Alternative splicing events are prognostic in hepatocellular carcinoma

Abstract

Alternative splicing events (ASEs) play a role in cancer development and progression. We investigated whether ASEs are prognostic for overall survival (OS) in hepatocellular carcinoma (HCC). RNA sequencing data was obtained for 343 patients included in The Cancer Genome Atlas. Matched splicing event data for these patients was then obtained from the TCGASpliceSeq database, which includes data for seven types of ASEs. Univariate and multivariate Cox regression analysis demonstrated that 3,814 OS-associated splicing events (OS-SEs) were correlated with OS. Prognostic indices were developed based on the most significant OS-SEs. The prognostic index based on all seven types of ASEs (PI-ALL) demonstrated superior efficacy in predicting OS of HCC patients at 2,000 days compared to those based on single ASE types. Patients were stratified into two risk groups (high and low) based on the median prognostic index. Kaplan-Meier survival analysis demonstrated that PI-ALL had the greatest capacity to distinguish between patients with favorable vs. poor outcomes. Finally, univariate Cox regression analysis demonstrated that the expression of 23 splicing factors was correlated with OS-SEs in the HCC cohort. Our data indicate that a prognostic index based on ASEs is prognostic for OS in HCC.

Keywords: TCGA; alternative splicing; consensus cluster; hepatocellular carcinoma; prognosis.

Figure 1

Diagram showing the seven types of ASEs. AA, alternate acceptor; AD, alternate donor; AP, alternate promoter; AT, alternate terminator; ES, exon skip; ME, mutually exclusive exons; RI, retained intron.

Figure 2

UpSet plot of OS-SEs and gene interaction network in HCC. (A) UpSet plot showing OS-SEs for HCC; (B) Gene interaction network showing all interactions between genes corresponding to the 500 most significant OS-SEs in HCC.

Figure 3

Comparison of the prognostic efficacy of the eight PIs for OS survival among HCC patients in the low and high risk subgroups. (A–I) Kaplan-Meier survival curves for patients in the low and high subgroups for each PI. Time-dependent ROC curves demonstrating the ability of each PI to predict patient survival after 2,000 days.

Figure 4

Cox regression analysis of OS-associated clinical features PI-ALL. (A) Univariate analysis; (B) Multivariate analysis.

Figure 5

Identification of three clusters of HCC patients that exhibited distinct ASE features and clinical outcomes using consensus clustering. (A) Cumulative distribution function for k = 2 to 10. (B) Relative change in the area under the CDF curve for k = 2 to 10. (C) Tracking plot for k = 2 to 10. (D–F) Consensus clustering matrix for k = 2, 3, and 4. (G) Heatmap of the consensus matrix. ^aP< 0.01, ^bP< 0.001. (H) Kaplan-Meier OS curves for the 343 HCC patients stratified by cluster.

Figure 6

Correlation analysis between splicing factor expression and OS-SEs. (A) Triangles represent the splicing factors and oval nodes represent the OS-SEs. Red ovals represent the OS-SEs that displayed a positive correlation with OS while the green ovals represent OS-SEs that exhibited a negative correlation with OS. The blue and red lines indicate negative and positive correlations, respectively. (B–K) Top 10 correlations between the genes corresponding to the OS-SEs and splicing factors according to P-value.

Figure 7

Enrichment analyses of the genes corresponding to the 500 most significant OS-SEs. (A) Bar graph showing the top 20 results from the enrichment analysis; (B) Enrichment analysis showing the gene networks and enrichment of various pathways. Colors correspond to different cluster IDs.

Figure 8

Overall study design.