Osteosarcoma (OS), a highly aggressive bone cancer, disproportionately affects children and adolescents. Despite its significant impact, therapeutic advancements have been limited since the 1986 introduction of combined surgical excision and chemotherapy. This slow progress is partly due to OS's rarity, hindering large clinical trials. Large animal models offer an alternative, and spontaneously occurring canine OS is considered an ideal model for human OS due to its higher prevalence in dogs, similar genetics and pathology, and immunocompetent status. However, species-specific reagent limitations have hampered the complete characterization of the canine OS TME, a complex environment composed of malignant osteoblasts, osteoclasts, fibroblasts, macrophages, lymphocytes, and other stromal and immune components. This TME creates an immunosuppressive environment that hinders antitumor immune responses. While increased macrophage abundance has unexpectedly been linked to reduced metastasis and enhanced survival in both humans and dogs, the underlying mechanisms remain poorly understood, highlighting the need for deeper investigation into OS pathobiology. Single-cell RNA sequencing (scRNA-seq) offers a powerful approach to overcoming species-specific reagent limitations and provides detailed transcriptomic information on individual cells within heterogeneous tissues. Recent human scRNA-seq studies of OS provide a valuable reference for cross-species cell type homology analysis. This study aimed to use scRNA-seq to perform a molecular dissection of the canine OS TME and to evaluate cell type transcriptomic homologies between canine and human OS.

The literature reveals a significant gap in understanding the complex interplay of cells within the osteosarcoma tumor microenvironment (TME). While studies have shown correlations between macrophage infiltration and clinical outcomes, the results are conflicting, with some suggesting a positive correlation with survival and others reporting a negative correlation. This inconsistency highlights the need for a more detailed understanding of the heterogeneity within the macrophage population and other immune cells within the TME. Previous studies have relied on bulk RNA sequencing or immunohistochemistry, limiting their ability to resolve the distinct cell populations and their interactions. The use of single-cell RNA sequencing (scRNA-seq) addresses these limitations by allowing for the identification of diverse cell types and their transcriptional states, providing a more comprehensive picture of the TME and its influence on tumor progression and response to therapy. This approach has been successfully employed in human osteosarcoma studies, revealing the complex heterogeneity of the TME, but the lack of comparable canine data has hindered direct cross-species comparisons. Therefore, a detailed single-cell analysis of canine osteosarcoma is crucial for validating the use of canine models in preclinical research and for identifying conserved therapeutic targets across species.

Six treatment-naïve dogs with spontaneously occurring primary appendicular osteosarcoma were included. Following amputation, tumor samples were processed for scRNA-seq within 30 minutes. Samples were digested with collagenase type II, passed through a 70-µm cell strainer, and subjected to density centrifugation using Ficoll Paque to enrich for live cells. Red blood cells were lysed using Ammonium-Chloride-Potassium lysis buffer. Single-cell suspensions were processed using the Chromium iX instrument (10x Genomics) with a target of 5000 cells per sample (two samples processed per dog for dogs 1 and 2). Libraries were prepared using a Chromium Next GEM Single Cell 3’ v3.1 Kit and sequenced on an Illumina NovaSeq 6000 sequencer. Cell Ranger (version 6.1.2) was used for read mapping and count matrix generation, aligning to the CanFam3.1 reference genome. Data filtering criteria included: 200 < nFeature_RNA < 5500, percent.mt < 12.5, and 100 < nCount_RNA < 75000. DoubletFinder was used for doublet removal. Seurat's alignment workflow was used for data integration, incorporating percent.MT as a latent variable to minimize mitochondrial read impact. Unsupervised clustering was performed using the clustree package to determine optimal parameters. Subclustering was done for major cell types (tumor/stroma, macrophage/monocyte, osteoclast, dendritic cell, and T cell) using Seurat's alignment workflow, and low-quality clusters were removed. Cell type classification relied on unsupervised clustering, gene set enrichment analysis (GSEA), and manual annotation based on canonical markers and reference mapping. Differential gene expression (DGE) analysis was conducted using pseudobulk conversion and DESeq2. CopyKAT and inferCNV were used for copy number variation (CNV) analysis. PySCENIC was used to predict active regulons, and CellChat was used to infer cell-cell interactions. For cross-species analysis, a publicly available human OS scRNA-seq dataset (GSE162454) was integrated with the canine dataset using Seurat's alignment workflow. Hierarchical clustering and Jaccard similarity index were used to evaluate cross-species cell type similarities. Differential gene expression analysis was performed to compare conserved and divergent transcriptional programs. Statistical analyses included Wilcoxon Rank Sum test, Benjamini-Hochberg correction for multiple comparisons, and permutation testing for CellChat.

