1 INTRODUCTION

Liver cancer is the third leading cause of cancer-related deaths and an important barrier to increasing life expectancy worldwide.¹ The 5-year survival rate of liver cancer is approximately 18%, and it is the second most lethal after pancreatic cancer.² Hepatocellular carcinoma (HCC) is the predominant subtype of primary liver cancer, accounting for approximately 90% of all cases.³ The molecular pathogenesis of HCC varies, and chronic hepatitis B virus (HBV) infection is the leading cause of HCC.⁴ Although hepatic resection is one of the most effective approaches for curative treatment, HCC still has high recurrence and low 5-year survival rates.⁵ Therefore, identifying a reliable prognostic model that provides survival estimates is indispensable for selecting individualized treatment strategies.⁶ Over the last few decades, many prognostic models have been proposed to address the need for survival surveillance.^{7, 8}

The comprehensive characterization of HCC samples has led to a better understanding of the molecular characteristics of HCC. For example, The Cancer Genome Atlas project, a next-generation sequencing-based study, uncovered multi-omics information of HCC.⁹ The Clinical Proteomic Tumor Analysis Consortium (CPTAC) project proposed a comprehensive proteomic analysis to identify key alterations and effective molecular biomarkers in HBV-related HCC.¹⁰ Furthermore, gene expression profiling-based models to predict the outcome of HCC have also been gradually used.^{6, 11} However, previous studies on HCC prognostic biomarker identification mainly focused on HCC itself, and the roles of molecular alterations in nontumor tissues in HCC prognosis W ignored. After hepatectomy, the remaining liver may also contribute to cancers due to the underlying liver disease. Hence, these nontumor tissues may also have a substantial impact on the prognosis of HCC.¹²

Here, we aimed to assess the prognostic value of proteins in HBV-infected nontumor liver tissues based on proteomic data. Two proteins related to overall survival (OS) were integrated into a gene signature for prognostic prediction. Our study provides a promising prognostic predictor that may benefit the clinical practice.

2 METHODS

3 RESULTS

3.1 Survival associated proteins in nontumor liver tissues

A total of 159 patients with HCC in the training cohort and 204 patients with HCC in the validation cohort were included in the analyses. The details of the characteristics are summarized in Table 1. The differential analysis identified 1275 proteins that were differentially expressed between HCC and paired nontumor liver tissues (Figure 1A,B). When compared with the nontumor liver tissues, 393 upregulated and 882 downregulated genes were identified in tumor tissues.

Table 1. Demographic and histopathologic characteristics of study patients

	Training cohort	Validation cohort
Age, median (range), years	54 (20–81)	50 (21–77)
Sex
Female	31	26
Male	128	178
ALT( > / ≤ 50 U/L)
High	47	83
Low	112	121
Main tumor size (>/≤5 cm)
Large	83	71
Small	76	132
NA	-	1
Multinodular
Yes	42	41
No	117	163
Cirrhosis
Yes	112	189
No	47	15
TNM staging
I–II	105	162
III–IV	54	42
BCLC staging
0 or A	68	159
B or C	91	45
AFP ( > / ≤300 ng/ml)
High	61	91
Low	98	110
NA	-	3

• Abbreviations: AFP, alpha fetoprotein; ALT, alanine aminotransferase; BCLC, barcelona clinic liver cancer; TNM, tumor node metastasis.

The association of 6478 proteins with OS was also evaluated in the tumor and nontumor tissues. The univariate Cox analysis revealed 1090 prognosis-related proteins in nontumor liver tissues. Among these proteins, 456 were risk factors and 634 were protective factors; and 2211 protein expression levels in the tumor tissues were found to be related to OS, including 1045 as risk factors and 1166 as protective factors.

We also explored the prognostic values of each protein extracted from tumor and nontumor tissues. The Spearman correlation analysis for genes in 159 pairs of liver tumor and nontumor liver tissues showed that prognostic values in tumor and nontumor liver tissues were significantly correlated (Figure 1C). Similar correlations were observed for 1275 differentially expressed proteins (Figure 1D).

3.2 Functional enrichment for prognosis-related proteins

A total of 1090 prognosis-related proteins in nontumor liver tissues were subjected to a functional enrichment analysis. Protective proteins were mainly enriched in metabolism-related pathways. The top three most significantly enriched pathways were “carbon metabolism,” “biosynthesis of cofactors,” and “peroxisome” (Figure 2A). And the top three most significantly enriched pathways of risky proteins were “endocytosis,” “salmonella infection,” and “spliceosome” (Figure 2B).

