Multi‐omics profiling identified two epithelioid sarcoma molecular subtypes with distinct signaling and immune characteristics

Epithelioid sarcoma (EpS) is an aggressive soft tissue sarcoma characterized by switch/sucrose non-fermentable (SWI/SNF)-related matrix-associated actin-dependent regulator of chromatin subfamily B member 1 (SMARCB1) loss [1]. Conventionally, EpS is histologically classified as either distal or proximal subtype, each exhibiting distinct clinical behaviors; these sometimes co-exist as “hybrid” EpS [2, 3]. Differential diagnosis from other SMARCB1-deficient tumors, such as extracranial extrarenal rhabdoid tumors (EERTs), can be challenging [4]. Beyond phenotypic diversity, EpS molecular heterogeneity remains poorly understood. To address this, we aimed to build a molecular classification and explore inter- and intra-tumor heterogeneity, using multi-omics profiling. We profiled 33 EpSs and 3 EERTs using whole-exome sequencing (WES), DNA methylation, and bulk RNA-sequencing (RNA-seq); single-cell RNA-sequencing (scRNA-seq) and 10x Genomics Visium spatial transcriptomics were performed on 8 and 4 EpSs, respectively (Figure 1A, Supplementary Materials and Methods). Cases were reviewed by two senior pathologists and classified by histology (14 distal, 14 proximal, 5 hybrid EpSs) (Supplementary Figure S1, Supplementary Tables S1-S2). Unsupervised RNA-seq clustering identified two EpS transcriptomic subtypes (Figure 1B): (i) distal-like (n = 20) — including all distal EpS, 2 proximal and 4 hybrid EpSs — enriched in cell adhesion, circulatory system development genes and extracellular matrix; (ii) proximal-like (n = 13) — comprising 12 proximal and 1 hybrid EpSs — enriched in synapse organization, response to wounding and macrophage activation (Supplementary Figure S2A, Supplementary Table S3). Gene set enrichment analysis revealed a significant enrichment in epithelial-mesenchymal transition (EMT) and ultraviolet response in distal-like EpS compared to proximal-like EpS (normalized enrichment score = 1.80 and 1.66, respectively; false discovery rate < 25%; p < 0.05) (Supplementary Figure S2B, Supplementary Table S4). We explored mechanisms of SMARCB1 loss in 31 EpSs and 3 EERTs, using WES, targeted next-generation sequencing, single-nucleotide polymorphism array, shallow whole-genome sequencing, DNA methylation and fluorescence in situ hybridization – depending on material availability (Supplementary Figure S3A, Supplementary Table S5). Biallelic SMARCB1 inactivation was identified in 16 (51.6%) of 31 EpSs (mostly homozygous deletions), without germline SMARCB1 alteration. Heterozygous loss-of-function alterations were found in 14/31 (45.2%) EpSs. SMARCB1 alteration types were similar across EpS transcriptomic subtypes and unrelated to SMARCB1 mRNA levels (Supplementary Figure S3B-D). Recurrent alterations, excluding SMARCB1, occurred in up to13% of EpSs (Supplementary Figure S3A, Supplementary Tables S6-S7). Median tumor mutational burden was 0.84 (range, 0.56-15.75) and 0.87 (range, 0.34-1.56) mut/Mb in distal-like and proximal-like EpS, respectively (Supplementary Figure S4A, Supplementary Table S8). Distal-like EpS harbored a higher number of arm-level copy number alterations (CNAs) than proximal-like EpS [median = 5 (range, 0-15) vs. 1 (range, 0-11), p = 0.041; Supplementary Figure S4B], though prior chemotherapy or disease stage may have influenced this result [5]. Immune deconvolution of bulk RNA-seq data showed significant enrichment in cytotoxic and CD8+ T cells, natural killer cells and M1 macrophages in distal-like EpS, whereas proximal-like tumors were enriched in M2 macrophages (Figure 1C, Supplementary Figure S5A). Immunohistochemistry (IHC) revealed that CD8+ T cells and CD163+ macrophages located at the tumor periphery in distal-like EpS with more frequent mature tertiary lymphoid structures, while macrophages were intratumoral in proximal-like EpS (Figure 1D, Supplementary Figure S5B-D). ScRNA-seq of 3 distal-like and 5 proximal-like EpS samples profiled 28,912 high-quality cells assigned to six main subpopulations according to the expression-based clustering, SMARCB1 expression, the canonical markers of known cell types, and inferred CNAs (Figure 1E, Supplementary Figure S6). Marker identification of myeloid cells, the most abundant immune subpopulation (Supplementary Figure S7A), showed predominant pro-tumoral macrophages expressing triggering receptor expressed on myeloid cells 2 (TREM2)+ in proximal-like EpS and inflammatory monocytes in distal-like EpS (Supplementary Figure S7B). To investigate cell-cell interactions, we independently integrated 6,542 single cells from distal-like EpS (Supplementary Figure S8) and 21,763 from proximal-like EpS (Supplementary Figure S9). CellPhoneDB analysis suggested distinct ligand-receptor interactions (Figure 1F): in distal-like EpS, inferred tumor-immune interactions mainly involved adhesion molecules (e.g., desmoglein 2 [DSG2]) or chemokine receptors (e.g., atypical chemokine receptor 2), consistent with spatial transcriptomics observations (Figure 1G, Supplementary Figure S10); proximal-like EpS was enriched in macrophage activation signaling (Figure 1G, Supplementary Figure S11). We next focused on the 14,320 tumor single cells and identified ten subclusters (Figure 1H, Supplementary Figure S12, Supplementary Table S9). One subcluster (EMT_5) gathered virtually all cells from distal-like samples and was enriched in genes related to extracellular matrix and cell adhesion. By contrast, subclusters from proximal-like samples were mostly patient-specific and linked to various biological processes. To assess the specificity of single-cell subclusters, we deconvoluted their signatures in bulk RNA-seq data (Supplementary Figure S13). EMT_5 was the most abundant signature in distal-like EpS, rare in proximal-like EpS and virtually absent from EERTs. We revalidated this in an independent dataset of 1,041 mesenchymal tumors. Again, distal-like EpS exhibited the highest EMT_5 signature score, while SMARCB1-deficient malignant rhabdoid tumors and SMARCA4-deficient undifferentiated tumors had the lowest (Figure 1I). Other sarcomas with mesenchymal-epithelial features (e.g., desmoplastic small round cell tumors) had low EMT_5 scores, highlighting the specificity of this signature towards distal-like EpS. Since DSG2 scored within the top five markers of the EMT_5 signature (Supplementary Figure S12A) and showed functional cell-cell interactions (Figure 1F), we asked whether its expression could serve as a IHC diagnostic biomarker for distal-like EpS. DSG2 was expressed in all distal-like tumor cells, while it was undetectable in all proximal-like tumors but one – the latter exhibiting a weak, focal staining – supporting its potential diagnostic value (Figure 1J, Supplementary Figure S14). We eventually explored the prognostic value of the EMT_5 signature. It was significantly associated with better overall survival and metastasis-free survival in two independent series (Figure 1K, Supplementary Figure S15). Epithelioid sarcoma is a rare, morphologically and clinically heterogeneous disease that lacks a molecular classification. By integrating bulk multi-omics and single-cell datasets from a large EpS cohort, combined with clinical annotation and pathological review, we propose a molecular classification comprising two transcriptomic subtypes: distal-like EpS, characterized by a prognostic, cancer cell-specific EMT signature and a T-cell-rich microenvironment; and proximal-like EpS, with distinct transcriptomic programs and an immunosuppressive, macrophage-rich microenvironment. The EMT signature is consistent with prior transcriptomic studies [5, 6]; however, these relied on supervised clustering based on histological subtypes. By contrast, our unsupervised analysis identified molecular-based subtypes, thereby allowing to reclassify 6 cases, originally categorized as “proximal” or “hybrid” histological subtypes, as distal-like EpS, which may carry therapeutic implications. Further, single-cell and spatial analyses enabled us to make significant advances. First, we confirmed that the EMT signature was specific to tumor cells, rather than resulting from stromal contamination in bulk RNA-seq data. Second, we identified that DSG2 might serve as a diagnostic marker for distal-like EpS. Third, we found that this scRNA-seq-derived EMT signature was prognostic, highly specific to the distal-like EpS, and virtually absent from other SWI/SNF-defective sarcomas. Whether this relates to the role of SMARCB1 in favoring EMT gene expression in preclinical models deserves further exploration [7, 8]. Finally, beyond confirming quantitative differences in immune populations recently reported in a smaller series of 18 EpSs [4], we additionally identify a specific spatial distribution and distinct functional cell-cell interactions using spatial transcriptomics and in silico prediction, which may have important therapeutic implications. Overall, this multi-omics single-cell dataset provides important molecular insights into the biology, diagnosis and prognostic of EpS, and should aid in developing new therapeutic strategies. Conceptualization, C.N., L.C-D., J.V. and S.P-V.; Methodology, C.N., L.C.D., J.V., C.H., R.B., J.X.J, D.P., J. M-P., A.V., A.B., L.L., E.R., C.R-S., L.L., C.A., M.G., J-Y.S., G.P., W.H.F., C.S-F., J.J.W., T.G.P.G, F.T., F.B. and S.P-V.; Pathological review: C.N, J-M.C; Software, C.N., L.C-D., J.V. and R. B.; Formal analysis, C.N. and L.C-D.; Investigation, C.N., L.C-D and J.V.; Resources, C.N., D.P., F.T., J-Y.B., F.B., B.V., R.B., A.LC., M.F., Ch.H, and S.P-V; Data curation, C.N., L.C-D and S.P-V;.; Writing – Original draft: C.N.; Writing – Review & editing: C.N., L.C-D., J.V., J.X.J., C.H. T.G.P.G., J.J.W., F.B., C.S-F. and