Cancer-Associated Fibroblast-Like Tumor Cells Remodel the Ewing Sarcoma Tumor Microenvironment

Intratumoral heterogeneity promotes cancer growth, metastasis, therapeutic failure, and drug resistance (1, 2). Cancer-associated fibroblasts (CAF) contribute to heterogeneity of epithelial-derived tumors by depositing extracellular matrix (ECM) proteins that dynamically remodel the tumor microenvironment (TME; ref. 3). In addition, genomic instability of tumor cells leads to clonal evolution that contributes to mutational heterogeneity of the tumor cell compartment (2). While genomic instability can drive tumor cell heterogeneity, more recent studies have highlighted the role of non-mutational epigenetic heterogeneity (4). Single-cell sequencing and spatial analyses demonstrate that tumor ecosystems are comprised of heterogeneous tumor cells with distinct epigenetic and phenotypic states (5, 6). Non-mutational tumor cell heterogeneity is particularly relevant to pediatric tumors which typically have low mutation rates but high transcriptional and phenotypic heterogeneity due to hijacking of developmental and epigenetic regulators (7). Phenotypically distinct tumor cell subpopulations differentially contribute to tumor growth, metastatic potential, and therapy response, highlighting the critical importance of studying tumor cell subpopulations in the context of contextually relevant ecosystems (8–10).

Ewing sarcomas (EwS) are histologically primitive, monomorphous bone and soft-tissue tumors that present across the lifespan, with a peak incidence in adolescence (11). Intensive cytotoxic therapy has improved prognosis for patients with localized tumors, but outcomes for patients with metastastic and relapsed EwS remain dismal (12). EwS tumors are initiated by chromosomal translocations between FET and ETS gene family loci, that generate chimeric in-frame fusion proteins, most commonly EWS::FLI1 (11). These fusions function as oncogenic transcription factors that cause widespread epigenomic reprogramming and aberrant activation and repression of hundreds of target genes (11, 13, 14). Recurrent mutations in other genes are uncommon and propagation of EwS tumors depends on continued expression of the fusion (11). Although intratumoral genomic instability is not a known feature of EwS, rare subpopulations of tumor cells that express low levels of EWS::FLI1 have been identified and these cells display enhanced metastatic properties in tumor models (15). In addition, modulation of EWS::FLI1-dependent gene signatures, without altering expression of the fusion itself, influences tumor cell state (reviewed in ref. 16). Genomic loss of STAG2 (17, 18), expression of tissue-specific transcription factors (19–21), activation of the Wnt/β-catenin pathway (22), and other cell-intrinsic and cell-extrinsic mechanisms (16), all influence transcriptional activity of EWS::FLI1 and affect the tumorigenic and metastatic properties of EwS cells. Importantly, despite different underlying molecular origins, EWS::FLI1-depleted cells universally show increased expression of mesenchymal lineage genes that are normally repressed by the fusion. The contribution of EWS::FLI1-low, mesenchymal-high state tumor cells to EwS pathogenesis has yet to be defined in the context of tumors in vivo.

Here, we set out to identify and characterize EWS::FLI1-low cells in EwS tumors in situ. Transcriptomic profiling of EwS cell lines and patient tumors identified NT5E (CD73) as a cell surface marker that consistently marks EWS::FLI1-low tumor cells in vitro and in vivo. Phenotypic, proteogenomic, and immunofluorescence profiling confirmed that CD73⁺ cells exist as spatially distributed cell subpopulations that are enriched along tumor borders and invasive fronts. Further, our studies showed that CD73⁺ cells deposit protumorigenic ECM proteins such as tenascin-C (TNC), collagens, and proteoglycans, proteins that are largely deposited by CAFs in carcinomas (23). This matrisomal gene program is normally repressed by EWS::FLI1, validating that CD73⁺ tumor cells have lost EWS::FLI1 activity. Despite this, we found that expression of the EWS::FLI1-activated gene signature is largely retained by CAF-like tumor cells, revealing the novel discovery that the activating and repressive properties of the fusion can be dissociated in individual tumor cells, creating cells with hybrid transcriptional states. Together these studies identify the existence of transcriptionally and phenotypically distinct subpopulations of EwS tumor cells that contribute to tumor heterogeneity and remodel the TME via deposition of pro-tumorigenic ECM.

EwS cell lines A673 (RRID:CVCL_0080), SKNMC (RRID:CVCL_0530), CHLA10 (RRID:CVCL_6583), CHLA9 (RRID:CVCL_M150), A4573 (RRID:CVCL_6245), TC32 (RRID:CVCL_7151), TC71 (RRID:CVCL_2213), RDES (RRID:CVCL_2169) as well as U2OS (RRID:CVCL_0042, osteosarcoma cell line) were obtained from ATCC and COG (https://www.childrensoncologygroup.org/) cell line repositories. A673, TC32, TC71, A4573, and SKNMC cell lines were maintained in RPMI1640 media (Gibco) supplemented with 10% FBS (Atlas Biologicals) and 2 mmol/L L-glutamine (Life Technologies). RDES was cultured in RPMI 1640 media supplemented with 15% FBS and 2 mmol/L L-glutamine. CHLA10 and CHLA9 were maintained in IMDM media (Fisher) supplemented with 20% FBS, 2 mmol/L L-glutamine, and 1X Insulin–Transferrin–Selenium–Ethanolamine (Gibco). U2OS was maintained in McCoy's 5A media (Gibco) supplemented with 10% FBS and 2 mmol/L L-glutamine. H7-MSCs (kind gift from Dr. Sweet-Cordero) were maintained in alpha-MEM supplemented with 10% FBS, 2 mmol/L L-glutamine, 1% antibiotic–antimycotic (Gibco). Cells were cultured at 37°C with 5% CO₂. Cells were all confirmed to be Mycoplasma-free and identities subject to short tandem repeat profiling confirmation every 6 months.

