Abstract IA02: Single-cell regulatory network analysis of therapy-associated cell state remodeling to inform combination strategies in osteosarcoma

Abstract

Intratumor heterogeneity and plasticity contribute to resistance, as coexisting tumor cell states exhibit non-overlapping drug sensitivities and divergent responses to therapy. Combination therapy design thus requires identifying vulnerabilities of treatment-persistent populations and measuring state-specific drug effects obscured by bulk response. We established a single-cell regulatory network framework using ARACNe/VIPER and OncoTarget/OncoTreat to identify combinations targeting Master Regulator (MR) dependencies of complementary cell states, with predicted combinations improving survival in vivo in DMG. Here, we extend this framework to osteosarcoma (OS), a genomically complex bone tumor with few actionable alterations, to target non-genetic mechanisms of treatment persistence contributing to relapse and metastatic progression. VIPER-inferred protein activity from scRNA-seq of 17 OS patient tumors identified four MR-defined states with distinct predicted vulnerabilities coexisting in nearly all tumors: a proliferative osteochondroprogenitor state; an osteogenic state with Notch/Hedgehog/TGFβ signaling; an invasive perivascular mesenchymal state; and a chondrogenic state with hypoxia-associated, inflammatory, and developmental plasticity programs that was enriched in post-chemotherapy tumors. To directly evaluate this remodeling, we performed snRNA-seq of 48 tumors (>600,000 cells; n=4/condition) across four treatment-naïve OS PDX models spanning the clinical spectrum of MAP responses, after vehicle, cisplatin/doxorubicin, or MAP (methotrexate/doxorubicin/cisplatin). PDX tumors recapitulated patient-derived states and the global pattern of proliferative depletion and chondrogenic expansion post-MAP. Within-model comparisons further resolved distinct treatment responses that paralleled clinical behavior: proliferative expansion in a model from a poor responder with early progression (intrinsic resistance); chondrogenic enrichment in two models from patients with early metastatic relapse (treatment persistence); and proliferative depletion without chondrogenic expansion in a model from a long-term survivor. We evaluated drugs predicted to invert the MR activity of individual states in an independent MAP-refractory OS PDX. Monotherapy did not improve tumor volume or survival, yet single-cell profiling validated their predicted complementary state-specific effects: ifosfamide depleted proliferative/osteogenic populations while enriching mesenchymal/chondrogenic states, whereas dinaciclib and valemetostat selectively depleted a replicative population bridging proliferative and chondrogenic states, limiting chondrogenic adaptation and nominating combinations under investigation. In summary, conserved OS regulatory states with distinct vulnerabilities are dynamically remodeled by treatment, and single-cell profiling reveals drug effects missed by bulk efficacy measures. Integrating longitudinal profiling with MR-based drug nomination can identify recurrent resistance mechanisms and inform evolution-guided combination strategies and trial design.