Collateral sensitivities with predictive signatures emerge in a novel model of evolved chemoresistance in osteosarcoma

Osteosarcoma (OS) is the most common primary malignant bone tumor in children and adolescents. Despite aggressive multimodal therapy, including surgery and chemotherapy with methotrexate, doxorubicin, and cisplatin (MAP), outcomes for patients with relapsed or refractory disease remain poor, with no standardized second-line regimen. We developed a clinically calibrated, temporally resolved in vitro model to examine how resistance and collateral drug responses evolve during treatment with methotrexate, doxorubicin, and cisplatin (MAP). We modeled the evolution of chemotherapy resistance in vitro by exposing an OS cell line, MG63.3, to clinically relevant cycles of MAP therapy. MAP resistance and collateral drug responses were quantified over time across five independent evolutionary replicates, and three solvent-treated replicates served as controls. We complemented this phenotypic screening with transcriptomic profiling to extract predictive biomarkers of collateral drug response that could be used to personalize second-line treatment in patients with refractory OS. MAP exposure produced progressive resistance to doxorubicin and methotrexate, whereas cisplatin sensitivity remained comparatively stable. Repeated temporal screening against 12 additional drugs and combinations revealed three broad patterns of collateral response: progressive resistance, progressive sensitivity, and non-linear or stochastic change. Etoposide resistance emerged consistently, while sensitivity developed toward palifosfamide-etoposide and gemcitabine-docetaxel. Transcriptomic profiling showed dynamic, replicate-specific evolutionary trajectories rather than a single uniform resistance state. To demonstrate one approach for how our large, paired dataset could be used to translate these findings into potentially actionable clinical tools, we extracted predictive gene expression signatures of collateral drug response. By applying a dynamic, clinically inspired model of chemotherapy resistance in OS, we generated a detailed temporal mapping of collateral drug responses that highlights potential therapeutic windows and can be used inform selection of second-line agents. This approach and the resulting dataset can ultimately support personalized medicine strategies for patients with relapsed or refractory OS.