System for study of diverse cellular composition in alveolar rhabdomyosarcoma Authors: Guak-Kim Tan, Kavya Kannan, Hollis Wright, Zu-Yuan Qian, Sonja D. Chen, Gillian M. St. John, Sofia M. Tosoni, Brian P. Rubin, Melvin Lathara, Charles Keller Published: 2026 ABSTRACT Alveolar rhabdomyosarcoma (ARMS) is an aggressive pediatric soft tissue sarcoma often driven by PAX3::FOXO1 or PAX7::FOXO1 fusions, for which therapy can be limited if metastatic. ARMS pathology shows pronounced cellular heterogeneity, including malignant subpopulations at distinct myogenic differentiation stages. Detailed subpopulation transcriptional programs and microenvironmental influences are of ongoing interest biologically and therapeutically. Using a genetically engineered mouse model and single-cell RNA sequencing (scRNA-seq), we characterized tumor composition and cellular states. Histology and immunohistochemistry confirmed diverse morphologies and differential expression of myogenic markers. scRNA-seq identified 17 clusters, including 11 malignant, immune and stromal subpopulations. Transcriptomic analysis revealed precursor-like cells expressed epithelial/EMT-like programs, while rhabdomyoblast-like cells upregulated extracellular matrix and mesenchymal regulators. Allografts in immunodeficient mice showed that immune context shapes cellular composition. Lineage-specific reporters and Symphony mapping validated spatial segregation of precursor and rhabdomyoblast-like cells. These findings demonstrate transcriptionally distinct malignant subpopulations within an immune-influenced microenvironment. These ARMS genetically engineered mouse model and allograft systems provides a platform to study tumor heterogeneity and drug responses.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1038/s41598-026-57841-3. Introduction Rhabdomyosarcoma (RMS) is the most common soft tissue sarcoma in children and adolescents1,2. Over the last four decades, the 3-year survival rate for children with high-risk RMS has persisted at ~ 20%3. RMS is thought to arise from myogenic or non-myogenic precursors but to have a phenotype of muscle precursor that fails to differentiate into normal muscle4. RMS is broadly classified into three main histological subtypes: embryonal (ERMS), alveolar (ARMS) and sclerosing/spindle-cell RMS. While ERMS is most common, ARMS is associated with significantly worse outcomes5. Despite advances in multimodal therapies, including surgery, chemotherapy and radiation, the prognosis for patients with ARMS remains poor, particularly in cases of metastatic or recurrent disease6,7. This highlights the urgent need for a deeper understanding of tumor biology and the identification of novel therapeutic targets. ARMS is characterized by specific chromosomal translocations, most notably the PAX3::FOXO1 and PAX7::FOXO1 fusion genes8. These genetic alterations drive the oncogenic processes that promote ARMS development and progression, contributing to its aggressive clinical behavior9–12. One of the presumed critical challenges in treating ARMS is its significant cellular heterogeneity13–15. Tumors are composed of a diverse array of cell populations with distinct phenotypic and functional characteristics15. This heterogeneity contributes to the tumor’s ability to adapt to and resist therapeutic interventions, leading to treatment failure and disease relapse16,17. Traditional bulk RNA sequencing approaches, while informative, fail to capture the complexity of cellular diversity within tumors. In contrast, single-cell RNA sequencing (scRNA-seq) allows for the profiling of gene expression at the single-cell level, enabling the identification of distinct cell populations within the tumor microenvironment. This technology has the potential to uncover rare cell types, delineate cellular hierarchies and reveal dynamic cellular states that contribute to tumor progression and therapeutic resistance18–21. scRNA-seq of human RMS has yielded several interesting findings, including the discovery of a mesenchymal-enriched, tumor-propagating subpopulation15, the presence of an immunosuppressive tumor niche in fusion-positive tumors, and the interplay between malignant cells and the tumor microenvironment22. In this study, we employ scRNA-seq to analyze ARMS tumors from a conditional mouse model of ARMS23. Through these analyses not otherwise possible in human tumors, we seek to gain insights into the cellular heterogeneity of ARMS, identify molecular signatures of distinct cell populations and explore how different immune backgrounds shape tumor cell subpopulation structure. This study aims to enhance understanding of ARMS biology, which may lay the groundwork for the development of more effective, targeted treatments. Materials and methods Mice All animal procedures were approved by the Institutional Animal Care and Use Committee (IACUC) of Children’s Cancer Therapy Development Institute, with protocol numbers 2016-001 (15–5446 