Integrate analysis of bulk and single-cell RNA sequencing demonstrates a transcriptional pattern characterized by GDF15 of tumor cells that predicts immunotherapy efficacy in non-small cell lung cancer

The advent of immunotherapy has transformed the landscape of lung cancer treatment. Nevertheless, the question of which populations may benefit from this approach remains unsolved. In this study, we designed a pipeline based on machine learning for processing the RNA-sequencing data from lung cancer patients treated with immune check point blockade therapy to identify the most important genes that predict the prognosis. The final model was developed by accelerated oblique random forests (AORSF) for its best performance on the training, test and 10-cross validation set. An intriguing phenomenon revealed by single-cell RNA sequencing data was that the prognostically unfavorable genes were predominantly expressed by a specific tumor cell that was characterized by GDF15, while CXCL9-positive macrophages expressed the most favorable genes. The specific tumor cell with the highest score of unfavorable genes, as calculated by the AUCell package, not only exhibited the feature of epithelial cell migration but also possessed a transcription factor indicating proliferation and the highest potency score of differentiation. Furthermore, the higher level of expression of GDF15 and the proportion of this specific tumor cell can both predict a worse overall survival in an external validation melanoma cohort treated with immune checkpoint blockade therapy. In conclusion, our study identified a specific tumor cell and its hub genes that affect the efficacy of immunotherapy and may represent a target for improving the outcomes of patients.