Predictive Value of The G1–G6 Transcriptomic Molecular Classication of Hepatocellular Carcinoma for its Biological Behavior and Clinicopathological Features in China

Background: Hepatocellular carcinoma (HCC) is one of the most common cancers worldwide and a large number of genetic alterations are involved in the carcinogenetic process. A G1–G6 transcriptomic classication was previously proposed in a French study, and Korean and Singaporean groups indicated its potential application in Asian HCC patients. However, the genomic proles of Chinese patients are distinct from patients of other regions, and therefore the suitability of this method in Chinese HCC patients has remained unknown. Materials/Methods: In this study, we tested the transcriptomic group classication from the French cohort on a cohort of HCC patients from China. a total of 107 HCC cases from China were selected for the G1– G6 transcriptomic molecular classication. The correlation between the G1–G6 molecular classication and clinicopathological features were analyzed. RNA sequencing and bioinformatics analysis were performed to screen related targets and molecular signaling pathways. Results: We investigated the G1–G6 signatures in 107 Chinese HCC patients. HCC cases from China (n=107) were distributed as follows: G1 (17.76%), G2 (1.87%), G3 (18.69%), G4 (9.35%), G5 (23.36%), and G6 (28.97%) groups. We observed concordance between the genetic proles and clinical features of Chinese HCC patients and French HCC patients. We found that the G1–G3 subgroups were associated with high serum alpha-fetoprotein (AFP) level, high copy number of hepatitis B virus (HBV) DNA, complex histopathological structure, macrovascular invasion, negative or weak Hep-Par1 expression, programmed death-ligand 1 expression, and liver cancer stemness. The G1 subgroup was mainly related to liver cancer stemness, and G3 subgroup showed the worst prognosis. The G5 and G6 subgroups were associated with activation of the Wnt/β-catenin pathway. Compared with the G1–G4 group, the G1–G3 group showed signicantly higher expression levels of regenerating family member 1 beta (REG1B), regenerating family member 3 gamma (REG3G), and inositol 1,4,5-trisphosphate receptor type 1 (ITPR1), and enriched calcium signaling pathway. Conclusions: Our results clarify the correlation between G1–G6 molecular classication and molecular markers and molecular signaling pathways in the Chinese HCC population and initially established a link between the phenotype and molecular characteristics. This study enhances our understanding of the heterogenicity of China HCC and indicates that the G1–G6 signatures can be used to identify potential therapeutic biomarkers against HCC patients in China. was used to analyze the correlation between the G1–G6 molecular classication and tumor biological behavior. After checking data for normal distribution and variance homogeneity, comparisons between two groups were evaluated using independent-sample t-tests or Mann–Whitney U test. For three or more groups, differences were statistically analyzed by one-way ANOVA or the non-parametric test (Kruskal– Wallis test). Kaplan–Meier and Cox proportional hazards survival regression analysis were used to evaluate the prognostic signicance of the G1–G6 molecular classications within 36 months. All p values were two-tailed, and P < 0.05 was considered statistically signicant.


Background
Hepatocellular carcinoma (HCC) is the most common malignancy worldwide, with more than half of the new HCC cases and deaths every year occurring in China [1]. The main causes of HCC include chronic hepatitis B virus (HBV) or hepatitis C virus (HCV) infection, alcoholic or non-alcoholic steatohepatitis, autoimmune hepatitis, and several metabolic diseases. However, the etiologies of HCC between the Euro-American area and Asia-Paci c region vary widely. HBV infection was endemic in China, where the prevalence rate of HBsAg is 7.2% [2], while HCV infection is common in western countries. In China, HCC only has a 5-year survival rate of 14.1% and a recurrence rate of about 70% [3]. Approximately 70-80% of HCC patients are diagnosed at an advanced stage and can receive only palliative care [4]. Although several promising drugs were developed over the last decades, these drugs have failed to meet clinical endpoints in phase III trials [5]. The failure of these trials is, at least in part, because of the lack of effective molecular markers or the minimal validation of known molecular markers in different populations.
Recently, researchers have been attempting to establish a molecular classi cation for HCC, which is prognostically informative, to accurately identify patients after curative resection who will bene t from additional early therapeutic interventions to prevent a recurrence [6,7]. Several studies from China focused on the correlation between molecular markers and the differentiation degree and invasion ability of HCC.
