Construction and validation of an autophagy-related long noncoding RNA signature for prognosis prediction in kidney renal clear cell carcinoma patients

Abstract

Purpose: The purpose of this study was to identify autophagy-associated long noncoding RNAs (ARlncRNAs) using the kidney renal clear cell carcinoma (KIRC) patient data from The Cancer Genome Atlas (TCGA) database and to construct a prognostic risk-related ARlncRNAs signature to accurately predict the prognosis of KIRC patients.

Methods: The KIRC patient data were originated from TCGA database and were classified into a training set and testing set. Seven prognostic risk-related ARlncRNAs, identified using univariate, lasso, and multivariate Cox regression analysis, were used to construct prognostic risk-related signatures. Kaplan-Meier curves and receiver operating characteristic (ROC) curves as well as independent prognostic factor analysis and correlation analysis with clinical characteristics were utilized to evaluate and verify the specificity and sensitivity of the signature in training set and testing set, respectively. Two nomograms were established to predict the probable 1-, 3-, and 5-year survival of the KIRC patients. In addition, the lncRNA-mRNA co-expression network was constructed and Gene Ontology (GO) enrichment analysis and Kyoto Encyclopedia of Genes and Genomes (KEGG) were used to identify biological functions of ARlncRNAs.

Results: We constructed and verified a prognostic risk-related ARlncRNAs signature in training set and testing set, respectively. We found the survival time of KIRC patients with low-risk scores was significantly better than those with high-risk scores in training set and testing set. ROC curves suggested that the area under the ROC (AUC) value for prognostic risk score signature was 0.81 in training set and 0.705 in testing set. And AUC values corresponding to 1-, 3-, and 5 years of OS were 0.809, 0.753, and 0.794 in training set and 0.698, 0.682, and 0.754 in testing set, respectively. We established the two nomograms that confirmed high C-index and accomplished good prediction accuracy.

Conclusions: We constructed a prognostic risk-related ARlncRNAs signature that could accurately predict the prognosis of KIRC patients.

Keywords: The Cancer Genome Atlas; autophagy; kidney renal clear cell carcinoma; long noncoding RNA; prognostic signature.

FIGURE 1

Construction and evaluation of prognostic‐related ARlncRNAs signature in training group and verification in testing group. Lasso regression analysis was performed to avoid overfitting in training group. (A) Lasso coefficient values and vertical dashed lines at the best log (lambda) value were displayed. (B) Lasso coefficient profiles of the prognostic lncRNAs. (C) Forest plot of multivariate cox regression analysis for seven prognostic‐related ARlncRNAs. The Hazard Ratio (HR) value and its 95% confidence interval with associated p‐value were showed. An HR of greater than 1 indicates that high gene expression was bad for the prognosis. These HRs greater than 1 were risk factors, which indicated that high expressions of lncRNAs were unfavorable for prognosis, while HRs less than 1 were protective factors, which indicated that high expressions of lncRNAs were favorable for prognosis. Kaplan–Meier survival curve for KIRC patients with high‐ and low‐risk scores in the training group (D) and testing group (G) ROC curves for the signature and its AUC value in training group (E) and testing group (H). ROC curves and their AUC value represented 1‐, 3‐, and 5‐year predictions in training group (F) and testing group (I)

FIGURE 2

The Kaplan–Meier (KM) survival curve of seven prognostic‐related autophagy‐associated lncRNA (ARlncRNAs). (A) The KM survival curves for survival times of AC008870.2 in the high‐ and low‐risk group; (B) The KM survival curves for OS of AC099850.3 in the high‐ and low‐risk groups; (C) The KM survival curves for OS of AC108449.2 in the high‐ and low‐risk groups; (D) The KM survival curves for OS of AL162586.1 in the high‐ and low‐risk groups; (E) The KM survival curves for OS of AL022328.2 in the high‐ and low‐risk groups; (F) The KM survival curves for OS of AL360181.2 in the high‐ and low‐risk groups; (G) The KM survival curves for OS of SPINT1‐AS1 in the high‐ and low‐risk groups

FIGURE 3

Evaluation of the prognostic signature in training group and verification in testing group. The risk Score distribution in high‐ and low‐risk score KIRC patients in training group (A) and testing group (B) scatter dot plot showed survival outcomes in high‐ and low‐risk KIRC patients in training group (C) and testing group (D). Heatmap showed the expressions of seven prognostic‐related autophagy‐associated lncRNAs (ARlncRNAs) in high‐ and low‐risk score KIRC patients in training group (E) and testing group (F)

FIGURE 4

Estimation of clinical Value of the prognostic‐related ARlncRNA Signature and clinicopathological variables in KRIC patients. (A) The forest plots for univariate Cox regression analysis showed that risk score, age, grade, AJCC stage, T stage, and N stage were prognostic risk‐related variables. (B) The forest plots for multivariate Cox regression analysis showed risk score, grade, and AJCC stage were independent prognostic factors. (C) Multivariate receiver operating characteristic (ROC) curve analysis showed predictive accuracy of the model: the AUC value of risk score was higher than other clinicopathological variables

FIGURE 5

The correlation analysis of the risk score from prognostic signature with clinicopathological characteristics in the KIRC patients. The correlation between risk score and clinicopathological characteristics stratified according to (A) age (< =60 years, n = 258 vs. >60 years, n = 249); (B) gender (FEMALE, n = 175 vs. MALE, n = 332); (C) tumor grades (G1 grade, n = 12 vs. G2 grade, n = 219 vs. G3 grade, n = 201vs. G4 grade, n = 75); (D) AJCC stages (stages I, n = 253 vs. stages II, n = 53 vs. stages III, n = 118 vs. stages IV, n = 83); (E) T stage (T1, n = 259 vs. T2, n = 65 vs. T3, n = 172 vs. T4, n = 11); (F) N stage (N0, n = 226 vs. N1, n = 15 vs. Nx, n = 266); (G) M stage (M0, n = 405 vs. M1, n = 78 vs. Mx, n = 24)

FIGURE 6

(A) Construction of a prognostic nomogram utilized risk score from the prognostic‐related ARlncRNAs signature and clinicopathological parameters clarified from multivariable Cox regression analysis to predict 1‐, 3‐, and 5‐year survival rate of KIRC patients. (B) Construction of a prognostic nomogram utilized risk score from the prognostic‐related ARlncRNAs signature and seven ARlncRNAs clarified from multivariable Cox regression analysis to predict 1‐, 3‐, and 5‐year survival rate of KIRC patients

FIGURE 7

Construction of a LncRNA‐mRNA co‐expression network and functional enrichment analysis. (A) Diagrammatic plot displayed the lncRNA‐mRNA co‐expression network contained 165 lncRNA‐mRNA pairs formed by 7 prognostic risk‐related ARlncRNAs and 97 mRNAs. (B) Sankey diagram showed the relationship between 7 prognostic risk‐related ARlncRNAs, 97 mRNAs, and risk types (risk or protective). (C–E) Gene Ontology (GO) analysis of target mRNAs, which were co‐expressed with seven prognostic risk‐related ARlncRNAs, revealed the enriched (C) biological processes, (D) cell components, and (E) molecular functions. (F) Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis of target mRNAs, which were co‐expressed with seven prognostic‐related ARlncRNAs, revealed the enriched signaling pathways

FIGURE 8

We compared the prognostic‐related ARlncRNAs signature with published predictive models in KIRC patients. The ROC curves showed that the present signature had higher prediction reliability and sensitivity than other published biomarkers