptional expression in the CD10⁺ population as well as increased proliferation of the CD10⁻ enMSC population. These indicate that OCCC both induces the transcriptional downregulation of CD10 and selectively enhances the growth of CD10⁻ enMSCs.

We next tested the functional significance of CD10 expression. We selected a CD10_low, cancer supportive enMSC sample and a CD10_high, cancer restrictive enMSC sample and sorted them into CD10⁺ and CD10⁻ populations. We then tested the impact of the CD10⁺ versus CD10⁻ enMSCs on OCCC growth. The CD10⁻ subset derived from either the CD10_low or CD10_high parent cells significantly increased OCCC cancer growth compared with the CD10⁺ subsets (from the CD10_low or CD10_high parent cells; Fig. 4D). Thus, the expression of CD10 rather than the phenotype of the parent cells drives the cancer supportive properties of enMSCs. Next, we created CD10 knockdown and CD10 overexpression enMSCs. We took the CD10⁺ subset of a native CD10_high, nonsupportive enMSC sample and created stable CD10 knockdown (KD) via lentiviral shRNA. This resulted in a 5-fold reduction in CD10 surface expression compared with scrambled control parent enMSCs (Fig. 4E; Supplementary Fig. S5). CD10 KD enMSCs versus control parent enMSCs significantly enhanced the growth and chemotherapy resistance of OCCC (Fig. 4E). The impact of CD10 KD was validated in an additional, independent CD10⁺ enMSC primary patient sample (Supplementary Fig. S5E).

We next overexpressed CD10 to determine if this would reverse the protumorigenic effects of the cancer supportive enMSCs. We used the CD10⁻ population of a native CD10_low, cancer supportive enMSC sample. We then used a lentiviral CD10 construct to overexpress CD10, resulting in a 3-fold increase in surface expression of CD10 (Fig. 4F; Supplementary Fig. S5). CD10 overexpressed (OE) enMSCs versus scrambled control parent enMSCs lost the ability to support OCCC growth and chemotherapy resistance (Fig. 4F). The impact of CD10 OE was validated in an additional, independent CD10⁻ enMSC primary patient sample (Supplementary Fig. S5F).

To determine the difference between CD10⁺ and CD10^– enMSCs and to identify critical pathways that mediate their impact on OCCC growth, we performed RNA-seq on the CD10⁺ population of CD10_high enMSCs from 3 independent patients and the CD10⁻ population of CD10_low enMSCs from an additional 5 independent patients. To assess the impact on cancer cells, we also performed RNA-seq on TOV-21G cancer cells cocultured with CD10⁻ enMSCs or CD10⁺ enMSCs (four independent cocultured samples per group). Figure 5 demonstrates the unsupervised hierarchical clustering of the CD10⁻ versus CD10⁺ enMSCs demonstrating clear separation by CD10 expression with distinct transcriptomic changes. A volcano plot of the most significantly DEGs demonstrates increased expression of 606 genes and decreased expression of 506 genes (Fig. 5B, i). Similarly, comparison of the cancer cells cocultured with CD10⁻ versus CD10⁺ enMSCs also demonstrated significant alterations in the transcriptome with 1,054 upregulated and 1,355 downregulated genes (Fig. 5A, ii and B, ii). Top DEGs were validated using qRT-PCR in independent enMSC and cocultured TC samples. Pathway analysis highlighted changes in metal/ion transport and regulation as the top differentially regulated pathways in cancer cells grown with CD10⁺ versus CD10⁻ enMSCs (Fig. 5D). In additiojn, a gene associated with protection against ferroptosis (iron-mediated cell death), ribonucleotide reductase regulatory subunit M 2 (RRM2), was significantly overexpressed in TCs grown with CD10⁻ enMSCs (Fig. 5C, ii; ref. 27). In line with this, Hephaestin (HEPH) was one of the top genes significantly overexpressed in the CD10- enMSC population. Hephaestin is a ferroxidase critical to iron export (28). Hephaestin converts Fe²⁺ to Fe³⁺ enabling iron to be transported out of the cell by ferroportin (FPN1; ref. 29).

