Gene coexpression network analysis identifies hubs in hepatitis B virus–associated hepatocellular carcinoma

Abstract

Background:

Hepatocellular carcinoma (HCC) is among the leading causes of cancer-related death worldwide. The molecular pathogenesis of HCC involves multiple signaling pathways. This study utilizes systems and bioinformatic approaches to investigate the pathogenesis of HCC.

Methods:

Gene expression microarray data were obtained from 50 patients with chronic hepatitis B and HCC. There were 1649 differentially expressed genes inferred from tumorous and nontumorous datasets. Weighted gene coexpression network analysis (WGCNA) was performed to construct clustered coexpressed gene modules. Statistical analysis was used to study the correlation between gene coexpression networks and demographic features of patients. Functional annotation and pathway inference were explored for each coexpression network. Network analysis identified hub genes of the prognostic gene coexpression network. The hub genes were further validated with a public database.

Result:

Five distinct gene coexpression networks were identified by WGCNA. A distinct coexpressed gene network was significantly correlated with HCC prognosis. Pathway analysis of this network revealed extensive integration with cell cycle regulation. Ten hub genes of this gene network were inferred from protein-protein interaction network analysis and further validated in an external validation dataset. Survival analysis showed that lower expression of the 10-gene signature had better overall survival and recurrence-free survival.

Conclusion:

This study identified a crucial gene coexpression network associated with the prognosis of hepatitis B virus–related HCC. The identified hub genes may provide insights for HCC pathogenesis and may be potential prognostic markers or therapeutic targets.

1. INTRODUCTION

Hepatocellular carcinoma (HCC) is one of the leading causes of cancer and related mortality. Hepatitis B is the most common cause of HCC in Taiwan and is one of the leading causes of HCC worldwide.^1,2 Multiple molecular signaling pathways have been identified in the pathogenesis of HCC.^3,4 However, many aspects of the underlying molecular mechanism in HCC development and progression remain unanswered.

Dysregulated genes in HCC may be inferred by comparing the expression of genes among different tissues and disease conditions.^5,6 The gene expression patterns of dysregulated genes in HCC have been reported to further classify HCC. These patterns can serve as tumor markers or can indicate survival.⁷ Biologically, dysregulated genes may interplay with each other and contribute to the pathogenesis of HCC. Inferring disease-driving or prognostic genes without considering other associated genes may overlook their coexpression traits. Thus, it is crucial to investigate the gene networks of closely related genes in the pathogenesis of HCC.

Weighted gene coexpression network analysis (WGCNA) has evaluated gene networks based on similarities of gene expression in similar biological conditions.⁸ WGCNA collects the highly coexpressed genes with correlation measurements into network modules. WGCNA can further estimate the shared gene levels between the two genes. The resulting coexpression network can be the subject to downstream analysis including the enrichment pathway, clinical correlation analysis, and prediction of candidate hub genes.⁸ WGCNA can identify prognostic genes or hub gene signatures in different cancer types such as breast, colorectal, and esophageal cancers.^9–11

This study examined the mRNA coexpression networks involved in HCC pathogenesis. Weighted gene coexpression networks were constructed and analyzed, and the hubs of the pivotal coexpression networks were inferred and further validated. This study utilized a systemic bioinformatic approach to identify gene networks and genes potentially correlated to HCC outcomes.

2. METHODS

2.1. Collection of HCC tissue

We collected HCC tissues from 50 patients with chronic hepatitis B and HCC. All patients were positive for serum HBsAg and negative for antibodies to hepatitis C virus (anti-HCV) and delta virus (anti-HDV). The use of these samples for the study of HCC was approved by the Institutional Review Board of Taipei Veterans General Hospital (IRB No. 97-09-17A and No. 2014-01-003B). The samples were preserved in liquid nitrogen before further processing. Freshly frozen tissues were processed to similar sizes. The tissues were homogenized, and the total RNA was extracted using TRIzol reagents (Invitrogen, USA).