PDX305 was generated from an abdominal soft-tissue metastasis from 17-year-old male patient with EwS. Cells were passaged as a patient-derived xenograft in NSG mice before cell line generation. PDX305 early passage cells were maintained in RPMI 1640 media (Gibco) supplemented with 10% FBS (Atlas Biologicals), 2 mmol/L l-Glutamine (Life Technologies), and 1% antibiotic–antimycotic. Poly(A)-capture RNA sequencing (RNA-seq) was performed on PDX305 cells as described in the RNA-seq sections.

EwS cells were trypsinized, washed with PBS, resuspended in 2% FBS in PBS, and stained with fluorophore-conjugated antibodies (Supplementary Table S1) for 30 minutes. on ice in the dark. Fluorophore-conjugated isotype controls and unstained controls were used for each experiment. Flow cytometry was performed on a BD Accuri C6 (10,000 events), with single-cell identification by forward scatter and side scatter. For FACS, cells were sorted on a BD FACSAria II. Analysis was conducted in FCS Express (De Novo Software).

Total RNA was extracted from cells using the RNeasy Mini Kit (Qiagen) and cDNA was generated using the iScript cDNA Synthesis Kit (Bio-Rad) according to the manufacturer's instructions. qPCR was performed using either Taqman Fast Universal PCR Master Mix (Applied Biosystems) assays or iTaq Universal SYBR-Green Supermix (Bio-Rad) on a Roche Light-Cycler 480 instrument (Roche Applied Science). Samples were run in technical triplicates and average Ct values were normalized to the geometric mean of two reference genes. The relative mRNA expression was calculated by the ddCt method. Primers and TaqMan assays.

A tumor tissue microarray (TMA) of EwS tumors was prepared by Seattle Children's Hospital Pathology. For IHC, TMA sections were pretreated with CC1 Tris buffer (pH 8) for 32 minutes, primary antibodies (see Supplementary Table S1) were incubated 1:100 for 32 minutes at 36°C. Staining was detected using OptiView DAB (Roche).

For immunofluorescence, cells were fixed with 4% PFA in PBS for 10 minutes then washed with PBS. Sorted CD73⁻/CD73⁺ cells were plated at 60,000 cells/well in 8-well glass bottomed chambers overnight. Cells were permeabilized with 10 minutes. 0.3% Triton-X then blocked with 0.2% BSA in PBS for 1 hour at room temperature. Primary antibodies were incubated for 1 hour at room temperature in 0.2% BSA (Supplementary Table S1). Alexa-fluor–conjugated secondary antibodies were incubated for 1 hour at room temperature in 0.2% BSA with 5% host serum (unless goat and donkey-derived secondary antibodies were used simultaneously, in which case host serum was not included). Signal was compared with matched species-specific IgG controls incubated at the same concentrations. For F-actin staining ActinGreen (ThermoFisher Scientific, #R37110) was used at 2 drops/mL in the secondary antibody solution. DAPI was included during secondary antibody incubations to mark nuclei at 1 μg/mL. Coverslips were mounted with ProLong Gold and allowed to cure overnight at room temperature in the dark. Slides were imaged on a Leica SPE confocal at 40× magnification or scanned using an Olympus slide scanner (BX61) at 10× and 20× magnification.

Formalin-fixed, paraffin-embedded (FFPE) tumor samples from xenografts or primary patient samples were deparaffinized, then antigen retrieval was carried out using Diva Decloaker in a Biocare Medical decloaking chamber (120°C for 30 seconds) and slides were allowed to cool to room temperature. Slides were blocked with 0.2% BSA for 1 hour at room temperature. Primary and secondary antibodies (Supplementary Table S1) were incubated in 0.2% BSA for 1 hour at room temperature. When both goat and donkey derived antibodies were used, donkey–anti-goat secondary (RRID:AB_2535853) was applied first, washed 3 × 10 minutes. with PBS, then additional secondaries were added to prevent cross-species reactivity. DAPI was included during secondary antibody incubations to mark nuclei at 1 μg/mL. Coverslips were mounted with Prolong Gold and allowed to cure overnight at room temperature in the dark, then slides were stored at 4°C until imaging. Slides were imaged on a Leica SPE confocal at 40× magnification or scanned on an Olympus slide scanner at 10× and 20× magnification.

Real-time cell analysis of cell migration and proliferation was monitored using a CIM-plate 16 or E-plate 16, respectively, on the xCELLigence DP system (Acea Bioscience). For proliferation, wells were coated with 0.2% gelatin and 100 μL complete media was added (5 × 10³ cells/well), and plates equilibrated for 1 hour at 37°C. For migration, before cell seeding, electrodes were coated with 0.2% gelatin and 50 μL complete media was placed in the upper chambers and 160 μL serum-free media was added to the lower chambers. 5 × 10⁴ cells/well were plated in the upper chamber in 100 μL media, then plates were equilibrated for 30 minutes at room temperature. Migration was evaluated up to 30 hours. Proliferation was evaluated up to 96 hours.

Spheroids were formed by plating EwS single cells at a density of 50,000 viable cells/mL in complete media in non-adherent 6 well plates overnight. Debris and single cells were removed by centrifuging samples for 10 seconds at 300 g. Rat tail collagen (Gibco) was prepared at 1.8 to 2 mg/mL in 0.08 N acetic acid, supplemented with 10X DMEM (final concentration 1X), neutralized with 1 mol/L NaOH and allowed to polymerize on ice for 45 to 60 minutes until fibers formed (24). Spheroids and collagen gels were mixed and plated as 80 μL droplets and allowed to polymerize for 30 minutes at 37°C, at which point 1 mL of complete IMDM media (20% FBS, 1X l-Glutamine, ITS-X) was added. 40 μL underlays of polymerized rat tail collagen were first applied to the bottom of a tissue culture–treated 24wp plate to prevent spheroids from attaching to or spreading along the plate surface. Spheroids were cultured 4 to 6 days then assessed by phase contrast microscopy and fixed for 15 minutes with 4% PFA in PBS. Invasion was scored as protrusive cells or multicellular strands extending from the spheroid border.