However, few studies have investigated the correlation between HCC molecular classi cation and pathological characteristics. A HCC classi cation method based on the expression of 16 genes was proposed by Boyault et al [8]. The authors classi ed tumors into six groups, G1-G6, using a minimal subset of 16 genes. This transcriptomic group classi cation was tested on cohorts from Singapore [9] and South Korea [10] with comparisons to European HCC patients. The results from these groups were generally in line with the results of Boyault et al, but there were still many inconsistencies. The Singapore study reported that the G1 subgroup had a higher proportion of patients with HBV in Singapore compared with European samples. HBV infection was found in the G2 subgroup of European HCC patients but not found in the G2 subgroup of Singapore HCC patients. In addition, the G6 subgroup was closely associated with satellite nodules in the European HCC population but not in the Singapore HCC population. The Singapore study also found no associations of clinical features with G4-G6 subgroups in the Southeast Asia HCC population. The probability of microvascular invasion (MVI) in the G3 subgroup was 5 times higher than other subgroups in Singapore samples. These differences suggested that the G1-G6 transcriptomic classi cation is not completely applicable to HCC patients in all regions of Asia. Furthermore, the genomic pro les of HCC patients are distinct in populations from China and other countries. For example, the TP53 mutation rate in Chinese populations is signi cantly lower than that in Korean and Singapore populations [11,12]. Therefore, it is valuable to validate the clinical relevance of the 16 gene HCC classi cation method in a Chinese population with HCC.
In this study, to determine whether the HCC molecular typing method established by the European team applies to Chinese HCC patients, a total of 107 HCC cases from China were selected for the G1-G6 transcriptomic molecular classi cation. The correlation between the G1-G6 molecular classi cation and pathological features, biological behaviors, serum marker levels, 3-year overall survival (OS) rate, and other indicators were analyzed in combination with the clinicopathological data of the HCC patients. We also detected and analyzed molecular markers and molecular signal pathways related to the G1-G6 molecular classi cation. RNA sequencing and bioinformatics analysis were performed to screen related targets and molecular signaling pathways.

HCC samples and clinical data
The use of human clinical samples was approved by the Medical Ethics Committee of Mengchao Hepatobiliary Hospital of Fujian Medical University. A total of 120 HCC specimens and paired adjacent noncancerous liver tissues were randomly selected from patients undergoing hepatectomy between January 2014 to December 2017. The patient follow-up information and clinicopathological characteristics were obtained from the biological sample bank and original pathology report. We also collected 12 fresh HCC tissues and paired adjacent noncancerous liver tissues for further analysis by RNA-seq. Data from The Cancer Genome Atlas (TCGA) database were obtained from TCGA portal sites (https://tcga-data.nci.nih.gov/tcga/tcgaDownload.jsp).

Quantitative real-time PCR (qPCR)
The qPCR procedures were performed as previously described [13]

Western blot
Tissues were harvested and lysed in cold extraction buffer (RIPA, Beyotime Biotechnology, Shanghai, China) as previously described [14]. Samples were centrifuged at 13,000 × g at 4°C for 15 min, and the protein concentration in the supernatants was measured using the Pierce BCA Protein Assay kit (Thermo Fisher Scienti c Inc., Waltham, IL, USA) according to the manufacturer's protocol. Equal amounts of protein samples were subjected to 10 % sodium dodecyl sulfate-polyacrylamide gel electrophoresis and subsequently electro-transferred to polyvinylidene di uoride membranes. The membranes were blocked with 5 % BSA diluted in Tris-buffered saline (20 mM Tris, 150 mM NaCl, pH 7.4) containing 0.05 % Tween-20 for 2 h and then incubated with primary antibodies at 4°C overnight. The primary antibodies are listed in Table 2. The membranes were washed ve times and then incubated with peroxidase-conjugated secondary antibodies (1:3,000, ab6721, Abcam, Cambridge, MA, USA) in blocking buffer for 1 h at RT.
Band intensities were quantitated by an enhanced chemiluminescence detection system using the SuperSignal™ West Pico Plus Kit (Thermo Fisher Scienti c Inc.). The protein density was quanti ed with Bio-Rad Image Lab software and ImageJ software. Tissue microarray (TMA) construction A total of 58 para n-embedded tissue samples from HCC patients (G1: 14 cases, G2: 1 case, G3: 14 cases, G4: 9 cases, G5: 10 cases, and G6: 10 cases) were randomly selected under a strati ed sampling procedure to generate TMAs. The donor wax blocks were made of the para n sections and stained by hematoxylin-eosin (HE) staining. To exclude tissues with necrotic or bleeding areas, the corresponding positions of cancer tissues and adjacent liver tissues were observed and marked under a microscope.
The recipient block was cast by melting conventional para n wax in molds for making blank blocks. The donor tissue blocks were transferred into the recipient wax wells and prepared with a 1.5 mm perforated needle. Next, we placed the wax blocks in the oven at 65°C for approximately 7-9 min and removed them at the semi-melted state; the blocks were cooled slightly at room temperature and moved to a refrigerator at 4°C. The freeze-thaw process was repeated so that the tissue core in the blocks and the wall of the pore were integrated.
The sections were treated with a 3 % peroxidase solution to block endogenous peroxidase. The sections were then incubated with 5 % BSA blocking solution to reduce the non-speci c background signal and false positives. Next, the sections were incubated overnight with the primary antibodies at 4°C. A section was incubated in antibody diluent alone without the primary antibody as a control. The sections were further incubated with ImmPress horseradish peroxidase anti-rabbit IgG antibody (Maixin Inc., Fujian, China) secondary antibodies. The immunoreactions were visualized using a DAB Kit (Maixin Inc.). Following counterstaining and mounting, digital images from the sections were created using a wholeslide scanner (KF-PRO-005-EX, KFBIO, Ningbo, China) and images were captured and analyzed using K-Viewer version 1.5.3.1 software (KFBIO).

RNA-seq and bioinformatics analysis
Total RNA was extracted from tissues using Trizol reagent (TransGen Biotech) according to the manufacturer's instructions. RNA quality was con rmed using an Agilent 2100 Bioanalyzer. Libraries for sequencing were created with the Illumina NEBNext® Ultra™ RNA Library Prep Kit or NEBNext® Ultra™ Directional RNA Library Prep Kit. Brie y, PCR ampli cation was performed to obtain the nal DNA library.
After the library was constructed, a Qubit2.0 Fluorometer was used for the preliminary quanti cation. RNA-seq was performed on the Illumina sequencing platform by Shanghai Jikai Company (Shanghai, China). Quality control indicated that the sequencing error rates and data ltering reads of the 24 samples were controlled within the acceptable range. RNA-seq, bioinformatics analysis, and TCGA database analysis were used to explore the target genes and molecular signaling pathways between the proliferation class (G1-G3, group H) and non-proliferation class (G4-G6, group L). We compared two or multiple gene expressions under different conditions using statistical methods, identi ed the speci c genes that correlated with the conditions, and analyzed the biological signi cance (quality control, matching, quantitative analysis process, signi cant difference analysis, and function of enrichment) of these speci c genes.

Statistical analysis
Statistical analysis was conducted by GraphPad Prism version 8 software (GraphPad Software, San Diego, CA, USA) and IBM SPSS Version 25 software. The chi-square test (Fisher exact probability method) was used to analyze the correlation between the G1-G6 molecular classi cation and tumor biological behavior. After checking data for normal distribution and variance homogeneity, comparisons between two groups were evaluated using independent-sample t-tests or Mann-Whitney U test. For three or more groups, differences were statistically analyzed by one-way ANOVA or the non-parametric test (Kruskal-Wallis test). Kaplan-Meier and Cox proportional hazards survival regression analysis were used to evaluate the prognostic signi cance of the G1-G6 molecular classi cations within 36 months. All p values were two-tailed, and P < 0.05 was considered statistically signi cant.

Results
Distribution of G1-G6 subgroups in HCC patients from Europe and China studies Gene expression analysis was performed on surgically resected HCC samples from Chinese patients who were grouped into G1-G6 transcriptomic categories according to the expression of 16 predictor genes. First, we isolated RNA samples from 120 HCC tumor samples for HCC classi cation, and 107 RNA samples with su cient tissue quantity and good quality were analyzed by qPCR. Data from these samples were analyzed according to the standard method proposed by the Europe group [8]. We found that the HCC cases from China (n=107) were distributed as follows: G1: n=19 (17.76%), G2: n=2 (1.87%), G3: n=20 (18.69%), G4: n=10 (9.35%), G5: n=25 (23.36%), and G6: n=31 (28.97%) subgroups.
Several studies have shown that no matter what molecular typing method is used, HCC can be de ned into two major groups, including a proliferation group and a non-proliferation group. Proliferation tumors are associated with aberrant activation of signaling pathways, while non-proliferative tumors display a well-differentiated phenotype [15,16]. Correspondingly, G1-G3 (combined with G1, G2, and G3) of the G1-G6 molecular typing can be classi ed as a proliferation group, while G4-G6 (combined with G4, G5, and G6) was classi ed as a non-proliferative group. Next, we pooled classes into the G1-G3 group and G4-G6 group. We found that the proportions of the G1-G3 group and G4-G6 group in the China cohort were 38.32% and 61.68%, respectively. In the European cohort, 35 I-II  4  2  3  0  6  9   III-IV  15  0  14  10  19  22   TNM stage  107  I-II  8  2  10  6  18  19   III-IV  11  0   To analyze the correlation between proliferation and clinicopathological characteristics of HCC patients based on G1-G6 molecular classi cation, we integrated the proliferative group (G1-G3) and nonproliferative group (G4-G6) by clinicopathological information. We found that the proliferative group (G1-G3) was correlated with high serum AFP level (P=0.002), high copy number of HBV DNA (P=0.047), complex histological subtype (P=0.010), macrovascular invasion (P=0.012), and negative or weak positive Hep-Par1 (P=0.012) ( Table 4). Using follow-up data of patients after hepatectomy, we also analyzed the correlation between HCC prognosis and G1-G6 molecular classi cation. As shown in Fig. 1C, the G1-G3 HCC patients showed a shorter 3-year OS than G4-G6 HCC patients (P=0.010). The recurrence-free survival of the G1-G3 group tended to be shorter than the G4-G6 group (P=0.072) ( Figure 1D). Further analysis showed that G1 patients tended to have a shorter 3-year OS than non-G1 patients (P=0.1867) ( Figure 1E) and G3 patients had the worst prognosis (P=0.0125) ( Figure 1F). Cox regression univariate analysis showed that the TNM stage (P=0.0002), BCLC stage (P=0.001), and G1-G6 molecular classi cation (P=0.012) were important factors that affected the overall prognosis of HCC patients. All factors with a p-value < 0.1 detected in univariate analyses were included in multivariate analyses, and the subsequent results showed that the G1-G6 molecular classi cation (P=0.035) is an independent prognostic risk factor for poor OS of HCC patients (Table 5). Predictor of HCC classi cation using qPCR A total of 19 para-cancer liver tissue samples were randomly selected to investigate the correlation between expression of the 16 genes in the proliferation group (G1-G3) compared with the nonproliferation group (G4-G6). Our results showed that the expression levels of AFP (P=0.0200), CDH2 (P=0.0022), HN1 (P<0.0001), NRAS (P=0.0096), PAK2 (P=0.0024), RAB1A (P=0.0020), and SAE1 (P<0.0001) were signi cantly higher in the proliferation group (G1-G3) than in the non-proliferation group (G4-G6) ( Fig. 2A). Conversely, the expression levels of LAMA3 (P=0.0059) and PAP (P=0.0495) were markedly higher in the non-proliferation group (G4-G6) than in the proliferation group (G1-G3) (Fig. 2B).
There were no signi cant differences in the expressions of the following genes between the proliferation group (G1-G3) and non-proliferation group ( PD-L1 is highly expressed in G1-G3 subgroups Several molecular targeted drugs have entered clinical trials as palliative and complementary treatments for HCC [17]. For example, the clinically approved antibodies targeting PD-1 or its ligand PD-L1 were shown to cause lasting responses in up to 25% of advanced HCC patients in two early trials. In HCC, the proportion of tumor cells expressing PD-L1 is approximately 15% [18]. Several studies showed in amed and membrane expression of PD-L1 in ancestral liver cancer HCC, while lymph epithelioma-like HCC is in amed by in ammatory cells that express PD-L1 in large numbers [16,19,20]. The expression pattern of PD-L1 has an important effect on HCC prognosis, which is poor if both tumor cells and macrophages express PD-L1 [21]. Screening of patients who may bene t from PD-1/PD-L1 inhibitor therapy is the most important clinical concern. The correlation between G1-G6 classi cation and tumor cell PD-L1 expression is not clear. Therefore, we examined the expression of the PD-L1 by IHC in a TMA. The TMA comprised 58 resected HCC specimens and 58 specimens of adjacent non-malignant and premalignant para-cancer liver tissue from the same patients. Staining was compared on consecutive tissue sections of the TMA to enable automated (unbiased) image analysis and direct comparison of the samples. The staining was quanti ed according to the percentage of area stained for each protein in comparison with isotype-matched control IgG (Fig. 3A). Representative images for the positive PD-L1 antibody staining are shown in Fig. 3B, along with histograms showing the pooled quanti ed data for the antigen (Fig. 3C). We found that the positive staining of PD-L1 in G1-G3 subgroups was signi cantly higher than in G4-G6 subgroups (P=0.028).
G5 and G6 HCC subgroups are associated with activation of the Wnt/β-catenin pathway signaling Aberrant activation of Wnt/β-catenin signaling plays a key role in HCC progression, with approximately half of HCC cases acquiring mutations in either CTNNB1 or AXIN1 [22]. Glutamine synthetase (GS) is classically overexpressed when the Wnt/β-catenin signaling pathway is activated due to CTNNB1 mutations and could be used as surrogate marker [23]. In our study, we found that the protein levels of βcatenin were substantially higher in G5 and G6 HCC subgroups than in G1-G4 subgroups ( Fig. 4A and B). Additionally, the mRNA expression level of GS in HCC liver tissues was signi cantly increased compared with levels in para-cancer liver tissues, and the G5 and G6 HCC subgroups showed signi cantly higher expression levels of GS than the G1-G4 subgroups (Fig. 4C). In line with these results, western blot analysis showed that protein levels of β-catenin and GS were higher in G5 and G6 subgroups than other subgroups (Fig. 4D). Further statistical analysis found that the ratio of GS expression to β-catenin expression in the G5 and G6 groups was approximately three-fold higher compared with levels in the G1-G4 subgroups, which suggests that the Wnt/β-catenin pathway is highly activated in G5 and G6 HCC. The activation level was highest in G6. These results suggested that G5 and G6 HCC subgroups were closely associated with the robust activation of Wnt/β-catenin signaling.
The G1 subgroup is associated with maintenance of tumor cell stemness Increasing studies have reported the presence of a small number of cells in tumor tissue with strong stemness [24,25]. We found that the G1-G3 HCC subgroups were closely associated with poor differentiation and high invasion of tumor cells. We further checked whether the proliferation group (G1-G3) was correlated with the stemness characteristics of HCC. Representative images for IHC of stemnessrelated molecular markers, EpCAM and Sox9, are shown in Fig. 5A and 5C. EpCAM staining was located at the cytomembrane and cytoplasm while SOX9 expression was identi ed in the nucleus. The positive expressions of EpCAM and Sox9 in the G1-G3 group were signi cantly higher than in the G4-G6 group ( Fig. 5B and D). The G1 subgroup showed the highest positive percentages of EpCAM and Sox9 compared with other subgroups (Fig. 5E). The G3 subgroup did not show a signi cant difference compared with other subgroups (Fig. 5F). qPCR analysis also showed a similar pattern as the IHC results ( Fig. 5G and H). These results suggested that the G1 subgroup is associated with the maintenance of stemness in tumor cells.
Transcriptome sequencing analysis for G1-G6 molecular classi cation of Chinese HCC patients The results of our study suggest that G1-G3 group can be used as an independent risk factor for evaluating HCC patients' prognosis and is signi cantly correlated with poor prognosis after hepatectomy.
To clearly understand the poor prognosis group based on G1-G6 molecular classi cation of Chinese HCC patients, 12 samples of HCC tissues (2 cases of each subgroup) and the corresponding adjacent liver tissues were randomly selected. The expression levels of genes related to differentiation, proliferation, and function of tumor cells were examined in these samples and analyzed by RNA-seq and bioinformatics. The error rates of the 24 samples were all within the acceptable range ( Supplementary   Fig. 1). The proportions of ltered reads of the 24 samples were all within the acceptable range, and the proportion of clean reads of each sample accounted for more than 90%, which suggested that the data were suitable for subsequent analysis (Supplementary Fig. 2).
The all HCC group (CA group) included 12 samples of HCC tissues, and the para-HCC group (pCA group) included 12 corresponding samples of adjacent liver tissues. High proliferative group (group H) includes 6 samples of G1-G3 cancer tissues, and low proliferative group (group L) includes 6 samples of G4-G6 cancer tissues. Compared with the pCA group, there were 3233 up-regulated genes and 1654 downregulated genes in the CA group, while 33,299 genes showed no changes between the two groups (Fig.   6A). Compared with group L (G4-G6 subgroup), there were 1,269 up-regulated and 1,445 down-regulated genes in group H (G1-G3 subgroup), and 33,692 genes showed no differential expression between the two groups (Fig. 6B). We also found that 108 genes overlapped between the H group (compared with group L) and CA (compared with group pCA) (Fig. 6C), and 27 genes were commonly down-regulated ( Fig. 6D). Among the 27 genes, four genes (REG1B, REG3G, C19orf18, and ITPR1) were signi cantly upregulated or downregulated in group H compared with group L ( Table 6). Analysis using TCGA database also con rmed that the expression levels of REG1B, REG3G, and ITPR1 were signi cantly increased in HCC liver tissues compared with normal liver tissues (Fig. 6E). Using the follow-up information database of TCGA patients, Kaplan-Meier analysis was conducted to analyze the prognosis of HCC patients. As shown in Fig. 6F, HCC patients with a high expression of ITPR1 showed a signi cantly shorter 3-year OS than HCC patients with low or medium expression of ITPR1. However, there were no statistical differences between HCC patients with high expressions of REG1B or REG3G and the corresponding control groups. These results suggested that ITPR1 may promote the occurrence and development of HCC and affect the prognosis of HCC patients. We next performed KEGG pathway analysis using the KEGG database and Clusterpro le software. Fisher exact analysis was used to test and calculate p-values, and the pathways with a p-value less than 0.05 were retained. The p-value represented the signi cance of enrichment of differential genes; a lower the pvalue indicates a greater correlation between the pathway and the differential genes. A Padj value less than 0.05 was taken as the threshold of signi cant enrichment. The KEGG pathway enrichment analysis revealed that the calcium signaling pathway, cGMP-PKG signaling pathway, renin secretion, neuroactive ligand-receptor interaction, and cell metabolism-related signaling pathways were signi cantly enriched in the G1-G3 subgroups (Fig. 7A-B).

Discussions
In this study, our results indicate that the G1-G6 signatures were generally in concordance between the genetic pro les of Chinese HCC patients and HCC patients in the original French study. Additionally, our data suggest that the G1-G3 group can be used as an independent risk factor for evaluating the prognosis of HCC patients, since the G1-G3 group was signi cantly correlated with poor prognosis after hepatectomy and the G3 group showed the worst prognosis among all subgroups. Furthermore, we demonstrated that the G5-G6 group was associated with activation of the Wnt/β-catenin pathway.
Moreover, we discovered that the G1-G3 group was associated with PD-L1 expression and the G1 subgroup is mainly related to liver cancer stemness. Importantly, we identi ed four genes (REG1B, REG3G, C19orf18, and ITPR1) that were signi cantly up-regulated or down-regulated in the proliferation group (G1-G3) compared with the non-proliferation group (G4-G6). Additionally, we found that the calcium signaling pathway, cGMP-PKG signaling pathway, renin secretion, neuroactive ligand-receptor interaction, and cell metabolism-related signaling pathways were signi cantly enriched in the G1-G3 subgroup.
Although our results were generally consistent with the studies from Europe and Singapore, there were still many inconsistencies. First, the distribution of the HCC subgroups in the Chinese cohort was signi cantly different from that in the original European cohort. In the European study, the G4 and G1 subgroups represented the largest and smallest subgroup, respectively. In our study, the G6 and G2 subgroups were the largest and smallest groups, respectively. Second, the Singapore research reported that the incidence of MVI in the G3 subgroup was markedly higher than in other subgroups. However, we did not nd a similar characteristic of the G3 subgroup of China HCC patients. Instead, we found that the incidence of MVI was signi cantly lower in the G5 subgroup than the other subgroups. Third, inconsistent with the European study but in line with the Singapore study, we did not nd any subgroups characterized with satellite nodules. Finally, our data showed that the G1-G3 subgroups had a shorter 3-year OS rate and the G3 subgroup has the worst prognosis, which was different from the studies in Europe and Singapore. These differences suggest that some of the clinical characteristics of HCC patients are distinct in the populations from China and other countries.
In this study, we found a high serum AFP level in the G1 group and a correlation between the G5-G6 group and Wnt/β-catenin pathway activation based on previous studies. Several studies showed that HCC with β-catenin mutations have slow proliferation, well-differentiated characteristics, and trabecular type false adenoid type histology, with cholestasis and intratumoral in ltrating immune de ciency [16,26].
β-catenin is a key molecular marker in the epithelial-mesenchymal transition. In physiological conditions, β-catenin binds cadherin and promotes the release of cadherin when β-catenin translocates to the nucleus [27]. In HCC, E-cadherin is reduced on the cell surface and N-cadherin is expressed once tumor cells undergo epithelial-mesenchymal transition [28]. It implicates that G5 and G6 subgroups may have some features as mentioned above.
We found several additional correlations in this study that had not been reported previously. For example, the G1-G3 group was associated with PD-L1 expression and liver cancer stemness. These ndings provide a new direction for the diagnosis and treatment of HCC. In the future, if G1-G3 patients were treated with targeted therapy for PD-1/PD-L1 and liver cancer stem cells, this may improve the survival and prognosis of a patient with highly malignant HCC with an extremely poor prognosis.
The pathogenesis of HCC is complex and involves multiple signaling pathways [29,30]. Signaling pathways involved in growth factors activated by intracellular calcium ions play a crucial role in various physiological and biochemical processes [31]. HCC encodes a variety of tyrosine kinases, including epidermal growth factor receptor (EGFR), whose functions are mainly in glucose and lipid metabolism, liver brosis, regeneration, and carcinogenesis [32,33]. Modica et al. [34] reported that EGF-dependent HCC proliferation may be affected by calcium ion, and intracellular dissociative calcium is a key regulator for EGFR signal propagation and apoptosis of tumor cells. Calcium not only promotes the proliferation of proliferating HCC cells but also induces the apoptosis of non-proliferating HCC cells, which might contribute to the reason why the proliferation group (G1-G3) has a poor prognosis. Considering the high hepatic activity of calcium and the overexpression of calcium channels in G1-G3 group, manipulation of calcium in ux in HCC cells may inhibit tumor metastasis.
ITPR1 encodes an intracellular receptor for type 1 inositol 1,4,5-triphosphate and mediates the release of calcium from the endoplasmic reticulum. A few recent studies indicated that ITPR1 affects tumor progression by regulating autophagy. For example, ITPR1 can protect renal cancer cells against natural killer cells by inducing autophagy [35,36]. In our study, we found that HCC patients with a high expression level of ITPR1 have a signi cantly lower 3-year OS rate than HCC patients with low or medium expression levels of ITPR1. Therefore, we speculate that ITPR1 may cooperate with calcium ions to promote the proliferation of proliferating HCC cells, thus leading to the increase of postoperative recurrence risk of HCC patients.
As a second messenger of cell information transmission, cGMP promotes cell division and inhibits cell differentiation. Our RNA-seq analysis showed that the cGMP-PKG signaling pathway was enriched in G1-G3 type HCC, which might contribute to the proliferative characteristics of the G1-G3 subgroups. In this study, we found that G1-G3 type HCC patients showed a high serum AFP level. A differential analysis based on the HCC transcriptome data from TCGA showed that the genes expressed in HCC tissues with high AFP expression are involved in neuroactive ligand-receptor interaction pathways [37]. Together, these results suggest that the proliferative characteristics of G1-G3 subgroups might be from the enrichment of neuroactive ligand-receptor interaction pathways.

Conclusions
In this study, we found that the G1-G6 transcriptomic molecular classi cation was applicable in the China HCC cohort regardless of the ethnic origin of patients. Through the results of these cohorts, we may be able to assertively establish the clinical and pathological relevance of the 16 gene score and use the classi cation system to develop therapeutic strategies for HCC patients worldwide in the future. Our results help clarify our understanding of the correlation between G1-G6 molecular classi cation and molecular markers and molecular signaling pathways in the Chinese HCC population and initially established a link between the phenotype and molecular characteristics. However, we mostly performed correlation analysis of G1-G6 molecular classi cation, focusing on the postoperative prognosis monitoring of HCC patients, and more studies are needed to fully clarify the molecular mechanism of HCC occurrence and development.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.