Introduction
Stress granules (SGs) are membrane-less cytoplasmic foci comprising mRNA, RNA-binding proteins, ribosomal subunits, and eukaryotic translation initiation factors.1 When eukaryotic cells are subjected to various environmental stimuli like hypoxia, acidosis, oxidative stress, and virus infection, SG formation attenuates protein synthesis to save energy, therefore this process is widely accepted to be a self-protective mechanism for cell survival.2 Notably, hypoxia, acidosis, and reactive oxygen species characterize the tumor microenvironment.3 By the formation of SGs, cancer cells could adapt to those hostile nonphysiological conditions for maintaining their continuous growth.4 Recently, increasing evidence indicates that SGs are closely related to cancer proliferation, metastasis, and chemotherapy.5 Thus, targeting SGs is considered a promising approach for cancer treatment.
To date, through SG enrichment and proteomic analysis, more than 400 SG-related proteins have been identified.6 It has been reported that the overexpression of some core SG proteins alone, for example, G3BP1, could induce SG formation even without stress conditions.7 As the key regulator of SG assembly, G3BP1 is upregulated in various cancers, like prostate, lung, and breast cancers.8 It is elucidated that the enhancement of G3BP1 translation induces SG formation, resulting in sarcoma metastasis and invasion.9 Along with this line, it is expected that when G3BP1 is knocked out, SGs would disassemble, accompanied by reduced sarcoma metastasis and invasion.10
As the most prevalent primary malignant liver tumor, hepatocellular carcinoma (HCC) is often diagnosed at a late stage in light of its insidious onset, which prevents patients from receiving radical treatment.11,12 Even worse, the treatment methods available at this stage are only effective in a few patients.13 However, if HCC is diagnosed in the early stage, there are more effective curative treatment options. It is thus vital to develop a tool for the early prognosis of HCC patients.14,15 The current predictive model relies on histopathological characteristics and cancer staging, which does not accurately predict the clinical outcome of early-stage patients.16 Therefore, it is essential to find one or more sensitive and specific prognostic indicators to detect HCC patients at the occurrence stage, guiding just-in-time therapeutic interventions to improve survival outcomes.
Recently, several gene signatures based on specific cell activities, such as autophagy and ferroptosis, have provided an early and good prognosis for individuals with HCC.17,18 As SG assembly promotes tumorigenesis through regulation of gene expression regulation and signal transduction,5,19 stress granule genes (SGGs) might be a potential target for early HCC detection. However, no studies have considered SGGs for predicting HCC patient clinical outcomes. It would be interesting to investigate whether SGGs could be used to construct a predictive gene panel to independently predict HCC development, which mutually compensates with other clinical characteristics.
Methods
Data
The International Cancer Genome Consortium (ICGC; https://dcc.icgc.org/ ) and The Cancer Genome Atlas (TCGA; https://portal.gdc.cancer.gov/ ) databases provided the gene expression and clinical data used in this study. Gene expression data of GSE36376, GSE14520, GSE10143, GSE54236, and GSE76427 were collected from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/ ); gene expression information of CHCC-HBV was from published research.20 Etiological information and baseline clinical characteristics of HCC patients that were available in the TCGA/GSE/CHCC-HBV/ICGC databases are summarized in Supplementary Tables 1 and 2, and 463 mammalian SGGs were retrieved from the Mammalian Stress Granules Proteome (MSGP) database (https://msgp.pt ).
Differential expression and functional enrichment of SGGs
The analysis of the differentially expressed SGGs between healthy and HCC liver tissue from TCGA cohorts was conducted in R. The cutoffs were: |log2 fold-change (FC)|>2 and adjusted p-value <0.05. GEO2R (https://www.ncbi.nlm.nih.gov/geo/geo2r/ ) was employed in evaluating the differential expression of SGGs in normal liver and HCC tissue from GEO cohorts following the criteria of |log2 (FC)|>0.6 and adjusted p-value <0.05. KEGG, GO, and GSEA analysis were applied as previously described.21
Prognostic risk model development
To identify the prognostic risk signature based on SGGs, univariate and multivariate Cox proportional hazards regression analyses were performed in R using the survival package. Patient risk scores were calculated as:
∑i=1nCoef i×EXP gene(i)
where Coef i stands for gene (i)’s Cox regression coefficient, and EXP gene (i) is the expression level of gene i. A median risk score was used as the cutoff to divide HCC patients into low- and high-risk score groups. Survival was described by the Kaplan-Meier method and log-rank analysis. Model sensitivity and specificity were evaluated by receiver operating characteristic (ROC) curve analysis.HCC samples and cell lines
HCC samples and para-tumor tissues with pathology confirmation were collected from HCC patients at Shandong Provincial Hospital Affiliated to Shandong First Medical University (Jinan, China) between the years 2021 and 2022. All experiments were approved and supervised by the Hospital’s Medical Ethics Committee and followed the latest version of the Declaration of Helsinki. Informed consent was obtained from all participants. Hep3B, Huh-7, HepG2, SK-Hep1, PLC, and L-02 cell lines from ATCC (https://www.atcc.org/ ) were cultured according to the manufacturing instructions. The specified lentivirus was introduced into Hep3B cells or Huh-7 cells followed by puromycin or blasticidin selection.
