Construction and Validation of a Tumor Microenvironment-Based Scoring System to Evaluate Prognosis and Response to Immune Checkpoint Inhibitor Therapy in Lung Adenocarcinoma Patients

Abstract

Background: Lung cancer is among the most dangerous malignant tumors to human health. Lung adenocarcinoma (LUAD) accounts for about 40% of all lung cancers. Accumulating evidence suggests that the tumor microenvironment (TME) is a crucial regulator of carcinogenesis and therapeutic efficacy in LUAD. However, the impact of tumor microenvironment-related signatures (TMERSs) representing the TME characteristics on the prognosis and therapeutic outcome of LUAD patients remains to be further explored.

Materials and methods: Gene expression files and clinical information of 1630 LUAD samples and 275 samples with immunotherapy information from different databases such as The Cancer Genome Atlas (TCGA), Gene Expression Omnibus (GEO) and Cancer Research Institute (CRI) iAtlas were downloaded and analyzed. Three hundred tumor microenvironment-related signatures (TMERS) based on a comprehensive collection of marker genes were quantified by single sample gene set enrichment analysis (ssGSEA), and then eight significant signatures were selected to construct the tumor microenvironment-related signature score (TMERSscore) by performing Least Absolute Shrinkage and Selection Operator (LASSO)-Cox analysis.

Results: In this study, we constructed a TME-based prognostic stratification model for patients with LUAD and validated it in several external datasets. Furthermore, the TMERSscore was found to be positively correlated with tumor malignancy and a high TMERSscore predicted a poor prognosis. Moreover, the TMERSscore of responders treated with Immune Checkpoint Inhibitor (ICI) therapies was significantly lower than that of non-responders, and the TMERSscore was positively correlated with the tumor immune dysfunction and exclusion (TIDE) score, implying that a low TMERSscore predicts a better response to ICI treatment and may provide independent and incremental predictive value over current biomarkers.

Conclusions: Overall, we constructed a TMERSscore that can be used for LUAD patient prognosis stratification as well as ICI therapeutic efficacy evaluation, supportive results from independent external validation sets showed its robustness and effectiveness.

Keywords: TIDE score; immune checkpoint inhibitor therapeutic response evaluation; lung adenocarcinoma; prognostic stratification; tumor microenvironment.

Figure 1

Flowchart of our research process.

Figure 2

Establishment of TMERSscore. (A) LASSO coefficient profiles of the 18 selected TMERSs. (B) Adjusting LASSO penalty parameters using 10-fold cross-validation. (C) Multivariate Cox regression analysis of the eight selected TMERSs for OS. (D) Kaplan-Meier survival curve of the TMERSscore in the training cohort. (E) ROC curves of TMERSscore predicting OS. *** p < 0.001.

Figure 3

Validation of the performance of TMERSscore to predict survival. (A,C,E,G) are the Kaplan-Meier survival curves of TMERSscore in GSE30219, GSE31210, GSE68465, and GSE72094, respectively. (B,D,F,H) are the ROC curves of TMERSscore predicting the OS at 1, 3, 5 years in GSE30219, GSE31210, GSE68465, and GSE72094 dataset. (I) DCA curves compared the clinical benefit of TMERSscore with other risk factors at 3 years. The none plot represents the assumption that no patients have 3-year survival, while all plots represent the assumption that all patients have 3-year survival at a specific threshold probability. The x-axis represents the threshold probabilities, and the y-axis measures the net benefit.

Figure 4

Expression patterns of DEGs contained in the eight selected TMERSs in different subgroups. (A,B) are volcano plots showing the differential expression status of DEGs in tumors compared to normal and in the high-risk group compared to the low-risk group, respectively. (C) Heatmap of the expression of DEGs between different subgroups. (D) Correlation plot showing the correlation between TMERSscore and the expression of DEGs.

Figure 5

Enrichment analysis results relevant to TMERSscore. (A) Gene ontology analyses of the DEGs. (B) KEGG analyses of the pathways of the DEGs. (C) GSEA of TMERSscore demonstrates the activation pathways of different subgroups.

Figure 6

Evaluation of tumor malignancy by TMERSscore. (A) Pairwise boxplot of paired normal and tumor samples in the GSE32863 dataset. (B) Pairwise boxplot of paired normal and tumor samples in the TCGA cohort. (C,E–G) Boxplots showing the difference between the TMERSscore at different AJCC pTNM Stage, T stage, N stage, and M stage, respectively. (D) Sankey diagram showing the relationship between high and low TMERSscore subgroups and AJCC Stages.

Figure 7

Evaluation of ICI therapeutic efficacy of LUAD patients by TMERSscore. (A) Boxplot demonstrating the difference in TMERSscore between responder and non-responder LUAD patients in the GSE126044 dataset receiving anti-PD-1 treatment. (B) Bar graph illustrating the percentage of clinical response to anti-PD-1 immunotherapy in the high and low TMERSscore groups for the GSE126044 dataset. (C) Correlations between TMERSscore values, TMB, PD-1, PD-L1, CTLA-4, CD80, CD86, TIDE score, IFNG, MSI.Expr.Sig, Merck18, Dysfunction, Exclusion, MDSC, CAF, TAM.M2 scores in the TCGA cohort. Different correlations between two signatures are indicated by different colors. * p < 0.05, ** p < 0.01, *** p < 0.001.

Figure 8

Evaluation of ICI therapeutic efficacy by TMERSscore in SKCM cohort. (A,B) Boxplots showing the difference in TMERSscore between responders and non-responders, and between CR/PR subgroups and PD/SD subgroups in the SKCM cohort receiving ICI treatment, respectively. (C) Kaplan-Meier survival curve of the TMERSscore in ICI-treated SKCM cohort. (D,F) Boxplots showing the difference between responders and non-responders of TMERSscore in the SKCM cohort for different ICI therapies, i.e., anti-CTLA4, anti-PD-1, anti-PD-1 combined anti-CTLA4, and between the CR/PR subgroup and the PD/SD subgroup, respectively. (E,G) Bar charts showing the percentage of responders vs. non-responders and CR/PR vs. PD/SD in different TMERSscore subgroups in the SKCM cohort, respectively. Responders include CR, PR, and PD response subgroups, and non-responders include SD response subgroups.