The tumor microenvironment (TME) plays an essential role in the occurrence and progression of malignancy. The potential prognostic TME-related biomarkers of lung adenocarcinoma (LUAD) remained unclear, which were investigated in this research. The RNA-sequencing profiles and corresponding clinical parameters were extracted from TCGA and GEO databases, based on which the stromal and immune scores were calculated through the ESTIMATE algorithm. Overlapping differentially expressed genes between stromal and immune score group were analyzed by the LASSO and Random Forrest algorithms and validated in cases from our center. And a prognostic 8-gene signature was constructed using Cox regression. The infiltration of 22 hematopoietic cell phenotypes was assessed by the CIBERSORT algorithms. We found that female, elder patients, and solid predominant subtype had obviously higher stromal and immune scores. And patients with early stage LUAD received a prominently higher immune score. A high stromal or immune score meant a good prognosis. Subsequently, eight TME-related prognostic genes (ATAD5, CYP4F3, CYP4F12, ESPNL, FXYD2, GPX2, NLGN4Y, and SERPINC1) were identified by both LASSO regression and Radom Forest algorithms. High 8-gene signature group exhibited worse overall survival. Furthermore, B cell naïve, plasma cells, T cell follicular helper, and macrophages M1 were prominently more in high signature group. Nevertheless, fewer T cells CD4 memory resting, monocytes, and dendritic cell resting were identified in the high signature group. The composition of the tumor microenvironment significantly affected the prognosis of LUAD patients. We provided a new strategy for the exploration of prognostic TME-related biomarkers and immunotherapy.
Keywords: Lung adenocarcinoma; Machine learning; Prognostic biomarker; Tumor microenvironment.