A) and 2016-002 (14–5460 A), and were conducted in accordance with the Guidelines for the Care and Use of Laboratory Animals. The conditional knock-in mouse model of ARMS, in which the Pax3-Foxo1 allele is activated in the skeletal muscle cell lineage via Myf6Cre (Myf6Cre Pax3P3Fm/P3Fm Trp53F2–10/F2–10; referred to hereafter as GEMM-ARMS) has been described previously23,24. These mice also carry Cre-activated reporters, eYFP to label Pax3::Foxo1 expressing cells and dsRED2 to label Myh2 expressing cells (Supplementary Table 1). Some tumors are derived from GEMM-ARMS mice, and some are allografts of GEMM-ARMS tumors into immunodeficient mice (Supplementary Table 1). NOD/SCID/IL2-receptor gamma-chain knockout (NOD.Cg-Prkdcscid Il2rgtm1Wjl; NSG; The Jackson Laboratory) mice lack mature T and B lymphocytes and natural killer (NK) cells, and exhibit defective macrophage function25. The SCID Hairless Outbred (Prkdcscid Hrhr; SHO; Charles River) mice are a triple-immunodeficient strain lacking T and B lymphocytes but retaining NK cells and macrophages. Histology and immunohistochemistry Hematoxylin and eosin (H&E)-stained histology slides were prepared from formalin-fixed GEMM-ARMS tumors at the Oregon Health & Science University Histology Shared Resource Core. Immunohistochemistry (IHC) for Desmin, Myogenin and Myh2 was performed by the laboratory of co-author BPR at the Cleveland Clinic Robert J. Tomsich Pathology and Laboratory Medicine Institute, Cleveland, OH. Interpretation of H&E and IHC were performed by pathologist co-author SC. Cell isolation and culture from GEMM-ARMS tumors Cells were isolated from GEMM-ARMS tumors as previously described26. Briefly, tumors were dissected from mice under sterile conditions and digested overnight at 4 °C in a 1% collagenase IV solution (17104019, Sigma-Aldrich, St. Louis, MO, USA) prepared in Dulbecco’s Modified Eagle Medium (DMEM; 11995-073, Thermo Fisher Scientific, Waltham, MA) supplemented with 10% fetal bovine serum (FBS; F8192, Sigma-Aldrich) and 1% penicillin-streptomycin (P4333-100ML, Millipore Sigma, Burlington, MA) (hereafter referred to as complete DMEM). The dissociated cells were passed through a 70 μm cell strainer and centrifuged at 200 xg for 5 min. The resulting cell pellet was resuspended in complete DMEM and transferred to a T75 tissue culture flask (12–565-349, Thermo Fisher Scientific), maintained at 37 °C in 5% CO₂. Cells were passaged every 3–4 days at a density of 1 × 106 cells per T75 flask. Mycoplasma testing was performed every 5 passages and after each experiment using the Promega PowerPlex16HS Assay (Madison, WI) at the University of Arizona Genetics Core, AZ. Intramuscular allograft tumor model in immunodeficient mice Six- to ten-week-old NSG or SHO mice were anesthetized with inhaled 2.5% of isoflurane delivered via anesthesia machine. To induce muscle injury, 0.05 ml of cardiotoxin (2.5–10 µM in sterile saline) was injected into the gastrocnemius muscle of the right hind limb. The contralateral leg served as a non-injured control. After 24 h, when the muscle environment was optimal for engraftment, 4 × 10⁶ cultured GEMM-ARMS tumor cells suspended in 50 µl of sterile phosphate-buffered saline (PBS; 20012-027, Gibco, Waltham, MA) were injected intramuscularly into the right gastrocnemius muscle using a 28G syringe. For some experiments, the tumor cell suspension was mixed with Matrigel (354262, Corning, San Diego, CA) prior to injection, with all preparations kept on ice to prevent premature solidification. Following injection, tumor growth was monitored and subsequently harvested for scRNA-seq, as detailed below. Sample preparation for scRNA-seq GEMM-ARMS and allografted immunodeficient mice were monitored twice weekly for tumor development. Once a tumor was detected, its size was measured using calipers three times per week and the ellipsoid formula: π/6 x length x width x height was used to determine tumor volume27. When tumor volume reached approximately 1.5 cm³, mice were euthanized by CO₂ asphyxiation followed by cervical dislocation. Lungs and tumors were collected during necropsy. Tumors were subdivided for downstream applications, including snap freezing for DNA/RNA sequencing, formalin fixation for histology, freezing for future cell culture and enzymatic digestion for scRNA-seq. Tumors were washed with 10 ml of chilled 1× Dulbecco’s phosphate-buffered saline (DPBS; 14200075, Thermo Fisher Scientific) and cut into 2–4 mm³ pieces. An enzyme mix was prepared in a gentleMACS C Tube (130-096-334, Miltenyi Biotec, Bergisch Gladbach, Germany) containing 2.35 ml of DMEM, 100 µl Enzyme H, 50 µl Enzyme R, and 12.5 µl Enzyme A from the gentleMACS dissociation kit (130-095-929, Miltenyi Biotec). Tumor pieces were added to the C Tube and mechanically dissociated using the gentleMACS Dissociator. The sample was incubated at 37 °C for 30 min and then mechanically dissociated again. After the cell suspension was