Plasmids and lentivirus preparation
Lentiviral constructs expressing short hairpin RNA (shRNA) against human KPNA2, MEX3A, WDR62, and SFN were purchased from Hedgehogbio (Shanghai, China). shG3BP1 lentiviral plasmid was acquired from Genechem (Shanghai, China). The human G3BP1 overexpression construct was generated by inserting the human G3BP1 sequence into the pLV3-CMV-EGFP backbone. Lentiviral particles were generated by transfecting target constructs, packaging, and enveloping plasmids into HEK293T cells; 48h later, the supernatants were collected and filtered utilizing 0.22 µm filters. All shRNA target sequences are placed in Supplementary Table 3.
RNA extraction and qPCR
Total RNA was isolated by TRIzol Reagents (#15596026; ThermoFisher Scientific, Waltham, MA, USA) as previously described.22 cDNA was synthesized with PrimeScript RT reagent kits (#RR047A; Takara, Shiga, Japan). Target gene expression was quantified by the 2−ΔΔCt method. Supplementary Table 4 includes the primer sequences.
Western blotting
RIPA buffer (#P0013B; Beyotime, Beijing, China) was used to prepare protein samples. Protein concentration was determined with a bicinchoninic acid (BCA) assay kit (BCA Kit, #K3001; Shenergy Biocolor, Shanghai, China), the normalized protein samples were separated by sodium dodecyl-sulfate polyacrylamide gel electrophoresis (SDS-PAGE) and transferred into the polyvinylidene difluoride (PVDF) membranes. After blocking with 5% milk, the membranes were incubated with primary antibodies overnight followed by incubation with horseradish peroxidase (HRP)-conjugated secondary antibodies. Supplementary Table 5 lists the primary antibodies.
Cell counting kit (CCK)8 assay
Cells were planted in a 96-well plate for 24 h at 1×104 cells per well, and 10 µL CCK8 (#CK04; Dojindo Laboratories, Kumamoto, Japan) was added to the culture medium for 90 m at the indicated time. Cell absorbance was measured at 450 nm.
Colony formation assay
A population of 250 cells was seeded per well in a 24-well plate. After 10 days of culture, the cells were fixed with 4% paraformaldehyde before staining with 0.1% crystal violet. Cell clones were photographed before statistical analysis.
5-ethynyl-2-deoxyuridine (EdU) staining assay
Cells were planted in a 96-well plate for 24 h at 1×104 cells per well and the EdU staining assay was performed following the kit manufacturer’s instructions. (#C10310; RiboBio, Guangzhou, Guangdong, China). Briefly, cells were treated with EdU at a dilution of 1:1,000 for 2 h at 37°C. After 4% paraformaldehyde fixation and 0.1% triton-100 permeabilization, the Apollo staining solution was applied for 30 m in the dark before Hoechst 33342 staining. Images were taken in three randomly chosen fields using an Axio Observer (Zeiss, Oberkochen, Germany) microscope.
Transwell migration assay
Aliquots of 200 µL serum-free medium containing 1×104 cells were planted in the upper chamber of the Transwell chamber (Corning Inc., Corning, NY, USA) and 800 µL of complete culture medium without cells was added to the lower chamber. 24 h later, 0.1% crystal violet (#G1062; Solarbio, Beijing, China) was used to stain the migrated cells. The number of migrated cells was counted after picture imaging.
