An Integrated Model Based on a Six-Gene Signature Predicts Overall Survival in Patients With Hepatocellular Carcinoma

Abstract

Background: Nowadays, clinical treatment outcomes of patients with hepatocellular carcinoma (HCC) have been improved. However, due to the complexity of the molecular mechanisms, the recurrence rate and mortality in HCC inpatients are still at a high level. Therefore, there is an urgent need in screening biomarkers of HCC to show therapeutic effects and improve the prognosis. Methods: In this study, we aim to establish a gene signature that can predict the prognosis of HCC patients by downloading and analyzing RNA sequencing data and clinical information from three independent public databases. Firstly, we applied the limma R package to analyze biomarkers by the genetic data and clinical information downloaded from the Gene Expression Omnibus database (GEO), and then used the least absolute shrinkage and selection operator (LASSO) Cox regression and survival analysis to establish a gene signature and a prediction model by data from the Cancer Genome Atlas (TCGA). Besides, messenger RNA (mRNA) and protein expressions of the six-gene signature were explored using Oncomine, Human Protein Atlas (HPA) and the International Cancer Genome Consortium (ICGC). Results: A total of 8,306 differentially expressed genes (DEGs) were obtained between HCC (n = 115) and normal tissues (n = 52). Top 5,000 significant genes were selected and subjected to the weighted correlation network analysis (WGCNA), which constructed nine gene co-expression modules that assign these genes to different modules by cluster dendrogram trees. By analyzing the most significant module (red module), six genes (SQSTM1, AHSA1, VNN2, SMG5, SRXN1, and GLS) were screened by univariate, LASSO, and multivariate Cox regression analysis. By a survival analysis with the HCC data in TCGA, we established a nomogram based on the six-gene signature and multiple clinicopathological features. The six-gene signature was then validated as an independent prognostic factor in independent HCC cohort from ICGC. Receiver operating characteristic (ROC) curve analysis confirmed the predictive capacity of the six-gene signature and nomogram. Besides, overexpression of the six genes at the mRNA and protein levels was validated using Oncomine and HPA, respectively. Conclusion: The predictive six-gene signature and nomograms established in this study can assist clinicians in selecting personalized treatment for patients with HCC.

Keywords: hepatocellular carcinoma; mRNA signature; overall survival; risk score; weighted gene co-expression network analysis.

Figure 1

Overall flowchart of this study.

Figure 2

ldentification of prognostic genes in hepatocellular carcinoma patients. (A) Volcano plot showing differentially expressed genes (DEGs) in hepatocellular carcinoma samples. (B) Clustering dendrogram of genome-wide genes in hepatocellular carcinoma samples. (C) Correlation between modules and traits. Absolute values of correlation coefficients between hepatocellular carcinoma status and modules greater than 0.15 were considered as hepatocellular carcinoma-related modules. (D) Module membership in nine hepatocellular carcinoma-related modules. The red module was the most significant module. (E, F) GO and KEGG analysis revealed the most significant biological process (BP), molecular function (MF), cellular component (CC), and pathways correlated with the high-risk group genes in the red module. (G) Volcano plot revealed DEGs in the red module.

Figure 3

Signature-based risk score is a promising marker in the training and validation cohorts. (A) The process of building the signature containing six genes most correlated with overall survival (OS) in the training set. The hazard ratios (HRs), 95% confidence intervals (CIs) calculated by univariate Cox regression, and the coefficients calculated by multivariate Cox regression using LASSO are shown. (B, C) Risk score distribution, survival overview, and heatmap for patients in the TCGA (B) and ICGC (C) datasets assigned to high- and low-risk groups based on the risk score.

Figure 4

Expression and survival analysis in training and validation datasets. (A, B) Kaplan–Meier overall survival (OS) curves for patients in the TCGA (A) and ICGC (B) datasets assigned to high- and low-risk groups based on the risk score. Patients with a high risk score exhibited poorer OS in the training and validation cohorts. (C, D) ROC curves showed the predictive efficiency of the risk signature for patients in the TCGA (C) and ICGC (D) datasets on the survival rate.

Figure 5

Cox regression analyses of the association between clinicopathological factors and OS. (A–F) Univariate/multivariate Cox regression analyses and heatmaps of the association between clinicopathological factors (including the risk score) and overall survival (OS) of patients in the TCGA (A, C, E) and ICGC (B, D, F) datasets.

Figure 6

The six-gene-based risk score is a promising marker for overall survival (OS) in subgroups. Subgroup analysis of OS based on pathological staging (A, B), grading (C, D), viral hepatitis (E), BMI (F–H), and age (I, J) of hepatocellular carcinoma (HCC) patients.

Figure 7

Differences in protein expression induced by six genes were verified in human tissue samples. (A–F) The mRNA expression levels of the six-gene signature in human cancers (conducted in Oncomine database). (G–L) Human Protein Atlas immunohistochemistry using anti-SQSTM1, anti-AHSA1, and anti-GLS antibodies. Normal liver (G–I) vs. tumor tissues (J–L).

Figure 8

Construction of a nomogram for survival prediction. (A) Nomogram combining signature with clinicopathological features. (B) Calibration plot showing that nomogram-predicted survival probabilities corresponded closely to the actual observed proportions.

Figure 9

The time-dependent receiver operating characteristic (ROC) and decision curve analysis (DCA) curves of the nomograms. Time-dependent ROC curve analysis evaluates the accuracy of the nomograms (A–C). The purple, red, yellow, green, or blue solid line represents the nomogram. The DCA curves can intuitively evaluate the clinical benefit of the nomograms and the scope of application of the nomograms to obtain clinical benefits (D–F). The net benefits (Y-axis) as calculated are plotted against the threshold probabilities of patients having 1, 3-, and 5-year survival on the X-axis. The gray dotted line represents the assumption that all patients have 1-, 3-, and 5-year survival. The black solid line represents the assumption that no patients have 1-, 3-, or 5-year survival. The red, blue, yellow, green, or purple solid line represents the nomograms.