A novel prognostic model for hepatocellular carcinoma based on 5 microRNAs related to vascular invasion

Abstract

Background: Hepatocellular carcinoma (HCC) is prevalent worldwide with a high mortality rate. Prognosis prediction is crucial for improving HCC patient outcomes, but effective tools are still lacking. Characteristics related to vascular invasion (VI), an important process involved in HCC recurrence and metastasis, may provide ideas on prognosis prediction.

Methods: Tools, including R 4.0.3, Funrich version 3, Cytoscape 3.8.2, STRING 11.5, Venny 2.1.0, and GEPIA 2, were used to perform bioinformatic analyses. The VI-related microRNAs (miRNAs) were identified using Gene Expression Omnibus HCC miRNA dataset GSE67140, containing 81 samples of HCC with VI and 91 samples of HCC without VI. After further evaluated the identified miRNAs based on The Cancer Genome Atlas database, a prognostic model was constructed via Cox regression analysis. The miRNAs in this model were also verified in HCC patients. Moreover, a nomogram was developed by integrating risk score from the prognostic model with clinicopathological parameters. Finally, a potential miRNA-mRNA network related to VI was established through weighted gene co-expression network analysis of HCC mRNA dataset GSE20017, containing 40 samples of HCC with VI and 95 samples of HCC without VI.

Results: A prognostic model of 5 VI-related miRNAs (hsa-miR-126-3p, hsa-miR-148a-3p, hsa-miR-15a-5p, hsa-miR-30a-5p, hsa-miR-199a-5p) was constructed. The area under receiver operating characteristic curve was 0.709 in predicting 5-year survival rate, with a sensitivity of 0.74 and a specificity of 0.63. The nomogram containing risk score could also predict prognosis. Moreover, a VI-related miRNA-mRNA network covering 4 miRNAs and 15 mRNAs was established.

Conclusion: The prognostic model and nomogram might be potential tools in HCC management, and the VI-related miRNA-mRNA network gave insights into how VI was developed.

Keywords: Co-expression network; Hepatocellular carcinoma; Overall survival; Prognostic model; Vascular invasion; microRNA.

Fig. 1

The workflow of the analysis process

Fig. 2

Identification of VI-related miRNAs in HCC and their functional enrichment analysis. Volcano map (a) and heatmap (b) of aberrantly expressed miRNAs between HCC samples with and without VI. GO analysis results showed the enriched biological processes, cell components, and molecular functions potentially associated with 31 downregulated miRNAs (c) and 6 upregulated miRNAs (d). KEGG analysis results showed the pathways potentially associated with 31 downregulated miRNAs (e) and 6 upregulated miRNAs (f)

Fig. 3

The survival analyses of HCC patients stratified by different expression levels of miRNAs. Kaplan–Meier survival curves indicated that there were 20 miRNAs significantly correlated with patients’ overall survival (OS) based on TCGA database

Fig. 4

Construction and validation of the prognostic model in HCC patients. (a) The univariate Cox regression analysis and Akaike information criterion (AIC) determined 5 VI-related miRNAs to build prognostic model, namely hsa-miR-126-3p, hsa-miR-148a-3p, hsa-miR-15a-5p, hsa-miR-30a-5p, hsa-miR-199a-5p. (b) Kaplan–Meier survival analysis showed that the survival time of patients with high-risk scores is significantly shorter than those with low-risk scores. The risk scores were calculated based on the 5-miRNA prognostic model. (c) Receiver operating characteristic (ROC) curve showed that the prognostic model has good accuracy in predicting 5-year overall survival of HCC patients from the TCGA database. (d) Distribution of risk scores of high- and low-risk HCC patients. Scatter plot (e) and bar plot (f) represented the correlation between survival time or survival state and risk score of HCC patients based on the prognostic model. (g) Heatmap showed that high-risk patients expressed lower levels of each miRNA biomarker

Fig. 5

The independent prognostic role of risk score in HCC. Forrest plots of the univariate (a) and multivariate (b) Cox regression analyses in HCC indicated that risk score might be an independent prognostic factor. The ROC curves showed the performance of the risk score and other clinicopathological parameters (age, gender, AJCC stage, and T-stage) in predicting 5-year overall survival of HCC patients (c) and patients with VI information available (d). (e, f) The ROC curves for two subgroups of patients in D

Fig. 6

Construction and validation of the prognostic nomogram with risk score as one of the parameters. (a) The nomogram used age, gender, T-stage, AJCC stage, and risk score to predict the 1-, 3-, and 5-year OS of HCC patients. The calibration plot to evaluate the concordance between the predicted and actual 1-year (b), 3-year (c), and 5-year (d) survival rates

Fig. 7

Real-time PCR validation of the expression levels of the 5 miRNAs in patient tissues. (a) hsa-miR-15a-5p, hsa-miR-30a-5p, hsa-miR-126-3p, hsa-miR-148a-3p, and hsa-miR-199a-5p were significantly downregulated in tissue samples from HCC patients with VI. (b) In HCC patients without VI, these 5 miRNAs' expression level was lower in patients with early metastasis/relapse

Fig. 8

Construction of the VI-related miRNA-mRNA regulatory network. (a) The soft threshold determination was based on scale-free topology criterion. (b) Modules with different colors were assigned by the Dynamic Tree Cut algorithm. (c) Heatmap showed the correlation between VI and the characteristic gene value of the module. Values outside of brackets are correlation coefficients, and values inside of brackets were gene significance level. (d) Scatterplot showed a highly significant correlation between gene significance for VI versus module membership in the salmon module. (e) The miRNA-mRNA network related to VI in HCC patients. Light blue diamonds represented mRNAs; purple ellipses represented miRNAs. (f) Protein–protein interactions illustrated the relationships between proteins