Immune-Related lncRNA Pairs as Prognostic Signature and Immune-Landscape Predictor in Lung Adenocarcinoma

Abstract

Background: Suppressive tumor microenvironment is closely related to the progression and poor prognosis of lung adenocarcinoma (LUAD). Novel individual and universal immune-related biomarkers to predict the prognosis and immune landscape of LUAD patients are urgently needed. Two-gene pairing patterns could integrate and utilize various gene expression data.

Methods: The RNA-seq and relevant clinicopathological data of the LUAD project from the TCGA and well-known immune-related genes list from the ImmPort database were obtained. Co-expression analysis followed by an analysis of variance was performed to identify differentially expressed immune-related lncRNA (irlncRNA) (DEirlncRNA) between tumor and normal tissues. Two arbitrary DEirlncRNAs (DEirlncRNAs pair) in a tumor sample underwent pairwise comparison to generate a score (0 or 1). Next, Univariate analysis, Lasso regression and Multivariate analysis were used to screen survival-related DEirlncRNAs pairs and construct a prognostic model. The Acak information standard (AIC) values of the receiver operating characteristic (ROC) curve for 3 years are calculated to determine the cut-off point for high- or low-risk score. Finally, we evaluated the relationship between the risk score and overall survival, clinicopathological features, immune landscape, and chemotherapy efficacy.

Results: Data of 54 normal and 497 tumor samples of LUAD were enrolled. After a strict screening process, 15 survival-independent-related DEirlncRNA pairs were integrated to construct a prognostic model. The AUC value of the 3-year ROC curve was 0.828. Kaplan-Meier analysis showed that patients with low risk lived longer than patients with high risk (p <0.001). Univariate and Multivariate Cox analysis suggested that the risk score was an independent factor of survival. The risk score was negatively associated with most tumor-infiltrating immune cells, immune score, and microenvironment scores. The low-risk group was correlated with increased expression of ICOS. The high-risk group had a connection with lower half inhibitory centration (IC50) of most chemotherapy drugs (e.g., etoposide, paclitaxel, vinorelbine, gemcitabine, and docetaxel) and targeted medicine-erlotinib, but with higher IC50 of methotrexate.

Conclusion: The established irlncRNA pairs-based model is a promising prognostic signature for LUAD patients. Furthermore, the prognostic signature has great potential in the evaluation of tumor immune landscape and guiding individualized treatment regimens.

Keywords: drug sensitivity; immune landscape; immune-related lncRNA pair; lung adenocarcinoma; signature.

Figure 1

Flow chart of the study.

Figure 2

Construction of a prognostic model using DEirlncRNA Pairs. The heatmap (A) and volcano plot (B) of identified differentially expressed immune-related lncRNAs (DEirlncRNAs). (C) A forest map showed 12 DEirlncRNA pairs identified by Multivariate Cox proportional hazard regression in the stepwise method.

Figure 3

Validation of the prognostic model. (A) The 1-, 3-, and 5-year ROC of the model suggested that all AUC values were over 0.77. (B) A comparison of ROC curves of 3-year with other common clinical factors showed the superiority of the riskscore. (C) Riskscore for 457 patients with LUAD; the maximum inflection point is the cut-off point obtained by the AIC. Risk scores (D) and survival outcome (E) of each case are shown. (F) Patients in the low-risk group experienced a longer survival time tested by the Kaplan–Meier test.

Figure 4

Correlation between riskscore and the clinical variables. (A) A strip illustration and scatter drawing showed that the (B) T stage, (C) N stage and (D) clinical stage were significantly related to the riskscore. (E) Univariate Cox hazard ratio analysis and (F) Multivariate Cox regression analysis of riskscore and other common clinical factors. ***p < 0.001.

Figure 5

Estimation of tumor-infiltrating cells by the prognostic model. Comparison of composition of (A–H) immune cells, namely, (A) CD8⁺ T cell, (B) CD4⁺ effector memory T cell, (C) NKT cells, (D) B cell, (E) macrophage, (F) myeloid dendritic cell, (G) granulocyte–monocyte progenitor cell, and (H) mast cell and (I) cancer associated fibroblast cell between the high risk and low-risk group. (J–L) Comparison of (J) stromal score, (K) immune score, and (L) microenvironment score between the high risk and low-risk groups.

Figure 6

Analysis of immune landscape between the high-risk and low-risk groups. (A) Overview of association among riskscore and immune cells and stromal cells shown by Spearman correlation analysis. (B–E) Comparison of expression level of (B) ICOS, (C) CTLA4, (D) CD274, and (E) PDCD1 levels. (F–L) Prediction of drug sensitivity (IC50) for chemotherapeutics such as (F) etoposide, (G) paclitaxel (H) vinorelbine, (I) gemcitabine, (J) docetaxel, (K) methotrexate, and (L) targeted therapy—erlotinib. **p < 0.01; ns, not significant.