This scRNA-seq analysis of six treatment-naïve canine osteosarcoma samples identified 41 distinct cell types, revealing significant heterogeneity within the TME. Key findings include: 1. **Tumor Cell Heterogeneity:** Nine distinct malignant osteoblast populations were identified, exhibiting transcriptional heterogeneity including an interferon (IFN) response signature in a subset of tumor cells. Fibroblasts showed an epithelial-mesenchymal transition (EMT) and angiogenesis signature, suggesting a role in promoting tumor growth. 2. **T Cell Subtypes:** Ten T cell clusters were identified, including a novel CXCL13+ follicular helper CD4 T cell population. Gene signatures were established for cytotoxic CD8 T cells, activated CD4 T cells, regulatory T cells (Tregs), and T cells exhibiting an IFN signature. 3. **Dendritic Cell Subtypes:** Five dendritic cell (DC) subtypes were identified, including mature regulatory DCs (mregDCs), a cell type not previously observed in canine peripheral blood. mregDCs showed enrichment for migration, regulatory, and maturation-associated gene signatures and were predicted to interact with T cells via PD-1/PD-L1 and CTLA4/CD80 interactions. 4. **Macrophage Heterogeneity:** Eight distinct macrophage/monocyte populations were identified, including activated TAMs, intermediate TAMs, and lipid-associated (LA) TAMs. C1QC LA-TAMs exhibited the strongest anti-inflammatory signature, while CD4+ monocytes showed the strongest pro-inflammatory signature. A spectrum of macrophage phenotypes was observed, consistent with human macrophage literature. 5. **Osteoclast Heterogeneity:** Four transcriptomically distinct osteoclast populations were identified, including a CD320+ osteoclast population not previously described in human or mouse tissues. Mature osteoclasts exhibited a gene signature associated with bone resorption and remodeling. 6. **Macrophage Marker Specificity:** Analysis of transcript abundances for widely used macrophage markers (MSR1, AIF1, CD163, and CD68) revealed varying degrees of specificity across myeloid cell types. CD163 showed the highest specificity for macrophages, while CD68 was expressed in TIMs, TAMs, and OCs, with high expression in mature OCs. This finding highlights potential inconsistencies in using these markers to assess macrophage infiltration in OS. 7. **Cell-Cell Interaction Analysis:** CellChat analysis predicted that fibroblasts, mature OCs, and endothelial cells had the strongest interactions. mregDCs and IFN-TAMs had the most interactions, and activated TAMs, IFN-TAMs, and C1QC LA-TAMs were predicted to be key contributors to shaping T cell-mediated immunity. 8. **Cross-Species Comparison:** Integration of the canine dataset with a human OS scRNA-seq dataset revealed a high degree of similarity in cell type gene signatures between species, with lymphocytes showing the strongest conservation. However, differences were observed in pDCs, mast cells, and monocytes.

This study provides the first comprehensive single-cell transcriptomic analysis of the canine OS TME, revealing remarkable cellular and molecular heterogeneity. The identification of numerous rare cell populations, such as mregDCs and IFN-γ-TAMs, opens new avenues for research. The conserved gene signatures between human and canine OS further validates the use of canine OS as a preclinical model for human OS research. The observed heterogeneity within the myeloid compartment, particularly TAMs, offers insights into conflicting findings in the literature regarding the prognostic significance of macrophage infiltration. The identification of potential surface markers for various cell types, based on the surfaceome database, will facilitate further functional studies using flow cytometry and other techniques. Moreover, the transcriptomic signatures presented can be used with deconvolution algorithms to analyze bulk RNA sequencing data from canine OS samples. This comprehensive resource will advance our understanding of OS biology and contribute to the development of novel immunotherapeutic strategies.

This study provides a comprehensive single-cell RNA sequencing atlas of the canine osteosarcoma tumor microenvironment, revealing significant heterogeneity and identifying novel cell populations. The high degree of conservation of cell type gene signatures between canine and human osteosarcoma strengthens the use of canine models for translational research. Future studies should focus on functional validation of identified cell types and their interactions, particularly the role of mregDCs and TAM subtypes in shaping antitumor immunity. Further investigation into the species-specific differences observed in certain cell types, such as mast cells and plasmacytoid dendritic cells, is also warranted.

While this study provides a detailed analysis of the canine OS TME, several limitations should be considered. The relatively small sample size (six dogs) may not fully capture the diversity of the TME. One sample showed a high proportion of neutrophils, potentially due to contamination. The reliance on human gene signatures for cell type annotation may have introduced bias and might have resulted in the misclassification of some canine-specific cell types. Further, the analysis was conducted at the transcript level, and protein expression may not always correlate with transcript abundance.