An Auto-RS Signature for Prognostic Stratification and Drug Sensitivity Prediction in Osteosarcoma.
AI interpretation is pending for this paper.
Open original publication →What the AI sees
Not AI summarized yet.
Research significance
Pending deeper interpretation.
Source abstract
BACKGROUND: Metastasis and poor chemotherapy response have stagnated therapeutic progress in osteosarcoma (OS) for the past three decades. Defining the transition from localized to metastatic OS before overt dissemination is fundamental for improving survival. However, effective early diagnostic tools remain scarce, largely due to limited exploitation of the metastasis-associated tumor microenvironment's own record of prior environmental and stress exposures encoded in cell-intrinsic transcriptional states. Here, we employed a supervised machine learning framework with iterative resampling and multi-stage model selection to identify molecular markers associated with metastasis in osteosarcoma and to develop a computational signature, Auto-RS. METHODS: Transcriptomic and clinical data from 139 OS patients with ≥5 years of follow-up were analyzed. A LASSO-Cox framework was applied to derive a gene expression-based risk score, Auto-RS, from which a nomogram integrating age and sex was generated for individualized prognosis. Model interpretability was assessed across six independent single-cell OS patient datasets, and drug sensitivity predictions were inferred by integrating Auto-RS with the Precily algorithm to uncover actionable therapeutic vulnerabilities. RESULTS: Auto-RS, constructed from the expression of four autophagy genes (BNIP3, MYC, PEA15, and SAR1A), served as an independent prognostic factor for overall survival (HR = 1.091; 95% CI, 1.047-1.136; p < 0.001). Time-dependent ROC analysis showed that Auto-RS was the most accurate single predictor (AUC = 0.88), exceeding metastasis (0.83), sex (0.45), and age (0.39). A basic prognostic model (BpM) incorporating metastasis status yielded a C-index of 0.741 (95% CI, 0.679-0.803). The addition of Auto-RS (CpM) improved discrimination (C-index = 0.788; 95% CI, 0.731-0.845), whereas a model without metastasis information (ApM) retained predictive ability (C-index = 0.709; 95% CI, 0.640-0.778). Single-cell analysis confirmed that Auto-RS features aligned with known metastatic trajectories, reflecting the transition from proliferative to invasive tumor states and highlighting coordinated programs among cancer-associated fibroblasts and immune cells. Drug sensitivity integration through Precily identified gemcitabine and cytarabine as FDA-approved agents predicted in silico to show greater sensitivity in the high-risk subgroup. CONCLUSIONS: We identified autophagy-mediated transcriptional 'stress fingerprints' that are tightly associated with OS metastasis. The Auto-RS signature, composed of BNIP3, MYC, PEA15, and SAR1A, enables early therapeutic stratification of patients independent of overt metastatic status. Moreover, Auto-RS delineates key molecular underpinnings of OS metastasis at single-cell resolution. As a practical laboratory tool, Auto-RS may represent a step toward improved risk stratification, where advances in metastasis prediction and therapeutic guidance converge to improve outcomes in OS.