centrifuged at 300 ×g for 30 s, the supernatant was aspirated, and the pellet resuspended in 10 ml DMEM. The suspension was filtered through a 70 μm MACS SmartStrainer and centrifuged again at 300 ×g for 7 min. The pellet was then resuspended in 1 ml chilled 1× Red Blood Cell Lysis Buffer (420301, BioLegend, San Diego, CA) and incubated at 4 °C for 10 min. After centrifugation at 300 ×g for 10 min at 4 °C, the pellet was resuspended in 5 ml chilled wash buffer. Cell counts were determined using a Countess Automated Cell Counter (Thermo Fisher Scientific), and cells were aliquoted at 6 × 105 cells in 500 µl. scRNA-seq was performed using the 10X Genomics Chromium X Series platform at the Oregon Health & Science University Gene Profiling/RNA and DNA Services Shared Resource and Massively Parallel Sequencing Shared Resource (MPSSR) cores, Portland, OR. The Chromium Next GEM Single Cell 3′ or 5’ Reagent Kits (10x Genomics) were used for single-cell capture, mRNA barcoding, reverse transcription into cDNA, and sequencing library preparation. Libraries were then sequenced on an Illumina platform according to the manufacturer’s recommendations. scRNA-seq data processing and annotation scRNA-seq data were pre-processed, normalized and clustered using Seurat v4 in R v4.1.228. Low-quality cells and low expressed genes were excluded prior to normalization. Highly variable genes were identified based on their expression relative to the mean variability of the dataset and used for clustering. Dimensionality reduction and visualization were performed using UMAP. Cluster annotation was performed with the SingleR package against built-in reference datasets and RMS cell line RNA-seq datasets29. Query datasets were subsequently projected onto the Seurat-derived reference using Symphony for reference-based mapping30. The sequencing data was deposited to the European Genome-phenome Archive (EGA) under the accession numbers: EGAD50000002241. Fluorescence Activated Cell Sorting (FACS) Cells dissociated and cultured from GEMM-ARMS mice expressing eYFP and dsRED2 reporters were counted and resuspended in 1× PBS containing 3% FBS at 1 × 10⁶ cells/ml and kept on ice until sorting. Sorting was performed using a BD FACSAria II (BD Biosciences, Franklin Lakes, NJ) equipped with a 100 μm flow nozzle, following the manufacturer’s specifications. A minimum of 1 × 106 sorted cells per sample were submitted to BGI (https://www.bgi.com/us/home) for bulk RNA sequencing. RNA-sequencing data acquisition Human RNA-sequencing data were obtained from the GTeX database and from the IRB-approved CuReFAST tumor banking system at the Children’s Cancer Therapy Development Institute (study # ccTDI-IRB-1). Results Histological examination reveals heterogeneity of tumors from the ARMS mouse model To investigate the cellular heterogeneity of ARMS, we first histologically examined three tumors derived from a genetically engineered mouse model (Myf6Cre Pax3P3Fm/P3Fm Trp53F2–10/F2–10) that recapitulates human ARMS (hereafter called GEMM-ARMS tumors)23,24. Hematoxylin and eosin (H&E) histological examination showed variable cell morphology (Fig. 1A-C). Each of the mouse models showed small round cells with minimal cytoplasm, round to ovoid nuclei and small nucleoli. Evidence of rhabdomyoblastic differentiation was noted by admixed slender strap cells containing eosinophilic cytoplasm with round to ovoid nuclei, as well as scattered rhabdoid cells with prominent eosinophilic cytoplasm, eccentric nuclei and small nucleoli (red arrowheads). Strap cells were best seen in mice U72739 and U73837 (black arrowheads), whereas rhabdoid morphology predominated in mouse U73834. Fig. 1Histology and immunohistochemistry of myogenic differentiation markers in tumors from a conditional knock-in mouse model of ARMS (GEMM-ARMS; Myf6Cre Pax3P3Fm/P3Fm Trp53F2–10/F2–10). (A–C) Representative H&E images showing variable rhabdomyosarcoma morphology, including strap cells (black arrowheads), scattered rhabdoid cells (red arrowhead) and small cells. (D–L) IHC demonstrates tumor cell heterogeneity across different stages of myogenic differentiation: early/intermediate (Desmin, D–F), intermediate (Myogenin, G–I) and late (Myh2, J–L). Panels (M, N) show magnified views of the boxed regions in (A, B), respectively. Scale bars: A–C, 50 μm; D, G, J, 200 μm; E-F, H-I, K-L, 100 μm. Histology and immunohistochemistry of myogenic differentiation markers in tumors from a conditional knock-in mouse model of ARMS (GEMM-ARMS; Myf6Cre Pax3P3Fm/P3Fm Trp53F2–10/F2–10). (A–C) Representative H&E images showing variable rhabdomyosarcoma morphology, including strap cells (black arrowheads), scattered rhabdoid cells (red arrowhead) and small cells. (D–L) IHC demonstrates tumor cell heterogeneity across different stages of myogenic differentiation: early/intermediate (Desmin, D–F), intermediate (Myogenin, G–I) and late (Myh2, J–L). Panels (M, N) show magnified views of the boxed regions