Skip to main content
NCBI home page
Journal of Cancer Research and Clinical Oncology logoLink to Journal of Cancer Research and Clinical Oncology
. 2023 Aug 7;149(15):14081–14100. doi: 10.1007/s00432-023-05239-3

Characterization of stem cell subtypes and prognostic signature in hepatocellular carcinoma

Chenjie Qiu 1,✉,#, Huili Wu 2,#, Wenxiang Shi 3,✉
PMCID: PMC11798313  PMID: 37548770

Abstract

Background

Cancer stem cells (CSCs) were linked to cancer aggressiveness and poor prognosis in patients with hepatocellular carcinoma (HCC).

Methods

We integrated two external HCC cohorts to develop the stem cell subtypes according to unsupervised clustering with 26 stem cell gene sets. Between the subtypes, differences in prognosis, clinical characteristics, recognized HCC subtypes, metabolic profile, immune-related features, somatic mutation, and drug sensitivity were examined. The prognostic signature was created, and validated by numerous cohorts, and used to assess the efficacy of immunotherapy and transcatheter arterial chemoembolization (TACE) treatment. The nomogram was developed based on the signature and clinical features. We further examined the function of KIF20A in HCC and proved that KIF20A had the potential to regulate the stemness of HCC cells through western blot.

Results

Low stem cell patterns, a good prognosis, positive clinical features, specific molecular subtypes, low metastatic characteristics, and an abundance of metabolic and immunological aspects were associated with Cluster 1, whereas Cluster 2 was the reverse. Chemotherapy and immunotherapy were more effective in Cluster 1. Cluster 1 and CTNNB1 and ALB mutation were more closely. Additionally, the prognosis, immunotherapeutic, and TACE therapy responses were all worse in the high-risk group. The nomogram could predict the survival probability of HCC patients. KIF20A was discovered to be overexpressed in HCC and was revealed to be connected to the stemness of the HepG2 cell line.

Conclusions

Two stem cell subgroups with different prognoses, metabolic, and immunological characteristics in HCC patients were identified. We also created a 7-gene prognostic signature and a nomogram to estimate the survival probability. The function of KIF20A in HCC stemness was initially examined.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00432-023-05239-3.

Keywords: Hepatocellular carcinoma, Stem cell, Molecular subtype, Prognostic signature, KIF20A

Introduction

The third most common cause of cancer-related fatalities worldwide is primary liver cancer, which is also the sixth most frequent tumor type. Hepatocellular carcinoma (HCC) is the most prevalent histologic type of primary liver cancer, accounting for more than 90% of cases (Sung et al. 2021; Villanueva 2019; Llovet et al. 2021). There are many risk factors for HCC, of which viral hepatitis, particularly infection with hepatitis B virus, is the most well-known cause in China (Tanaka et al. 2011). The symptoms of HCC are nonspecific in the early stage, so it is frequently detected at an advanced stage. Radical surgery, chemoradiotherapy, targeted therapy, immunotherapy, and transcatheter arterial chemoembolization (TACE) are the main forms of treatment for HCC (Llovet et al. 2021). Despite substantial advancements in the effectiveness of systemic therapy, the majority of HCC patients have a dismal prognosis. Therefore, finding a suitable method for HCC prognosis prediction and early diagnosis is essential.

Cancer stem cells (CSCs), which were first identified in acute myeloid leukemia, function as leukemia-initiating cells and are indicated by CD34+ CD38− surface marker status (Lapidot et al. 1994; Bonnet and Dick 1997). The ability of CSCs to self-renew, differentiate, and regenerate all of the characteristics of tumors allows them to drive tumor heterogeneity (Lee et al. 2022; Yamashita and Wang 2013). According to the previous studies, CSCs are linked to malignant behaviors, such as the start of tumorigenesis, resistance to chemotherapy and radiotherapy, tumor recurrence, and metastasis( Batlle and Clevers 2017; Prager et al. 2020). The extensive capacity to promote malignant phenotypes makes CSCs promising as an important therapeutic target in tumor research (Pardal et al. 2003).

Stephanie identified and characterized liver CSCs with a CD133 surface phenotype that possessed higher proliferation efficiency than traditional CSCs in vitro and vivo (Ma et al. 2007). Patients with elevated CD133 expression had worse overall survival and tumor recurrence rates (Song et al. 2008). The main surface markers of CSCs in HCC are CD24, CD44, CD90, CD117, CD133, and EpCAM, and high expression of these markers is associated with increased aggressiveness of HCC and a worse prognosis for patients (Lee et al. 2022). Yang et al. found that tumor angiogenesis and a poor prognosis were linked to CK19, ABCG2, CD133, nestin, and CD44 in HCC (Yang et al. 2010). Due to the close relationship between HCC and CSCs, we concentrated on classifying HCC patients based on stem cell signatures to assess prognosis.

In this study, we used a public database to gather 26 stem cell-related gene sets, and we used the ssGSEA algorithm to assess the relative expression of stem cell markers in large sample cohorts of HCC patients. Unsupervised clustering analysis separated the 578 HCC patients into two distinct stem cell subtypes. We compared the two subtypes’ prognosis, clinical parameters, association with other recognized HCC subtypes, metabolic profile, malignant phenotype, immune infiltration features, immunomodulators, immunotherapy and chemotherapy response, and somatic mutation. We next developed and validated the prognostic signature based on differential gene expression analysis between the two HCC subtypes using several datasets, and we assessed the signature’s effectiveness in the immunotherapy and TACE treatment cohorts. Multiple databases were used to investigate the key role of KIF20A in HCC, and the hypothesis that KIF20A regulates stemness was confirmed in vitro using the HepG2 cell line.

Materials and methods

Data collection

Transcriptome data, clinical features, and somatic mutation data of HCC were retrieved from the TCGA database (https://portal.gdc.cancer.gov/), and the expression profile of TCGA-LIHC in FPKM format was converted to TPM format for further data merging. GEO database (http://www.ncbi.nlm.nih.gov/geo) was used to download information about the clinical traits and gene expression profile for GSE14520 and GSE54236. Transcriptome profile of CHCC-HBV cohort was acquired from the NODE database (https://www.biosino.org/node), and the related clinical data were taken from the previous study (Gao et al. 2019). We also used the ICGC database (https://dcc.icgc.org/) to download the clinical information for the LIRI-JP cohort as well as the gene expression matrix. The combat function of the R package “sva” was used for batch effects elimination and expression profile data combination.

Six immunotherapy cohorts were used to study immunotherapy response, including five immune checkpoint blockade (ICB) cohorts and one adoptive T-cell treatment cohort. The ICB cohorts were as follows: the advanced clear cell renal cell carcinoma (KIRC) cohort from Braun’s study (Braun et al. 2020), the lung cancer cohort from GSE135222 dataset, and the three melanoma cohorts, namely VanAllen’s study (Allen et al. 2015), Nathanson’s study (Nathanson et al. 2017) and GSE91061 dataset. GSE100797 dataset was used to download the data for the adoptive T-cell treatment cohort. GSE104580 was a cohort containing the information on the therapeutic response to TACE in HCC patients. The details of all cohorts in this study are listed in Table S1.

Stem cell gene sets’ acquisition and stem cell subtypes’ identification

StemChecker (http://stemchecker.sysbiolab.eu/), a database that supplied the human or mouse signature associated with stemness, including TF target genes, expression profiles, RNAi screens, literature curation, and computationally derived, was used to derive 26 stem cell gene sets. We used ssGSEA to quantitatively assess the stemness characteristics of 578 HCC samples from TCGA-LIHC and GSE14520 based on the gene sets. Then, using the 26 kinds of stemness enrichment scores of the HCC samples, stem cell subtypes were identified using the R package “ConsensusClusterPlus”, and the k value was determined by consensus CDF and delta area change.