Cells were sorted on the basis of CD73 positivity and 540,000 (CHLA10) or 720,000 (A673) CD73⁻ or CD73⁺ single cells were resuspended in PBS, diluted 1:1 in Matrigel (Corning), and injected subcutaneously into NSG mice (005557, Jackson Laboratory). Tumor formation and growth was measured by calipers every other day. All mice were euthanized at end point when the tumor of one animal reach 2-cm diameter in any direction.

GeoMx Digital Spatial Profiling (NanoString) was carried out according to the manufacturer's instructions. Tumor specimens were generated as follows: subcutaneous injection of 4 × 10⁵ CHLA10 cells in 50% Matrigel, primary tumors collected 54 days postinjection; tail vein injection of 7 × 10⁵ CHLA10 cells injected in PBS, tumors collected 28 days postinjection. Tumors were fixed for 24 hours in 10% neutral buffered formalin before paraffin embedding. Three sections of FFPE tumor tissues (1 from subcutaneous tumor, 2 from 2 different tail vein–injected mice) were sectioned and adhered to a single slide. Hybridization of the Human next-generation sequencing (NGS) Whole Transcriptome Atlas probe library was performed according to the manufacturer's instructions overnight at 37°C. Each probe is tagged with a photocleavable unique molecular identifier for NGS-based readout. Slides were blocked for 30′ before staining with morphology markers (Supplementary Table S1), and SYTO DNA dye. After slide scanning, 12 regions of interest (ROI) comprising > 1,000 SYTO+ nuclei each were selected, UV-cleaved, and collected for NGS sequencing to quantify probe counts. Library preparation and Illumina NGS were carried out by NanoString. The NanoString GeoMx NGS data analysis pipeline was used (version 2.3.4) to assess probe counts, including probe and count-level QC steps using the GeoMxTools R package (RRID:SCR_023424). Gene level count data were normalized using the quartile 3 (Q3) method comparing the upper quartile of counts in each segment with the geometric mean of negative control probes. Highly variable genes were identified by ranking genes by coefficient of variation across all 12 ROIs (CV = standard deviation/mean).

Cells were lysed with RIPA buffer (Fisher Scientific) supplemented with protease and phosphatase inhibitors (Sigma). Western blot was performed using the Bio-Rad Mini-PROTEAN Tetra System. Following transfer, nitrocellulose membranes were blocked in Odyssey Blocking Buffer (LI-COR) for 1 hour. Membranes were washed and incubated rotating overnight at 4°C with a primary antibody. Membranes were then washed three times in TBST for 5 minutes each and incubated with a secondary antibodies (LI-COR IRDye 700CW or 800CW; 1:10,000) for 1 hour. Membranes were imaged on a LI-COR Odyssey scanner.

Cellular Indexing of Transcriptomes and Epitopes sequencing (CITE-seq) was performed as described previously (25). 500,000 cells were resuspended in PBS + 1% FBS and spun at 300 g for 10 minutes at 4°C and then resuspended in 25 μL staining buffer (BioLegend, #420201). 2.5-μL human TruStain FcX (BioLegend, #42230) was added per sample and incubated at 4°C for 10 minutes. Hash-Tag Antibodies, used to label the samples and minimize batch effects, were added at 250 ng/sample (0.5 μL/sample) and incubated at 4°C for 30 minutes. Cells were washed twice in 1-mL staining buffer and spun at 300 g for 5 minutes at 4°C. Samples were pooled 1:1 at 1,000 cells/μL and libraries were generated using the 3′ V3 10X Genomics Chromium Controller following the manufacturer's protocol (CG000183). Final library quality was assessed using the Tapestation 4200 (Agilent) and libraries were quantified by Kapa qPCR (Roche). Pooled libraries were then subjected to paired-end sequencing according to the manufacturer's protocol (Illumina NovaSeq 6000). Bcl2fastq2 Conversion Software (Illumina) was used to generate de-multiplexed Fastq files and the CellRanger (3.1) Pipeline (10X Genomics) was used to align reads and generate count matrices. Seurat's FindMarkers function was used to identify differentially expressed genes in NT5E expressing versus NT5E negative cells, which were ranked by fold change. Further quality control and analyses were performed in R (see Lawlor Lab Github for code).

Gene ontology and gene set enrichment analysis (GSEA) for selected gene lists was performed using MSigDB, Enrichr (26), Metascape (27), escape and dittoSeq (Bioconductor, see supplementary R notebooks on GitHub).

Poly(A)-capture RNA-seq was performed on PDX305 and FACS sorted CD73^−/+ A673 and CHLA10 cells (RNA isolated from cell pellets after sorting). Libraries were prepared with NEBNext Ultra II RNA Library Prep kit and Paired end 150 bp sequencing was performed on a Novaseq600 by Novogene. Adapter trimming was performed using Trim Galore (RRID:SCR_011847, Babraham Institute). Trimmer reads were aligned to GRCh38 using the STAR aligner (RRID:SCR_004463; ref. 28). Counts and differential expression analysis were calculated by DESeq2 v1.18.1 (RRID:SCR_015687) using FDR-adjusted P values < 0.05.

Log₂ expression values for tumor microarray datasets (GSE34620, GSE142162, GSE17679) were downloaded from the R2 Genomics Analysis and Visualization Platform. Only EwS patient tumor samples were included.

Statistical tests were conducted in GraphPad Prism 9 (RRID:SCR_002798). Error bars = SEM unless otherwise specified. Unpaired t tests were used unless otherwise specified. ns, P > 0.05; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. BioRender.com was used to prepare experimental schematics.