Given alterations in metal transport and handling in OCCC grown with CD10⁻ enMSCs and the upregulation of HEPH in the CD10⁻ enMSCs, we investigated iron regulation in the TME as a potential mechanism of CD10⁻ enMSC-mediated OCCC growth promotion.

We hypothesized that increased HEPH in the CD10⁻ enMSCs results in iron export, which increases iron in the extracellular TME for use by OCCC cells to support growth. Iron is critical to cancer cell growth, and OCCC may have unique requirements for iron given its association with hypoxia and endometriosis (6, 30, 31). We first investigated the expression of iron regulatory genes in enMSCs. We validated increased protein levels of HEPH in CD10⁻ versus CD10⁺ enMSCs via Western blot analysis (Fig. 6A and B). Consistent with increased iron export capacity, we also demonstrated increased FPN1 in CD10⁻ versus CD10⁺ enMSCs (Fig. 6A and B).

We also demonstrated significantly increased levels of the iron storage protein, ferritin (FT), in CD10⁻ versus CD10⁺ enMSCs. FT is composed of two subunits, ferritin H (FTH) and ferritin L (FTL). FTH is the subunit with ferroxidase activity necessary to store iron within the ferritin complex whereas FTL is the main storage subunit (32). The levels of FTH were not significantly different whereas levels of FTL were increased 2-fold in CD10⁻ versus CD10⁺ enMSCs. Interestingly, when we evaluated TCs with and without coculture with CD10⁻ versus CD10⁺ enMSCs, we identified a significant shift in the ratio of FTH versus FTL. Coculture of TCs with CD10⁻ enMSCs resulted in a 2.6-fold increase in FTL but −2.5-fold decrease in FTH. Coculture with CD10⁺ enMSCs demonstrated nonsignificant changes in FTH and FTL. We then quantified the levels of transferrin receptor (TFR; the main mechanism of cellular iron uptake; ref. 33) in OCCCs with and without coculture with CD10⁻ and CD10⁺ enMSCs. Interestingly, coculture with CD10⁻ and CD10⁺ enMSCs led to an increase in TC TFR compared with TCs grown alone. As FT and TFR are canonically inversely related via posttranslational regulation based on intracellular iron stores, this finding was unexpected (34). To more definitively assess levels of intracellular iron, we next measured the LIP (the amount of free or accessible iron within the cell) in CD10⁻ versus CD10⁺ enMSCs and in TOV-21G cells before and after coculture with enMSCs. The TOV-21G LIP is strikingly increased (2.2-fold increase) after coculture with CD10⁻ enMSCs and decreased after coculture with CD10⁺ enMSC with a −3-fold decrease in LIP (Fig. 6C). CD10⁻ enMSC have a baseline higher LIP compared with CD10⁺ enMSCs and there is a proportional decrease in the LIP of CD10⁻ enMSCs after coculture with TOV-21G cells, which corresponds to the rise in LIP of TOV-21G cells (Fig. 6D). This provides evidence that CD10⁻ enMSCs donate iron to OCCC cells.

To further test this hypothesis, we used the iron chelator, deferoxamine (DFO; Fig. 6E). We treated OCCC cells grown alone or with CD10⁻ versus CD10⁺ enMSCs with increasing doses of DFO. DFO was added to standard serum-supplemented media, which contains iron and transferrin. Chelation of iron with DFO decreased the proliferation of OCCC cells grown alone, demonstrating the necessity of iron for OCCC growth (Fig. 6E). At low levels of DFO, CD10⁻ enMSCs were able to rescue OCCC cells from the growth inhibiting effects of DFO, consistent with donation of iron by CD10⁻ enMSCs. However, increased doses of DFO blocked CD10⁻ enMSC-mediated OCCC growth promotion, indicating high levels of iron chelation can overcome the effects of CD10⁻ enMSCs. This was not due to toxicity of DFO on the enMSC population. Collectively, this indicates CD10⁻ enMSCs support OCCC growth through enhancing intracellular iron.