Prediction of various known HCC subtypes

Nearest Template Prediction (NTP) was a module in GenePattern for prediction sample molecular subtypes based on a list of marker genes. Gene profiles of TCGA-LIHC and GSE14520 were uploaded for HCC subtypes prediction based on the published studies, including Boyault’s subtypes (G1&G2, G3, G5&G6), Chiang’s subtypes (CTNNB1, Interferon, Polysomy, Proliferation, Unannotated), Desert’s subtypes (ECM/STEM, Periportal, Perivenous), Yang’s subtypes (C1, C2, C3), and Hoshida’s subtypes (S1, S2, S3) (Boyault et al. 2007; Chiang et al. 2008; Desert et al. 2017; Yang et al. 2020a; Hoshida et al. 2009). Additionally, TCGA classification (iCluster1, iCluster2, iCluster3), immune-related subtypes (immune subtypes and TME subtypes) were derived from the previous studies (Cancer Genome Atlas Research Network 2017; Bagaev et al. 2021).

Estimation of metabolism pattern, classic biological processes, and tumorigenesis functional states

The Kyoto Encyclopedia of Genes and Genomes (KEGG) database was used to identify metabolism-related pathways, and the R package “KEGGREST” was used to download the human main metabolism pathways (amino acid, carbohydrate, cofactors and vitamins, energy, glycan, lipid, and xenobiotics metabolism). The hallmarks of classic biological processes (BP) were obtained from the previous study (Mariathasan et al. 2018). CancerSEA (http://biocc.hrbmu.edu.cn/CancerSEA/), a single-cell level database, sought to thoroughly decode various functional states of cancer cells, and from this database, we obtained the hallmark profiles of phenotypes associated with malignant tumors. GSVA algorithm was used to assess the metabolic pathways and malignant phenotype, whereas ssGSEA algorithm was used to estimate BP.

Tumor microenvironment (TME) evaluation

578 HCC samples were examined with the R package “ESTIMATE” to determine the stromal, immune, estimate scores, and tumor purity. Additionally, the immune enrichment score of HCC samples was calculated using the ssGSEA algorithm based on 29 types of immune cells and functions. The remaining five algorithms, like EPIC, MCPCOUNTER, QUANTISEQ, TIMER, and XCELL, were also applied to evaluate the immune cells infiltration. The TIP database (http://biocc.hrbmu.edu.cn/TIP/) and GSVA algorithm were used to evaluate the seven steps of the cancer-immunity cycle, consisting of release of cancer cell antigens, cancer antigen presentation, priming and activation, trafficking of immune cells to tumors, infiltration of immune cells into tumors, recognition of cancer cells by T cells, and killing of cancer cells.

Immunomodulators and immunotherapy prediction

The TISIDB database (http://cis.hku.hk/TISIDB/) yielded five different categories of immunomodulators, including chemokines, receptors, MHC molecules, immunoinhibitors, and immunostimulators. The TCIA database (https://tcia.at/home), providing thorough immunogenomic analysis for solid tumors from the TCGA database, was used to compare the immunotherapy response between the HCC subtypes based on PD1 and CTLA4 treatment. The TIDE database (http://tide.dfci.harvard.edu/) can estimate various biomarkers to predict a patient’s response to immunotherapy based on the expression profiles. We used the database to forecast the TCGA cohort’s immunotherapy response.

Tumor mutation burden (TMB) and mutation-related parameters

The top 30 genes with the highest mutation rates for the two HCC subtypes were displayed in waterfall plots using the R package “maftools”. TMB was estimated and compared between the two subtypes in the TCGA cohort. The previous study also yielded information on mutation-related parameters, such as DNA damage measures [copy number variation (CNV) burden, aneuploidy score, loss of heterozygosity (LOH), homologous recombination deficiency (HRD), and intra-tumor heterogeneity (ITH)], malignant tumor behavior (wound healing, proliferation), etc. (Thorsson et al. 2018).

GSEA and chemotherapy drugs’ sensitivity

Utilizing the aforementioned gene sets, GSEA was carried out using the biological process and KEGG pathways retrieved from MSigDB, with the criteria being p < 0.05, q < 0.25, and |NES|> 1. By measuring the half maximum inhibitory concentration (IC50), the chemotherapy sensitivity was assessed. Through the R package “pRRophetic” based on the GDSC database (https://www.cancerrxgene.org/), we compared the IC50 of various chemotherapy drugs for HCC between the two subtypes.

Construction of the stem cell signature and the nomogram

We first conducted a differential expression analysis between the two stem cell subtypes in the TCGA cohort to build the stem cell signature in HCC to predict the prognosis of patients. To identify the differentially expressed (DE) genes associated with patients’ OS, univariate Cox regression analysis was conducted. The R package “glmnet” was used to further screen fewer genes and avoid overfitting through lasso regression based on the DE prognosis-related genes, and tenfold cross-validation was utilized. Finally, the independent genes were found using multivariate Cox analysis, and the stem cell signature was created. The risk score was calculated as the sum of the expression levels of the included genes multiplied by their corresponding coefficients.

Additionally, we confirmed the stem cell signature’s predictive significance in four independent HCC cohorts, including GSE14520, LIRI-JP, CHCC-HBV, and GSE54236. To evaluate the prognostic usefulness of the signature in both the TCGA training cohort and external validation cohorts, survival analysis and time-ROC analysis at 1, 2, and 3 years were used. Univariate and multivariate Cox analyses were applied to evaluate whether the signature was an independent risk factor for HCC patients. In addition, the nomogram was conducted based on the signature and the clinical features.

Exploration of gene expression-based stemness index (mRNAsi)

The stemness characteristics of the TCGA-LIHC cohort were retrieved from the previous study (Malta et al. 2018). We examined the relationship between mRNAsi and clinical variables, stem cell subtypes, and stem cell signature. Using the R package “survminer”, the prognostic significance of mRNAsi in HCC was also investigated.

Expression and prognosis analysis of KIF20A in HCC

In the TCGC-LIHC cohort, the expression of KIF20A mRNA in normal and tumor tissues was compared. To validate the expression difference between adjacent and HCC tissues, we also used the HCCDB (http://lifeome.net/database/hccdb/home.html), an integrative molecular database of HCC. KIF20A protein expression was evaluated in the UALCAN database (http://ualcan.path.uab.edu/), while the HPA database (https://www.proteinatlas.org/) provided the typical immunohistochemistry pictures in normal and HCC tissues. The association between KIF20A mRNA expression and immune subtypes, molecular subtypes, and clinical factors was examined using the TISIDB. The prognostic value of KIF20A in HCC patients was evaluated using Kaplan–Meier (KM) curves, which were taken from the KM plotter database (http://kmplot.com/analysis/), and ROC curve.

Correlation between KIF20A and stem-related markers

Human HCC cell line HepG2 was cultured in Dulbecco’s modified Eagle’s medium (DMEM), supplemented with 10% fetal bovine serum (FBS) and 1% penicillin–streptomycin, and cultured in 5% CO2 incubator at 37 ℃. Small interfering RNA (siRNA) against KIF20A (si-KIF20A) and the negative control (NC) were synthesized by HanBio Technology (Shanghai, China). The primer and siRNA sequences were derived from the previous studies (Nakamura et al. 2020; Wu et al. 2021) and are shown in Table S2. According to the protocol, Lipofectamine 3000 (Invitrogen, USA) was used to transfect siRNA into HepG2 cell growing in 6-well plates at 70–80% confluence, and the knockdown effectiveness of KIF20A expression was assessed by Real-Time Polymerase Chain Reaction (RT-PCR) and Western blot (WB), which were previously described (Qiu et al. 2021; Shi et al. 2020). The primary antibodies and secondary antibodies used in this study are listed in Table S3.

Statistical analysis

The correlations between continuous variables were examined using the Pearson test or Spearman test. The differences of continuous variables between two groups were compared using the Mann–Whitney U test. The differences between category variables were examined using Chi-square and Fisher’s exact tests. The prognostic significance between the two groups was described using the KM method and the log-rank test. All statistical analyses were conducted using R software, and p < 0.05 was considered a statistical difference.

