A Long Non-coding RNA Signature to Improve Prognostic Prediction of Pancreatic Ductal Adenocarcinoma

Abstract

Background: Pancreatic ductal adenocarcinoma (PDAC) remains one of the most aggressive solid malignant tumors worldwide. Increasing investigations demonstrate that long non-coding RNAs (lncRNAs) expression is abnormally dysregulated in cancers. It is crucial to identify and predict the prognosis of patients for the selection of further therapeutic treatment. Methods: PDAC lncRNAs abundance profiles were used to establish a signature that could better predict the prognosis of PDAC patients. The least absolute shrinkage and selection operator (LASSO) Cox regression model was applied to establish a multi-lncRNA signature in the TCGA training cohort (N = 107). The signature was then validated in a TCGA validation cohort (N = 70) and another independent Fudan cohort (N = 46). Results: A five-lncRNA signature was constructed and it was significantly related to the overall survival (OS), either in the training or validation cohorts. Through the subgroup and Cox regression analyses, the signature was proven to be independent of other clinic-pathologic parameters. Receiver operating characteristic curve (ROC) analysis also indicated that our signature had a better predictive capacity of PDAC prognosis. Furthermore, ClueGO and CluePedia analyses showed that a number of cancer-related and drug response pathways were enriched in high risk groups. Conclusions: Identifying the five-lncRNA signature (RP11-159F24.5, RP11-744N12.2, RP11-388M20.1, RP11-356C4.5, CTC-459F4.9) may provide insight into personalized prognosis prediction and new therapies for PDAC patients.

Keywords: lncRNA; overall survival; pancreatic ductal adenocarcinoma; prognosis; signature.

Figure 1

Flow chart of the study. The study was carried out in TCGA and Fudan lncRNA dataset of PDAC patients. The TCGA training cohort was used to identify prognostic lncRNAs. The LASSO regression model was used to establish a prognostic signature based on the prognostic lncRNAs. The prognosis analysis was validated in the TCGA and Fudan validation cohort, respectively.

Figure 2

Kaplan–Meier analyses of the overall survival (OS) based on the 5-lncRNA signature. (A) TCGA training cohort (N = 107); (B) TCGA validation cohort (N = 70); (C) Entire TCGA cohort (combined training and validation patients, N = 177); (D) Fudan validation cohort (N = 46). The tick marks on the Kaplan–Meier curves represent the censored subjects. The differences between the two curves were determined by the two-side log-rank test. The number of patients at risk is listed below the survival curves.

Figure 3

Forest plot summary of analyses of overall survival (OS). Univariate and multivariate analyses based on the 5-lncRNA signature and clinical covariates in the entire TCGA cohort (A,B) and Fudan validation cohort (C,D). The blue solid squares represent the hazard ratio (HR), and the red transverse lines represent 95% confidence intervals (CI). All P-values were calculated using Cox regression hazards analysis.

Figure 4

Kaplan–Meier survival analysis to assess the independence of the 5-lncRNA signature from the TNM stage, histological grade, and MSS status. The patients from the entire TCGA were stratified into subgroups. The 5-lncRNA signature was applied to the TNM stage II and III patients (A), histological grade I&II patients (B), histological grade III&IV patients (C), MSS status patients (D), separately. The number of patients at risk is listed below the survival curves. The tick marks on the Kaplan–Meier curves represents the censored subjects. Two-sided log-rank test was adopted to determine the differences between the two curves.

Figure 5

Receiver operating characteristic (ROC) analysis of the sensitivity and specificity of the overall survival (OS) prediction by the 5-lncRNA risk score, histologic grade, TNM stage and all combined risk factors in the entire TCGA cohort (A; N = 177) and the Fudan validation cohort (B; N = 46). As shown, the 5-lncRNA risk score combined with other factors shows a better prediction of OS either in the TCGA cohort or Fudan validation cohort.

Figure 6

Enriched functions and pathways of the top 1,000 significantly differentially expressed genes (DEGs) in high vs. low risk PDAC patients in the TCGA dataset. The interaction network was generated with the Cytoscape plug-in ClueGO and CluePedia. Functions and pathways of up-regulated DEGs (A), down-regulated DEGs (B). The size of the nodes shows the term significance after Bonferroni correction. The significant term of each group is highlighted.