2.2. Differential gene expression analysis

Conventional 3′ expression array experiments were performed for extracted RNA from HCC tissues. The microarray experiment was performed by Affymetrix U133 Plus 2.0 platform. Array hybridization was performed according to the manufacturer’s instructions. The raw data of microarray were preprocessed with R statistical programming language.¹² The RMA algorithm from the ‘affy’ package¹³ of the Bioconductor was used to obtain the normalized gene expression values. The top 2000 significantly changed probe sets were taken for further analyses.

2.3. Weighted gene coexpression network analysis

WGCNA was performed to construct clustered coexpressed gene modules across the samples. Weighted gene coexpression network analysis was performed to identify the relationship and correlation between genes across the samples.⁸ First, Pearson’s correlation and topological overlap measurement were performed to construct coexpression modules. Pearson’s correlation coefficient was used to illustrate the similarities between the two probe sets. A topological overlap matrix (TOM) was applied to determine whether each pair of nodes is connected to each other or not, which makes networks less sensitive to spurious connections. In the signed TOM correlation network, negative correlation nodes were considered to be nonconnected. High positive and high negative correlation nodes were considered to be connected in the unsigned TOM correlation network. The module eigengenes (ME) were derived as the first principal component of each module and were further utilized for clinical correlation analysis.

2.4. Functional annotation and enrichment analysis

Functional annotation and enrichment analysis were applied to each coexpression network. The inferred differentially expressed genes (DEGs) involved in each coexpression network served as an input for functional annotation. Clustering analysis of functional annotation and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment could identify the biological processes involved in the coexpression network. These analyses were performed with the g:Profiler web server.¹⁴

2.5. Statistical analysis

The clinical demographics of HCC patients were analyzed. Independent sample t tests were used to evaluate differences in the continuous variables that passed the normality test; variables that failed the normality test were tested with an independent two-group Mann-Whitney U test. The Chi-square test was performed for the discrete variables. Univariate and multivariate Cox regression analyses were performed to evaluate the influence of clinical and tumor features including the MEs inferred by WGCNA. The median value of MEs for each module served as a cutoff to define a higher or lower ME of each module. Statistically significant parameters in univariate analysis were further subjected for multivariate analysis. The Kaplan-Meier method was applied to compare the cumulative incidence of overall and disease-free survivals. The difference in overall and disease-free survival distributions between groups was compared by the log-rank test. Statistical significance in all analyses was considered as denial of a two-tailed p < 0.05. All statistical analyses were performed using R.¹²

2.6. Construction of protein-protein interaction network and validation of hub genes

Network analysis was performed to construct a protein-protein interaction (PPI) network of the coexpression network. The Cytoscape was used to analyze and visualize the interaction networks.¹⁵ First, the network output from WGCNA was utilized as the input for the Cytoscape. The maximum neighborhood component method was further applied with cytoHubba, a Cytoscape app, to identify the most connected hub genes.

To validate the identified hub genes in this study, the tumorous and nontumorous HCC samples of the Cancer Genome Atlas Liver Hepatocellular Carcinoma (TCGA-LIHC) cohort¹⁶ were used as a validation set. The expression of genes of interest was examined in tumor and nontumor samples, and the correlation between each gene was also studied. A higher or lower expression of genes defined by group median was compared with overall survival and disease-free survival in each hub gene as well as the combined 10 hub gene set. The analysis was performed with the enhanced version of the GEPIA (Gene Expression Profiling Interactive Analysis) web server (GEPIA2).¹⁷

3. RESULTS

3.1. Clinical profiles and demographics of HCC patients

This study recruited 50 patients with HCC and HBV infection who underwent surgical resection. The median age of these patients was 50 years (range, 37-61 years old). Most were male (44 patients, 88%), and 42% of the patients had liver cirrhosis (n = 21). Five patients (10%) had ascites. Most patients (n = 48, 96%) were Child-Pugh classification A. More details of the patient profiles are presented in Table 1.

Table 1.

