Identification and validation for biological characteristics and prognosis of tumor microenvironment subtypes in patients with lung adenocarcinoma.

Abstract

e21006 Background: Immunotherapy has brought a great revolution to treatment of patients with lung adenocarcinoma (LUAD). However, the infiltrating cells in the tumor microenvironment (TME) could influence immunotherapy outcomes. This study aimed to investigate the TME subtypes based on 50 immune, stromal, and stem cell types in TME. Methods: The xCell algorithm was used to infer the enrichment scores in TME. We performed the k-mean consensus clustering to identify TME subtypes for LUAD patients. The ESTIMATE, ABSOLUTE, LUMP, and CPE algorithms were used to compare tumor purity and immune infiltration of TME subtypes. The genomic mutations and copy number variants were also calculated by “maftools” R package and GISTIC 2.0. We established a neural network model, based on the TCGA dataset and verified it in GEO datasets by the Kaplan-Meier curve and Cox regression analyses, to prove the stability of TME subtypes. Using prophetic algorithm, ridge regression model was established to predict the sensitivity of drugs in all datasets. In addition, TIDE score and submap analyses were used to speculate the potential response of patients in immunotherapy. Results: In our study, 50 types of cells, including immune (including lymphocyte and myeloid cells), stem, and stromal cells in TME, were included for analysis. Based on cell scores, 504 LUAD patients were divided into four different subtypes (C1, C2, C3 and C4). In univariable Cox regression, the hazard ratio (HR) value of C2 vs. C1 was 2.066 (95%CI: 1.313 - 3.250), the HR value of C3 vs. C1 was 1.821 (95%CI: 1.067- 3.108), and that of C4 vs. C1 was 2.313 (95%CI: 1.461 - 3.662). Then, a neural network model was established (the area under the curves in the testing cohort: 0.911) to validate our conclusions in external sets and similar results were observed. Through pseudotime trajectory, we discovered there were potential connections among clusters from C1 to C4. The C2 and C3 were the intermediate forms of C1 and C4, and thus, the survival, immunological, and metabolic status of C2 and C3 may be intermediate between C1 and C4. By comparing the cell infiltration and immune gene expression of the four clusters, we found that C1 and C3 may have a similar immune phenotype which was opposed to that of C2 and C4. Meanwhile, the enrichment scores of cell proliferation-related pathways were progressively increased in the four clusters, which may be related to survival outcomes of them. We noticed that C1 harbored a unique EGFR mutation compared to other clusters, which may be sensitive to erlotinib. Patients of C4 had lower TIDE scores, which indicated that these patients may get higher benefits from immunotherapy. Conclusions: In conclusion, we identified four TME subtypes with distinct clinical outcomes and biological features, including tumor immunity, TME status, tumor mutations, and responses to chemotherapy and immunotherapy.