Immunofluorescence
Cells grown on glass coverslips were treated with sodium arsenite (#S7400; Sigma-Aldrich, St Louis, MO, USA) for the indicated time. After 4% PFA fixation, cells were permeabilized with 0.1% Triton X-100 and then blocked with goat serum working solution (#ZLI-9056; ZSGB-Bio, Beijing, China), followed by incubation of indicated primary antibodies overnight at 4°C. On the next day, the coverslips were incubated with a fluorescent second antibody (#A21429 or #A11034; ThermoFisher Scientific) for one hour at room temperature. Then the slides were mounted with ProLong Gold Antifade mountant (#P36970; ThermoFisher Scientific) after extensive wash with PBS. The images were captured by a confocal microscope (TCS SP8; Leica, Mannheim, Germany).
Statistical analysis
The code and parameters used in this investigation were analyzed as previously described.21 All experimental procedures and assays were repeated at least three times. The graph data were reported as means ± standard deviation. Student’s t-test or one-way analysis with Bonferroni’s correction was used to determine the significance of differences between or among groups. R and the GraphPad Prism programs were used to generate data graphics, and p-values <0.05 were considered significant (*p<0.05; **p<0.01; ***p<0.001; ns, not significant).
Results
Differentially expressed SGGs in normal liver and HCC tissue
Principal component analysis found that 463 SGGs had expression profiles that were distinct in HCC compared with other cancer types such as breast cancer and lung adenocarcinoma (Fig. 1A), suggesting that they were specific to the carcinogenesis of HCC. To identify the distinct SGGs in HCC, a TCGA-HCC cohort including 50 normal liver samples and 374 HCC samples was analyzed, revealing 34 differential SGGs (Fig. 1B). Interestingly, only the expression of the ADAMTS13 gene was decreased in HCC, whereas the remaining 33 genes were all upregulated (Fig. 1C and Supplementary Table 6). The results led us to wonder whether SGGs were prone to be highly expressed, which predisposed SG formation in HCC. To verify this hypothesis, the expression of the 463 SGGs in normal and HCC tissue was analyzed in another six HCC cohorts. Expectedly, the number of highly expressed SGGs was dramatically more than that of low-expression SGGs in each indicated HCC cohort, and this discrepancy was most pronounced in the TCGA-HCC cohort (Fig. 1D). Additionally, in this TCGA-HCC cohort, we observed 200 SGGs whose expression was significantly correlated with overall survival (OS), and 196 of them all exhibited a significant disadvantageous association between gene expression and OS (Supplementary Table 7). Taken together, the results implied that highly expressed SGGs promoted more SG formation in HCC which led to an unfavorable outcome in HCC patients.
Functional enrichment analysis of differentially expressed SGGs in HCC
To further understand the functions of the 34 differential SGGs listed above, GO and KEGG analyses were carried out in the TCGA-HCC cohort. GO analysis revealed 30 significantly enriched GO terms using the differential SGGs, including 10 biological processes, 10 cellular components, and 10 molecular functions (Fig. 2A). KEGG analysis found that the above differential SGGs were enriched in DNA replication, the cell cycle, and gap junctions (Fig. 2B).
To establish functional links among the differential SGGs, we constructed a protein-protein interaction (PPI) network with Cytoscape software. It was shown that CDK1 as the hub gene connected 15 differential SGGs (Fig. 2C–D). The biological contribution of CDK1 in HCC was investigated by carrying out a single-gene GSEA analysis. As shown in Figure 2E, in addition to a positive correlation with the cell cycle pathway, genes involved in RNA degradation were upregulated in the CDK1 high-expression HCC group, which was consistent with the role of SGGs in mRNA degradation.23
Construction of a prognostic signature using SGGs
To evaluate the correlation of the 34 differential SGGs with the OS of HCC patients, univariate Cox regression analysis was performed in the TCGA-HCC cohort. Twenty-two differentially expressed SGGs were found to be significantly correlated with the OS of HCC patients (Fig. 3A). To accurately predict patient prognosis, we further constructed a prognostic signature using multivariate Cox analysis based on the above 22 genes. Interestingly, four SGGs, KPNA2, MEX3A, WDR62, and SFN, were recognized as potential independent risk variables with p-values <0.05 (Table 1). Then, using the expression of the risk genes and regression coefficients, a four-SGG-based risk signature was created. Patient risk scores were as 0.096×expression amount of SFN+0.677×expression amount of KPNA2-0.649×expression amount of WDR62+0.331×expression amount of MEX3A. TCGA-HCC patients were stratified by the median score into high-risk and low-risk groups. Kaplan-Meier survival curves (log-rank p<0.001) indicated that the high-risk group had a poor prognosis and the low-risk group had a favorable outcome (Fig. 3B). The expression of the four-SGG signature, risk score scattering, and patient survival in the high- and low-risk groups are shown in Figure 3C–E.