Clinical and pathological features of patients with hepatocellular carcinoma

Variable	Total (n = 50)
Patient demographics
Age	50.3 (37.1-61.2)
Male, n (%)	44 (88.0)
Albumin, g/dL	4.0 (3.8-4.3)
Total bilirubin, mg/dL	0.85 (0.6-1.18)
ALT, U/L	38.5 (30.3-54.8)
AST, U/L	41.0 (27.3-63.8)
Creatinine, mg/dL	0.9 (0.8-1.1)
Platelet, 1000/μL	174.5 (134.25-222.0)
PT INR	1.00 (0.98-1.07)
Cirrhosis, no (%)	21 (42.0)
Follow-up, mo	62.1 (28.5-131.3)
Child-Pugh class (A/B)	48/2
Presence of ascites (%)	5 (10.0)
HBeAg (positive/negative)	2/38
Tumor factors
Tumor size, cm	4.5 (3.6-6.5)
AFP, ng/mL (<100/≥100)	26/24
Tumor grading (well/moderate/poor differentiation)	6/31/9
HCC pattern (solitary/multiple)	28/22
Macrovascular invasion (Y/N)	13/37
Microvascular invasion (Y/N)	35/15
Recurrence (Y/N)	31/19
Metastasis (Y/N)	13/37
Cancer stage
BCLC stage (A/B/C/D)	23/8/19/0
CLIP (0/1/2/3/4/5)	17/17/6/8/1/1
Okuda (1/2)	38/12
Tumor stage I/II/III/IV	21/9/17/3

Values are presented as median (interquartile range, 25th-75th percentile).

AFP = alpha-fetoprotein, ALT = alanine aminotransferase; AST = aspartate aminotransferase; BCLC = Barcelona clinic liver cancer; CLIP = Cancer of the Liver Italian Program; HCC = hepatocellular carcinoma; HBeAg = Hepatitis B e antigen; NR = international normalized ratio; PT = prothrombin time.

The median size of the resected tumor is 4.5 cm (interquartile range [IQR], 3.6-6.5 cm). The tumor-node-metastasis (TNM) staging of HCC according to the seventh edition of the American Joint Committee on Cancer showed that 21 patients were stage I, 9 were stage II, 17 were stage III, and 3 were stage IV. The median follow-up after surgery was 62.1 months (IQR, 28.5-131.3 months). Thirty-one patients (62%) had recurrence during follow-up after surgical resection. Thirteen patients (26%) had distant metastasis either at the time of receiving resection or during follow-up. The details are presented in Table 1.

3.2. Gene expression analysis and WGCNA analysis

There were 1649 significant DEGs identified via comparison of gene expression arrays of tumorous and nontumorous tissue from HCC patients. Those DEGs were subjected to a construct coexpressed gene network by WGCNA. The tumorous and nontumorous samples were distinctly separated and clustered by WGCNA (Fig. 1A). The DEGs were further classified into five coexpressed gene network modules: turquoise, yellow, blue, brown, and gray (Fig. 1B).

Fig. 1.

Construction of coexpressed gene networks with weighted gene coexpression network analysis. A, The tumor (pink boxed) and nontumor (green boxed) samples were distinctly separated and clustered. B, In total, five coexpression gene networks were inferred by the analysis.

The ME is defined as the first principal component: It summarized and represented the expression profile of all genes in each coexpression network (module). The inferred MEs of each module were further correlated with TNM tumor stage of HCC. Two gene coexpression networks (turquoise and yellow modules, Fig. 1B) were positively correlated with TNM tumor stage (p < 0.001 and p < 0.05, respectively; Fig. 2A); one gene coexpression network (brown module, Fig. 1B) showed negative correlation with TNM tumor stage (p = 0.01; Fig. 2A).

Fig. 2.

Clinical correlation of the gene coexpression networks. A, Correlation between coexpressed gene modules and stage of hepatocellular carcinoma (HCC). Here, X axis is the tumor-node-metastasis stage, and the Y axis is the module eigengene (ME). (B) Kaplan-Meier analysis showed that the coexpressed gene network turquoise was significantly correlated with HCC prognosis.

