Characterization of Tumor Microenvironment in Lung Adenocarcinoma Identifies Immune Signatures to Predict Clinical Outcomes and Therapeutic Responses

Abstract

Background and objective: Increasing evidence has elucidated the clinicopathological significance of individual TME component in predicting outcomes and immunotherapeutic efficacy in lung adenocarcinoma (LUAD). Therefore, we aimed to investigate whether comprehensive TME-based signatures could predict patient survival and therapeutic responses in LUAD, and to assess the associations among TME signatures, single nucleotide variations and clinicopathological characteristics.

Methods: In this study, we comprehensively estimated the TME infiltration patterns of 493 LUAD patients and systematically correlated the TME phenotypes with genomic characteristics and clinicopathological features of LUADs using two proposed computational algorithms. A TMEscore was then developed based on the TME signature genes, and its prognostic value was validated in different datasets. Bioinformatics analysis was used to evaluate the efficacy of the TMEscore in predicting responses to immunotherapy and chemotherapy.

Results: Three TME subtypes were identified with no prognostic significance exhibited. Among them, naïve B cells accounted for the majority in TMEcluster1, while M2 TAMs and M0 TAMs took the largest proportion in TMEcluster2 and TMEcluster3, respectively. A total of 3395 DEGs among the three TME clusters were determined, among which 217 TME signature genes were identified. Interestingly, these signature genes were mainly involved in T cell activation, lymphocyte proliferation and mononuclear cell proliferation. With somatic variations and tumor mutation burden (TMB) of the LUAD samples characterized, a genomic landscape of the LUADs was thereby established to visualize the relationships among the TMEscore, mutation spectra and clinicopathological profiles. In addition, the TMEscore was identified as not only a prognosticator for long-term survival in different datasets, but also a predictive biomarker for the responses to immune checkpoint blockade (ICB) and chemotherapeutic agents. Furthermore, the TMEscore exhibited greater accuracy than other conventional biomarkers including TMB and microsatellite instability in predicting immunotherapeutic response (p < 0.001).

Conclusion: In conclusion, our present study depicted a comprehensive landscape of the TME signatures in LUADs. Meanwhile, the TMEscore was proved to be a promising predictor of patient survival and therapeutic responses in LUADs, which might be helpful to the future administration of personalized adjuvant therapy.

Keywords: lung adenocarcinoma; signature; survival; therapeutic response; tumor microenvironment.

Figure 1

Characterization of TME in LUADs and distinct patterns of TME subtypes. (A) Cellular interaction of tumor microenvironment (TME) cell types in LUADs; (B) Barplot showing the specific 23 immune fractions represented by various colors in each TMEcluster; (C) Unsupervised clustering of TME cell types and histologic subtypes for LUAD patients.

Figure 2

Construction and validation of the TMEscore in different LUAD datasets. (A) Kaplan-Meier curves of high- and low-TMEscore subgroups in the entire TCGA cohort; (B) Alluvial diagram showing the relationships among TME subtypes and TMEscore subgroups as well as clinical outcomes; (C) Forest plot showing the prognostic value of TMEscore in different datasets.

Figure 3

Mutation profiles of different TMEscore subgroups. (A, B) Waterfall plots exhibiting the mutation profiles of patients with high/low-TMEscore in which various colors with annotations at the bottom represented the different mutation types and tumor mutation burden; (C, D) Boxplots showing the mutation frequency of the 10 most frequently mutated genes in high- and low-TMEscore subgroups.

Figure 4

Visualization of mutational signatures and TME signatures in TME subgroups. (A, B) Mutation spectra showing the 96 substitution classification defined by the substitution class and sequence context immediately 3’ and 5’ to the mutated base in the high- and low-TME subgroups. (C, D) Barplot showing the differential mutation signatures between the high- and low-TME subgroups. The y-axis indicates exposure of 96 trinucleotide motifs to overall signature. The plot title indicates best match against validated COSMIC signatures and cosine similarity value along with the proposed etiology. (E) Unsupervised analysis and hierarchical clustering of the 217 TME signature genes and their associations with clinicopathological characteristics.

Figure 5

Predictive value of TMEscore as a biomarker for immunotherapy. (A, B) Violin plot showing TIDE score and PD-L1 level between the high- and low-TMEscore subgroups; (C, D) Violin plot showing TMEscores in groups with different microsatellite instability (MSI) status and with different TMB levels; (E) ROC curves to compare the accuracy of TMB, MSI and TMEscore in predicting responses to immunotherapy.

Figure 6

Predictive value of TMEscore as a biomarker for chemotherapy. (A) Violin plot comparing IC50 of cisplatin between the high- and low-TMEscore subgroups; (B) Violin plot comparing IC50 of gemcitabine between the high- and low-TMEscore subgroups.