3.3 Construction and definition of the LGPI

Of the 1275 differentially expressed proteins, 330 of the 1275 proteins in nontumor liver tissues were identified as prognostic proteins (p < 0.05). To prevent overfitting, the top 10 most significant proteins (according to p values) were further subjected to a multivariate Cox analysis (Figure 3A). We then constructed the LGPI consisting of two genes (DAO and MME) with the following formula: LGPI = (−0.87) × protein expression level of DAO + (−0.65) × protein expression level of MME. The LGPI was Z-score normalized for further analysis. The optimal cut-off point was identified as 0.44 (Figure 3B). Patients with HCC were divided into high- and low-risk subgroups based on the optimal cut-off points. In the training cohort, 49 patients were placed in the high-risk group, and 110 patients were placed in the low-risk group. A significant survival difference was observed between the high-and low-risk groups (HR = 3.877, 95% confidence interval [CI]: 2.127–7.066, Log-rank p < 0.001) (Figure 3C). In the validation cohort, we calculated the LGPI score for each patient based on the proposed formula. Then, the LGPI Z-score was normalized. Patients were divided into a high-risk group (n = 69) and a low-risk group (n = 135) based on the cut-offs identified in the training cohort. A significant survival rate difference was observed (HR = 1.706, 95% CI: 1.056–2.754; Log-rank p = 0.018) (Figure 3D).

3.4 Validation of the LGPI as an independent prognostic factor

In the multivariate analyses, LGPI remained an independent prognostic factor in the training and validation cohorts. In the training cohort, LGPI was an independent prognostic factor after adjusting for clinicopathological factors (hazard ratio [HR] = 3.399, 95% confidence interval [CI]: 1.827–6.325; p < 0.001) (Table 2). In the validation cohort, LGPI was also an independent prognostic factor after adjusting for clinicopathological factors (HR = 1.689, 95% CI: 1.051–2.714; p = 0.030) (Table 2). We performed subgroup analyses to observe survival differences between the high- and low-risk groups when considering different clinicopathological parameters in the training (Figure 4A) and validation (Figure 4B) cohorts. To leverage clinical and molecular information, we also combined two HCC staging systems, tumor node metastasis (TNM) and Barcelona clinic liver cancer (BCLC), with LGPI group strategies (Figure 5).

Table 2. Multivariate analysis of liver gene prognostic index with survival

Parameters	Training cohort		Validation cohort
	HR (95% CI)	p value	HR (95% CI)	p value
Age	0.96 (0.93–0.99)	0.002	0.997 (0.974–1.021)	0.811
Sex (male/female)	1.191 (0.611–2.325)	0.608	1.444 (0.646–3.227)	0.370
ALT ( > / ≤ 50 U/L)	1.326 (0.749–2.348)	0.333	0.761 (0.466–1.240)	0.273
Main tumor size (>/≤5 cm)	2.135 (0.870–5.239)	0.098	0.603 (0.325–1.118)	0.108
Multinodular (yes/no)	1.226 (0.499–3.015)	0.657	0.487 (0.264–0.897)	0.021
Cirrhosis (yes/no)	1.2623 (0.656–2.426)	0.485	2.176 (0.515–9.190)	0.290
TNM staging (III-IV/I-II)	0.737 (0.384–1.415)	0.360	1.717 (1.144–2.579)	0.009
BCLC staging (B or C/0 or A)	0.616 (0.115–3.304)	0.705	2.210 (1.519–3.216)	<0.001
AFP ( > / ≤300 ng/ml)	1.932 (1.058–3.527)	0.032	1.413 (0.879–2.271)	0.154
LGPI (high/low)	3.399 (1.827–6.325)	<0.001	1.689 (1.051–2.714)	0.030

• Note: Age, TNM staging, BCLC staging were coded as continuous variable. Specifically, TNM stage was coded as I = 1, II = 2, III = 3, IV = 4. BCLC staging was coded as 0 = 1, A = 2, B = 3, C = 4. The risk factors of sex, ALT, main tumor size, multinodular, cirrhosis, AFP and LGPI are male, >5 cm, yes, yes, >300 ng/ml and high-risk group.
• Abbreviations: AFP, alpha fetoprotein; ALT, alanine aminotransferase; BCLC, barcelona clinic liver cancer; LGPI, liver gene prognostic index; TNM, tumor node metastasis.

4 DISCUSSION

Patients with HCC are at risk of recurrence, even after undergoing a complete hepatectomy. Effective prognostic biomarkers are indispensable for determining patients with poor survival. Most previous studies on prognostic factor identification of HCC mainly focused on HCC itself but greatly downplayed the role of nontumor tissues. We developed a prognostic signature for HCC based on the expression levels of two proteins in nontumor liver tissues.

