Integrative Single-cell and Spatial Transcriptomic Analysis of Osteosarcoma Reveals Conserved and Distinct Ecosystems...

Osteosarcoma is a heterogeneous malignancy, exhibiting significant variability among patients, individual cancer cells within a tumor, and the stromal cells that compose primary and metastatic lesions. To facilitate the study of this complex disease, we compiled a unique cross-species single-cell transcriptomic dataset totaling over a million cells/nuclei from human specimens, canine specimens, patient-derived xenografts/PDX, and syngeneic mouse models at both primary (bone) and metastatic (lung) sites. Using a rigorous process for multi-species alignment and annotation, we identified six conserved tumor cell transcriptional states organized along hierarchical differentiation trajectories from progenitor to differentiated phenotypes. Parallel analysis of tumor-associated cells identified conserved macrophage, fibroblast, and endothelial populations that exhibit species- and site-specific reprogramming. Validation by mapping cell types using spatial transcriptomics revealed structured neighborhood architectures that were reproduced across multiple samples. Cell-cell interaction analysis revealed similarities and differences in tumor-host networks across primary and metastatic sites and across species. This analysis enabled pathway-specific assessment of tumor-host communication fidelity across osteosarcoma model systems relative to humans, revealing canine osteosarcoma as a more faithful model. Metastatic lung lesions, counterintuitively, exhibited more intense and complex extracellular matrix (ECM) signaling than primary bone tumors. A key example was tumor-derived fibronectin (FN1), which engages integrin and syndecan receptors on lung epithelial cells, driving a pathological mesenchymal and profibrotic state that promotes fibrotic niche formation and metastatic lung colonization. Together, this cross-species resource delineates both conserved and divergent tumor microenvironment programs, demonstrates how model-aware analyses uncover previously unrecognized tumor-host interactions, and underscores the need for therapies that co-target tumor heterogeneity and its supportive metastatic niche.