Lineage-associated tumor cell states predict outcome and guide immunotherapeutic target selection in pediatric osteosarcoma

Abstract

Osteosarcoma is a highly heterogeneous primary bone malignancy that predominantly affects adolescents and young adults, and robust prognostic biomarkers and effective targeted therapies remain lacking. While single-cell transcriptomics has begun to resolve tumor composition, a tumor cell-intrinsic framework linking transcriptional states to clinical outcome and therapeutic targeting is not established. Here, we present a single-cell transcriptomic analysis of pediatric osteosarcoma, integrating 22 samples across 18 patients spanning six diagnostic, six post-treatment, and ten metastatic lesions profiled using complementary sequencing platforms. Using non-negative matrix factorization, we identify recurrent lineage-associated transcriptional programs that define osteoblastic-like, chondroblastic-like, and fibroblastic-like tumor cell states. These programs show overlapping activity across tumor cells, indicating that tumor cell states are not strictly discrete. Strikingly, combined osteoblastic-like and chondroblastic-like program activity is associated with poorer overall survival, a finding validated across two independent bulk RNA-sequencing cohorts. These tumor cell states are further linked to distinct immune microenvironmental compositions. Mapping candidate immunotherapeutic targets at single-cell resolution reveals variable expression across tumor cell states, supporting state-informed strategies for CAR T cell targeting. Together, our findings establish a tumor cell state framework that links osteosarcoma heterogeneity to clinical outcome and provides a basis for patient stratification and the development of state-informed therapeutic strategies in pediatric osteosarcoma.