Table 1Regression coefficients of the potential independent risk genes
ID | Coefficient | HR | HR. 95L | HR. 95H | p-value |
---|
SFN | 0.096 | 1.101 | 1.007 | 1.203 | 0.035 |
KPNA | 0.677 | 1.968 | 1.438 | 2.694 | <0.001 |
WDR62 | −0.649 | 0.522 | 0.308 | 0.884 | 0.016 |
MEX3A | 0.331 | 1.393 | 1.086 | 1.786 | 0.009 |
Prognostic prediction power of the four-SGG signature
The association between various clinical features and OS was further examined by univariate and multivariate regression analysis. Univariate regression analysis showed that risk score, pathological stage, and, T stage were significantly associated with the OS of HCC patients (Fig. 4A). However, multivariate Cox regression analysis found that only the risk score was significantly associated (Fig. 4B).
To assess the efficacy of the SGG signature-based prognostic method, ROC curves were built. As shown in Fig. 4C, the risk score had a greater area under the curve (AUC, 0.776) than any of other clinical parameters, implying that it had the best performance of the constructed SGG model to predict the OS of HCC patients. To investigate the hidden explanation for this good performance of the four-SGG signature, the correlations of the expression of the four SGGs and HCC clinic characteristics were evaluated. As shown in Figure 4D and Supplementary Figure 1, the expression of KPNA2, MEX3A, and WDR62 but not SFN were positively correlated with pathological stage, T stage, and tumor grade, even though all four genes were upregulated in HCC (Supplementary Fig. 2).24 Kaplan-Meier survival analysis found that all four SGGs individually indicated poor OS and disease-free survival of HCC patients (Supplementary Fig. 3), which further explained the reliability of the four-SGG signature-dependent prognosis.
Validation of the four-SGG signature performance in the testing group
We further validated the effectiveness of the four-SGG signature in another HCC cohort from the ICGC database.25 After stratifying 203 HCC patients into high- and low-risk groups based on the median risk score, we found that patients in low-risk groups had longer OS than those in high-risk groups (Supplementary Fig. 4A). Risk score scattering, patient survival condition, and four-SGG signature expression in both groups are shown in Supplementary Figure 4B–D. The results confirmed the robust effectiveness of the four-SGG signature for predicting HCC prognosis.
Highly expressed KPNA2, MEX3A, WDR62, and SFN were involved in HCC development
Considering the strong prognostic power of the four-SGG signature in HCC, we wanted to experimentally validate whether these four proteins were involved in HCC development. To do this, the expression level of the genes was determined in HCC tissue and adjacent normal liver tissue. As shown in Figure 5A–C, compared with matched adjacent normal tissues, KPNA2, MEX3A, WDR62, and SFN were all significantly upregulated in HCC tissue both at the mRNA and protein levels. Expectedly, KPNA2, MEX3A, and WDR62 were predominantly overexpressed in several HCC cell lines including Huh-7, Hep3B, HepG2, SK-Hep1, and PLC compared with L-02 normal human liver cells (Fig. 5D–E and Supplementary Fig. 5A). It was noted that the SFN expression pattern in HCC cell lines was not consistent with its expression in HCC tissues. This discordance was likely because of differences between HCC cell lines and HCC tumors.26
To further prove the oncogenic potential of these four genes in HCC, we first individually knocked them down in Hep3B and Huh7 cells (Fig. 6A–C and Supplementary Fig. 5B). It was then demonstrated that absence of KPNA2, MEX3A, WDR62, or SFN expression significantly impaired cell proliferation as documented by the CCK8 assay (Fig. 6D and Supplementary Fig. 5C), colony formation assay (Fig. 6E and Supplementary Fig. 5D), EdU staining (Fig. 6F). In addition, the Transwell migration assays showed that the migratory capacity of HCC cells was significantly weakened by KPNA2, MEX3A, WDR62, or SFN silencing (Fig. 6G). Overall, the results showed that high expression of KPNA2, MEX3A, WDR62, and SFN contributed to the proliferation and migration of HCC cells.