All experiments conducted in mice were approved by the Institutional Animal Care and Use Committee of Seattle Children's Research Institute.

Sequencing data generated by this study have been deposited to NCBI Gene Expression Omnibus (RRID:SCR_005012) (GSE236289) database. Analyses in R will be available upon publication at https://github.com/LawlorLab/CD73_CAFlike.

To study EWS::FLI1-low tumor cells in patient tumors and tumor xenografts in vivo, we first sought to define a cell surface marker that would permit their identification and isolation from bulk tumor cells. Given that EWS::FLI1 normally represses mesenchymal lineage genes (28, 29), we reasoned that expression of mesenchymal genes could be used to discriminate EWS::FLI1-low cells. Single-cell transcriptomic studies recently showed that EwS cells exist along a transcriptional axis that partially reflects a mesenchyme development trajectory (19, 30). Using these data, we confirmed that activation of a mesenchymal development signature highly correlates with loss of EWS::FLI1-dependent gene repression (Fig. 1A). Next, we intersected the two gene signatures to identify transcripts that are normally repressed by EWS::FLI1 (31) and expressed by mesenchymal-high state tumor cells (19) Fifty-two transcripts were included in the overlap, including NT5E which encodes the cell surface protein CD73 (Fig. 1B; Supplementary Table S2). NT5E/CD73 (5′-nucleotidase ecto) is a GPI-anchored enzyme that is highly expressed by normal mesenchymal stem cells (MSC; ref. 32), and we and others have previously shown that its expression is transcriptionally inhibited by EWS::FLI1 (19, 28).

Having identified CD73 as a potential marker of EWS::FLI1-low cells, we next performed flow cytometry on a panel of eight established and one early passage PDX-derived EwS cell lines. Human MSC and osteosarcoma cell line U2OS were used as positive and negative controls, respectively. All cells were grown in standard tissue culture conditions (see Methods section). Surface expression of CD73 by EwS cells was highly heterogeneous both within and between cell lines (Fig. 1C). Cell lines ranged from fewer than 1% (RDES) to nearly 100% (PDX305) CD73⁺ cells. CD73⁺ fractions in the remaining EwS lines ranged from 2% to 42% and we observed remarkable reproducibility among biologic replicates demonstrating that each model grows in a preferred homeostatic and heterogeneous state. CD73⁺ cells were always present in large numbers in A673, CHLA10, and TC32 cultures (ranging from ∼15% to 50%), whereas CD73⁺ cells rarely comprised more than 5% of RDES, SK-N-MC, and TC71 (Fig. 1C).

Next, we compared mesenchymal properties of CD73-high (>30% CD73⁺) and CD73-low (< 5% CD73⁺) cell lines. As shown (Supplementary Fig. S1A), CD73-low cell lines grew as rounded cells and formed multicellular clusters with few membrane protrusions. In contrast, cell lines with a high proportion of CD73+ cells were flattened and spindle-shaped with F-actin rich membrane protrusions (Supplementary Fig. S1A). Cell lines with rare CD73⁺ cells were almost entirely non-invasive in 3D collagen I matrices, whereas CD73-high cell lines were robustly invasive (Supplementary Fig. S1B). FACS was then used to isolate CD73⁺ and CD73⁻ cell fractions from the same A673 and CHLA10 cell cultures (Supplementary Fig. S1C). Isolated CD73⁻ cell fractions grew as round cells that formed multicellular clusters with extensive cell–cell contacts and generated few membrane protrusions (Fig. 1D). Conversely, syngeneic CD73⁺ cells displayed mesenchymal morphologies and generated extensive F-actin rich membrane protrusions (Fig. 1D). In addition, sorted CD73⁺ cells migrated faster than corresponding CD73⁻ cells from the same culture (Fig. 1E; Supplementary Fig. S1D). Significantly, sorted cell populations reproducibly shifted towards their baseline mixed populations over 1 week, demonstrating that EwS cells are plastic and interconvert between CD73⁺ and CD73⁻ states (Supplementary Fig. S1E). Thus, CD73 expression is remarkably plastic and heterogeneous both within and between EwS models and surface expression of CD73 reproducibly marks subpopulations of EwS cells that harbor mesenchymal properties.

Having established that CD73 surface expression marks EwS cells that have lost EWS::FLI1-mediated repression of mesenchymal development genes, we next measured EWS::FLI1 to affirm that CD73⁺ cells are, in fact, EWS::FLI1-low cells. Unexpectedly, we detected no reproducible depletion of EWS::FLI1 transcript (Fig. 1F) or protein (Fig. 1G) in CD73⁺ vs. CD73⁻ cells. In addition, proliferation (Supplementary Fig. S1F) and in vivo tumorigenic capacities (Supplementary Fig. S1G) of FACS-sorted cells were equivalent or more pronounced in CD73⁺ than syngeneic CD73⁻ cells, corroborating the observation that CD73⁺ cells had not downregulated expression of the oncogenic fusion. To determine if the relative transcriptional activity of EWS::FLI1 was altered in CD73⁺ cells we performed RNA-seq on FACS-sorted cell populations of both A673 and CHLA10 models (Supplementary Table S3). This transcriptomic profiling confirmed that CD73⁺ EwS cells express high levels of the EWS::FLI1-repressed gene signature (Fig. 1H). Surprisingly, however, this loss of EWS::FLI1-mediated gene repression was not accompanied by a corresponding loss of fusion-dependent gene activation (Fig. 1I and J). CD73⁺ EwS cells displayed only modest loss of the EWS::FLI1-activation signature, no significant difference in expression of 78 direct EWS::FLI1 activated targets (ref. 30; Fig. 1I and J), and no loss of expression of the cell-cycle program that is positively regulated by the fusion (Supplementary Fig. S1H). Consistent with the observed mesenchymal phenotypes of CD73⁺ EwS cells, differential gene expression analysis identified the epithelial–mesenchymal transition (EMT) signature as their most enriched hallmark signature compared with CD73⁻ cells (Supplementary Fig. S1I). Thus, these studies together demonstrated that CD73 surface expression identifies subpopulations of tumor cells that have de-repressed the EWS::FLI1-repressed signature but that apparently retain expression of the fusion and much of its tumorigenic activation signature.