To test if increased extracellular iron alone (without the presence of enMSCs) alters OCCC growth, we added increasing doses of FAC (0.05–0.2 μmol/L) to TOV-21G cells grown alone in standard serum-supplemented media and measured proliferation over time. Interestingly, FAC did not significantly enhance the proliferation of OCCC and at higher levels it proved toxic. We concurrently assessed the changes in canonical iron import and storage proteins (TFR, FTL, and FTH) and the LIP. FAC treatment decreased TFR and increased FTL and FTH with no significant change in the LIP consistent with expected canonical regulation of iron storage proteins and maintenance of a stable LIP (Supplementary Fig. S6).

The above data indicates CD10⁻ enMSCs enhance OCCC growth at least in part through increasing the LIP. However, labile intracellular iron can be toxic, leading to increased reactive oxygen species, lipid peroxidation, and ferroptosis. Indeed, as demonstrated above, OCCC cells grown with CD10⁻ enMSCs upregulate RRM2, which is a ferroptosis-responsive gene. On the basis of this, we hypothesized that CD10⁻ enMSCs induce a unique susceptibility of OCCC to ferroptosis. To test this, we treated OCCC cells alone or with CD10⁻ versus CD10⁺ enMSCs with the ferroptosis inducer, erastin (35). Erastin, even at doses as low as 0.5 μmol/L, significantly decreased the number of OCCC cells grown with CD10⁻ enMSCs whereas only the highest dose of erastin (10 μmol/L) significantly decreased the numbers of OCCC cells grown alone. CD10⁺ enMSCs provided protection against erastin at the 10 μmol/L dose (Fig. 7A). The IC₅₀ of erastin in OCCC cells (TOV-21G) alone was 7.2 μmol/L versus 1.1 μmol/L in OCCC cells + CD10⁻ enMSCs. We repeated these experiments with another ferroptosis inducer, DHA (36). DHA treatment resulted in a 2- to 3-fold increase in OCCC cell death when grown with CD10⁻ enMSCs (IC₅₀ 1.2 μmol/L) versus OCCC cells grown alone (IC₅₀ 5.1 μmol/L) or with CD10⁺ enMSCs (IC₅₀ 9.5 μmol/L) consistent with CD10⁻ enMSC enhancing the sensitivity of OCCC cells to ferroptosis (Fig. 7B).

To verify the induction of ferroptosis, we performed BoDIPY 581/591 C11 staining followed by flow cytometry to detect lipid peroxidation, which is a hallmark of ferroptosis (21). After treatment with 1 μmol/L erastin, OCCC cells grown with CD10⁻ enMSC demonstrated significantly more lipid peroxidation compared with OCCCs grown alone or with CD10⁺ enMSCs consistent with increased induction of ferroptosis (Fig. 7C). We also performed immunofluorescence to detect 4-HNE, which is a derivative of lipid peroxidation (37). Consistent with the BoDIPY analysis, 4-HNE staining was significantly increased in OCCCs with CD10⁻ enMSC versus OCCCs alone or OCCCs with CD10⁺ enMSCs treated with low doses of erastin (1 μmol/L; Fig. 7D).

We then performed an in vivo study with TOV-21G cells with CD10⁻ versus CD10⁺ enMSCs injected into the ovarian bursa of NSG mice. TCs were luciferase labeled. Tumors were allowed to engraft for 5 days. Low-dose erastin treatment (10 mg/kg daily) or vehicle control was started on day 5 and continued daily for 20 days. Tumor burden was followed with IVIS imaging and mice were followed for survival. Erastin significantly decreased tumor growth and improved the survival of mice with CD10⁻ enMSC containing tumors (Fig. 7E and F). Mice with CD10⁻ enMSC containing tumors treated with erastin had equivalent survival to mice with CD10⁺ enMSC containing tumors (Fig. 7F). This indicates erastin reversed the protumorigenic impact of CD10⁻ enMSCs and ultimately turned the growth promoting effects of iron donation into a therapeutic liability significantly sensitizing OCCC cells to ferroptosis. At time of sacrifice, we resected tumors and verified enMSCs were present in all groups at both the primary and metastatic sites. We also compared the CD10 surface expression of isolated enMSCs from the CD10⁻ groups. Interestingly, the CD10 expression on enMSCs increased with erastin treatment compared with control treatment (Fig 7G). Thus, erastin both induced ferroptosis in TCs grown with CD10⁻ enMSCs and altered the CD10 expression on enMSCs, resulting in a higher proportion of CD10⁺ enMSCs, both of which may contribute to the overall survival benefit of erastin treatment.

