Machine learning developed an immune evasion signature for predicting prognosis and immunotherapy benefits in lung adenocarcinoma.

Abstract

Background: Lung adenocarcinoma (LUAD) is one of the most common cancers worldwide and a major cause of cancer-related deaths. The advancement of immunotherapy has expanded the treatment options for LUAD. However, the clinical outcomes of LUAD patients have not been as anticipated, potentially due to immune escape mechanisms.

Methods: An integrative machine learning approach, comprising ten methods, was applied to construct an immune escape-related signature (IRS) using the TCGA, GSE72094, GSE68571, GSE68467, GSE50081, GSE42127, GSE37745, GSE31210 and GSE30129 datasets. The relationship between IRS and the tumor immune microenvironment was analyzed through multiple techniques. In vivo experiments were performed to investigate the biological roles of the key gene.

Results: The model developed by Lasso was regarded as the optional IRS, which served as an independent risk factor and had a good performance in predicting the clinical outcome of LUAD patients. Low IRS-based risk score indicated higher level of NK cells, CD8⁺ T cells, and immune activation-related functions. The C-index of IRS was higher than that of many developed signatures for LUAD and clinical stage. Low risk score indicated had a lower tumor escape score, lower TIDE score, higher TMB score and higher CTLA4&PD1 immunophenoscore, suggesting a better immunotherapy response. Knockdown of PVRL1 suppressed tumor cell proliferation and colony formation by regulating PD-L1 expression.

Conclusion: Our study developed a novel IRS for LUAD patients, which served as an indicator for predicting the prognosis and immunotherapy response.