Functional autophagy gene set signature and state classification reveal a link between autophagy induction, lysosomal activity, and poor prognosis in glioblastoma

Autophagy is an essential mechanism for maintaining cell homeostasis and, when dysregulated, is related to various pathologies. In cancer, it functions as a double-edged sword, either as a tumor suppressor or as a tumor promoter. Elevated GFP-LC3 puncta and increased levels of autophagy-related proteins, commonly interpreted as markers of high autophagic activity, are often associated with poor prognosis in tumors. However, current assessments of autophagy largely rely on isolated cellular and protein-level observations. While transcriptomic approaches offer broader insights, they often utilize large gene signatures without adequately accounting for the dynamic nature of autophagic flux. In this study, we propose a functional gene set signature for autophagy based on the rational mapping of flux stages, enabling the inference of distinct autophagy states. Using a supervised machine learning approach, we developed two indices - AutoIndex and LysoIndex - that effectively distinguish between cells treated with autophagy inducers and inhibitors. Applying this framework, we analyzed autophagy states across four cancer types, comparing tumor samples to matched normal tissues and evaluating their association with patient survival. Among these, glioblastoma (GBM) exhibited the strongest correlation between inferred autophagy states and clinical outcomes. GBM samples characterized by high LysoIndex and low AutoIndex--corresponding to a transcriptomic induction state--were linked to significantly poorer survival. Compared to lower-grade gliomas, GBM showed elevated lysosomal activity, which may contribute to its more aggressive behavior. Additionally, GBM cell lines treated with temozolomide displayed signatures consistent with autophagy induction, supporting our models predictive relevance. These results reinforce the understanding and monitoring of autophagy, and suggest that, in the case of GBM, its negative influence may represent a potential target for novel therapeutic interventions.