3.3. Identification of a gene coexpression network correlated with HCC prognosis

The five gene coexpression modules were further subject to correlation analyses with clinical factors. The results of the univariate analysis of factors associated with the overall survival of HCC patients are summarized in Table 2. The univariate analysis identified significant factors for inferior overall survival: a tumor size ≥5 cm, macrovascular or microvascular invasion, and higher ME of two coexpression networks (turquoise and yellow modules).

Table 2.

Univariate Cox regression analysis for lower overall survival of patients with hepatocellular carcinoma

Variable	Univariate analysis
	Crude HR	95% CI	p
Age ≥50	1.145	0.481-2.729	0.7592
Male	3.502	0.469-26.12	0.2216
Child-Pugh class B	2.771	0.635-12.09	0.1750
Tumor size ≥5 cm	1.208	1.092-1.335	0.0002
Tumor number ≥3	1.876	0.756-4.656	0.1749
AFP ≥100 ng/mL	1.795	0.744-4.333	0.1932
Macrovascular invasion	3.332	1.370-8.103	0.0079
Microvascular invasion	12.80	1.711-95.72	0.0130
Higher ME turquoise	2.571	1.036-6.380	0.0417
Higher ME yellow	2.697	1.082-6.718	0.0332
Higher ME blue	0.568	0.239-1.352	0.2011
Higher ME brown	0.671	0.282-1.596	0.3669
Higher ME gray	1.586	0.667-3.772	0.2967

Median value of MEs within each module is used as a cutoff to define higher or lower ME.

AFP = alpha-fetoprotein; HR = hazard ratio; ME = module eigengene.

The crude hazard ratio of a higher ME of the coexpression genes from the turquoise module was 2.571 (95% CI, 1.036-6.38) for overall survival; the yellow module had a higher ME of 2.697 (95% CI, 1.082-6.718). Higher MEs of all five modules were not significantly associated with HCC recurrence in the univariate analysis (Supplementary Table 1, http://links.lww.com/JCMA/A162). Despite the two strong tumor factors including tumor size larger than 5 cm and microvascular invasion, the multivariate Cox regression analysis suggested that a higher ME of the turquoise module was an independent predictor for overall survival of HCC (Table 3). The adjusted hazard ratio of higher ME in the turquoise module was 3.475 (95% CI, 1.2403-9.735). The yellow module was not significantly correlated with overall survival in the multivariate analysis. These findings were also confirmed with a Kaplan-Meier method analysis which showed that a lower ME of the turquoise module (ME_turquoise) was correlated with a better prognosis (Log-rank test, p < 0.05, Fig. 2B).

Table 3.

Multivariate Cox regression analysis of demographic features with clinical outcomes of patients with hepatocellular carcinoma

Variable	Multivariate analysis
	Adjusted HR	95% CI	p
Recurrence
Tumor size ≥5 cm	1.152	1.047-1.267	0.0038
Microvascular invasion	4.858	1.520-15.53	0.0077
Higher ME turquoise	1.428	0.684-2.979	0.3428
Higher ME yellow	1.132	0.519-2.471	0.7553
Overall survival
Tumor size ≥5 cm	1.267	1.111-1.445	0.0004
Microvascular invasion	13.46	1.467-123.5	0.0215
Higher ME turquoise	3.475	1.240-9.735	0.0178
Higher ME yellow	1.199	0.453-3.177	0.7145

Median value of MEs within each module is used as a cutoff to define higher or lower ME.

HR = hazard ratio; ME = module eigengene.