Results

Identification of two stem cell subtypes in HCC

We gathered information for a total of 578 HCC samples from the TCGA and GEO databases to determine the probable stem cell expression pattern in HCC. We eliminated the batch effect among various datasets. The distribution of gene expression in TCGA-LIHC and GSE14520 is shown before and after batch effect correction (Fig. 1A, B). The scores of each HCC sample were then quantitatively evaluated using the ssGSEA algorithm based on the stem cell gene sets. Clustering analysis was used to determine two distinct stem cell subtypes (Fig. 1C). According to principal component analysis (PCA) and t-distributed stochastic neighbor embedding (tSNE) methods, the sample distribution across the two subtypes was considerably different (Fig. 1D). The heatmap of the stem cell scores of the two subtypes showed that Cluster 2 had higher stem cell scores than Cluster 1 (Fig. 1E). In addition, KM analysis demonstrated that Cluster 1 in the combination cohort, TCGA cohort, and GSE14520 cohort had longer overall survival (OS) and recurrence-free survival (RFS) than Cluster 2 (Fig. 1F–G). The above results suggested that high stem cell scores may be related to a worse prognosis.

Fig. 1.

Fig. 1

Open in a new tab

Identification of stem cell subtypes based on the transcriptome data of HCC samples in TCGA and GEO datasets. A The distribution of the gene expression in TCGA-LIHC and GSE14520 before the batch effect correction. B The distribution of the gene expression after combination the expression profiles. C Consensus clustering matrix when k = 2. D PCA and tSNE showing the sample distribution difference between the two stem cell subtypes. E The expression of 26 stem cell gene sets between the two subtypes. Cluster 1 group had longer OS time (F) and longer RFS time (G) than Cluster 2 in the combination, TCGA and GSE14520 cohorts, respectively

The correlation of stem cell subtypes with clinical parameters and known HCC subtypes

Based on the clinical parameters of HCC, we examined the connection between the two subtypes and relevant traits. In the TCGA cohort, Cluster 1 was associated with old age (age > 55), small tumor size (T1-2), low pathological stage (stage I–II), low histologic grade (G1-2), and low-serum AFP level (AFP ≤ 300 ng/ml), while Cluster 2 was associated with young age (age ≤ 55), large tumor size (T3-4), high pathological stage (stage III–IV), high histologic grade (G3-4), and high-serum AFP level (AFP > 300 ng/ml). In the GSE14520 cohort, the subtypes were related to serum AFP levels (Fig. 2A, B).

Fig. 2.

Fig. 2

Open in a new tab

Correlation between stem cell subtypes and clinical parameters and various known HCC subtypes. A The distribution of clinical characteristics and HCC subtypes in our stem cell subtypes in TCGA cohort. B The distribution of clinical characteristics and HCC subtypes in our stem cell subtypes in GSE14520 cohort

We also contrasted our stem cell subtypes with several reported HCC subtypes. In the TCGA cohort, Cluster 1 was related to iCluster 2, inflammatory and TGF-beta dominant, F and IE/F types, while Cluster 2 was related to iCluster 1, iCluster 3, IFN-gamma dominant, lymphocyte deficient and wound healing, D and IE types (Fig. 2A). In the GSE14520 cohort, Cluster 1 exhibited a lower metastatic signature score than Cluster 2 (Fig. 2B). In addition, the outcomes of the NTP module subtype prediction revealed that Cluster 1 was significantly associated with Boyault’s G5 and G6, Chiang’s CTNNB1 class, Desert’s Periportal type, Yang’s metabolism subtype C1, and Hoshida’s S3 type, while Cluster 2 was significantly associated with Boyault’s G3 type, Chiang’s proliferation and interferon classes, Desert’s ECM/STEM type, Yang’s metabolism subtype C2-3, and Hoshida’s S1-2 type in both cohort (the TCGA and GSE14520 cohorts) (Figs. 2A, B, S1A, B).

The correlation of HCC subtypes with metabolic characteristics and malignant behaviors

Eighty-four metabolism-relevant gene signatures were acquired to further investigate the connection between the various HCC subtypes and metabolic pathways. The relative scores of each patient were then estimated using the GSVA method. The findings showed that, with the exception of GPI-anchor, N-glycan, O-glycan, steroid biosynthesis, and sphingolipid metabolism, practically all seven types of metabolism pathways were enriched in Cluster 1 (Fig. 3A). Additionally, from the previous study and the single-cell database, signatures of tumor malignant activity, such as traditional BP and tumorigenesis functional states, were collected. Cluster 2 had strong enrichment of signatures related to the cell cycle, FGFR3-related genes, homologous recombination, and mismatch repair (Fig. 3B). Cluster 2 was also connected to apoptosis, DNA damage, DNA repair, and proliferation via the CancerSEA database (Fig. 3C).

Fig. 3.

Fig. 3

Open in a new tab

Correlation of the subtypes with metabolic characteristics and malignant behaviors. A Heatmap of the metabolism pattern (amino acid, carbohydrate, cofactors, energy, glycan, lipid, and xenobiotics) between the two stem cell subtypes according to the KEGG database. B Comparison of traditional biological processes between the two subtypes. C The tumorigenesis functional states in the two subtypes

Comparison of the immune landscape of the HCC subtypes

Across multiple tumor types, it was discovered that the immune landscape varied greatly in terms of molecular subtypes or prognostic signatures. As a result, we concentrated on the immune cells and functions in the HCC subtypes using a variety of commonly used methods. Cluster 1 had higher stromal and ESTIMATE scores, while Cluster 2 had higher tumor purity (Fig. 4A). We then calculated the infiltration of immune or stromal cell scores using six additional methods. Most immune or stromal cells, especially B cells, cancer-associated fibroblasts (CAFs), macrophages, etc., were highly abundant in Cluster 1 (Fig. 4B). Cancer immunity cycles and immunomodulators were tightly associated with immune cell infiltration; thus, we also examined these associations in the subtypes. Cancer antigen presentation (step 2) and killing of cancer cells (step 7) were upregulated in Cluster 1, and trafficking of T cells to tumors (step 4) and infiltration of T cells into tumors (step 5) were down-regulated in Cluster 1 (Fig. 4B). Moreover, there were observable differences between the HCC subtypes in the expression of the majority of immunomodulators (Fig. 4C).

Fig. 4.

Fig. 4

Open in a new tab

Immune pattern between the two stem cell subtypes. A Immune, stromal, estimate scores, and tumor purity calculated by ESTIMATE between the two subtypes. B Heatmap of the immune pattern (cancer-immunity cycle and immune cells) between the two subtypes estimated by EPIC, MCPCOUNTER, QUANTISEQ, ssGSEA, TIMER, and XCELL. C The expression levels of five kinds of immunomodulators (chemokines, receptors, MHC molecules, immunoinhibitors, and immunostimulators) between the two subtypes

Using the TCIA and TIDE databases, we also examined the immunotherapy response between various subtypes in the TCGA cohort. The results demonstrated that Cluster 1 had a better immunotherapy response than Cluster 2, regardless of anti-PD1 and/or anti-CTLA4 treatment status (Fig. S2A). The typical distribution of patients based on immunophenoscores (high vs. low) is shown in Fig. S2B. Additionally, we discovered that patients in Cluster 1 had higher immunotherapy response rates, MSI indices, CAF infiltration levels, and dysfunction scores, while patients in Cluster 2 had higher TIDE scores, MDSC and TAM M2 infiltration levels, and exclusion scores (Fig. S2C).

Somatic variations between the two HCC subtypes in the TCGA cohort

We retrieved somatic mutation data from the TCGA database to compare the gene mutations of the two subtypes. There was no statistically significant variation in TMB between the two subtypes (Fig. 5A). High TMB in the TCGA cohort was linked to a shorter OS based on the best cut-off value (Fig. 5B). People with high TMB in Cluster 2 might have a worse prognosis than those with low TMB in Cluster 1 (Fig. 5C). The top 30 genes with the highest mutation frequency in Clusters 1 and 2 are displayed in waterfall plots (Fig. 5D). CTNNB1 and ALB mutation rates were higher in patients in Cluster 1, while TP53, MUC4, and AXIN1 mutation rates were higher in patients in Cluster 2 (Fig. 5E, F). In addition, Cluster 2 had a higher segment number, alteration frequency, aneuploidy score, CTA score, HRD score, and ITH score and a lower IFN-gamma response (Fig. 5G). Cluster 2 was also had high scores for proliferation and wound healing (Fig. 5H).

