Integrated transcriptomic and deep learning analyses define acRG signature for prognosis and therapeutic options in osteosarcoma

Abstract

Osteosarcoma is a highly malignant bone tumor characterized by significant clinical heterogeneity and poor prognostic outcomes. In this study, we integrated single-cell RNA sequencing data to characterize the landscape of genes related to N⁴-acetylcytidine (ac4C) modification. We calculated an acRG activity score and found that activity was predominantly enriched in osteoblastic cells. Using a machine learning framework, we constructed a six-gene prognostic signature termed acRGBS. This signature served as a robust and independent predictor of overall survival across multiple independent cohorts.

Further analysis revealed that high acRGBS scores were associated with an immunosuppressive tumor microenvironment and increased immune evasion potential. Among the hub genes identified, S100A13 was highlighted as a potential driver of tumor progression. Functional experiments demonstrated that S100A13 knockdown significantly inhibited cell proliferation and reduced xenograft tumor growth in vivo. Mechanistically, S100A13 physically interacts with the mitochondrial calcium uniporter (MCU) to maintain mitochondrial calcium homeostasis. Silencing S100A13 led to mitochondrial calcium overload, which exacerbated oxidative stress and induced cell apoptosis. Finally, we showed that S100A13 knockdown significantly enhanced the anti-tumor efficacy of cisplatin treatment. These findings suggest that S100A13 is a promising therapeutic target and a valuable biomarker for risk stratification in patients with osteosarcoma.