The observation of potential dissociation between EWS::FLI1-activated and -repressed signatures in FACS-sorted populations of CD73⁺ EwS cells led us to more deeply investigate the relationship between phenotype and gene expression in individual cells. To achieve this, we exploited single-cell proteogenomic profiling using CITE-seq (25). Transcriptomes and cell surface CD73 protein expression were measured in over 12,000 cells from eight established EwS cell lines and one early passage PDX-derived line (Fig. 2A; Supplementary Fig. S2A). Expression of NT5E/CD73 was profoundly heterogeneous within and between models and a positive correlation between mRNA and surface protein expression suggested that the CD73 status of EwS cells is largely transcriptionally determined (Fig. 2B and C). The transcriptomes of 1,760 NT5E⁺ cells were next compared with 10,466 NT5E⁻ cells across all cell lines and then in each cell line independently to identify markers of NT5E⁺ cells (Supplementary Table S4). As shown (Supplementary Fig. S2B; Fig. 2D), and consistent with data from RNA-seq of FACS-sorted cells, gene ontology analysis of the 200 most differentially upregulated genes in NT5E⁺ cells identified the EMT gene signature across all nine EwS models.

EMT is a complex and highly conserved developmental process that is frequently hijacked during carcinoma pathogenesis (33). To better understand how an EMT signature could be differentially activated in sarcoma cells that are, by definition, mesenchymal (34), we explored the identity of the EMT signature genes that were upregulated in CD73⁺ cells. Forty-six Hallmark EMT signature genes were more highly expressed by NT5E⁺ cells and we noted that many of them encode for ECM and matrisome-associated proteins (ref. 35; Fig. 2E; Supplementary Fig. S2C). These included numerous collagens, integrins, fibrillin 1 (FBN1), fibronectin (FN), secreted protein acidic and cysteine rich (SPARC), and tenascin-C (TNC; Supplementary Fig. S2D). The ECM of carcinomas is normally remodeled by CAFs, and we noted that 36 of the 200 most upregulated genes in NT5E+ EwS cells were recently identified as markers of diverse CAF subsets across multiple tumor types (e.g., ACTA2, COL1A1, and THY1). In addition, 24 NT5E-marker genes are included among 69 genes that specifically marked ECM-rich CAFs (dCAFs) that are associated with poor prognosis (36). Quantitative RT-PCR of FACS-sorted cells validated higher expression of ECM-related genes in CD73⁺ cell populations (Supplementary Fig. S2E), and immunofluorescence confirmed increased deposition of TNC protein (Supplementary Fig. S2F). Thus, these findings suggested that the transcriptional and phenotypic properties of CD73⁺ EwS tumor cells are similar to ECM-producing CAFs (3, 37).

Having established that the mesenchymal-high state of CD73⁺ EwS cells is characterized by significant and reproducible upregulation of matrisomal gene programs, we next evaluated expression of EWS::FLI1-activated and -repressed gene signatures in these cells. Consistent with our observations in CD73⁺ vs. CD73⁻ sorted cell populations (Fig. 1H–J), single NT5E⁺ cells showed only a modest decrease in expression of the EWS::FLI1-activated signature (Fig. 2F), but a marked upregulation of EWS::FLI1-repressed and ECM genes (Fig. 2G and H). These findings demonstrate a partial decoupling of EWS::FLI1-dependent gene activation and gene repression in NT5E⁺ cells, with a significant and reproducible re-activation of mesenchymal and ECM genes that are normally repressed by the fusion.

Given our finding of extensive intratumoral heterogeneity in vitro, we next investigated transcriptional and phenotypic profiles of EwS tumor cell states in vivo. Local and disseminated CHLA10 tumor xenografts were generated in immunocompromised mice by subcutaneous and tail vein injections, respectively. Tumors were harvested at humane endpoints and subjected to hematoxylin and eosin (H&E) staining and immunofluorescence (Fig. 3A–F). As shown, CD73⁺ tumor cells were detected in all tumors with a notable increase in frequency in tail vein–derived tumors compared with subcutaneous masses (Fig. 3B–F). Spatial heterogeneity of CD73⁺ cell subpopulations was also apparent within each tumor (Fig. 3). In the tail vein–derived tumors, we observed the highest CD73 expression in tumor cells in peri-necrotic regions, tumor borders, and in invasive foci that had infiltrated the retroperitoneal cavity (Fig. 3E and F; Supplementary Fig. S3A). In keeping with our in vitro findings, tumor cell–derived ECM deposition was evident in regions adjacent to the CD73⁺ tumor cells (Fig. 3B–F). Similar spatial heterogeneity in CD73 expression and proximity to ECM deposition was observed in additional A673 and PDX305 subcutaneous xenografts (Supplementary Fig. S3B and C).

We next performed integrated digital spatial profiling and whole human transcriptome analysis to define the transcriptional states of phenotypically distinct tumor cells in these in vivo TMEs (Fig. 4A). Ki67 and cytoskeletal protein vimentin were used as morphology markers to identify differentially proliferative and mesenchymal regions, respectively, and CD31 was used to localize vasculature (Fig. 4B). ROIs were chosen to represent a broad range of proliferation (by Ki67), mesenchymal identity (by Vimentin expression), and nearby microenvironment structures (e.g., tumor–stromal borders, vasculature, and necrosis). We then selected four spatially distinct ROIs from each tumor for capture and NGS. Between 1,400 and 6,500 cells were captured from each of the 12 ROIs which were distributed in tumor interiors (e.g., ROI1, ROI5, ROI11), along tumor–stromal borders (e.g., ROI2, ROI7), and in invasive tumor foci that were surrounded by necrotic tissue (e.g., ROI9, ROI10; Fig. 4B). Normalized MKI67 and VIM gene expression in the captured ROIs (Supplementary Table S5) corroborated the observed spatial differences in Ki67 and vimentin immunofluorescence (Supplementary Fig. S4A and B). As shown, tumor cells that were captured from subcutaneous and tail vein–derived focal liver tumors generally expressed higher levels of MKI67 than cells captured from an infiltrating retroperitoneal tumor (Supplementary Fig. S4A and B). In contrast, expression of VIM was generally higher in disseminated tumors and heterogeneity among ROIs was apparent in all three tumors (Supplementary Fig. S4A and B).