Fig. 5.

Fig. 5

Open in a new tab

Mutation status between the two stem cell subtypes in the TCGA cohort. A Comparison of TMB between Cluster 1 and Cluster 2. B Survival difference between the two subtypes. C KM survival analysis among four groups divided by stem cell subtypes and TMB. D Oncoprint of the 30 genes with the highest mutation rates in Cluster 1 and Cluster 2, respectively. E Comparison of mutation rate of TP53, CTNNB1, MUC4, AXIN1, and ALB between the two subtypes. F Histograms of mutation status of the above five genes in Cluster 1 and Cluster 2. Violin plots of DNA damage measures, CTA score, IFN-gamma response (G), and malignant tumor features (H) between the two subtypes

Enrichment analysis and drug sensitivity analysis of the HCC subtypes

GSEA was carried out on the basis of BP and KEGG gene sets to clarify the potential mechanistic differences between the two HCC subtypes. In contrast to Cluster 2, which was associated with meiotic pathways such as the cell cycle and nuclear division, Cluster 1 was mostly enriched in pathways related to metabolism (Fig. 6A). In addition, KEGG analysis yielded similar results: Cluster 1 was linked to metabolic processes and Cluster 2 was linked to tumor development and progression (Fig. 6B).

Fig. 6.

Fig. 6

Open in a new tab

Enriched pathways analysis and chemotherapeutic response prediction for the two stem cell subtypes. A Enrichment analysis of biological processes in Cluster 1 and Cluster 2. B KEGG pathways enriched in the two subtypes. C Chemotherapeutic drug sensitivity analysis based on IC50 using pRRophetic

To evaluate the drug sensitivity of the two HCC subtypes, we also determined the IC50 of various chemotherapeutic agents. The IC50 values for cisplatin, doxorubicin, gemcitabine, paclitaxel, vinblastine, and other common HCC treatments were lower in Cluster 1, indicating that Cluster 1 patients were more responsive to chemotherapy drugs (Fig. 6C).

Construction of the prognostic signature based on the HCC stem cell subtypes

We then created a prognostic signature based on the stem cell subtypes discovered in the TCGA cohort to differentiate the prognosis of HCC patients. First, 269 stem cell-related genes with differential expression were discovered by contrasting Clusters 1 and 2 (Table S4). A total of 182 genes related to OS were then chosen using univariate Cox regression (Table S5). Using LASSO analysis, a gene set of 13 genes was further tested (Fig. 7A). Finally, the 7-gene signature was obtained using multivariate Cox analysis, and the coefficient values of these genes are shown (Fig. 7B). The heatmap of the signature revealed that Cluster 2 had higher expression levels of signature genes (Fig. 7C). Moreover, a general positive correlation between the seven genes was discovered, according to correlation analysis (Fig. 7D).

Fig. 7.

Fig. 7

Open in a new tab

Identification of a stem cell signature for prediction of OS in the TCGA cohort. A LASSO regression model of the prognostic related DE genes in the TCGA cohort. B Coefficient values of the selected seven genes using multivariate Cox analysis. C Heatmap of the expression of the seven genes between the two stem cell subtypes. D Correlation of the seven genes in the TCGA cohort. E PCA plot of the sample distribution between the high- and low-risk groups. F KM analysis of the stem cell signature in the TCGA training cohort. G ROC analysis of the signature at 1, 2, and 3 years in the training cohort. H Univariate and multivariate Cox analyses showing the signature as an independent risk factor for HCC patients. I Nomograms used to predict clinical outcomes after 1, 3, and 5 years based on the signature and clinical features. J Calibration curve of 1, 3, and 5 years. K Comparison of risk scores stratified by stem cell subtypes, gender, AFP level, T stage, tumor differentiation degree, and tumor stages

We determined the high-risk and low-risk groups using the optimal cut-off value of the risk score. The two groups’ distributions were distinct (Fig. 7E). Patients at high risk had noticeably shorter OS according to KM analysis (Fig. 7F), and ROC analysis showed that the area under the curve (AUC) values of the signature at 1, 2, and 3 years were 0.779, 0.744, and 0.736, respectively, indicating that the signature had good prognostic value (Fig. 7G). Simultaneously, stratified prognosis analysis based on clinical factors, such as age, sex, T stage, N stage, tumor grade, tumor stage, cirrhosis, and AFP level, was carried out. The results revealed that the prognostic signature was effective in separating the high- and low-risk groups according to OS (Fig. S3A–H). Additionally, univariate and multivariate Cox analyses showed that the signature was an independent risk factor for HCC (Fig. 7H). We then constructed a nomogram that incorporated both clinical features and the signature to predict outcomes (Fig. 7I). To validate our nomogram, we analyzed its calibration diagram, which demonstrated that the predicted OS was highly consistent with the observed OS at 1, 3, and 5 years (Fig. 7J). According to the correlation between the signature and clinical factors, the risk score was higher in Cluster 2, female patients, and patients with AFP > 300 ng/ml. In addition, the risk score increased with increasing tumor T stage, tumor grade, and tumor stage (Fig. 7K).

Validation of the signature with four external datasets

To validate the prognostic predictive value of the signature, we collected four independent external datasets. We determined the risk score for each patient in the four cohorts using the formula of the signature. The findings revealed that in the validation cohorts, patients in the high-risk group were more likely to have shorter OS (Fig. 8A–D). We examined the predictive value of the signature by time-dependent ROC analysis, and the AUC values at 1, 2, and 3 years showed that the signature could successfully predict the prognosis of HCC patients (Fig. 8E–H). We also discovered that in these validation cohorts, the risk group identified by the signature may function as an independent risk factor (Fig. 8I–L). The association between the signature and the tumor volume doubling time, which was determined using the imaging data from the GSE54236 cohort, was further investigated. The outcome showed that the rate of tumor growth increased with the risk score (Fig. 8M).

Fig. 8.

Fig. 8

Open in a new tab

Validation of the stem cell signature in the four independent external datasets. A–D KM curves showing the OS differences between the high- and low-risk groups in GSE14520, LIRI-JP, CHCC-HBV, and GSE54236 cohorts, respectively. E–H Time-dependent ROC curves indicating the significance of the signature in predicting the 1-, 2-, and 3-year OS in the above four cohorts. I–L Univariate and multivariate Cox analyses demonstrating the signature as an independent risk factor for HCC patients in the external cohorts. M KM analysis of the calculated tumor doubling time between the high- and low-risk groups in GSE54236 cohort

The predictive value of the signature in immunotherapy and TACE treatment

The role of the stem cell signature in anticipating the immunotherapy response in several malignancies, including KIRC, melanoma, and lung cancer, was subsequently investigated. We found that the high-risk group had a much lower immunotherapy response rates than the low-risk group (Fig. S4A–G). Additionally, the prognosis was worse in the high-risk group, which may be connected to the lower response rates (Fig. S4A–G). Information on the therapeutic response to TACE in HCC patients was contained in the GSE104580 dataset. Our findings demonstrated that patients who responded to TACE treatment had lower risk scores, indicating a better prognosis (Fig. S4H). The signature’s AUC value was 0.711, showing that it may accurately predict how HCC patients will respond to TACE treatment (Fig. S4I).

Relationship between the stem cell subtypes, stem cell signature, and mRNAsi

To determine the connection between mRNAsi and the stem cell subtypes and the signature, the mRNAsi values of the TCGA-LIHC cohort were obtained from a previous study. We found that as the mRNAsi increased, the proportion of patients in Cluster 2 gradually increased, whereas the proportion of patients in Cluster 1 steadily declined. The low-risk group primarily had low mRNAsi values, while the high-risk group primarily had high mRNAsi values (Fig. S5A). The boxplots allowed us to confirm that high mRNAsi values were more common in Cluster 2 and the high-risk group (Fig. S5B, C). The results of the correlation analysis revealed that the risk score and mRNAsi were strongly positively correlated (Fig. S5D). Furthermore, according to the optimal cut-off value of mRNAsi, which was 0.42 (Fig. S5E), we discovered that HCC patients with low mRNAsi had longer OS than patients with high mRNAsi (Fig. S5F).

Expression difference, prognostic value, and stemness role of KIF20A in HCC

KIF20A was identified as the most important gene in the signature, as it had the highest coefficient among the seven genes, and we therefore verified the role of KIF20A in HCC and its potential link with stemness in HCC. When compared to that in normal tissues, the expression of KIF20A mRNA was higher in HCC tissues (Fig. 9A, B). To confirm the differences in KIF20A mRNA expression between adjacent and HCC tissues, we also analyzed data from the HCCDB, and considerable overexpression of KIF20A was found in virtually all datasets (Fig. 9C). The KIF20A protein level in HCC was subsequently discovered to be comparable to the mRNA level (Fig. 9D). The HPA database yielded typical immunohistochemistry results, showing that KIF20A was strongly expressed in HCC (Fig. 9E). The link between KIF20A expression and subtypes and clinical characteristics demonstrated that KIF20A expression was significantly different across these subtypes, and that KIF20A expression increased as tumor grades and stages increased (Fig. 9F). Using KM and ROC methods, the predictive value of KIF20A was examined. The findings revealed that HCC patients had a poor prognosis (OS, PFS, RFS, and DSS) due to elevated KIF20A expression (Fig. 9G). The AUC values of KIF20A at 1, 2, and 3 years were 0.763, 0.701, and 0.656, respectively, showing that KIF20A was useful for predicting the prognosis of HCC (Fig. 9H).

Fig. 9.

Fig. 9

Open in a new tab

The role of KIF20A in HCC. A The mRNA expression of KIF20A in normal and tumor tissues in the TCGA-LIHC cohort. B The mRNA expression of KIF20A in paired normal and tumor tissues. C The mRNA expression differences of KIF20A between adjacent and HCC tissues in the HCCDB. D The protein expression difference of KIF20A in HCC through the UALCAN. E The typical immunohistochemistry images in HCC from the HPA. F The relationship between KIF20A and immune subtypes, molecular subtypes, tumor grades, and tumor stages from the TISIDB. G The prognostic differences, such as OS, PFS, RFS, and DSS, in HCC according to the KIF20A expression from the KM plotter. H ROC curves of KIF20A at 1, 2, and 3 years in HCC. I The knockdown efficiency of KIF20A in HepG2 cell using RT-PCR. J Western blot showed that the expression of stem-related markers beta-catenin, GATA4, TBX1, NKX2.5, CTNT, and SOX2 was down-regulated in si-KIF20A transfected HepG2 cell. K Western blot showed that the expression of stem-related markers CD13, CD44, EpCAM, and CD133 was down-regulated in si-KIF20A transfected HepG2 cell

Since KIF20A was overexpressed in HCC, we transfected HepG2 cells with siRNA to inhibit KIF20A expression. The RT‒PCR and WB results demonstrated that si-KIF20A-1 and si-KIF20A-2 effectively decreased KIF20A mRNA and protein expression in HepG2 cells (Fig. 9I, J). Next, we used WB to examine the protein expression of stem-related markers, such as beta-catenin, GATA4, TBX1, NKX2.5, CTNT, SOX2, CD13, CD44, EpCAM, and CD133. The results revealed that the levels of these markers decreased after transfection with si-KIF20A, indicating that KIF20A may have an impact on the stem characteristics of HCC cells (Fig. 9J, K).

Discussion

Tumor heterogeneity is one of the important characteristics of malignant tumors, which includes the heterogeneity of the origin of tumor cells and the tumor microenvironment. Tumor heterogeneity is an important factor for poor prognosis and treatment resistance of patients (Yang et al. 2022). HCC is considered to be one of the most heterogeneous malignant tumors. Gao et al. found that even in the intra-tumor of the same HCC patient, the proportion of gene mutation in tumor tissues at different spatial locations also differed significantly (Gao et al. 2017). With the rapid development of high-throughput sequencing technology, this intra-tumor heterogeneity has been assessed at the molecular level. For example, Xue et al. identified five subtypes of immune microenvironment, called TIMELASER system (Xue et al. 2022). Although the molecular classification of HCC is gradually being established, it is still not as complete as that of breast or bladder cancer, and there is no correlation between the different molecular classification system. In addition, no studies have classified HCC based on stem cell gene sets. Therefore, we used specific stem cell gene sets to identify our new HCC subtypes.

Two HCC subtypes with distinct stem cell properties were identified based on the gene sets of stem cells using a public database. Similar investigations have been carried out in other malignancies, such as bladder cancer, gastric cancer, and colorectal cancer (Tang et al. 2020; Xiang et al. 2021; Zheng et al. 2022), but no pertinent investigation has yet been carried out in HCC. We discovered Cluster 1 with low stem cell scores and Cluster 2 with high stem cell scores through clustering analysis. We investigated the prognostic value, metabolic patterns, clinical correlations, immune infiltration, somatic mutation, and drug sensitivity of the two subtypes.

Correlation analysis with known HCC subtypes showed that the Cluster 2 subtype was related to Boyault’s G3 subclass, which was typically characterized by high TP53 mutation frequency and overexpressed genes associated with the cell cycle (Boyault et al. 2007). TP53 mutation was associated with high chromosome instability and a poor prognosis in HCC patients (Laurent-Puig et al. 2001; Liu et al. 2012). Cluster 1 was dominated by the G5 and G6 subclasses and presented a high mutation rate for the CTNNB1 gene encoding beta-catenin. Chiang’s classification also indicated that Cluster 1 was enriched for the CTNNB1 class, with a high CTNNB1 mutation frequency (Chiang et al. 2008). Desert identified four HCC subclasses, namely, ECM, STEM, PP, and PV types, and the latter two were considered well-differentiated, nonproliferative HCC subtypes. Our results suggested that the PP and PV subclasses accounted for the majority of Cluster 1, indicating a more favorable outcome for this cluster. The PP type was linked to less aggressiveness, better survival, and lower recurrence, and the PV type showed a high frequency of predicted CTNNB1 mutations (Desert et al. 2017). Yang et al. classified HCC patients into three metabolism-related subtypes with various metabolic activities, AFP levels, and prognoses (Yang et al. 2020a). Cluster 1 was primarily dominated by Yang’s C1 type, with favorable survival, and Cluster 2 was mainly dominated by Yang’s C3 type, with poor survival. In addition, Hoshida et al. also found three HCC subclasses with different clinical characteristics and activated pathways. The S1 class was related to activation of the WNT signaling pathway, the S2 class was associated with the MYC and AKT pathways, and the S3 class was related to small tumor size, well-differentiated tumors, and low AFP level (Hoshida et al. 2009). Cluster 1 in our study matched Hoshida’s S3 group, indicating that Cluster 1 has a favorable prognosis.

Tumors are thought to have a significantly dysregulated metabolism. Analysis of metabolism in malignancies could clarify the pathogenesis and reveal potential targets for clinical treatment. For instance, obesity can induce intestinal microbial metabolites to promote HCC (Yoshimoto et al. 2013). Ma et al. found that dexamethasone could successfully treat HCC by switching the metabolism of tumor cells from glycolysis to gluconeogenesis (Ma et al. 2013). Metformin could prevent HCC by inhibiting pathways driving hepatic lipogenesis (Bhalla et al. 2012). Metformin-treated HCC patients had an improved prognosis (Schulte et al. 2019). Moreover, taking statins could lower the risk of HCC (McGlynn et al. 2015; Shi et al. 2014). Our research revealed that the two HCC subtypes differed significantly in terms of metabolism. Patients in Cluster 1 are more likely to benefit from metabolic therapy, because Cluster 1 had greater enrichment of metabolism signatures than Cluster 2.

The TME is considered a major factor in the development of tumors and is closely linked to their formation, metastasis, and survival (Hinshaw and Shevde 2019). Tumor immunosuppressive therapies have been shown to increase patient survival times but have poor effects on cold tumors (Oura et al. 2021). Transformation of cold tumors into hot ones or identification of the two tumor types can provide significant perspectives for designing effective anticancer strategies (Zhang et al. 2022). In this investigation, two distinct HCC subtypes with various levels of immune infiltration were found. In particular, Cluster 1 exhibited higher immune cell infiltration, indicating that Cluster 1 may include hot tumors with improved immunotherapy responsiveness. Kurebayashi et al. found that there are three immune subtypes of HCC, and high infiltration of B cells was related to a favorable prognosis, which was consistent with our results (Kurebayashi et al. 2018). Tumor cells and CAFs could upregulate CD163 and CD206 and downregulate IL-6 to induce M2 polarization. Both M2 macrophages and CAFs could promote the proliferation and invasion of HCC (Chen et al. 2021). However, CAFs were found to be more abundant in Cluster 1, which contradicted the good prognosis of this group and the results of the previous study. The possible reason is that Cluster 1 membership was associated with early tumor stage, low AFP levels, and high metabolic activity, and the prognosis might be the result of comprehensive factors.

We also constructed a prognostic signature based on the differentially expressed genes between the two HCC subtypes. KIF20A was identified as one of the hub genes in HCC based on multiple bioinformatics methods, and in vitro experiments demonstrated that knockdown of KIF20A could increase the activities of caspase-3 and caspase-9, suppress cell proliferation by inhibiting the G1/S transition during the cell cycle, and enhance the chemotherapy sensitivity of cisplatin and sorafenib in HCC cells (Wu et al. 2021; Li et al. 2020; Lu et al. 2018). We found that KIF20A could affect the stem characteristics of HCC cells through in vitro experiments. Shi et al. found that Gli2 could directly activate the transcription of FOXM1, further inducing the transcription of KIF20A through the MMB complex. The Gli2-KIF20A axis is important for HCC progression and could be an independent biomarker for HCC prognosis (Shi et al. 2016). EPO and its receptor are involved in angiogenesis in human HCC (Ribatti et al. 2007). Numerous studies have found that EPO can be used in multiple prognostic signatures in HCC (Guo et al. 2022; Huang et al. 2021). EPO is highly expressed in HCC tissues, related to a poor prognosis, and promotes HCC cell proliferation (Miao et al. 2017; Yang et al. 2015). Ke et al. found that the relationship between EPO and worse prognosis might be mediated by respiratory dysfunction secondary to mitochondrial DNA mutations (Ke et al. 2017). In addition, human constitutive androstane receptor (hCAR) is down-regulated in HCC, and hCAR could inhibit the expression of EPO by suppressing the expression and activation of HNF4a, a vital transcription factor for EPO (Li et al. 2022; Sun et al. 2015). Silencing of MAGEA6 inhibits the stemness maintenance and self-renewal of HCC stem cells and suppressed tumor progression of HCC cells through the AMPK signaling pathway (Pineda et al. 2015; Guo et al. 2019). SFN plays an important role in biological processes, such as the cell cycle, signal transduction, proliferation, apoptosis, migration, invasion, differentiation, and metabolism (Chen and Yang 2013). SFN, which could be regulated by TP53, is related to the DNA damage response, and it could arrest the cell cycle in the G2/M phase and maintain genome stability (Chan et al. 1999; Ravi et al. 2011; Meng et al. 2011). SFN serves as an oncogene in HCC according to computational and experimental analyses and was found to be related to tumor grade and a poor prognosis (Yang et al. 2020b; Reis et al. 2015).

This study still has certain limitations. The majority of the results are drawn from public databases, and there are not enough pertinent studies or prospective clinical cohorts to validate them. To verify the classification accuracy and dependability of the prognostic signature, we need to collect additional HCC tissues. The importance of PFN2 and SLC1A7 may also be investigated in the future, because there are few studies on them in HCC.

Conclusions

In conclusion, two stem cell subtypes of HCC were identified by unsupervised clustering, and the prognosis, TME and metabolic features, somatic mutation distribution, and therapeutic response were significantly different between the two subtypes of HCC. According to the stem cell-related genes, a seven-gene prognostic signature was constructed and validated, and it had significant predictive value for immunotherapy and TACE treatment. KIF20A is significantly upregulated in HCC and has the potential to impact the stemness of HCC cells.

Supplementary Information

Below is the link to the electronic supplementary material.

Supplementary file1 (DOCX 2340 KB) (2.3MB, docx)
Supplementary file2 (XLSX 51 KB) (51.5KB, xlsx)

Acknowledgements

The authors would like to thank all the public databases included in this study.

Abbreviations

AUCs

Area under curves

BP

Biological processes

CSCs

Cancer stem cells

CAFs

Cancer-associated fibroblasts

CNV

Copy number variation

DE

Differentially expressed

DSS

Disease-special survival

DMEM

Dulbecco’s modified Eagle’s medium

FBS

Fetal bovine serum

HCC

Hepatocellular carcinoma

HRD

Homologous recombination deficiency

hCAR

Human constitutive androstane receptor

ICB

Immune checkpoint blockade

ITH

Intra-tumor heterogeneity

KM

Kaplan–Meier

KEGG

Kyoto encyclopedia of genes and genomes

LOH

Loss of heterozygosity

NTP

Nearest template prediction

NC

Negative control

OS

Overall survival

PCA

Principal component analysis

PFS

Progression-free survival

RT-PCR

Real-time polymerase chain reaction

RFS

Recurrence-free survival

siRNA

Small interfering RNA

tSNE

T-distributed stochastic neighbor embedding

TACE

Transcatheter arterial chemoembolization

TME

Tumor microenvironment

TMB

Tumor mutation burden

WB

Western blot

Author contributions

CQ: supervision, project administration, data curation, and writing—original draft. WS: formal analysis, data curation, investigation, writing—original draft. HW: investigation, and writing—original draft.

Funding

Not applicable.

Availability of data and materials

The data are available from the corresponding author for reasonable requests.

Declarations

Conflict of interest

The authors declare that they have no competing interests.

Ethical approval and consent to participate

It was approved by Changzhou Hospital of Traditional Chinese Medicine’s Research Ethics Committee.

Informed consent

Not applicable.

Footnotes

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Chenjie Qiu and Huili Wu contributed equally to this work.

Contributor Information

Chenjie Qiu, Email: qiuchenjie2020@163.com.

Wenxiang Shi, Email: swx1995@sjtu.edu.cn.