KPNA2, WDR62, MEX3A, and SFN contribute to HCC development dependent on their involvement in SG regulation
SG formation could be regarded as a protective mechanism for tumors against harsh environments such as hypoxia, acidity, and reactive oxygen species.4 As SGs are promoted in vitro by sodium arsenite, which induces oxidative stress, and protein misfolding,27 we next asked whether KPNA2, MEX3A, WDR62, and SFN were recruited to SGs and involved in their formation. To test this hypothesis, we first used the SG marker, G3BP1,28 to establish a cell line that stably expressed G3BP1-GFP, Hep3B-G3BP1. As shown in Figure 7A and Supplementary Figure 6, sodium arsenite-induced G3BP1 puncta with optimal condition of 500 µM arsenite for 30 m without affecting cell viability. Interestingly, KPNA2, MEX3A, and WDR62, but not SFN appeared in these large SG foci at the same time. More significantly, after KPNA2, MEX3A, WDR62, or SFN were knocked down in Hep3B-G3BP1 cells, arsenite treatment failed to induce SG puncta in any of those cells compared with normal GFP-G3BP1-expressing cells. (Fig. 7B–C). The underlying mechanism might be the favorable role of these four proteins on G3BP1 mRNA stability, as their depletion significantly reduced G3BP1 mRNA expression (Fig. 7D).
SG participates in a wide variety of signaling pathways controlling cancer cell proliferation and invasion.29 Therefore, it was expected that the knockdown of G3BP1 from SGs would disrupt their fundamental structure and weaken their tumor-promoting activity. In line with this, G3BP1 knockdown (Supplementary Figs. 7A–B and 8A–B) not only significantly decreased SG formation under arsenic stress conditions (Supplementary Figs. 7C and 8C) but also significantly inhibited HCC cell proliferation and migration (Supplementary Figs. 7D–E and 8D–E). Collectively, the results revealed that KPNA2, MEX3A, WDR62, and SFN had protumorigenic activity through regulation of G3BP1-dependent SG formation.
Discussion
SGs are intracellular compartments formed in response to disadvantageous factors, and the pathogenesis of several diseases including cancer is associated with aberrantly active SG formations.29 Several key tumorigenic drivers, such as mTOR, HDAC6, and RAS oncogene have been shown to promote SG formation.5 The evidence shows that SGs function to orchestrate various oncogenic signals and external stimuli to mediate cancer cell proliferation, invasion, and metastasis. Therefore, it can be expected that SGs might be a promising prognostic and therapeutic target for cancer. To date, many SG-associated proteins have been identified, but no studies have used SGGs as a marker of clinical outcomes of HCC patients. Because we found the vast majority of SGGs were upregulated and associated with poor prognosis in HCC, it was interesting to investigate whether SGG-based methods acted as an independent prognostic tool to accurately predict the outcomes of HCC.
We found that 34 of 463 SGGs were differentially expressed in TCGA-HCC cohorts with a threshold of |log2 (FC)|>2. Surprisingly, only one differentially expressed gene was downregulated, and the remaining 33 were upregulated. We analyzed the expression pattern of the 463 SGGs in six other HCC cohorts and observed that many more differentially expressed SGGs were upregulated, which was similar to the findings in the TCGA-HCC cohort. To discover biomarkers of HCC prognosis, Cox regression models were applied for survival analysis in this study. Survival analysis is used to analyze the length of time between a start point (e.g., the detection of cancer) and an endpoint (e.g., mortality).30 The effects of predictor variables such as age, sex, and tumor stage on the occurrence of a specific event such as death can be evaluated by Cox regression analysis.31 Univariate Cox regression analysis evaluates the association of single factors with patient survival. Multivariate Cox regression simultaneously evaluates the effects of different factors on patient survival. Hazard ratios (HRs) reflect the probability of a particular event such as death associated with specific predictor variable(s) over time.30 Following univariate Cox regression and multivariate Cox regression analyses, a four-SGG signature consisting of KPNA2, MEX3A, WDR62, and SFN was established using data extracted from the TCGA-HCC cohort. The performance of our signature in prognostic prediction was further validated in another HCC cohort, which further demonstrated its effectiveness and reliability. Because the deterioration of liver function contributes to poor HCC outcomes, in addition to gene expression-based predictive indexes, clinical indexes reflecting liver function, such as bilirubin, albumin, and alkaline phosphatase levels have also been reported to be independent biomarkers for HCC prognosis.32,33 Thus, the integrative evaluation of gene expression and biochemical parameter-based predictive methods would be more precise in HCC prognosis.
