Validation of miRNA prognostic power in hepatocellular carcinoma using expression data of independent datasets

Abstract

Multiple studies suggested using different miRNAs as biomarkers for prognosis of hepatocellular carcinoma (HCC). We aimed to assemble a miRNA expression database from independent datasets to enable an independent validation of previously published prognostic biomarkers of HCC. A miRNA expression database was established by searching the TCGA (RNA-seq) and GEO (microarray) repositories to identify miRNA datasets with available expression and clinical data. A PubMed search was performed to identify prognostic miRNAs for HCC. We performed a uni- and multivariate Cox regression analysis to validate the prognostic significance of these miRNAs. The Limma R package was applied to compare the expression of miRNAs between tumor and normal tissues. We uncovered 214 publications containing 223 miRNAs identified as potential prognostic biomarkers for HCC. In the survival analysis, the expression levels of 55 and 84 miRNAs were significantly correlated with overall survival in RNA-seq and gene chip datasets, respectively. The most significant miRNAs were hsa-miR-149, hsa-miR-139, and hsa-miR-3677 in the RNA-seq and hsa-miR-146b-3p, hsa-miR-584, and hsa-miR-31 in the microarray dataset. Of the 223 miRNAs studied, the expression was significantly altered in 102 miRNAs in tumors compared to normal liver tissues. In summary, we set up an integrated miRNA expression database and validated prognostic miRNAs in HCC.

Figure 1

Overview of the statistical pipeline and sample numbers in the included datasets (A). Pathological characteristics of datasets including stage (B), metastasis (C), and gender (D).

Figure 2

Slightly over 5% of all miRNAs were measured in each study. Overlap of detected miRNAs in each dataset (A) and aggregate summary (B).

Figure 3

Top five best performing miRNAs based on expression in the tumor tissue using data of the TCGA dataset. Kaplan-Meier survival plots, p value vs. cutoff plots and beeswarm plots for hsa-miR-149 (A) and hsa-miR-139 (B) and Kaplan-Meier plots for hsa-miR-3677 (C), hsa-miR-550a (D), and hsa-miR-212 (E).

Figure 4

Best performing miRNAs when utilizing expression ratios of tumor and matched normal samples using data of the GSE31384 dataset. Kaplan-Meier survival plots and p value vs. cutoff plots for hsa-miR-584 (A) and hsa-miR-31 (B) and Kaplan-Meier plots for hsa-miR-105 (C), hsa-miR-29c (D), and hsa-miR-146b-3p (E). Ranked HR values for all investigated miRNAs for overall survival (F) and disease-free survival (G).

Figure 5

miRNAs significant in at least two studies (A) and miRNAs consistently down/up regulated in three datasets (B–F) – tumor vs. normal expression in TCGA, GSE10694 and GSE6857 datasets.

Figure 6

Heatmap showing correlation coefficients across all miRNAs significant in the survival analysis in the TCGA (A) and GSE31384 (B) datasets.