S.P-V.; Visualization, C.N., L.C-D. and S.P-V; Supervision, T.G.P.G., F.B., F.T. and S.P-V.; Project administration, C.N. and S.P-V.; Funding acquisition, J-Y. S., T.G.P.G., and S.P-V. We dedicate this research to patients and their families. We thank patients and families for their donations to the Gustave Roussy Sarcoma Tumor Board and Research program, Gustave Roussy Sarcoma Donors, Association “Un élan pour Lucas” and Family L. We also thank FIGHT KIDS CANCER (FKC) Representatives KickCancer, Imagine for Margo, Foundation Kriibskrank Kanner and CRIS Cancer Foundation for awarding the 2024 FIGHT KIDS CANCER & St. Baldrick's Foundation Arceci Innovation Award to Sophie Postel-Vinay. The independent selection process of this Innovation Award, funded by FKC, was administered by the St Baldrick's Foundation. This work was funded by program grants to Sophie Postel-Vinay from National Institute of Health and Medical Research Action Thématique Incitative sur Programme-Avenir Group / La Ligue Contre le Cancer 2018; Fondation Association pour la Recherche sur le Cancer (PGA1-RF20190208576), European Research Council (ERC) (TargetSWitch 101077864), Cancéropôle Ile-de-France (2017-1-EMERG-72 and 2020-1-EMRG-28-IGR-1), La Ligue contre le Cancer Val de Marne (Subvention Recherche Scientifique 2021 et 2022), Agence Nationale pour la Recherche (RHU Condor ANR-21-RHUS-0010), and supported by the National Institute for Health Research University College London Hospitals Biomedical Research Centre. This work was further funded by program grants to Gustave Roussy from Institut National du Cancer (INCa-DGOS-Inserm_12551 SIRIC2 and INCa-DGOS-Inserm-ITMO Cancer_18002 SIRIC EpiCURE), as well as the “ Programme émergent d’épigénétique” funded by The Fondation Gustave Roussy. The laboratory of Thomas G.P. Grünewald acknowledges funding by the SMARCB1 association, the Barbara and Wilfried Mohr foundation, and the European Union (ERC, CANCER-HARAKIRI, 101122595). For Sophie Postel-Vinay and Thomas G.P. Grünewald ERC grants, views and opinions expressed are, however, those of the authors only and do not necessarily reflect those of the European Union or the European Research Council. Neither the European Union nor the granting authority can be held responsible for them. Jia Xiang Jin acknowledges support by scholarships of the Rudolf and Brigitte Zenner Foundation and the German Academic Scholarship Foundation. Julien Vibert is supported by a fellowship from the Inserm-Foundation Bettencourt. All experiments were performed in accordance with the Declaration of Helsinki. The study was approved by the Institut Gustave Roussy Review Board (IRB number: 2024-316). All the patients signed an informed consent to allow the use of tumor residual samples for scRNA-seq analysis. Archived samples were used according to the French legal framework of MR-004 reference methodology set up by the French Data Protection Authority (CNIL) (Number I14112002202020 Health Data Hub). Sophie Postel-Vinay is a principal investigator or sub-investigator of clinical trials from Abbvie, Agios Pharmaceuticals, Amgen, Argen-X Bvba, Arno Therapeutics, Astex Pharmaceuticals, Astra Zeneca, Aveo, Bayer Healthcare Ag, Bbb Technologies Bv, Blueprint Medicines, Boehringer Ingelheim, Bristol Myers Squibb, Celgene Corporation, Chugai Pharmaceutical Co., Clovis Oncology, Daiichi Sankyo, Debiopharm S.A., Eisai, Eli Lilly, Exelixis, Forma, Gamamabs, Genentech, Inc., GlaxoSmithKline, H3 Biomedicine, Inc, Hoffmann La Roche Ag, Innate Pharma, Iris Servier, Janssen Cilag, Kyowa Kirin Pharm. Dev., Inc., Loxo Oncology, Lytix Biopharma As, Medimmune, Menarini Ricerche, Merck Sharp & Dohme Chibret, Merrimack Pharmaceuticals, Merus, Millennium Pharmaceuticals, Nanobiotix, Nektar Therapeutics, Novartis Pharma, Octimet Oncology Nv, Oncoethix, Onyx Therapeutics, Orion Pharma, Oryzon Genomics, Pfizer, Pharma Mar, Pierre Fabre, Roche, Sanofi Aventis, Taiho Pharma, Tesaro Inc., and Xencor. SPV has participated in advisory boards for Merck KGaA and has received laboratory research funding for work unrelated to this project from AMGEN and Hoffman-LaRoche. All data generated or analyzed during this study are included in this published article and its supporting information files. The Data Access Committee (DAC) and the policies associated with that have been submitted and validated by EGA are as follows: DAC: EGAC50000000552; Policy: EGAP50000000502. Scripts for the figures are available on request at https://github.com/LeoColmet-Daage/swisarc-manuscript-figures. The RNA-seq processing pipeline is available at https://github.com/nf-core/rnaseq/tree/master. The WES processing pipeline is available at https://github.com/nf-core/sarek/tree/master. Single-cell processing pipeline is available at https://github.com/gustaveroussy/single-cell. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.