Through binding to the target protein nuclear localization signal, KPNA2 was initially thought to mediate protein nucleus importation. Later, it was found to be localized in SGs and involved in SG assembly.34 KPNA2 is highly expressed in various tumors and has tumor-promoting activity mediated by nuclear transport of cancer-associated cargo proteins such as c-Myc, PLAG1, and SMARCC1.35 The oncogenic function of KPNA2 relating to SG formation demonstrated by us provides an alternative mechanism to explain its tumorigenic activity. WDR62 is localized in the nucleus and high expression in prostate and gastric cancer has been associated with decreased patient survival time.36,37 Although WDR62 was reported to be recruited to SGs to promote JNK activation, whether it participated in SG formation has not been assessed because of technological limitations.38 Consistent with previous findings, we detected recruitment of WDR62 to SGs; but found that WDR62 was required for SG assembly probably by stabilization of G3BP1 mRNA. We also found for the first time that WDR62 was upregulated and served as a diagnostic marker in HCC. MEX3A, an RNA-binding protein, is overexpressed in multiple cancers like pancreatic, breast, and lung.39 It was reported to promote cancer proliferation by stabilizing CDK6 or LAMA2 mRNA.40 Bioinformatic analysis and experimental validation confirmed that MEX3A was overexpressed in HCC. It was also detected in SG and was required for SG formation, which provides another novel explanation of its oncogenic activity. As a member of the 14-3-3 family, SFN participates in diverse cellular activities such as protein stability, mitotic translation, and cell death.41 Similar to our results, SFN expression was previously found to be increased in HCC and to have predictive accuracy when used in HCC diagnostics.42 However, pathogenic mechanisms of SFN in HCC were not explored in those studies. Although SFN has already been defined as a component of SG by SG purification and proteomic analysis,43 our results did not detect colocalization of SFN and SGs. Interestingly, the formation of SGs was still significantly attenuated after silencing SFN expression. The observed SG formation defects could account for the downregulation of the G3BP1 in mRNA levels, with a similar reason for the effects of KPNA2, MEX3A, and WDR62 knocking down on SG formation. Given the pivotal role of G3BP1 in SGs, the detailed molecular mechanisms underlying these four SGGs in G3BP1 regulation require further investigations.
The clinical information summarized from different datasets used in this research (Supplementary Table 8) corroborates the finding that more than 90% of HCC appears in patients with cirrhosis, which is attributed to various chronic liver diseases such as chronic hepatitis B virus (HBV) or hepatitis C virus (HCV) infection, alcohol abuse, and fatty liver disease.44 These different HCC etiologies are associated with distinct genetic alterations or gene expression modulations that contribute to HCC development.45,46 For example, TP53 and ACVR2V, two tumor suppressors, are most frequently mutated in HBV and NASH-related HCC, respectively.45,47 Moreover, substantial differences in hepatocarcinogenesis have been observed in HBV and HCV-infected HCC.48 It has often been reported that NASH-HCC or NAFLD-HCC, have unique molecular features that drive HCC progression.47,49 The results support the view that HCC is a highly heterogeneous disorder with complex etiologies, and HCC without any chronic liver diseases is rare in the clinic.50 Therefore, defining the intrinsically oncogenic mechanisms of HCC, which is not ascribed to any risk factors, is not easily evaluated in practice. An alternative approach to finding unique HCC-related genes was likely to compare tumor and adjacent liver tissues from HCC patients with specific chronic liver diseases such as HBV infection or NAFLD. TCGA database, well-known for its comprehensive cancer information worldwide, is widely used in gene expression and prognostic analysis. Although the TCGA-HCC cohort has a relatively large sample size (n=374), the different pathological types make the sample number in each pathological classification relatively small, which is not conducive to subsequent biomarker analysis under the specific etiologic condition. Therefore, many gene expression-based HCC prognosis studies frequently analyze the differences between HCC and noncancer samples without pathological classification in advance when adopting the TCGA-HCC database; after verification in another HCC cohort, the reliability and accuracy of these prognostic gene signatures for HCC could be conclusively determined.17,51
In this study, using similar strategies, we developed a four-SG-related gene signature in HCC prognosis and progression. The study has several limitations. First, the sample size was not large enough owing to resource limitations that may limit the generalization of our results. Second, information such as tumor etiologies, causes of death, and baseline clinical characteristics was lacking in the available datasets. That limited subgroup analysis based on different clinical features. An HCC database with a larger sample size, detailed clinical information, and specific etiologies should be used to investigate SGG-related HCC prognostic signatures, which might facilitate the establishment of more precise prognosis biomarkers suitable for particular HCC subtypes.