Having confirmed that CHLA10 cells display heterogeneity of VIM expression in vivo, we next performed unsupervised analyses of whole transcriptomic data to interrogate spatial and biologic heterogeneity in an unbiased fashion. First, principal components analysis (PCA) of the 12 ROI-transcriptome dataset showed that subcutaneous tumor ROIs clustered tightly and separately from ROIs captured from disseminated tumors despite the fact that all tumors were generated from CHLA10 cells (Fig. 4C). Transcriptomes of ROIs captured from tail vein–derived tumors also demonstrated considerably more heterogeneity than subcutaneous ROIs. This was especially notable when comparing tumor cells captured from tumor interiors to those captured from invasive tumor foci (ROI9 and ROI10; Fig. 4C). The transcriptomic profile of tumor cells in ROI10 was distinct from all other ROIs and, significantly, targeted analysis of the data reveals that these tumor cells expressed the highest level of NT5E (Fig. 4D). Next, we sought to identify genes with the most spatial heterogeneity. To achieve this, we again took an unbiased approach and identified the most variably expressed genes across all 12 ROIs as determined by coefficient of variance (38). Limiting analysis to the top 10% most differentially expressed transcripts generated a list 1,339 genes and gene ontology analysis revealed these to be highly enriched for ECM-related gene programs (Fig. 4E). Unsupervised hierarchical clustering of these 1,339 genes clearly separated ROIs captured from subcutaneous and disseminated tumors and showed that NT5E and an ECM organization gene signature were markedly upregulated in tumor cells captured from peri-necrotic foci (Fig. 4F; ROI9 & ROI10). One notable exception to this trend was ROI2, a region along the tumor–stromal interface of the subcutaneous tumor that highly expressed ECM genes but not NT5E/CD73. Thus, intratumoral spatial heterogeneity of EwS cell phenotypes and transcriptional profiles is evident in vivo. NT5E and matrisomal gene programs are, in general, more highly expressed by tumor cells in distant sites than those that are growing in localized masses and expression of ECM-encoding genes is particularly enriched along tumor borders and invasive foci.

Finally, we compared expression of EWS::FLI1-activated and -repressed genes in tumor cells captured from the 12 spatially distributed ROIs. As predicted from our in vitro data, EwS cells that highly expressed the ECM gene signature showed loss of EWS::FLI1-dependent gene repression (Fig. 4G). However, loss of the repressive signature was not always accompanied by loss of EWS::FLI1-mediated gene activation, and hybrid transcriptional states were uncovered (Fig. 4H). Transcriptomes of EwS cells isolated from the core of the subcutaneous tumor (ROIs 1, 3, & 4) were typical of EWS::FLI1-high cells (activated genes ON/repressed genes OFF). Likewise, the ROI10 transcriptional signature was fully compatible with EWS::FLI1-low cells (activated genes OFF/repressed genes ON). In contrast, transcriptomes of tumor cells captured from the remaining regions displayed features of both EWS::FLI1-high and -low cells. This is particularly evident for ROIs 9, 11, and 12 wherein EwS cells were found to express high levels of both EWS::FLI1-activated and -repressed genes. Thus, these studies demonstrate dissociation of EWS::FLI1 activating and repressive transcriptional activity in EwS tumor regions in vivo and suggest that tumor subpopulations can upregulate ECM signatures without downregulating expression of EWS::FLI1-activated genes. Although it is possible that hybrid ROIs were comprised of mixed populations of EWS::FLI1-high and EWS::FLI1-low cells, when combined with our identification of transcriptionally hybrid states in vitro (Fig 2), these data lead us to conclude that hybrid EwS cells are present in tumors in vivo in and these cells acquire mesenchymal properties while retaining EWS::FLI1-dependent gene activation.

There is growing evidence that cells in transitional EMT cell states, rather than fully epithelial or mesenchymal cells, drive growth and metastatic progression of carcinomas (39, 40). Having identified NT5E/CD73 as a marker of transcriptionally hybrid and phenotypically distinct CAF-like tumor cell subpopulations in EwS models, we sought to investigate the presence of these cells in primary patient tumors. Analysis of three EwS gene expression datasets (GSE34620, GSE142162, GSE17679; refs. 18, 41, 42) revealed marked inter-tumor variability in NT5E expression (Fig. 5A). Next, we identified genes that were strongly correlated with NT5E (R > 0.6) across patient tumors and discovered a striking overlap among the independent cohorts (Fig. 5B). Hallmark EMT and ECM-related gene ontology signatures were highly enriched in these 217 shared NT5E-correlated genes (Supplementary Table S6), corroborating the enrichment of these signatures in NT5E+ EwS cells in tumor models (Fig. 5C). Integrating the top 200 markers of NT5E+ EwS cells in vitro (Fig. 2E; Supplementary Table S3) with the 217 genes NT5E-correlated genes in patient tumors in vivo, generated a list of 27 genes that were highly reproducibly upregulated alongside NT5E in EwS tumor cells and these include numerous EMT and ECM genes, most of which are normally repressed by EWS::FLI1 (refs. 31, 43, 44; Fig. 5D). Consistent with this observation, expression of the 28 gene signature by individual patient tumors correlated positively with expression of both ECM organization and EWS::FLI1-repressed gene signatures (Fig. 5E; Supplementary Fig. S5A and B).