References

  1. Bagaev A, Kotlov N, Nomie K, Svekolkin V, Gafurov A, Isaeva O et al (2021) Conserved pan-cancer microenvironment subtypes predict response to immunotherapy. Cancer Cell 39:845-865 e7 [DOI] [PubMed] [Google Scholar]
  2. Batlle E, Clevers H (2017) Cancer stem cells revisited. Nat Med 23:1124–1134 [DOI] [PubMed] [Google Scholar]
  3. Bhalla K, Hwang BJ, Dewi RE, Twaddel W, Goloubeva OG, Wong KK et al (2012) Metformin prevents liver tumorigenesis by inhibiting pathways driving hepatic lipogenesis. Cancer Prev Res (phila) 5:544–552 [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Bonnet D, Dick JE (1997) Human acute myeloid leukemia is organized as a hierarchy that originates from a primitive hematopoietic cell. Nat Med 3:730–737 [DOI] [PubMed] [Google Scholar]
  5. Boyault S, Rickman DS, de Reynies A, Balabaud C, Rebouissou S, Jeannot E et al (2007) Transcriptome classification of HCC is related to gene alterations and to new therapeutic targets. Hepatology 45:42–52 [DOI] [PubMed] [Google Scholar]
  6. Braun DA, Hou Y, Bakouny Z, Ficial M, Sant’ Angelo M, Forman J et al (2020) Interplay of somatic alterations and immune infiltration modulates response to PD-1 blockade in advanced clear cell renal cell carcinoma. Nat Med 26:909–918 [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Cancer Genome Atlas Research Network (2017) Electronic address wbe, Cancer Genome Atlas Research N. Comprehensive and integrative genomic characterization of hepatocellular carcinoma. Cell 169:1327-1341 e23 [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Chan TA, Hermeking H, Lengauer C, Kinzler KW, Vogelstein B (1999) 14-3-3Sigma is required to prevent mitotic catastrophe after DNA damage. Nature 401:616–620 [DOI] [PubMed] [Google Scholar]
  9. Chen L, Yang B (2013) 14-3-3 sigma is a useful immunohistochemical marker for diagnosing ovarian granulosa cell tumors and steroid cell tumors. Int J Gynecol Pathol 32:156–162 [DOI] [PubMed] [Google Scholar]
  10. Chen S, Morine Y, Tokuda K, Yamada S, Saito Y, Nishi M et al (2021) Cancer-associated fibroblast-induced M2-polarized macrophages promote hepatocellular carcinoma progression via the plasminogen activator inhibitor-1 pathway. Int J Oncol. 10.3892/ijo.2021.5239 [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Chiang DY, Villanueva A, Hoshida Y, Peix J, Newell P, Minguez B et al (2008) Focal gains of VEGFA and molecular classification of hepatocellular carcinoma. Cancer Res 68:6779–6788 [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Desert R, Rohart F, Canal F, Sicard M, Desille M, Renaud S et al (2017) Human hepatocellular carcinomas with a periportal phenotype have the lowest potential for early recurrence after curative resection. Hepatology 66:1502–1518 [DOI] [PubMed] [Google Scholar]
  13. Gao Q, Wang ZC, Duan M, Lin YH, Zhou XY, Worthley DL et al (2017) Cell culture system for analysis of genetic heterogeneity within hepatocellular carcinomas and response to pharmacologic agents. Gastroenterology 152:232-242 e4 [DOI] [PubMed] [Google Scholar]
  14. Gao Q, Zhu H, Dong L, Shi W, Chen R, Song Z et al (2019) Integrated proteogenomic characterization of HBV-related hepatocellular carcinoma. Cell 179(561–577):e22 [DOI] [PubMed] [Google Scholar]
  15. Guo JC, Yang YJ, Zhang JQ, Guo M, Xiang L, Yu SF et al (2019) microRNA-448 inhibits stemness maintenance and self-renewal of hepatocellular carcinoma stem cells through the MAGEA6-mediated AMPK signaling pathway. J Cell Physiol 234:23461–23474 [DOI] [PubMed] [Google Scholar]
  16. Guo C, Tang Y, Yang Z, Li G, Zhang Y (2022) Hallmark-guided subtypes of hepatocellular carcinoma for the identification of immune-related gene classifiers in the prediction of prognosis, treatment efficacy, and drug candidates. Front Immunol 13:958161 [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Hinshaw DC, Shevde LA (2019) The tumor microenvironment innately modulates cancer progression. Cancer Res 79:4557–4566 [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Hoshida Y, Nijman SM, Kobayashi M, Chan JA, Brunet JP, Chiang DY et al (2009) Integrative transcriptome analysis reveals common molecular subclasses of human hepatocellular carcinoma. Cancer Res 69:7385–7392 [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Huang C, Zhang C, Sheng J, Wang D, Zhao Y, Qian L et al (2021) Identification and validation of a tumor microenvironment-related gene signature in hepatocellular carcinoma prognosis. Front Genet 12:717319 [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Ke S, Chen S, Dong Z, Hong CS, Zhang Q, Tang L et al (2017) Erythrocytosis in hepatocellular carcinoma portends poor prognosis by respiratory dysfunction secondary to mitochondrial DNA mutations. Hepatology 65:134–151 [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Kurebayashi Y, Ojima H, Tsujikawa H, Kubota N, Maehara J, Abe Y et al (2018) Landscape of immune microenvironment in hepatocellular carcinoma and its additional impact on histological and molecular classification. Hepatology 68:1025–1041 [DOI] [PubMed] [Google Scholar]
  22. Lapidot T, Sirard C, Vormoor J, Murdoch B, Hoang T, Caceres-Cortes J et al (1994) A cell initiating human acute myeloid leukaemia after transplantation into SCID mice. Nature 367:645–648 [DOI] [PubMed] [Google Scholar]
  23. Laurent-Puig P, Legoix P, Bluteau O, Belghiti J, Franco D, Binot F et al (2001) Genetic alterations associated with hepatocellular carcinomas define distinct pathways of hepatocarcinogenesis. Gastroenterology 120:1763–1773 [DOI] [PubMed] [Google Scholar]
  24. Lee TK, Guan XY, Ma S (2022) Cancer stem cells in hepatocellular carcinoma - from origin to clinical implications. Nat Rev Gastroenterol Hepatol 19:26–44 [DOI] [PubMed] [Google Scholar]
  25. Li X, Huang W, Huang W, Wei T, Zhu W, Chen G et al (2020) Kinesin family members KIF2C/4A/10/11/14/18B/20A/23 predict poor prognosis and promote cell proliferation in hepatocellular carcinoma. Am J Transl Res 12:1614–1639 [PMC free article] [PubMed] [Google Scholar]
  26. Li Z, Kwon SM, Li D, Li L, Peng X, Zhang J et al (2022) Human constitutive androstane receptor represses liver cancer development and hepatoma cell proliferation by inhibiting erythropoietin signaling. J Biol Chem 298:101885 [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Liu J, Ma Q, Zhang M, Wang X, Zhang D, Li W et al (2012) Alterations of TP53 are associated with a poor outcome for patients with hepatocellular carcinoma: evidence from a systematic review and meta-analysis. Eur J Cancer 48:2328–2338 [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Llovet JM, Kelley RK, Villanueva A, Singal AG, Pikarsky E, Roayaie S et al (2021) Hepatocellular carcinoma. Nat Rev Dis Primers 7:6 [DOI] [PubMed] [Google Scholar]
  29. Lu M, Huang X, Chen Y, Fu Y, Xu C, Xiang W et al (2018) Aberrant KIF20A expression might independently predict poor overall survival and recurrence-free survival of hepatocellular carcinoma. IUBMB Life 70:328–335 [DOI] [PubMed] [Google Scholar]
  30. Ma S, Chan KW, Hu L, Lee TK, Wo JY, Ng IO et al (2007) Identification and characterization of tumorigenic liver cancer stem/progenitor cells. Gastroenterology 132:2542–2556 [DOI] [PubMed] [Google Scholar]
  31. Ma R, Zhang W, Tang K, Zhang H, Zhang Y, Li D et al (2013) Switch of glycolysis to gluconeogenesis by dexamethasone for treatment of hepatocarcinoma. Nat Commun 4:2508 [DOI] [PubMed] [Google Scholar]