Conclusion
We established a four-SGG signature that accurately predicted survival outcomes of HCC patients. All four genes facilitated SG formation that contributed to HCC development. To our knowledge, this is the first study to decipher the molecular mechanisms underlying a predictive signature in HCC by the combination of bioinformatic analysis and experimental validation. This novel SGG signature might assist in the clinical diagnosis and treatment of HCC.
Supporting information
Supplementary Table 1
Liver function of patients in the TCGA-HCC database.
(DOCX)
Supplementary Table 2
ALT and AFP levels of patients in GSE 14520 database.
(DOCX)
Supplementary Table 3
shRNA targets.
(DOCX)
Supplementary Table 4
Primer sequences.
(DOCX)
Supplementary Table 5
Primary antibodies.
(DOCX)
Supplementary Table 6
Thirty-four SGGs differentially expressed in HCC and normal tissue.
(DOCX)
Supplementary Table 7
Association of the gene expression of 200 SGGs with OS in the TCGA-HCC cohort.
(DOCX)
Supplementary Table 8
Etiology of HCC patients in the TCGA/GSE/CHCC-HBV/ICGC database.
(DOCX)
Supplementary Fig. 1
Relationship between SFN gene expression levels and tumor grade.
SFN gene expression levels in different pathological stages (A), T stages (B), and tumor grades (C). ns, no significance.
(JPG)
Supplementary Fig. 2
KPNA2, MEX3A, WDR62, and SFN gene expression level in HCC.
Diagram prepared with the Expression DIY tab in the GEPIA2.24 Dark colors indicate increased gene expression and vice versa.
(JPG)
Supplementary Fig. 3
OS of HCC patients with different KPNA2, WDR62, MEX3A, or SFN expression.
The OS (A) and disease-free survival (B) of HCC patients with elevated or low KPNA2 expression. (C–D) OS (C) and disease-free survival (D) of HCC patients with elevated or low WDR62 expression. (E–F) OS (E) and disease-free survival (F) of HCC patients when MEX3A expression is high or low. (G–H) The OS (G) and disease-free survival (H) of HCC patients when SFN expression is high or low. OS, overall survival.
(JPG)
Supplementary Fig. 4
(A) Kaplan-Meier analysis of the OS in the ICGC-HCC cohort. High- and low-risk individuals with HCC are distributed utilizing the median risk score. (B–C) the distribution of risk scores and patient survival status in the high- and low-risk groups. (D) The expression of the four-SGG signature in the high- and low-risk groups.
Prognostic performance of the four-SGG signature in an ICGC-HCC cohort.
(JPG)
Supplementary Fig. 5
KPNA2, MEX3A, WDR62, and SFN knockdown mitigated cell proliferation in Huh7 cells.
(A) mRNA levels of KPNA2, MEX3A, WDR62, and SFN in L-02 human normal liver cells and different human hepatoma cell lines. (B) KPNA2, MEX3A, WDR62, and SFN shRNA knockdown in Huh7 cells. (C) The impacts of KPNA2, MEX3A, WDR62, and SFN deletion on Huh7 cell growth were evaluated by CCK8 assay. (D) Colony formation assays were conducted in Huh7 cells after KPNA2, MEX3A, WDR62, or SFN knockdown. *p<0.05; **p<0.01; ***p<0.001; ****p<0.0001; ns, not significant.
(JPG)
Supplementary Fig. 6
Temporal dependency of SG formation in arsenite-treated Hep3B-GFP-G3BP1cells.
SG formation by liver cell imaging after 500 µM arsenite. Scale bar, 25 µm.
(JPG)
Supplementary Fig. 7
Effect of G3BP1 knockdown on SG assembly and tumor cell progression in Hep3B cells.