Many previous studies developed linear models to predict prognosis for patients with HCC.^{15, 16} For example, Wu et al. recently proposed a pyroptosis-related long noncoding gene signature for HCC, which could be used to predict the survival rate of patients with HCC.¹⁷ Fang et al. analyzed the significance of m6A RNA methylation regulators in HBV-related HCC and proposed a prognostic indicator based on their expression levels.¹⁸ However, the molecular prognostic value of adjacent nontumor liver tissues is less documented. Previous studies also have shown that molecular biomarkers in nontumor tissues are related to HCC progression, metastasis, and inferior survival rates.^{6, 19} HBV infection is a predominant risk factor for the onset and development of HCC. After resection, the presence of HBV continues to influence residual liver tissue, with the potential for recurrence. An hepatitis B surface antigen (HBsAg) level >200 IU/ml is an independent predictor of late recurrence, while an HBsAg level >50 IU/ml is an independent predictor of very late recurrence and late mortality.²⁰ Therefore, the molecular profiles of HBV-related liver tissues may also have prognostic value. Gene expression profiles in HBV-related liver tissues should be explored further as guides for HCC prognosis.

We identified proteins in nontumor liver tissues that were related to OS for HCC. Although the number of proteins in nontumor tissues with prognostic value was lower than that in tumor tissues, there were significant positive relationships between the prognostic values of protein levels in tumor and nontumor tissues. Hence, genes in nontumor liver tissues may also actively participate in HCC prognosis. Functional enrichment analysis showed that protective proteins in the liver tissues were markedly enriched in several metabolism-related pathways. A previous study based on the Korean National Health Insurance Service database showed that metabolic risk factor burden was associated with an increased risk of HCC, non-HCC cancer, and all-cause mortality in patients with chronic hepatitis B.²¹ Our findings suggest that liver metabolism is the main cause of poor HCC prognosis. However, many problems regarding liver metabolism and HCC prognosis remain, which need to be resolved.

Notably, two proteins, DAO and MME, were included in the prognostic signature development. DAO is a gene that encodes the peroxisomal enzyme d-amino acid oxidase. The plasma DAO level is an independent factor for readmission for HBV-related recurrent hepatic encephalopathy. The high expression of DAO in HCC tissues has also been recognized as a protective factor.²² A previous study suggested that MME may be important in combination with hepatitis C virus-related HCC.²³ However, the role of MME in HBV-associated HCC remains unclear. The LGPI was based on two proteins and showed moderate performance in the training and validation cohorts for OS prediction. Subgroup analyses showed that the prognostic value of LGPI in some subgroups of the validation cohort was not very good, which may be due to differences in basic characteristics between the training and validation cohorts. Future larger cohorts are needed to validate our results.

However, there are several limitations of this study. First, the retrospective nature of our study limited the prognostic value of LGPI. The LGPI could be further validated in a future prospective study. Second, although the gene functional enrichment analysis suggested that liver tissue metabolism status may be related to the prognosis of patients with HBV-related HCC, future in vivo or in vitro studies are needed for validation. Third, the validation cohort only provided gene expression in RNA levels rather than protein levels, and immunochemistry tests to validate our results are needed in the future. Furthermore, the complex nature and batch effects of different detection platforms should be noted.

In conclusion, the LGPI algorithm showed molecular factors in liver nontumor tissues have promising prognostic value for HCC. Moreover, it allows for a simple but accurate clinical description, and future prospective studies are needed to validate the prediction performance of the LGPI for HCC prognosis.

AUTHOR CONTRIBUTIONS

All authors fulfilled the ICMJE authorship criteria and agree to be accountable for all aspects of this study. Peng Lin participated in the study design, data acquisition, statistical analysis, data interpretation, and drafting of the manuscript. Dong-Yue Wen participated in statistical analysis, data interpretation, and drafting of the manuscript. Jin-Shu Pang participated in the statistical analysis, data interpretation, drafting of the manuscript. Wei Liao participated in statistical analysis, data interpretation, and drafting of the manuscript. Yu-Ji Chen participated in statistical analysis, data interpretation, and drafting of the manuscript. Yun He participated in study design, statistical analysis, data interpretation, and manuscript revision. Hong Yang participated in study design, statistical analysis, data interpretation, and manuscript revision. All authors approved the final version of the manuscript.

CONFLICTS OF INTEREST

The authors declare no conflicts of interest.