in (A, B), respectively. Scale bars: A–C, 50 μm; D, G, J, 200 μm; E-F, H-I, K-L, 100 μm. ARMS is a malignancy presumed to derive from the skeletal muscle lineage but with dysregulated myogenic differentiation31. Therefore, we next examined the GEMM-ARMS tumors for myogenic differentiation markers at different stages: early/intermediate (Desmin), intermediate (Myogenin) and late (Myh2)32 (Fig. 1D-L). By immunohistochemistry (IHC), all models showed diffuse cytoplasmic positivity for Desmin (Fig. 1D-F) with a subset of tumor cells having nuclear positivity for Myogenin and Myh2 (Fig. 1G-L). While our GEMM-ARMS tumors exhibited positive staining for the ARMS diagnostic markers Desmin, Myogenin, and Myh2, variations in their distribution and intensity within the tumors suggest the presence of distinct subpopulations of cells with varying degrees of aberrant myogenic differentiation, which may contribute to the complexity of this aggressive malignancy. scRNA-seq analysis reveals notable presence of immune cells within ARMS tumors To provide insights into the cellular composition and heterogeneity of ARMS tumor microenvironment, tumors from the same three GEMM-ARMS mice in Fig. 1 were enzymatically dissociated individually, and the released cells were subjected to scRNA-seq analysis as detailed in Materials and Methods. The single-cell data obtained from tumor sample of mouse U72739 was first assembled and mapped as the reference dataset using Seurat33. scRNA-seq data from the other two tumor samples were then projected onto the U72739 UMAP. Results show that cell atlas datasets from three GEMM-ARMS samples could be “anchored” and integrated (Fig. 2A), indicating a shared biological state across the three different biological samples. Fig. 2UMAP visualization of cell subtypes in mouse GEMM-ARMS tumors. (A) Single-cell RNA-seq profiles from three GEMM-ARMS tumors (mouse IDs U72739, U73837 and U73834) are shown. U72739 was used to assemble the reference dataset with Seurat, and data from U73837 and U73834 were projected onto this UMAP. (B) Malignant and non-malignant cells were annotated using SingleR, which classified single-cell transcriptomes by comparison with bulk transcriptomes from the Human Primary Cell Atlas. (C) Inferred copy number variation (CNV) profiles from RNA-seq were used to distinguish malignant from non-malignant clusters, corroborating the SingleR-based classification. UMAP visualization of cell subtypes in mouse GEMM-ARMS tumors. (A) Single-cell RNA-seq profiles from three GEMM-ARMS tumors (mouse IDs U72739, U73837 and U73834) are shown. U72739 was used to assemble the reference dataset with Seurat, and data from U73837 and U73834 were projected onto this UMAP. (B) Malignant and non-malignant cells were annotated using SingleR, which classified single-cell transcriptomes by comparison with bulk transcriptomes from the Human Primary Cell Atlas. (C) Inferred copy number variation (CNV) profiles from RNA-seq were used to distinguish malignant from non-malignant clusters, corroborating the SingleR-based classification. ARMS tumors inherently contain non-malignant cells34. To distinguish malignant versus non-malignant cells in GEMM-ARMS tumors, we employed the R package SingleR to annotate single cells by comparing their transcriptomic profiles to reference bulk transcriptomes derived from “normal” cells on the Human Primary Cell Atlas (https://data.humancellatlas.org)35. The majority of cells (> 80%) were presumed malignant, whereas most presumed normal cell types were identified as macrophages, B cells, T cells, and NK cells, as well as (vascular) endothelial cells (Fig. 2B). Copy number variations (CNVs), which involve amplifications or deletions of genomic regions, are prevalent in malignant cells and play a pivotal role in tumorigenesis and disease progression36. Presumed normal versus malignant cells were also identified by comparing RNA-seq “inferred” CNV profiles among cell clusters (Fig. 2C, see also Supplementary Fig. 1), reinforcing the findings presented in Fig. 2B. scRNA-seq analysis reveals distinct muscle differentiation states in ARMS tumors We next used SingleR to probe the subpopulations of cells within the tumor cell clusters identified in Fig. 2B. Using unsupervised clustering analysis, a total of 17 major clusters were identified from the dissociated GEMM-ARMS samples (Fig. 3A). The cell types, identified based on enriched known markers (e.g., CD68 and CD163 for macrophages), include the normal cells detected in Figs. 2B and 11 major tumor cell sub-clusters. This finding corroborates our IHC analysis in Fig. 1, which implies the presence of diverse tumor subpopulations characterized by varying myogenic differentiation marker expressions. The gene expression of GEMM-ARMS tumors was further visualized using Loupe Browser (10x Genomics), revealing that distinct ARMS cell clusters exhibit differential expression of myogenic differentiation markers. Some