(A–B) The silent efficiency of shG3BP1 was verified by qPCR (A) and immunofluorescence (B) in Hep3B cells. Scale bar, 10 µm. (C) The number of stress granules per cell was evaluated under the treatment of arsenite for 30min in G3BP1 knockdown Hep3B cells. Scale bar, 10 µm. (D) The effect of G3BP1 deletion on Hep3B cell proliferation was evaluated by CCK8 assay. Data are presented relative to 24 h. (E) Cell migration was tested in Hep3B cells by Transwell assay after G3BP1 knockdown. Scale bar, 100 µm. *p<0.05; **p<0.01; ***p<0.001; ****p<0.0001.
(PDF)
Supplementary Fig. 8
Effect of G3BP1 knockdown on SG assembly and tumor cell progression in Huh7 cells.
(A–B) The silent efficiency of shG3BP1 was verified by qPCR (A) and immunofluorescence (B) in Huh7 cells. Scale bar, 10 µm. (C) The number of stress granules per cell was evaluated after arsenite treatment for 30 m in G3BP1 knockdown Huh7 cells. Scale bar, 10 µm. (D) Effect of G3BP1 deletion on Huh7 cell proliferation evaluated by CCK8 assay. Data are relative to 24 h. (E) Cell migration assayed by Transwell assay after G3BP1 knockdown in Huh7 cells. Scale bar, 100 µm. *p<0.05; **p<0.01; ***p<0.001; ****p<0.0001; ns, not significant.
(PDF)
Abbreviations
- AUC:
area under the curve
- BCA:
bicinchoninic acid
- BRCA:
breast cancer
- CCK:
cell counting kit
- EdU:
5-ethynyl-2-deoxyuridine
- G3BP1:
ras GTPase-activating protein-binding protein 1
- GEO:
gene expression omnibus
- GO:
gene ontology
- GSEA:
gene set enrichment analysis
- HBV:
hepatitis B virus
- HCC:
hepatocellular carcinoma
- HCV:
hepatitis C virus
- HRP:
horseradish peroxidase
- HRs:
hazard ratios
- ICGC:
international cancer genome consortium
- IF:
immunofluorescence
- KEGG:
kyoto encyclopedia of genes and genomes
- KIRC:
kidney renal clear cell carcinoma
- LIHC:
human liver hepatocellular carcinoma
- LUAD:
lung adenocarcinoma
- LUSC:
lung squamous cell carcinoma
- MSGP:
mammalian stress granules proteome
- N/A:
not available
- NAFLD:
nonalcoholic fatty liver disease
- OS:
overall survival
- OV:
ovarian cancer
- PPI:
protein-protein interaction
- PRAD:
prostate adenocarcinoma
- PVDF:
polyvinylidene difluoride
- qPCR:
quantitative real-time polymerase chain reaction
- ROC:
receiver operating characteristic
- ROS:
reactive oxygen species
- RBPs:
RNA-binding proteins
- SDS-PAGE:
sodium dodecyl-sulfate polyacrylamide gel electrophoresis
- SGs:
stress granules
- SGGs:
stress granule genes
- TCGA:
the cancer genome atlas
- WB:
western blot
Declarations
Ethical statement
All experiments involving human tissues were approved and supervised by the Medical Ethics Committee of Shandong Provincial Hospital Affiliated to Shandong First Medical University (approval number: SWYX: NO. 2019-202.) according to the latest version of the Declaration of Helsinki. Informed consent was obtained from all participants.
Data sharing statement
Data supporting the findings of this study are included in the supplementary information files.
Funding
This research was funded by the National Natural Science Foundation of China (32200557 to DW), the “Outstanding University Driven by Talents” Program and Academic Promotion Program of Shandong First Medical University (2019LJ007 to JZ), the Natural Science Foundation of Shandong Province (ZR2022QH271 to DW), and the Postdoctoral Innovative Projects of Shandong Province (SDCX-ZG-202203047 to DW).
Conflict of interest
The authors have no conflict of interests related to this publication.
Authors’ contributions
Study concept and design (DW, JZ), acquisition of data (DW, XF), analysis and interpretation of data (DW, ML, XF), drafting of the manuscript (DW, ML), critical revision of the manuscript for important intellectual content (DW, JZ), generation of the tables and figures (DW, ML), and study supervision (DW, JZ). All authors have contributed significantly to this study and approved the final manuscript.