Extracellular vesicles proteomics-based machine-learning model predicts immunotherapy response in NSCLC

Lung cancer remains the most common cause of cancer mortality worldwide. While the introduction of immune checkpoint inhibitors has changed the treatment landscape in non-small cell lung cancer, primary and acquired resistance continue to limits their long-term efficacy. The identification of reliable, minimally invasive biomarkers to guide immunotherapy response remains an urgent clinical need. In this study, we performed an in-depth proteomic profiling of plasma-derived small extracellular vesicles collected prior to treatment in 65 non-small cell lung cancer patients receiving pembrolizumab, aiming to uncover predictive molecular signatures. Mass spectrometry analysis initially detected over two thousand plasma extracellular vesicles proteins, including reported extracellular vesicle proteins, thereby validating our isolation and detection methodology. Among these, a four-proteins panel (MUC1, MUC5B, MUC5AC, and ANPEP) was associated with T cell dysfunction and an immunosuppressive tumor phenotype. This signature correlated with systemic inflammatory markers, particularly the platelet-to-lymphocyte ratio, and accurately predicted immunotherapy outcomes. Based on these findings, we stablished a plasma-derived score that strongly correlated with both progression-free and overall survival, providing a promising non-invasive tool for patient stratification and personalized immunotherapy.