clusters showed upregulation of early and intermediate markers Desmin, Myod1 and Myogenin, while others display low levels of the late marker Myh1 (Fig. 3B). These findings suggest that ARMS cells, whether phenotypically or by mimicry, are arrested at different stages of muscle-lineage commitment, highlighting the heterogeneous cellular composition of the tumor microenvironment. Fig. 3Tumor cell heterogeneity in GEMM-ARMS tumors. (A) Unsupervised clustering of single-cell RNA-seq of GEMM-ARMS tumors using SingleR identified 17 major clusters assigned based on signature gene expression. (B) UMAP visualization of myogenic differentiation marker expression across tumor clusters shows heterogeneous patterns. Some clusters upregulate early/intermediate markers (Desmin, Myod1, Myogenin), while others exhibit low levels of the late marker Myh1, indicating that cells from GEMM-ARMS tumors were arrested at different stages of muscle-lineage commitment. Tumor cell heterogeneity in GEMM-ARMS tumors. (A) Unsupervised clustering of single-cell RNA-seq of GEMM-ARMS tumors using SingleR identified 17 major clusters assigned based on signature gene expression. (B) UMAP visualization of myogenic differentiation marker expression across tumor clusters shows heterogeneous patterns. Some clusters upregulate early/intermediate markers (Desmin, Myod1, Myogenin), while others exhibit low levels of the late marker Myh1, indicating that cells from GEMM-ARMS tumors were arrested at different stages of muscle-lineage commitment. Host immune environment shapes cellular heterogeneity in GEMM-ARMS tumor allografts As shown earlier in Fig. 2B, GEMM-ARMS tumors contain a mixture of malignant and non-malignant cells, with the majority of non-malignant cells exhibiting immune cell-like gene expression. To establish a more tumor cell–enriched model, we isolated cells from GEMM-ARMS tumors, expanded them in vitro, and then orthotopically implanted them into two immunodeficient mouse strains, NSG and SHO, which differ in the extent of immune cell deficiency25,37. In the allograft study, one presumes that only cells with tumorigenic potential (i.e., ARMS tumor cells in our case) are capable of forming tumor nodules after implantation38. Since both strains are commonly used for allograft and xenograft studies39,40, this study also enabled us to examine how the host immune microenvironment, specifically the presence or absence of certain immune cell types, affects tumor cellular composition. In NSG mice, which are deficient in mature B and T lymphocytes, NK cells and macrophages, approximately half of the original cell clusters observed in GEMM-ARMS tumors were absent in the allograft (Fig. 4A, B). The missing cell populations were predominantly non-malignant cell types, including adipocytes, astrocyte-like cells, epithelial cells, neuron-like cells, microglia, dendritic cells and immune-related cell clusters such as basophils, and B and T cells (Fig. 4C, D). Notably, despite the lack of macrophages and NK cells in NSG mice, clusters with transcriptomic profiles resembling these cell types were still detected in the allografts. Fig. 4UMAPs of cell subtypes from GEMM-ARMS tumor cell allografts in immunodeficient NSG mice. (A) Single-cell RNA-seq UMAP of GEMM-ARMS tumors (Fig. 3A) used as a reference atlas. (B) UMAP of allograft tumors generated by injecting GEMM-ARMS-derived cultured cells into NSG mice, which lack mature B and T lymphocytes, NK cells and macrophages, with clusters mapped onto the reference atlas using the Symphony R package. (C, D) Predicted cell type specificity of clusters in (A) and (B), respectively, determined using the SingleR R package. UMAPs of cell subtypes from GEMM-ARMS tumor cell allografts in immunodeficient NSG mice. (A) Single-cell RNA-seq UMAP of GEMM-ARMS tumors (Fig. 3A) used as a reference atlas. (B) UMAP of allograft tumors generated by injecting GEMM-ARMS-derived cultured cells into NSG mice, which lack mature B and T lymphocytes, NK cells and macrophages, with clusters mapped onto the reference atlas using the Symphony R package. (C, D) Predicted cell type specificity of clusters in (A) and (B), respectively, determined using the SingleR R package. In contrast, in SHO mice, which lack mature B and T lymphocytes but retain functional macrophages and NK cells, B and T cell-like clusters remained detectable in the allografts. Furthermore, most of the non-malignant cell clusters, except for four: adipocytes, astrocyte-like cells, pro-B cells and basophils, were still present (Fig. 5A, B). Fig. 5UMAPs of cell subtypes from GEMM-ARMS tumor cell allografts in immunodeficient SHO mice. (A) Single-cell RNA-seq UMAP of GEMM-ARMS tumors (Fig. 3A) used as a reference atlas. (B) UMAP of allograft tumors generated by injecting GEMM-ARMS-derived cultured cells into SHO mice, which lack mature B and T lymphocytes but retain functional macrophages and NK cells. Clusters were mapped onto the reference