The ME_turquoise was further correlated with the demographic features of the HCC patients (Table 4). A higher ME_turquoise was not significantly associated with general patient characteristics including age, gender, cirrhosis, Child-Pugh class, or ascites. Versus the lower ME_turquoise, HCC with the presence of a higher ME_turquoise was significantly correlated with larger tumors (≥5 cm, 60% vs 28%), higher alpha-fetoprotein (AFP) levels (≥100 ng/mL, 68% vs 36%), and more advanced tumor stage (stage III + IV, 56% vs 24%). A higher ME_turquoise was not significantly associated with multiple HCCs or macrovascular/microvascular invasion.

Table 4.

Correlation of higher or lower module eigengene of turquoise module (ME_turquoise) with demographic features of patients with hepatocellular carcinoma

Variable	Higher MEturquoise		p
	Yes	No
Patient demographics
Age, median (IQR)	46.2 (26.0-59.3)	52.0 (38.0-61.3)	0.477
Male, n (%)	22 (88)	22 (88)	1.000
Cirrhosis, n (%)	12 (48)	9 (36)	0.567
Child-Pugh class (A/B)	24/1	24/1	1.000
Presence of ascites (%)	4 (16)	1 (4)	0.346
Tumor factors
Tumor size ≥5 cm, n (%)	15 (60)	7 (28)	0.046
AFP ≥100 ng/mL, n (%)	17 (68)	9 (36)	0.048
Solitary/multiple HCC	11/14	17/8	0.154
Macrovascular invasion (Y/N)	7/18	6/19	1.000
Microvascular invasion (Y/N)	20/5	15/10	0.217
TNM stage I + II vs III + IV	11/14	19/6	0.043

AFP = alpha-fetoprotein; HCC = hepatocellular carcinoma; IQR = interquartile range; ME = module eigengene; TNM = tumor-node-metastasis.

3.4. Functional annotation and gene set enrichment analysis

This study further analyzed the biological processes related to the DEGs in the turquoise module correlated with HCC prognosis. Functional annotation analysis showed that the DEGs were significantly enriched in protein-binding activities via their molecular function. The DEGs were also extensively integrated with cell cycle regulation during the biological process including cell cycle, cell division, mitosis, DNA metabolism, and DNA repair/replication (Fig. 3A). KEGG pathway analysis of this gene coexpression network showed that the DEGs were majorly enriched in the cell cycle (Fig. 3). The detailed results of the enrichment analysis are shown in the supplementary Figure 1, http://links.lww.com/JCMA/A156.

Fig. 3.

Functional annotation and pathway enrichment analysis of the turquoise coexpressed gene network. BP = biological process; CC = cellular component; GO = gene ontology; MF = molecular function.

3.5. Identification of hub genes in the PPI network and validation of hub genes

The turquoise module of the prognostic gene coexpression network was further subjected to Cytoscape to construct a PPI network. The constructed PPI network is composed of 425 nodes (DEGs) and 12,522 edges (Fig. 4A). The mostly highly connected ten hub genes in the PPI network were identified as follows: ANLN, ASPM, CCNB1, CDK1, CDKN3, CENPF, ECT2, KIF4A, NEK2, and TOP2A (10 nodes and 45 edges, Fig. 4B).

Fig. 4.

Protein-protein interaction (PPI) network. A, The prognosis associated with the turquoise coexpressed gene network was utilized to construct a PPI network. Orange nodes are hub genes. B, The most connected cluster is composed of ten hub genes: ANLN, ASPM, CCNB1, CDK1, CDKN3, CENPF, ECT2, KIF4A, NEK2, and TOP2A. Color, MNC of nodes.

This study further utilized an external dataset from the TCGA-LIHC cohort as the validation set. The validation set is composed of 369 tumorous and 50 nontumorous HCC samples. Cell cycle regulators such as CDK1, CCNB1, CDKN3, and CENPF were all relatively abundant in the tumor samples (Fig. 5A). Higher expression of CDK1, CCNB1, and CENPF was associated with a significantly worse overall survival (Fig. 5B). Cytokinesis regulators including CDK1, ECT2, and KIF4A were upregulated in tumor samples (Fig. 5A) and associated with poor overall survival (Fig. 5B). Other hub genes such as ANLN, ASPM, NEK2, and TOP2A were upregulated in tumor samples compared to nontumor samples (Fig. 5A). Higher expression of ANLN, ASPM, NEK2, and TOP2A was associated with inferior overall survival (Fig. 5B). Only CDKN3 showed borderline significance toward overall survival (Log-rank p = 0.058, Fig. 5B).

Fig. 5.

Validation of hub genes in the TCGA-LIHC dataset. A, Gene expression of each of the 10 hub genes. The hub genes were upregulated in tumor samples vs nontumor samples. *p < 0.01l; red, tumor group; gray, nontumor group; TPM, transcripts per million. B, Kaplan-Meier analysis of each hub gene for overall survival. * p < 0.01; ** <0.001. C, Survival analysis showed that lower expression of the 10-gene signature had better overall survival. D, A lower expression in the 10-gene signature had better recurrence-free survival. The blue line is the lower expression group, and the red line is the higher expression group. Dotted line, 95% CI.

We further combined the coexpressed hub genes as a 10-gene signature to explore the gene expression signature in the validation set. The validation set was further defined into two groups according to the median expression levels of the 10-gene signature. Survival analysis showed that lower expression of the 10-gene signature had better overall survival (Log-rank p < 0.001, Fig. 5C). The lower expression group also had better recurrence-free survival in the TCGA-LIHC dataset (Log-rank p < 0.005, Fig. 5D).

4. DISCUSSION

HCC remains a major threat in cancer-related death worldwide. The survival of advanced HCC has recently improved from 10 months to over 2 years with advances in systemic treatment for HCC.^1,18,19 However, the recurrence of HCC after curative surgical resection or local ablation remains an important and unsolved issue in HCC-related death.²⁰ Sorafenib was the first approved effective systemic treatment for advanced HCC. It failed to curb recurrence or prolong survival after surgical resection or local ablation.²¹ Although other attempts in adjuvant settings have been made, none has yet to produce any clinical benefits.¹ Thus, it is crucial to investigate the underlying molecular pathogenesis of HCC via a systemic approach that provides a broader view of potential prognostic or therapeutic targets.

HCC has a diverse pathogenesis via different pathways. HCC may escape treatments including radiotherapy, chemotherapy, multikinase inhibitor, or immune therapy.¹ Here, we investigated the aggressiveness of HCC with consideration of interplays at the gene level including the concept of gene coexpression. This study identified a distinct coexpressed gene network correlated to HCC prognosis. A set of hub genes was inferred. These genes were further validated with a larger validation public dataset and were upregulated in HCC. Most of these genes were associated with poor prognosis in HCC, and the 10-gene signature correlated well with overall survival and recurrence-free survival. Thus, our systemic bioinformatic approach offers additional dimensions and insights into HCC pathogenesis.

Cytokinesis is a critical step in the cell cycle. Errors in DNA repair lead to increases in DNA replication stress and chromosomal instability,²² which in turn leads to cytokinesis failure and development of cancer.^22,23 The hub genes including CDK1, CCNB1, CENPF, CDKN3, TOP2A, ANLN, ECT2, and KIF4A identified here are involved in cytokinesis. CDK1, ECT2, and KIF4A regulate cytokinesis.²⁴ CDK1 is reported to interact with CCNB1,²⁵ CENPF,²⁶ or ANLN²⁷ to disrupt cell division in promoting cancer development. Additionally, overexpression of CDK1 promotes HCC progression and is associated with poor prognosis via cell cycle-dependent phosphorylation of hTERT.²⁸

