Development of a 5-Gene Signature to Evaluate Lung Adenocarcinoma Prognosis Based on the Features of Cancer Stem Cells

Abstract

Cancer stem cells (CSCs) can induce recurrence and chemotherapy resistance of lung adenocarcinoma (LUAD). Reliable markers identified based on CSC characteristic of LUAD may improve patients' chemotherapy response and prognosis. OCLR was used to calculate mRNA expression-based stemness index (mRNAsi) of LUAD patients' data in TCGA. Association analysis of mRNAsi was performed with clinical features, somatic mutation, and tumor immunity. A prognostic prediction model was established with LASSO Cox regression. Kaplan-Meier Plotter (KM-plotter) and time-dependent ROC were applied to assess signature performance. For LUAD, univariate and multivariate Cox analysis was performed to identify independent prognostic factors. LUAD tissues showed a noticeably higher mRNAsi in than nontumor tissues, and it showed significant differences in T, N, M, AJCC stages, and smoking history. The most frequently mutated gene was TP53, with a higher mRNAsi relating to more frequent mutation of TP53. The mRNAsi was significantly negatively correlated with immune score, stromal score, and ESTIMATE score in LUAD. The blue module was associated with mRNAsi. The 5-gene signature was confirmed as an independent indicator of LUAD prognosis that could promote personalized treatment of LUAD and accurately predict overall survival (OS) of LUAD patients.

Figure 1

Flowchart of research design.

Figure 2

mRNAsi and clinical characteristics of LUAD. (a) mRNAsi differences between neoplastic and normal tissues. (b) Differences in mRNAsi between female and male LUAD patients. (c) mRNAsi difference in LUAD patients with ge > 65 and age ≤ 65. (d) mRNAsi differences between LUAD patients at different T stages. (e) mRNAsi differences between LUAD patients at different N stage. (f) mRNAsi analysis of M1 stage, M2 stage, and M3 stage patients. (g) mRNAsi differences among the four AJCC stage. (h) Differences in mRNAsi among grouped patients according to smoking.1 stands for life-long nonsmokers (fewer than 100 cigarettes during lifetime), 2 stands for current smokers (includes daily smokers and nondaily smokers or occasional smokers), 3 stands for current reformed smokers for >15 years, 4 stands for current reformed smokers for ≤15 years, and 5 stands for current reformed smokers; duration not specified = 5.

Figure 3

Associations of mRNAsi with mutations. (a) An overview of the mutant map in LUAD; only the 10 genes with the highest mutation frequency are shown here. (b) mRNAsi differences between TP53 mutant (MT) and TP53 wild-type (WT) samples. (c) Differences in mRNAsi between TTN mutant LUAD samples and TTN wild-type LUAD samples. (d) mRNAsi differences between MU16 mutant LUAD samples and MU16 wild-type LUAD samples. (e) mRNAsi difference in patients with CSMD3 mutant and CSMD3 wild-type. (f) mRNAsi was compared between RYR2 mutant LUAD and RYR2 wild-type LUAD patients. (g) mRNAsi in LUAD patients with mutant LRP1B was compared with that in LUAD patients without mutant LRP1B. (h) Difference analysis was used to compare the difference in mRNAsi between samples with and without mutations in USH2A. (i) Violin plots showed mRNAsi between LUAD samples with and without LRP1B mutations. (j) Differences between XIRP2 wild-type samples and XIRP2 mutant samples Violin diagram of mRNAsi.

Figure 4

Clinical data and mutation trends for LUAD samples with different mRNAsi.

Figure 5

Correlation of mRNAsi with molecular subtypes and tumor immunity. (a) mRNAsi differences between LUAD samples classified according to CIMP. (b) mRNAsi differences among molecular subtypes identified by iCluster. (c) Correlativity between mRNAsi and stromal score of LUAD samples in TCGA. (d) Pertinent analysis between mRNAsi and immune score of LUAD samples in TCGA. (e) Correlation analysis between mRNAsi and ESTIMATE score of LUAD samples in TCGA.

Figure 6

Identification of gene module associated with mRNAsi. (a) Analysis of network topology for various soft-thresholding powers. (b) Hierarchical clustering tree bases on the topological overlap dissimilarity. (c) Correlation between 14 gene modules and gender, age, T stage, N stage, M stage, AJCC stage, smoking, and mRNAsi. (d) Go analysis of the blue modules. (e) KEGG analysis of blue module.

Figure 7

Construction of 5-gene signature on account of mRNAsi-related genes. (a) In TCGA training set, distribution of the risk score, survival data, and the mRNA expression of prognosis signature. (b) Survival curves of LUAD patients in a TCGA training set. (c) ROC analysis for OS prediction in TCGA training set.

Figure 8

Internal and external verification of 5-gene signature. (a) Distribution of the risk score, survival data, and the mRNA expression of prognosis signature in different cohorts. (b) Survival curves of patients with LUAD in different cohorts in the high-risk and low-risk groups. (c) Time-dependent ROC analysis for OS prediction in four cohorts.

Figure 9

Correlation between risk score and each clinicopathologic feature. A t-test or one-way ANOVA determined the correlation between risk score and age (a), gender (b), T stage (c), N (d), M stage (e), AJCC stage (f), and smoking (g), respectively.

Figure 10

Kaplan-Meier stratification survival analyses in TCGA-LUAD data set, including age ≥ 65 (a), age ≤ 65 (b), male (c), female (d), T1 (e), T2 (f), T3-T4 (g), M0 (h), N0-N1 (i), N2-N3 (j), stage I-II (k), and stage III-IV (l).

Figure 11

Five-gene signature outperformed the other three signatures in predicting the performance of the OS. (a) Kaplan-Meier curve of prognosis in patients with TCGA-LUAD predicted by 8-gene signature. (b) ROC curve of the 8-gene signature for 1-, 3-, and 5-year OS. (c) Kaplan-Meier curve of prognosis in patients with TCGA-LUAD predicted by 3-gene signature. (d) ROC curve of the 3-gene signature for 1-, 3-, and 5-year OS. (e) Kaplan-Meier curve of 3-gene signature developed by Cheng Yue et al. for predicting prognosis of patients with TCGA-LUAD.(f) ROC curve of the 3-gene signature developed by Yue et al. for 1-, 3-, and 5-year OS. (g) Kaplan-Meier curve for predicting the OS of TCGA-LUAD patients based on the 6-gene risk model developed by Wang et al. (h) ROC curve analysis showing the prognostic prediction efficiency of the risk model. (i) Kaplan-Meier curve for predicting the OS of TCGA-LUAD patients based on the 7-gene risk model developed by Al-Dherasi et al. (j) ROC curve analysis showing the prognostic prediction efficiency of the 7-gene risk model.