This analysis of bulk tumor RNA profiling confirmed inter-tumor heterogeneity of NT5E expression and the relationship between high NT5E and expression of EWS::FLI1-repressed matrisomal genes. To assess intratumoral heterogeneity and the potential existence of distinct CAF-like tumor cell subpopulations we evaluated CD73 and ECM protein expression in tumor sections from archived patient biopsies. We first stained a tumor TMA with an anti-CD73 antibody using the EWS::FLI1-induced protein NKX2–2 as a tumor marker. CD73⁺ cells were detected in only 7 of 24 tumor cores (20 unique patients; Supplementary Table S7) and the frequency of CD73⁺ tumor cells varied from rare to extensive (Supplementary Fig. S6A–C). Staining of adjacent sections of the TMA with an antibody against TNC confirmed inter- and intra-tumoral heterogeneity of TNC and deposition of this protumorigenic ECM protein was most robust in regions adjacent to CD73⁺ tumor cells (Supplementary Fig. S6D).

Given the observed spatial heterogeneity of CD73⁺ tumor cells and tumor-derived ECM in EwS xenografts (Fig. 3; Supplementary Fig. S3) and TMA cores (Supplementary Fig. S6B and C), we next stained full biopsy sections from two tumors that were included in the TMA. One of these (#20) showed extensive CD73⁺ cells on the TMA core while the other (#35) was scored negative (Supplementary Table S7). Whole sections of both biopsies showed the presence of large acellular regions that denoted areas where core punches had been previously harvested for inclusion on TMAs (Fig. 6A and B). Staining for the CD99 tumor marker showed that biopsy #20 included EwS tumor embedded within adjacent non-tumor tissue (Fig. 6A) while biopsy #35 was predominantly comprised of viable tumor (Fig. 6B). Importantly, CD73⁺ tumor cells within CD99⁺ regions were detected in both tumors, including biopsy #35 which was devoid of CD73⁺ cells on the TMA (Fig. 6A and B). Spatial heterogeneity of ECM proteins TNC and biglycan (BGN) was also evident between and within the tumors and both proteins were more highly expressed adjacent to CD73⁺ tumor cells (Fig. 6A and B). Thus, EwS tumor cells are both transcriptionally and phenotypically heterogeneous in vitro and in vivo and spatial co-localization of CD73⁺ tumor cells with pro-tumorigenic ECM suggests that EWS::FLI1-low and EWS::FLI1-high/low hybrid tumor cells function as CAF-like cells (Fig. 6C).

Accruing evidence suggests that tumor cell heterogeneity is a universal feature of all cancers and that even genetically homogeneous tumors develop extensive cell diversity through epigenetic processes (4, 5). Phenotypically distinct tumor cell subpopulations can contribute in different ways to tumor progression as they may have very different proliferation rates, metastatic potential, and differential responses to treatment (5). EwS are genetically quiet tumors that are driven by a single fusion oncoprotein (11). They show little evidence of clonal genetic diversity or evolution and they are histologically classified as homogeneous, small round blue cell tumors (12). Here, we have identified that EwS tumor cells exist in a previously uncharacterized spectrum of cell states that differentially express mesenchymal gene programs. In particular, our findings show that transcriptionally distinct subpopulations of highly mesenchymal tumor cells are present in EwS cell lines in vitro, xenograft models in vivo, and in patient tumors in situ. These cells can be identified by cell surface expression of CD73 and they are distinguished by their upregulation of matrisomal protein-encoding genes and their deposition of ECM. The distinct capacity of these cells to remodel the local TME through deposition of pro-tumorigenic ECM leads us to label them “CAF-like” EwS cells (Fig. 6C).

The critical importance of ECM remodeling to tumor growth, invasion, and metastatic progression is well established (45–47). In epithelial-derived tumors ECM is largely deposited by CAFs, nonmalignant stromal cells that most often arise from resident tissue fibroblasts or other mesenchymal cells and that are activated to secrete pro-tumorigenic matrisomal proteins in response to cues from the tumor cell compartment (3). More rarely, subpopulations of carcinoma cells themselves acquire the capacity to deposit ECM and promote pro-tumorigenic remodeling of the local TME and there is evidence that tumor cell–derived ECM confers a more aggressive tumor state (47, 48). The critical contribution of ECM and nonmalignant stromal cells to sarcomas is less well studied but posttranslational remodeling of collagen by tumor cells plays an essential role in mediating metastatic progression of soft-tissue sarcomas (49). The contribution of ECM remodeling to growth and progression of EwS has not been deeply investigated, though clinico-pathologic correlative studies suggested that a stroma-rich TME influences EwS gene expression and patient prognosis (50). EwS cells and tumors can deposit fibrillar and non-fibrillar proteins (51–53) and we previously showed that activation of canonical Wnt/beta-catenin and TGFβ signaling upregulates secretion of pro-tumorigenic and -angiogenic matrisomal proteins by EwS cells, including collagens and TNC (54, 55). In the current work, we have identified discrete subpopulations of EwS cells that deposit pro-tumorigenic ECM proteins in vivo. These cells are largely CD73⁺ and they are enriched on tumor borders, peri-necrotic regions, and in invasive foci. Importantly, these cells were rarely found on our examination of small TMA tumor cores. Diagnostic EwS tumor specimens are normally obtained from needle or small open biopsies of viable tumor interiors (56). Thus, these CD73⁺ tumor cells and tumor cell–derived ECM can be missed when histologic and molecular analyses are limited to small tumor specimens. As our understanding of the critical importance of tumor heterogeneity and tumor: TME cross-talk grows, so too does our appreciation that TMA cores may not be adequate for studies of tumor biology that aim to dissect tumors as spatially dynamic ecosystems (57).

