A Three-Gene Classifier Associated With MicroRNA-Mediated Regulation Predicts Prostate Cancer Recurrence After Radical Prostatectomy

Abstract

Background and objective: After radical prostatectomy (RP), prostate cancer (PCa) patients may experience biochemical recurrence (BCR) and clinical recurrence, which remains a dominant issue in PCa treatment. The purpose of this study was to identify a protein-coding gene classifier associated with microRNA (miRNA)-mediated regulation to provide a comprehensive prognostic index to predict PCa recurrence after RP.

Methods: Candidate classifiers were constructed using two machine-learning algorithms (a least absolute shrinkage and selector operation [LASSO]-based classifier and a decision tree-based classifier) based on a discovery cohort (n = 156) from The Cancer Genome Atlas (TCGA) database. After selecting the LASSO-based classifier based on the prediction accuracy, both an internal validation cohort (n = 333) and an external validation cohort (n = 100) were used to examined the classifier using survival analysis, time-dependent receiver operating characteristic (ROC) curve analysis, and univariate and multivariate Cox proportional hazards regression analyses. Functional enrichment analysis of co-expressed genes was carried out to explore the underlying moleculer mechanisms of the genes included in the classifier.

Results: We constructed a three-gene classifier that included FAM72B, GNE, and TRIM46, and we identified four upstream prognostic miRNAs (hsa-miR-133a-3p, hsa-miR-222-3p, hsa-miR-1301-3p, and hsa-miR-30c-2-3p). The classifier exhibited a remarkable ability (area under the curve [AUC] = 0.927) to distinguish PCa patients with high and low Gleason scores in the discovery cohort. Furthermore, it was significantly associated with clinical recurrence (p < 0.0001, log rank statistic = 20.7, AUC = 0.733) and could serve as an independent prognostic factor of recurrence-free survival (hazard ratio: 1.708, 95% CI: 1.180-2.472, p < 0.001). Additionally, it was a predictor of BCR according to BCR-free survival analysis (p = 0.0338, log rank statistic = 4.51).

Conclusions: The three-gene classifier associated with miRNA-mediated regulation may serve as a novel prognostic biomarker for PCa patients after RP.

Keywords: biochemical recurrence; clinical recurrence; microRNA; prostate cancer; protein-coding gene classifier; radical prostatectomy.

Figure 1

Network of the miRNA-gene pairs involving genes from both miRWalk 3 and the correlation analysis.

Figure 2

Construction and assessment of the three-gene calssifier associated with microRNA-mediated regulation. (A) Process of variable selection in least absolute shrinkage and selector operation (LASSO) regression. (B) Cross validation in LASSO regression. (C) Violin plot for the classifier. (D) Receiver operating characteristic (ROC) curve of the classifier’s ability to predict the Gleason score. (E) Decision tree based on the classification and regression tree (CART) algorithm. (F) Comparison of prediction accuracy between the LASSO-based classifier and decision tree-based classifier. (G) Heatmap of genes included in LASSO-based classifier and corresponding miRNAs.

Figure 3

Scatter plots of the four miRNA-gene pairs’ expression and corresponding linear regression lines for (A) hsa-miR-133a-3p-FAM74B, (B) hsa-miR-222-3p-FAM74B, (C) hsa-miR-1301-3p-GNE, and (D) hsa-miR-30c-2-3p-TRIM46.

Figure 4

Examination of each gene included in the three-gene classifier and corresponding miRNA’s prognostic ability. (A–C) Recurrence-free survival (RFS) analyses for FAM74B, GNE, and TRIM46 in the internal validation cohort. (D–G) RFS analyses for hsa-miR-133a-3p, hsa-miR-222-3p, hsa-miR-1301-3p, and hsa-miR-30c-2-3p in the internal validation cohort.

Figure 5

Examination of the three-gene classifier’s prognostic ability. (A) Recurrence-free survival (RFS) analysis for the classifier in the internal validation cohort. (B) Time-dependent receiver operating characteristic (ROC) curve analysis for the classifier and clinical factors in the internal validation cohort. (C) RFS analysis for the classifier in the discovery cohort. (D) RFS analysis for the classifier in the subgroup of the internal validation cohort with a Gleason score of 7. (E) BCR-free survival (BCRFS) analysis for the classifier in the external validation cohort. (F) BCRFS analysis for the classifier in the subgroup of the external validation cohort with a Gleason score of 7.

Figure 6

Functional enrichment analysis of the three genes’ co-expressed genes. (A) Significantly enriched Biological Process (BP) GO terms. (B) Significantly enriched Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways.