Identification of a novel metabolism‑related prognostic signature in hepatocellular carcinoma through bioinformatics analysis and validation through experimental studies

Background: Given that metabolic reprogramming has been recognized as an essential hallmark of cancer cells, this study sought to investigate the potential prognostic values of metabolism-related genes(MRGs) for hepatocellular carcinoma (HCC) diagnosis and treatment. Methods: The metabolism-related genes sequencing data of HCC samples with clinical information were obtained from the International Cancer Genome Consortium(ICGC) and The Cancer Genome Atlas (TCGA). The differentially expressed MRGs were identified by Wilcoxon rank sum test. Then, univariate Cox regression analysis were performed to identify metabolism-related DEGs that related to overall survival(OS). A novel metabolism-related prognostic signature was developed using the least absolute shrinkage and selection operator (Lasso) and multivariate Cox regression analyses. Furthermore, the signature was validated in the TCGA dataset. Finally, the expression levels of hub genes were validated in cell lines by Western blotting (WB) and quantitative real-timePCR (qRT-PCR). Results: A total of 178 differentially expressed MRGs were detected between the ICGA dataset and the TCGA dataset. We found that 17 MRGs were most significantly associated with OS by using the univariate Cox proportional hazards regression analysis in HCC. Then, the Lasso and multivariate Cox regression analyses were applied to construct the novel metabolism-relevant prognostic signature, which consisted of six MRGs. The prognostic value of this prognostic model was further successfully validated in the TCGA dataset. Further analysis indicated that this signature could be an independent prognostic indicator after adjusting to other clinical factors. Six MRGs (FLVCR1, MOGAT2, SLC5A11, RRM2, COX7B2, and SCN4A) showed high prognostic performance in predicting HCC outcomes. Finally, hub genes were chosen for validation and the expression of FLVCR1, SLC5A11, and RRM2 were significantly increased in human hepatocellular carcinoma cell lines when compared to normal human hepatic cell line, which were in agreement with the results of differential expression analysis. Conclusions: In summary, our data provided evidence that the metabolism-based signature could serve as a reliable prognostic and predictive tool for overall survival in patients with HCC.

detected using the Wilcoxon test method [12]. |logFC|>1 and adjusted P<0.05 were considered as significant. The common differentially expressed MRGs were identified between the ICGC database and the TCGA database by used FunRich software. These intersection MRGs were selected for further analysis.

Functional enrichment analysis of MRGs
Gene Ontology (GO) [13]and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis [14] were performed by "clusterprofiler" R package and "enrichplot" R package in R software [15]. Functional categories with false discovery rate(FDR) of less than 0.05 were considered statistically significant.

Survival-associated MRGs
We collected clinical information of 232 HCC patients in the ICGC database. A total of 3 patients who were followed for less than 2000 days and more than 30 days were excluded to avoid the interference of irrelevant factors. Survival analysis was performed on 229 patients. To screen out MRGs associated with the prognosis in HCC patients, univariate Cox analysis was implemented by the R "survival" package, and data were visualized using forest plots. Only differentially expressed MRGs a P value < 0.001 were were screened for subsequent analyses.

Construction of metabolism-related signature for HCC
HCC patients in ICGC dataset were used for constructing the COX prognostic signature, and patients in TCGA dataset were used for validating the signature. Lasso and multivariate Cox regression analyses were performed to construct a prognostic model [16]. To avoid the prognostic signature overfitting and remove highly related survival-related MRGs, Lasso Cox regression was carried out using R "survival" and "glmnet" package. MRGs detected via Lasso algorithm were evaluated by step wise multivariate Cox regression analysis. By weighting the estimated Cox regression coefficients,the modle of tumor-related metabolism genes risk was constructed [17].
The prognostic metabolism-related gene signatures were shown as risk score = Ʃ (βi × Expi), where βi, the coefficients, represented the weight of the respective signature and Expi represented the expression value. Based on the risk score formula, patients were assigned into low-risk group and high-risk group with the median risk score as the cutoff point. The Kaplan-Meier (K-M) survival curve was used the log-rank test to evaluate the differences in survival rate between the two groups. Furthermore, the receiver operating characteristic (ROC) curve was implemented by R "survivalROC" package [18] and the corresponding area under the ROC curve (AUC) was measured to assess the sensitivity and specificity of the metabolism-related signature.

Validation of metabolism-related signature
To verify the prognostic value of metabolism-related signature, we used the TCGA database as the validation cohort. The same formula was used to calculate the risk scores for each patient. Survival and ROC curve analyses were implemented as described above. In addition, Univariate and multivariate analyses were used to estimate the effect of risk scores on overall survival and the clinicopathologic features.
We also explored the correlation between the expression of these MRGs and several clinical features. For further validation of our analysis, The Human Protein Atlas (HPA) online database (http://www.proteinatlas.org/) was applied to identify the expression of these MRGs at a translational level [19].

Cell culture
Human normal hepatocyte cell line (LO2) and HCC cell lines (HepG2, Hep3B, HLF and PLC/PRF/5) obtained from the laboratory were maintained in the DMEM medium (Gibco, Wuhan, China) with 10% heat-inactivated fetal bovine serum (FBS, GibcoBRL) and antibiotics of 1% streptomycin and penicillin at 37° C in an atmosphere of 5% CO2.

Western Blotting Analysis
The protein extraction, and western blot analysis were performed as described previously [21].

Statistical analysis
All statistical analyses were performed by version 3. data were considered to be statistically significant with P value <0.05. and 381 upregulated genes) were extracted from in the ICGC database (Fig.1a, 1c), 251 differentially expressed MRGs (consisting of 36 downregulated and 215 upregulated genes) were extracted from in the TCGA database( Fig.1b, 1d). Finally, the common differentially expressed MRGs were identified in the two databases, a total of 178 MRGs(consisting of 28 downregulated and 150 upregulated genes) (Fig.1e) were selected for subsequent analysis.

GO, KEGG and PPI analysis of metabolism-related DEGs
To evaluate the potential molecular mechanisms of MRGs in HCC, the 178 differentially expressed MRGs were further analyzed by GO functional annotation and KEGG pathway enrichment. The results showed the top 10 biological processes GO terms, cellular component GO terms, molecular function GO terms (Fig. 2a), and the top 10 KEGG pathway terms (Fig. 2c). The correlation between the intersection genes and the top 5 biological processes, including organic anion transport, organic acid transport, carboxylic acid transport, lipid catabolic process, and monovalent inorganic cation transport is shown in Figure 2b. The KEGG analysis showed that the intersection genes were associated with material metabolism, especially arachidonic acid metabolism.

Prognostic values of survival-related MRGs
To better define the characteristics of survival-related MRGs, we explored the

Construction of metabolism-related prognostic signature for HCC
To establish the metabolism-associated prognostic signature, the Lasso regression and multivariate Cox proportional hazards regression analyses were conducted. The Lasso regression analysis was applied to exclude genes that may be highly correlated with other genes (Fig.4a, 4b). Then, a prognostic signature model was established based on Then, the risk score of each patient was calculated according to this prognostic modle.
Based on the median risk score, 229 HCC patients were classified into a high risk group Although univariate Cox analysis indicated that pathologic stage, T stage and our model were markedly associated with OS (Fig.6f, p<0.001), after the multivariate analysis, only metabolism-related prognostic signature(p<0.001) can be used as an independent prognostic factor (Fig. 6g).These results demonstrated that this prognostic model exhibited great applicability and stability in predicting the prognosis of HCC patients.
To validate the MRGs in this model, the protein expression levels were analyzed using the HPA database. The results showed that FLVCR1, SLC5A11,MOGAT2, and RRM2 protein levels matched their mRNA expression levels (Fig.7). However, representative images of the SCN4A and COX7B2 protein levels were not available in the HPA database.

Clinical value of prognostic signature
To further evaluate the clinical value of MRGs, the relationship between MRGs prognostic indicators and clinical features were investigated, and the results indicated that FLVCR1, MOGAT2, RRM2, SCN4A, and COX7B2 were differentially expressed in patients with various clinical features (Fig.8). To validate the clinical value of the metabolism-related prognostic signature, the association between the risk score and clinical characteristics were subsequently assessed, and the results demonstrated that high risk scores were positively associated with histological grade, and survival status in patients with HCC (Fig.8).

Validation of hub genes by WB and qRT-PCR
In order to explore diagnostic biomarkers or therapeutic targets which may play more significant roles in promoting HCC progression, we chose three genes(FLVCR1, SLC5A11 and RRM2) as our candidate biomarkers for WB and qRT-PCR validation in cell lines. As illustrated in Fig. 9, the expression levels of FLVCR1, SLC5A11 and RRM2 were significantly increased in human hepatocellular carcinoma cell lines compared with LO2, which were consistent with the results of bioinformatics analysis, indicating that the results were compelling.

Discussion
As a typical metabolic disease, the metabolic changes and regulatory mechanism of hepatocellular carcinoma are very complex, which has not been fully defined. HCC is different from normal liver tissue in glucose metabolism, lipid metabolism and protein metabolism, and also different from other tumors. These metabolic differences provide a theoretical basis for therapeutic strategies targeting tumor metabolism for HCC. Till date, there is no report on MRGs-based prognostic signatures for HCC patients. There is an urgent need to determine reliable metabolic biomarkers and predictive models to predict the prognosis of HCC.
In Six metabolism related genes which constitute the prognosis model were identified as potential biomarkers of HCC. SLC5A11, an inositol-specific sodium-dependent glucose cotransporter responsible for inositol uptake, has been reportedly related to anaplastic thyroid carcinoma, but the relationship between SLC5A11 and HCC has not been described previously [22,23]. In this study, our results indicated SLC5A11 may

Conclusions
In this study, we assessed the metabolism-related genes expression profiles based on ICGC database and TCGA database. Moreover, a novel metabolism-related prognostic model was constructed, which could be as as an independent prognostic predictor for