Our single-cell studies of EwS cell transcriptomes and spatial profiling studies of tumors in vivo generated the unanticipated discovery that, although they expressed high levels of matrisomal genes that are normally repressed by EWS::FLI1, CD73⁺ EwS cells retained expression of the fusion and were largely not “EWS::FLI1-low”. While we did identify invasive cell clusters that were consistent with a fully EWS::FLI1-low state (e.g., Figure 4H; ROI10), CAF-like EwS cells most often existed in hybrid EWS::FLI1 transcriptional states. These cells retained expression of the fusion-activated gene signature alongside high expression of genes that are normally repressed by EWS::FLI1. Thus, these EwS cells gained mesenchymal properties, including the capacity to deposit ECM, without losing proliferative or tumorigenic potential. Notably, we also observed transcriptional heterogeneity among CAF-like populations in vivo, including cells that expressed NT5E but not TNC (Supplementary Fig. S2D), and deposition of TNC protein in xenograft regions without detectable CD73 expression (Fig. 3B). Bone fide CAFs exist as transcriptionally distinct subsets and dynamically switch between these different states (58, 59). Further studies will be needed to determine if CAF-like subpopulations of EwS cells can likewise be further subdivided into multiple transcriptional states with different functions in TME remodeling.

These observations lead us to conclude that EwS cell states are more nuanced than first appreciated. While more migratory and metastatic EWS::FLI1-low cells were first presumed to express lower levels of the fusion (15), it is now evident that the activity of the fusion can be influenced by cell intrinsic and extrinsic factors that modulate its transcriptional output (16) including context-dependent transcription factors (19–21), chromatin remodelers (17, 18), and nuclear condensates (14, 60). Our current work adds further complexity, revealing that the activating and repressing properties of the fusion can be decoupled in individual tumor cells. This decoupling may be possible because EWS::FLI1 gene activation and repression occur through distinct mechanisms requiring the activity of different partners. Gene repression can occur by multiple means including direct EWS::FLI1 binding, upregulation of transcriptional repressors, and recruitment of HDACs (16). Further, recent studies have identified mechanisms that antagonize this repression, including Wnt-induced sensitization to TGFβ signaling (22, 55), chromatin reorganization after STAG2 loss (17, 18), and activity of developmental transcription factors such as HOXD13 (19). Whether these or other mechanisms facilitate de-repression of NT5E and ECM genes and generation of a CAF-like state remain to be elucidated, and this is a key gap in knowledge that the field must now fill.

Tumor cell heterogeneity and cell plasticity are key factors in tumor evolution (4) and metastatic progression of carcinoma models is, at least in part, driven by transitional or hybrid cells that exist along a continuum between epithelial and mesenchymal states (39, 40). In addition, interactions between dynamic tumor cells and an equally dynamic TME create complex tumor ecosystems that support tumor growth, enable metastatic progression, and promote resistance to current treatment approaches (61–63). Given this, the tumor: TME interface and ECM remodeling processes are increasingly being investigated as novel targets on which to focus basic and translational cancer research efforts (23, 64–67). Here we have reported the existence of CAF-like EwS cells that contribute to ECM remodeling, and we propose that these cells are likely to play critical roles in mediating tumor progression and influencing treatment response. A deeper understanding of CAF-like EwS cells is now needed, including determination of their prevalence, origins, fundamental biology, and impact on neighboring tumor cells as well as the immune and nonimmune TME. Elucidating the dynamic and highly heterogeneous nature of EwS ecosystems will afford novel insights into tumor biology that will guide development and rational integration of TME-targeted therapeutics into EwS treatment regimens.

E.R. Lawlor reports grants from NIH, 1Million 4 Anna, Sam Day Foundation, AACR/Stand Up to Cancer, Sam Waxman Foundation, and Lafontaine U-Can-Cer Vive Foundation during the conduct of the study as well as grants from Cure Childhood Cancer Foundation outside the submitted work. No disclosures were reported by the other authors.

E.D. Wrenn: Conceptualization, resources, data curation, software, formal analysis, funding acquisition, investigation, visualization, writing–original draft, writing–review and editing. A.A. Apfelbaum: Conceptualization, resources, data curation, software, formal analysis, funding acquisition, investigation, visualization, writing–original draft, writing–review and editing. E.R. Rudzinski: Resources, investigation. X. Deng: Resources, investigation. W. Jiang: Methodology. S. Sud: Methodology. R.A. Van Noord: Methodology. E.A. Newman: Methodology. N.M. Garcia: Investigation. A. Miyaki: Investigation, visualization. V.J. Hoglund: Investigation. S.S. Bhise: Investigation. S.B. Kanaan: Investigation. O.G. Waltner: Software, formal analysis. S.N. Furlan: Conceptualization, software, formal analysis, investigation. E.R. Lawlor: Conceptualization, supervision, funding acquisition, project administration, writing–review and editing.

The authors thank members of the Lawlor lab for helpful discussion, and the staff at the Fred Hutchinson and Seattle Children's Research Institute flow cytometry, comparative medicine, and bioinformatics cores. Grant and gift support for this work is gratefully acknowledged and was provided by the following sources: NIH/NCI R01 CA215981 (to E.R. Lawlor), F31CA247104 (to A.A. Apfelbaum); research supported by the 2022 AACR-QuadW Foundation Sarcoma Research Fellowship in Memory of Willie Tichenor, Grant Number 22–40–37-WREN (to E.D. Wrenn); Sam Day Foundation (to E.R. Lawlor); the 1M4Anna Foundation (to E.R. Lawlor and S.N. Furlan); Lafontaine U-Can-Cer-Vive Foundation (to E.R. Lawlor). This research was supported by the Flow Cytometry Shared Resource of the Fred Hutch/University of Washington/Seattle Children's Cancer Consortium (P30 CA015704).

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.