Here we identify a novel subset of enMSCs that selectively support OCCC growth. These OCCC-supportive enMSCs are marked by decreased CD10 expression. This subset of CD10⁻ enMSCs have increased iron export proteins, resulting in reciprocal increases of intracellular iron within OCCC cells. Thus, iron regulation is a critical mechanism driving CD10⁻ enMSC support of OCCC growth. Importantly, this also implies a unique therapeutic vulnerability as CD10⁻ enMSC increase the susceptibility of OCCC to the ferroptosis inducers erastin and DHA. Therefore, the growth promoting effects of CD10⁻ enMSCs can be weaponized to kill OCCC.

Our work further links the specific endometriotic microenvironment with OCCC growth and supports the unique role of endometriosis-derived stromal cells in the pathophysiology of OCCC. It is intriguing that CD10⁻ enMSCs support OCCC but not HGSC, further highlighting the specificity of unique microenvironments for selective cancer growth. These findings also emphasize the fundamental differences in ovarian cancer histologic subtypes, underlining the need to study each subtype independently. It should be noted that we used the TOV-21G cell line as our main OCCC model, however results were repeated and confirmed in RMG1 and OVISE cell lines. Although RMG1 and OVISE are largely considered representative of clear cell histology, they lack some classic clear cell features (RMG1 lacks ARID1a mutation) and this may be why, in some experiments, the impact of CD10⁻ enMSCs are more pronounced in the TOV-21G line (38, 39).

It is also interesting that not all endometriosis derived MSCs support OCCC growth. Identification of CD10 loss as a marker of MSCs with cancer supportive properties points to both a potential biomarker of “at risk” endometriosis tissue and begins to identify potential mechanisms for the stromal support of OCCC. We also demonstrate that OCCC acts upon enMSCs to induce a protumorigenic phenotype. OCCC cells alter the proportion of CD10⁺ versus CD10⁻ enMSCs, thus converting a baseline CD10_high enMSC population into a CD10_low enMSC population. This corresponds with the acquisition of OCCC supportive properties. Indeed, our in vitro work demonstrated CD10_high enMSCs restrict OCCC growth, however this was not pronounced in vivo. Upon further evaluation, after in vivo growth with OCCC, the initial CD10_high enMSC population demonstrated a significant decrease in CD10 expression such that only 20% to 30% of enMSC continued to express CD10. The loss of CD10 expression likely altered the phenotype of the enMSC, thus leading to growth promotion rather than suppression. CD10 OE and KD experiments confirmed the functional importance of CD10, however it is possible that other cell features, intrinsic to the enMSC or as a result of CD10 expression, also contribute to the impact of enMSCs on OCCC. Further work is necessary to delineate the full impact of CD10 on enMSC function.

The ability of OCCC cells to alter the surrounding stromal population to enhance the prevalence of the CD10⁻ subset of MSCs is similar to our findings in HGSC, where cancer cells convert normal MSCs into cancer supportive CA-MSCs. We demonstrate that OCCC cells induce a selective growth advantage of the CD10⁻ enMSC population but also transcriptionally downregulate CD10 expression in CD10⁺ enMSCs. We also demonstrate direct coculture of enMSCs with OCCCs is the most effective means of CD10 loss implying direct cell interactions mediate the alterations in CD10. The mechanism underlying these alterations has yet to be discovered and will be the focus of future work. However, it is clear that the loss of CD10 expression is functionally important conveying protumorigenic properties to enMSCs.

