Ewing Sarcoma and Osteosarcoma Have Distinct Immune Signatures and Intercellular Communication Networks

Abstract

Purpose: Ewing sarcoma and osteosarcoma are primary bone sarcomas occurring most commonly in adolescents. Metastatic and relapsed disease are associated with dismal prognosis. Although effective for some soft tissue sarcomas, current immunotherapeutic approaches for the treatment of bone sarcomas have been largely ineffective, necessitating a deeper understanding of bone sarcoma immunobiology.

Experimental design: Multiplex immunofluorescence analysis of immune infiltration in relapsed versus primary disease was conducted. To better understand immune states and drivers of immune infiltration, especially during disease progression, we performed single-cell RNA sequencing (scRNAseq) of immune populations from paired blood and bone sarcoma tumor samples.

Results: Our multiplex immunofluorescence analysis revealed increased immune infiltration in relapsed versus primary disease in both Ewing sarcoma and osteosarcoma. scRNAseq analyses revealed terminally exhausted CD8+ T cells expressing co-inhibitory receptors in osteosarcoma and an effector T-cell subpopulation in Ewing sarcoma. In addition, distinct subsets of CD14+CD16+ macrophages were present in Ewing sarcoma and osteosarcoma. To determine pathways driving tumor immune infiltration, we conducted intercellular communication analyses and uncovered shared mechanisms of immune infiltration driven by CD14+CD16+ macrophages and unique pathways of immune infiltration driven by CXCL10 and CXCL12 in osteosarcoma.

Conclusions: Our study provides preclinical rationale for future investigation of specific immunotherapeutic targets upon relapse and provides an invaluable resource of immunologic data from bone sarcomas.

Figure 1.. Multiplex immunofluorescence analysis in paired patient samples reveals an evolution of immune infiltration upon bone sarcoma progression.

a-d, Multiplex immunofluorescence for CD4, CD8, CD68, CD20, FoxP3, and DAPI was performed on local samples of Ewing sarcoma and osteosarcoma both from primary disease and relapse. Longitudinal samples (primary, relapse 1, relapse 2) from four patients: two with Ewing sarcoma and two with osteosarcoma are shown. e-f, Workflow of multiplex immunofluorescence analysis for CD4, CD8, CD68, CD20, FoxP3, and DAPI performed on paired patient primary and relapse samples of Ewing sarcoma from the Children’s Oncology Group Biorepository (AEWS20B1-Q). Quantification of immune cell infiltrate following multiplexed IHC analysis in 11 primary bone Ewing tumors as compared to lung relapse samples in the same 11 patients. *=p<0.05 using a student’s t-test. g, Data from the samples in (f) plotted per paired patient sample. P=primary bone Ewing sarcoma, R=lung relapse of Ewing sarcoma.

Figure 2.. Identification of canonical cell types in the peripheral blood and tumor microenvironment of patients with pediatric bone sarcomas and healthy controls.

Live immune cells (i.e. all CD45+ cells) were sorted from peripheral blood mononuclear cells from healthy pediatric blood donors, Ewing sarcoma and osteosarcoma patients and from single-cell suspensions of tumor tissue samples from Ewing sarcoma and osteosarcoma patients. Samples were subjected to droplet-based single-cell RNAseq and canonical cell types were identified (Methods). a, Schematic depicting the workflow employed to generate single-cell RNAseq data from healthy donors and patients. b, Heatmap depicting the top differentially expressed genes across single cells from each inferred canonical immune population. Major canonical lineage markers are indicated. The heatmap is downsampled to 10% of all cells for visualization. c, UMAPs across sample groups showing the inferred cell types. The top row shows PBMC samples from healthy donors (6793 cells), Ewing sarcoma (3156 cells) and osteosarcoma (7480 cells) and Ewing sarcoma patients. The bottom row shows tumor-infiltrating CD45+ cell samples from Ewing sarcoma (5194 cells) and osteosarcoma (6409 cells). d, Quantification of the frequencies of major immune lineages across samples from PBMC. Each dot represents an individual patient sample. e, Quantification of the frequencies of major immune lineages from tumor infiltrating CD45+ cells between Ewing and osteosarcoma. Each dot represents an individual patient sample.

Figure 3.. Analysis of St. Jude’s PeCan and the TCGA TARGET-OS datasets reveals a distinct immune infiltrate composition between primary Ewing sarcoma and osteosarcoma and association between myeloid cells and survival in osteosarcoma