Anillin (ANLN) is an actin-binding protein and is involved in cytokinesis. Upregulation of ANLN is associated with the development and worse prognosis of HCC.²⁹ CDKN3 promotes the proliferation of HCC in vitro.³⁰ ECT2 promotes worse outcomes and recurrence of HCC via Rho/ERK.³¹ KIF4A is one of the kinesin superfamily of proteins, and overexpression of KIFA promotes migration and metastasis of HCC cell lines.³² DNA topoisomerase II alpha (TOP2A) is a nuclear enzyme essential for transcription, replication, and mitotic chromosome structure.³³ Dysregulation of TOP2A leads to chromosome instability and tumorigenesis,³⁴ and overexpression of TOP2A in HCC is associated with younger age, vascular invasion, and worse prognosis.³⁵ Cytokinesis failure can lead to increased polyploidization and formation of cancer stem-like cells with increased tumor aggressiveness.³⁶ HCC with a highly polyploid phenotype is correlated with poor prognosis.³⁷ AFP expression is a feature of HCC cancer stem cell.³⁸ Larger tumors and elevated serum AFP level are associated with recurrence and worse prognosis in HCC.^2,39 Therefore, cell cycle deregulation may underlie the association with tumor size, serum AFP levels, and prognosis-related gene clusters.

Other genes among the identified hubs in this study have been associated with HCC progression. For example, NEK2 is a regulator of mitosis, and overexpression of NEK2 is associated with HCC invasion through the epithelial-mesenchymal transition⁴⁰; it promotes sorafenib resistance via regulation of β-catenin.⁴⁰ The assembly factor for spindle microtubules (ASPM) is essential for mitotic spindle function and promotes cancer progression via Wnt/β-catenin signaling pathway.^41,42 Overexpression of ASPSM has been associated with invasiveness and progression of HCC.⁴³ However, the linkage of these genes to other genes in the gene coexpression network remains unclear. These trends in coexpression seen here imply unknown molecular mechanisms and require further study to clarify.

Our study has several limitations. First, the exploration dataset only contained 50 HCC patients, which was relatively small dataset. The results from the exploration set needed to be validated in an additional validation set. We further applied a larger external validation set from the TCGA-LIHC cohort to validate the hub genes and found that those genes are associated with the progression and prognosis of HCC. The exploration dataset only included patients with chronic hepatitis B without other causes of HCC. The biological process associated with the hub genes identified here were mostly related to the cell cycle. Indeed, those genes were also reported not only in HCC but also in other cancers.^26,28,41 The findings implied that the coexpressed hub genes were linked to common pathways associated with the development and progression of cancer. Finally, the datasets included in this study offer gene expression profiling of HCC at the RNA expression level. RNA expression is generally correlated with the expression of the corresponding protein, but considerable variation can be found between RNA and protein levels.⁴⁴ Differentially expressed mRNAs have a better correlation at the level of mRNA and protein expression than nondifferentially expressed mRNAs.⁴⁵ The protein expression of the identified hub genes was assessed via immunohistochemistry data from the Human Protein Atlas project⁴⁶ and confirmed with positive expression in HCC tissues (Supplementary Figure 2, http://links.lww.com/JCMA/A156).

In conclusion, our study evaluated the gene networks underlying HBV-associated HCC through systems and bioinformatic approaches. This study identified a pivotal coexpressed gene network associated with HCC prognosis. Hub genes in this gene network were elucidated and further validated in an external validation dataset as prognostic markers. Thus, we proposed that the identified hubs may be potential prognostic biomarkers and therapeutic targets for HCC.

ACKNOWLEDGMENTS

This study was supported by grants from the Ministry of Health and Welfare (MOHW110-TDU-B-211-144019, MOHW111-TDU-B-221-014007), the Ministry of Science and Technology (MOST 109-2314-B-010-033-MY2, MOST 109-2622-B-010-002-CC2, MOST 110-2314-B-A49A-505, MOST 110-2314-B-075-010-MY3, MOST 111-2314-B-A49-047) of the Republic of China (Taiwan) and the Cancer Progression Research Center, National Yang-Ming University from The Featured Areas Research Center Program within the framework of the Higher Education Sprout Project by the Ministry of Education (110CRC-T205, 111W31205) of the Republic of China (Taiwan).

APPENDIX A. SUPPLEMENTARY DATA

Supplementary data related to this article can be found at http://links.lww.com/JCMA/A156.