The identification of altered iron regulation mediating the stromal support of OCCC is critically important. Although the mechanism governing total body iron homeostasis is well-defined, very little is known regarding the handling of iron within the tumor microenvironment. This may be of particular importance in the endometriosis microenvironment given the accumulation of iron due to continued menstrual cycling and thus the handling of iron may be vital to the prevention or development of OCCC. Here we present a new paradigm with CD10⁻ enMSCs donating iron to OCCC to mediate cancer cell growth. Interestingly, the donated iron appears to remain in the LIP rather than sequestered in ferritin. This may be due to the altered FTH versus FTL ratio with a relative decrease in FTH, which is the subunit necessary for iron oxidation to enable iron storage. The finding that increasing extracellular iron alone is insufficient to enhance the OCCC LIP and cell proliferation further supports the importance of CD10⁻ enMSC alteration of TC iron handling. OCCC cells respond to increased extracellular iron alone (without the presence of CD10⁻ enMSCs) with canonical increases in both FTL and FTH, decreased TFR and maintenance of a stable LIP without appreciable change in cell proliferation. This indicates that not only are the CD10⁻ enMSCs exporting iron, they are also altering iron handling within the TC, which ultimately increases the LIP and cell proliferation.

This increase in labile iron appears necessary for enMSC-mediated OCCC growth, however this leaves the OCCC cells susceptible to ferroptosis. Indeed, decreased FTH is associated with ferroptosis (40). Even relatively low doses of ferroptosis inducers, erastin or DHA, significantly decreased the survival of OCCC cells grown with CD10⁻ enMSCs in vitro with an IC₅₀ of 1.1 and 1.2 μmol/L respectively (compared with 7.2 and 5.1 μmol/L with OCCC cells alone). Other groups have investigated the cytotoxicity of ferroptosis inducers in cancer cells potentially susceptible to ferroptosis such as gastric cancer (erastin IC₅₀ of 14 μmol/L; ref. 41) and hepatocellular cancer (DHA IC₅₀ values of 20–40 μmol/L; ref. 42). More importantly, ferroptosis induction in the presence of CD10⁻ enMSCs lead to dramatic decreases in tumor growth and increases in mouse survival in vivo. Importantly, the majority of MSCs associated with OCCC (100% of the cases we have sampled, Fig. 1D) are CD10_low thus ferroptosis inducers represent a promising therapeutic avenue for OCCC. Importantly, in vivo erastin treatment also impacted the phenotype of the enMSCs increasing the proportion of CD10⁺ enMSCs. This indicates inducing ferroptosis may have a dual effect of both enhancing TC death and altering the protumorigenic stroma thus magnifying the potential therapeutic benefit.

In addition, the loss of CD10 stromal expression may mark endometriosis tissue at risk for malignant transformation and may have important implications for risk stratification and prevention strategies. Future work is needed to explore the relationship between CD10 stromal loss and malignant transformation within the endometriotic microenvironment.

Collectively, this work provides novel insights into the stromal support of OCCC growth, and specifically identifies stromal-mediated iron regulation within endometriosis as an important driver of OCCC biology. This work also demonstrates the potential to capitalize on stromal-mediated iron regulation to increase OCCC susceptibility to ferroptosis.

A.M. Vlad reports grants from UPMC during the conduct of the study; grants from NIH, Merck, OCRA, and DOD-OCRP outside the submitted work; and also has a patent for Application 16/798,137 pending. L.G. Coffman reports grants from The Brenda and Roger Gibson Clear-Cell Cancer Research Fund during the conduct of the study and other support from Immunogen outside the submitted work. No disclosures were reported by the other authors.

H.I. Atiya: Conceptualization, formal analysis, investigation, writing–original draft, writing–review and editing. L. Frisbie: Formal analysis, investigation, writing–review and editing. E. Goldfeld: Investigation, writing–review and editing. T. Orellana: Investigation, writing–review and editing. N. Donellan: Resources, investigation, methodology. F. Modugno: Resources, methodology. M. Calderon: Investigation, writing–review and editing. S. Watkins: Investigation, writing–review and editing. R. Zhang: Conceptualization, investigation, writing–review and editing. E. Elishaev: Resources, investigation. T.R. Soong: Formal analysis, investigation, writing–review and editing. A. Vlad: Resources, methodology, writing–review and editing. L. Coffman: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft, project administration, writing–review and editing.

L.G. Coffman was supported by the Brenda and Roger Gibson Clear-Cell Cancer Research Fund and grant K08CA211362-04. H.I. Atiya was supported by The HERA Ovarian Cancer Foundation Outside the Box Research Grant.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.