Publicly available pediatric Ewing sarcoma and osteosarcoma datasets were utilized in conjunction with our scRNAseq data to infer frequencies of tumor infiltrating immune cells and the relationships between immune populations, tumor mutation burden and survival outcomes. a, Schematic depicting the analysis workflow for data from St. Jude’s PeCan database and The Cancer Genome Atlas’s TARGET-OS database. b, Tumor mutation burden was significantly higher in osteosarcoma versus Ewing sarcoma (median 0.67 mutations per megabase versus 0.13 mutations per megabase, p<0.0001 by rank sum test). c, Tumor mutation burden was significantly correlated with ImmuneScore across osteosarcoma and Ewing sarcoma as quantified from bulk RNAseq data using ESTIMATE. d, Cox proportional hazards survival analysis showing that osteosarcoma patients with higher levels of ImmuneScore had better overall survival. e, Cox proportional hazards analysis showing that osteosarcoma patients with higher levels of monocytic lineage infiltrated had better overall survival. f, Heatmap showing infiltration levels of individual immune cell subsets in primary osteosarcoma and Ewing sarcoma inferred by CIBERSORTx using our single-cell RNAseq data to derive the CIBERSORTx signature matrix (Methods). CD8+ T cell frequencies were higher in Ewing sarcoma versus osteosarcoma, whereas CD14+CD16+ macrophages levels were higher in osteosarcoma. g, ImmuneScore from ESTIMATE was significantly correlated monocytic lineage abundance from MCPcounter and CD14+CD16+ macrophage frequency from CIBSERSORTx in osteosarcoma patients from TARGET-OS. h, Cox proportional hazards survival analysis showing that higher levels of CD14+CD16+ macrophages were associated with better overall survival in osteosarcoma patients from TARGET-OS.

Figure 4.. Analysis of CD8+ T cells reveals more effector-like populations in Ewing sarcoma versus osteosarcoma and a subset of exhausted CD8+ T cells in both pediatric sarcomas.

CD8+ T cells from PBMC and the TME were bioinformatically isolated from all immune cells and were re-analyzed to evaluate heterogeneity across samples and tissues. a, UMAPs show unsupervised clustering of CD8+ T cells from PBMC and tumor-infiltrating CD45+ cells from all sample types. b, Heatmap showing the top differentially expressed genes across clusters from (a). c, Enrichment of sample types across clusters from (a). d, Gene set enrichment analysis showing progenitor exhaustion states and tissue residence signatures. Cluster 3 is associated with terminal exhaustion, while clusters 0 and 4 show Tex intermediate signatures. e-h, Leading edge analysis showing the top differentially expressed genes for key gene signatures and clusters from (d). Cluster 3 shows hallmarks of terminal exhaustion, including co-expression of inhibitory receptors TIGIT, LAG3, PDCD1 (gene for PD1) and HAVCR2 (gene for TIM3).

Figure 5.. Dissection of myeloid cell states reveals diverse functional enrichment and heterogeneity across major macrophage states.

Monocytes and macrophages from PBMC and the TME were bioinformatically isolated from all immune cells and were re-analyzed to evaluate myeloid cell heterogeneity across samples and tissues. a, UMAPs showing all myeloid cells across PBMC (top row) and the TME (bottom row) colored by the inferred canonical myeloid state governed by expression of CD14, CD16 and complement receptors. Monocytes were enriched in PBMC, but we also present at lower frequencies in the TME. b, Gene set enrichment analysis using hallmark gene sets from the molecular signatures database. CD14+CD16+ macrophages were enriched for gene sets associated with immune responses including IL6 JAK STAT signaling, TNFa signaling and inflammatory response. c, Same UMAPs as (a) but colored by unsupervised cluster agnostic of cell states. A total of 12 clusters were identified, with 5 clusters consisting mostly of monocyte states, 5 clusters consisting mostly of macrophage states and 2 clusters shared between monocytes and macrophages. d, Associations between clusters from (c) and tissue and sample types. e, Top 50 differentially expressed genes across clusters from (c) with considerable heterogeneity across CD14+CD16+ macrophage states.

Figure 6.. Analysis of intercellular communication in recurrent Ewing sarcoma and osteosarcoma reveals conserved and distinct drivers of immune cell infiltration.

Inference of downstream signaling activities of ligands produced by CD14+CD16+ macrophages in recurrent disease reveals two major clusters of chemokines that drive immune infiltration in recurrent disease. a, Frequencies of CD14+CD16+ macrophages and immune infiltration inferred by ImmuneScore are correlated across 3 datasets of recurrent osteosarcoma and Ewing sarcoma. b, Schematic of NicheNet-based inference of ligands produced by CD14+CD16+ macrophages and the downstream signaling pathways they activated. c, Putative ligands produced by CD14+CD16+ macrophages from our scRNAseq data and the downstream pathways putative activated by these ligands (Methods). d, Results from linear modeling of the relationship between immune infiltration, disease type (e.g. Ewing sarcoma versus osteosarcoma) and ligand-activated gene set scores. The linear model coefficient for each ligand that was significantly associated with ImmuneScore is shown. Ligands in (d) drive infiltration independent of disease. e, CXCL10 and CXCL12 downstream signaling pathways are elevated in osteosarcoma versus Ewing sarcoma and are associated with ImmuneScore (p=0.035 for CXCL10; p=0.021 for CXCL12). f, Downstream signaling pathways activated by CD14+CD16+ ligands can be grouped into 3 clusters. g, Cluster 1 and 2 (from f) scores are correlated with ImmuneScore.