atlas using the Symphony R package. (C, D) Predicted cell type specificity of clusters in (A) and (B), respectively, determined using the SingleR R package. UMAPs of cell subtypes from GEMM-ARMS tumor cell allografts in immunodeficient SHO mice. (A) Single-cell RNA-seq UMAP of GEMM-ARMS tumors (Fig. 3A) used as a reference atlas. (B) UMAP of allograft tumors generated by injecting GEMM-ARMS-derived cultured cells into SHO mice, which lack mature B and T lymphocytes but retain functional macrophages and NK cells. Clusters were mapped onto the reference atlas using the Symphony R package. (C, D) Predicted cell type specificity of clusters in (A) and (B), respectively, determined using the SingleR R package. Transcriptional heterogeneity of Pax3::Foxo1 expressing and rhabdomyoblast-like cells in GEMM-ARMS tumors To better characterize the cellular composition and molecular features of GEMM-ARMS tumors, eYFP and dsRED2 fluorescent reporters were introduced by breeding into the GEMM-ARMS mouse background to mark distinct ARMS-associated cell populations. The eYFP reporter labels cells expressing the Pax3::Foxo1 fusion gene, a known oncogenic driver in ARMS that promotes tumorigenesis by blocking normal myogenic differentiation, thereby maintaining an early, less differentiated precursor-like subpopulation41,42. In parallel, dsRED2 labels Myosin Heavy Chain 2 (Myh2) expressing rhabdomyoblast-like cells, which are more differentiated malignant cells in ARMS. These cells phenotypically resemble late-stage muscle precursors of normal skeletal muscle development43,44. In zebrafish rhabdomyosarcoma models, such differentiated (matured) tumor cells have different roles and abilities to cause disease progression45. In ARMS, rhabdomyoblasts fail to complete terminal differentiation into mature muscle fibers, and their presence is a key histopathological feature of the disease14,15. We performed flow cytometric sorting based eYFP and dsRED2 expression, with selection criteria and gating strategy shown in Supplementary Tables 1 and Supplementary Fig. 2–3. The sorted tumor cells were then used as a query set for Symphony mapping30, enabling projection of their single-cell transcriptomic profiles onto a UMAP reference atlas generated from GEMM tumors (Fig. 3A). The mapping analysis revealed that Pax3::Fox1 expressing cells (eYFP+; n = 4) and Myh2 expressing rhabdomyoblast-like cells (dsRED2+; n = 3) are positioned within cell type clusters 3 and 12, respectively (Fig. 6A), both of which are annotated as malignant ARMS tumor cell populations (Fig. 6B). Fig. 6UMAP projection of fluorescent reporter-labelled subpopulations in GEMM-ARMS tumors. (A) Flow cytometry-sorted tumor cells expressing eYFP (Pax3::Foxo1 expressing precursor-like cells) or dsRED (Myh2 expressing rhabdomyoblast-like cells) were projected onto the reference UMAP atlas generated in Fig. 3A. eYFP+ cells (shown as squares) localize primarily to cluster 3, whereas dsRED+ rhabdomyoblasts (shown as circles) map to cluster 12. (B) Reference cell type annotation, as shown in Fig. 2A, indicates that both eYFP + and dsRED+ cells belong to malignant ARMS tumor populations. UMAP projection of fluorescent reporter-labelled subpopulations in GEMM-ARMS tumors. (A) Flow cytometry-sorted tumor cells expressing eYFP (Pax3::Foxo1 expressing precursor-like cells) or dsRED (Myh2 expressing rhabdomyoblast-like cells) were projected onto the reference UMAP atlas generated in Fig. 3A. eYFP+ cells (shown as squares) localize primarily to cluster 3, whereas dsRED+ rhabdomyoblasts (shown as circles) map to cluster 12. (B) Reference cell type annotation, as shown in Fig. 2A, indicates that both eYFP + and dsRED+ cells belong to malignant ARMS tumor populations. Analyses of the most highly and differentially expressed genes (DEGs) further revealed distinct transcriptional profiles for eYFP + and dsRED2 + sorted cell populations (Supplementary Table 2; see also Supplementary Data Excel file for full DEG lists for all cell clusters). Cluster 3 (Pax3::Foxo1 expressing cells; eYFP+) was enriched for epithelial and epithelial-mesenchymal transition (EMT)-associated genes, including Epcam, Claudin family members, Krt16 and Krt18, accompanied by downregulation of Cdh1 (E-cadherin). This profile aligns with recent evidence suggesting that RMS may arise from non-myogenic progenitors or may represent a transitional EMT-like state46,47, potentially contributing to cell migration and tumor invasion. Cluster 12 (Myh2 expressing rhabdomyoblast-like cells; dsRED2+) instead showed upregulation of fibroblast- and mesenchymal-associated genes, notably extracellular matrix components (Col1a1, Col15a1, Fn1, Fmod, Lum, Dcn and Bgn) and mesenchymal regulators (Twist1, Osr1 and Mmp3), consistent with roles in ECM production, fibrosis or stromal remodeling within the tumor microenvironment48,49. Together, these results demonstrate that Pax3::Foxo1 expressing cells and rhabdomyoblast-like cells are positioned within transcriptionally distinct malignant subpopulations in GEMM-ARMS tumors. Their divergent epithelial/EMT-like and fibroblast/mesenchymal-like programs indicate that ARMS tumor cells can adopt multiple phenotypic states, potentially reflecting functional specialization during tumor progression. These findings provide a foundation for understanding intratumoral heterogeneity in ARMS. Finally, to evaluate how well GEMM tumors recapitulate clinical conditions, we performed bulk RNA-seq on 11 human RMS cell lines and 3 human or mouse skeletal muscle cell lines as controls (Supplementary Fig. 4 A). Using Symphony, these transcriptomic profiles were projected onto the GEMM-derived reference UMAP (Supplementary Fig. 4B, C), revealing that human ARMS cell lines predominantly aligned with mouse tumor cluster 2 (Supplementary Fig. 4D–F). In contrast, normal skeletal muscle cell lines localized to cluster 12, annotated as a fibroblast-like population by SingleR based on fibroblast- and mesenchymal-associated gene expression (Supplementary Fig. 4D-F). Discussion In the context of childhood cancer, understanding the cellular and molecular heterogeneity of ARMS is particularly important. Pediatric cancers often exhibit distinct biological characteristics compared to adult cancers, and the developing tissues in children may respond differently to oncogenic mutations50–52. Moreover, additional molecular heterogeneity has been shown to underlie tumor aggression and therapy resistance in some cases53,54. The long-term side effects of aggressive treatments are also a significant concern in pediatric oncology, emphasizing the need for therapeutic approaches that minimize harm to healthy tissues55. However, before advancing such treatments, a deeper understanding of ARMS tumor biology is essential, as the complex tumor microenvironment and tumor cell composition remains poorly characterized. Previous studies have shown that ARMS tumors exhibit a heterogeneous cellular composition comprising primarily malignant cells driven by the characteristic PAX3::FOXO1 or PAX7::FOXO1 fusion oncogenes56. Overall, ARMS tumors exhibit cellular and histopathological features reminiscent of developing skeletal muscle57–59. The tumor cells encompass a spectrum of differentiation states, ranging from early myogenic progenitor-like populations to more differentiated rhabdomyoblasts that express muscle lineage markers such as Desmin, Myogenin and Myosin Heavy Chain 2 (Myh2), not unlike fusion negative embryonal rhabdomyosarcoma zebrafish models45. Additionally, the tumor microenvironment contains various non-malignant stromal components, including fibroblasts, immune cells and endothelial cells, which may influence progression and therapy response, as reported elsewhere22,34,60. Curiously, however, the human cell lines that are commonly studied represent only one of the more than a dozen tumor cell subpopulations. Recent scRNA‑seq studies in rhabdomyosarcoma reveal developmental hierarchies mirroring normal myogenesis, with stem‑like and proliferative progenitor populations driving tumor growth and therapeutic resistance14,15. Importantly, in fusion-positive RMS, RAS-pathway inhibition can redirect malignant cells toward more differentiated, less proliferative states34. Single-cell profiling also uncovers a strongly immunosuppressive microenvironment, characterized by M2‑polarized macrophages and a NECTIN3-TIGIT axis that may promote T cell exhaustion, and shows that malignant cell differentiation states strongly predict patient survival22. A limitation of our study is that while we identified the same immune cell components also described for human RMS in DeMartino et al.22, we did not perform functional studies of immune cell populations in the GEMM and allograft models. Our analysis of GEMM-ARMS tumors reveals a highly heterogeneous tumor microenvironment composed of multiple malignant subpopulations alongside diverse non-malignant cells, including immune and stromal components. Histological and immunohistochemical analyses confirmed the presence of distinct myogenic differentiation states within the tumors, consistent with the known biology of ARMS as a malignancy often arising from skeletal muscle lineage cells with disrupted differentiation4,32. scRNA sequencing further elucidated this heterogeneity, uncovering 17 major clusters including 11 malignant tumor subpopulations and a range of non-malignant immune and stromal cells – a surprising finding. Importantly, integration of scRNA-seq data across biological replicates demonstrated consistent cellular states, reinforcing the robustness of our model and analysis. The predominance of immune cells such as macrophages, B and T lymphocytes, and NK cells within the tumor microenvironment highlights the significant immune infiltration in ARMS, consistent with recent reports emphasizing the role of the immune milieu in sarcomas34,60. Our allograft studies in immunodeficient mice with differing immune deficiencies underscore the critical influence of host immune context on tumor cell composition. Tumors grown in highly immunocompromised NSG mice exhibited loss of approximately half of the original cell clusters, predominantly non-malignant populations, while SHO mice retaining macrophages and NK cells preserved most clusters. This finding could suggest that engraftment of certain non-malignant populations is dependent on or interacts closely with specific immune subsets61,62. Alternatively, our ex vivo cell cultures may have preferentially selected for specific tumor cell subpopulations. The persistence of immune-related gene expression and non-malignant cell types in the tumor allografts may reflect three possible explanations: (i) tumor immune mimicry, in which cancer cells adopt phenotypic or transcriptional features of immune cells to evade immune surveillance61, (ii) transdifferentiation of tumor cells along alternative lineages within the tumor microenvironment62,63 or (iii) incomplete purification of tumor cells. Notwithstanding the latter technical issue, these findings highlight complex cellular interactions within the tumor microenvironment and underscore the importance of the immune system in shaping tumor cell heterogeneity. By using My6Cre-driven fluorescent reporters to label Pax3::Foxo1 expressing precursor-like cells and Myh2 expressing rhabdomyoblast-like cells within the skeletal muscle cell lineage, combined with Symphony mapping, we identified two transcriptionally distinct malignant subpopulations. Pax3::Foxo1 expressing cells exhibited an epithelial/EMT-like feature characterized by upregulation of Epcam, Claudins and Krt genes with downregulation of Cdh1, reflecting a transitional state potentially linked to tumor progression and invasion46,47. Myh2 expressing rhabdomyoblast-like cells displayed a fibroblast/mesenchymal-like transcriptional profile enriched for extracellular matrix and mesenchymal regulatory genes such as Col1a1, Fn1, Twist1 and Mmp3, implicating this subpopulation in fibrosis and microenvironment remodeling48,49. To further validate and extend our scRNA-seq findings, proteomic analyses could provide complementary evidence for EMT-like and progression-associated programs by confirming protein-level pathway activation. Prior studies in colorectal64, breast65 and pancreatic66 cancers have shown upregulation of ECM and EMT-associated proteins such as collagens, laminins, EPCAM and fibronectin. In RMS, proteomic profiling of RMS-derived exosomes has identified enrichment of adhesion- and ECM-remodeling proteins, further supporting protein-level regulation of tumor progression67. Applying such multi-omics strategies in RMS could help confirm the functional significance of these subpopulations and identify potential therapeutic targets. We also caution that while we draw parallels between muscle differentiation and tumor cell subpopulations, the Pax3::Foxo1/eYFP+ expressing tumor cells being ‘primitive’ and muscle stem cell-like, versus Myosin heavy chain 2 (Myh2)/DsRED2 expressing tumor cells being more akin to maturing muscle myoblasts, these two cell types do not necessarily come from the other. The cell states adopted by a cancer cell subpopulation may simply be phenocopy. Our findings reveal mesenchymal-enriched, progenitor-like subpopulations, a feature that has also been reported in recent human RMS studies15,34. However, different studies have used distinct nomenclature to classify RMS subpopulations, including ‘myoblast-like’68, ‘mesenchymal-like’, ‘muscle stem cell (MuSC)-like’ and ‘myogenin+ cycling progenitors’34. In the future, integrating existing RMS scRNA-seq datasets could help identify shared signature gene programs and clarify how these subpopulations relate across different models and RMS subtypes. Together, these findings highlight the phenotypic plasticity of ARMS tumor cells and the complex interactions within their microenvironment. The coexistence of distinct malignant subpopulations with divergent differentiation and microenvironmental remodeling programs may contribute to intratumoral heterogeneity, therapeutic resistance and disease progression. Future studies dissecting the functional roles of these subpopulations and their crosstalk with immune and stromal cells will be critical to identify novel therapeutic vulnerabilities in ARMS. Our GEMM-ARMS model, combined with advanced single-cell and cell labeling via Cre-driven fluorescent reporter approaches, provides a valuable functional (experimental) platform for elucidating ARMS biology and the dynamic tumor-immune microenvironment. This set of tools for understanding cellular diversity and plasticity in ARMS is hoped to inform future studies that address tumor cell heterogeneity and stromal microenvironmental cross-talk as mechanisms of therapeutic resistance. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2