  32. Malta TM, Sokolov A, Gentles AJ, Burzykowski T, Poisson L, Weinstein JN et al (2018) Machine learning identifies stemness features associated with oncogenic dedifferentiation. Cell 173:338-354 e15 [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Mariathasan S, Turley SJ, Nickles D, Castiglioni A, Yuen K, Wang Y et al (2018) TGFbeta attenuates tumour response to PD-L1 blockade by contributing to exclusion of T cells. Nature 554:544–548 [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. McGlynn KA, Hagberg K, Chen J, Graubard BI, London WT, Jick S et al (2015) Statin use and risk of primary liver cancer in the Clinical Practice Research Datalink. J Natl Cancer Inst 107:djv009 [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Meng L, Hsu JK, Tsai RY (2011) GNL3L depletion destabilizes MDM2 and induces p53-dependent G2/M arrest. Oncogene 30:1716–1726 [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Miao S, Wang SM, Cheng X, Li YF, Zhang QS, Li G et al (2017) Erythropoietin promoted the proliferation of hepatocellular carcinoma through hypoxia induced translocation of its specific receptor. Cancer Cell Int 17:119 [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Nakamura M, Takano A, Thang PM, Tsevegjav B, Zhu M, Yokose T et al (2020) Characterization of KIF20A as a prognostic biomarker and therapeutic target for different subtypes of breast cancer. Int J Oncol 57:277–288 [DOI] [PubMed] [Google Scholar]
  38. Nathanson T, Ahuja A, Rubinsteyn A, Aksoy BA, Hellmann MD, Miao D et al (2017) Somatic mutations and neoepitope homology in melanomas treated with CTLA-4 blockade. Cancer Immunol Res 5:84–91 [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Oura K, Morishita A, Tani J, Masaki T (2021) Tumor immune microenvironment and immunosuppressive therapy in hepatocellular carcinoma: a review. Int J Mol Sci 22:5801 [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Pardal R, Clarke MF, Morrison SJ (2003) Applying the principles of stem-cell biology to cancer. Nat Rev Cancer 3:895–902 [DOI] [PubMed] [Google Scholar]
  41. Pineda CT, Ramanathan S, Fon Tacer K, Weon JL, Potts MB, Ou YH et al (2015) Degradation of AMPK by a cancer-specific ubiquitin ligase. Cell 160:715–728 [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Prager BC, Bhargava S, Mahadev V, Hubert CG, Rich JN (2020) Glioblastoma stem cells: driving resilience through chaos. Trends Cancer 6:223–235 [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Qiu C, Shi W, Wu H, Zou S, Li J, Wang D et al (2021) Identification of molecular subtypes and a prognostic signature based on inflammation-related genes in colon adenocarcinoma. Front Immunol 12:769685 [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Ravi D, Chen Y, Karia B, Brown A, Gu TT, Li J et al (2011) 14-3-3 sigma expression effects G2/M response to oxygen and correlates with ovarian cancer metastasis. PLoS ONE 6:e15864 [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Reis H, Putter C, Megger DA, Bracht T, Weber F, Hoffmann AC et al (2015) A structured proteomic approach identifies 14-3-3Sigma as a novel and reliable protein biomarker in panel based differential diagnostics of liver tumors. Biochim Biophys Acta 1854:641–650 [DOI] [PubMed] [Google Scholar]
  46. Ribatti D, Marzullo A, Gentile A, Longo V, Nico B, Vacca A et al (2007) Erythropoietin/erythropoietin-receptor system is involved in angiogenesis in human hepatocellular carcinoma. Histopathology 50:591–596 [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Schulte L, Scheiner B, Voigtlander T, Koch S, Schweitzer N, Marhenke S et al (2019) Treatment with metformin is associated with a prolonged survival in patients with hepatocellular carcinoma. Liver Int 39:714–726 [DOI] [PubMed] [Google Scholar]
  48. Shi M, Zheng H, Nie B, Gong W, Cui X (2014) Statin use and risk of liver cancer: an update meta-analysis. BMJ Open 4:e005399 [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Shi C, Huang D, Lu N, Chen D, Zhang M, Yan Y et al (2016) Aberrantly activated Gli2-KIF20A axis is crucial for growth of hepatocellular carcinoma and predicts poor prognosis. Oncotarget 7:26206–26219 [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Shi W, Lu J, Li J, Qiu M, Lu Y, Gu J et al (2020) Piperlongumine attenuates high calcium/phosphate-induced arterial calcification by preserving P53/PTEN signaling. Front Cardiovasc Med 7:625215 [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Song W, Li H, Tao K, Li R, Song Z, Zhao Q et al (2008) Expression and clinical significance of the stem cell marker CD133 in hepatocellular carcinoma. Int J Clin Pract 62:1212–1218 [DOI] [PubMed] [Google Scholar]
  52. Sun W, Kosyna FK, Jelkmann W, Depping R (2015) Prolyl-4-Hydroxylase 2 Potentially Contributes to Hepatocellular Carcinoma-Associated erythrocytosis by maintaining hepatocyte nuclear factor-4alpha expression. Cell Physiol Biochem 37:2257–2264 [DOI] [PubMed] [Google Scholar]
  53. Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A et al (2021) Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 71:209–249 [DOI] [PubMed] [Google Scholar]
  54. Tanaka M, Katayama F, Kato H, Tanaka H, Wang J, Qiao YL et al (2011) Hepatitis B and C virus infection and hepatocellular carcinoma in China: a review of epidemiology and control measures. J Epidemiol 21:401–416 [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Tang C, Ma J, Liu X, Liu Z (2020) Development and validation of a novel stem cell subtype for bladder cancer based on stem genomic profiling. Stem Cell Res Ther 11:457 [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. Thorsson V, Gibbs DL, Brown SD, Wolf D, Bortone DS, Ou Yang TH et al (2018) The immune landscape of cancer. Immunity 48:812-830 e14 [DOI] [PMC free article] [PubMed] [Google Scholar]
  57. Van Allen EM, Miao D, Schilling B, Shukla SA, Blank C, Zimmer L et al (2015) Genomic correlates of response to CTLA-4 blockade in metastatic melanoma. Science 350:207–211 [DOI] [PMC free article] [PubMed] [Google Scholar]
  58. Villanueva A (2019) Hepatocellular carcinoma. N Engl J Med 380:1450–1462 [DOI] [PubMed] [Google Scholar]
  59. Wu C, Qi X, Qiu Z, Deng G, Zhong L (2021) Low expression of KIF20A suppresses cell proliferation, promotes chemosensitivity and is associated with better prognosis in HCC. Aging (albany NY) 13:22148–22163 [DOI] [PMC free article] [PubMed] [Google Scholar]
  60. Xiang R, Song W, Ren J, Wu J, Fu J, Fu T (2021) Identification of stem cell-related subtypes and risk scoring for gastric cancer based on stem genomic profiling. Stem Cell Res Ther 12:563 [DOI] [PMC free article] [PubMed] [Google Scholar]
  61. Xue R, Zhang Q, Cao Q, Kong R, Xiang X, Liu H et al (2022) Liver tumour immune microenvironment subtypes and neutrophil heterogeneity. Nature 612:141–147 [DOI] [PubMed] [Google Scholar]
  62. Yamashita T, Wang XW (2013) Cancer stem cells in the development of liver cancer. J Clin Invest 123:1911–1918 [DOI] [PMC free article] [PubMed] [Google Scholar]
  63. Yang XR, Xu Y, Yu B, Zhou J, Qiu SJ, Shi GM et al (2010) High expression levels of putative hepatic stem/progenitor cell biomarkers related to tumour angiogenesis and poor prognosis of hepatocellular carcinoma. Gut 59:953–962 [DOI] [PubMed] [Google Scholar]
  64. Yang Z, Sun B, Zhao X, Shao B, An J, Gu Q et al (2015) Erythropoietin and erythropoietin receptor in hepatocellular carcinoma: correlation with vasculogenic mimicry and poor prognosis. Int J Clin Exp Pathol 8:4033–4043 [PMC free article] [PubMed] [Google Scholar]
  65. Yang C, Huang X, Liu Z, Qin W, Wang C (2020a) Metabolism-associated molecular classification of hepatocellular carcinoma. Mol Oncol 14:896–913 [DOI] [PMC free article] [PubMed] [Google Scholar]
  66. Yang L, Zhang Z, Sun Y, Pang S, Yao Q, Lin P et al (2020b) Integrative analysis reveals novel driver genes and molecular subclasses of hepatocellular carcinoma. Aging (albany NY) 12:23849–23871 [DOI] [PMC free article] [PubMed] [Google Scholar]
  67. Yang C, Zhang S, Cheng Z, Liu Z, Zhang L, Jiang K et al (2022) Multi-region sequencing with spatial information enables accurate heterogeneity estimation and risk stratification in liver cancer. Genome Med 14:142 [DOI] [PMC free article] [PubMed] [Google Scholar]
  68. Yoshimoto S, Loo TM, Atarashi K, Kanda H, Sato S, Oyadomari S et al (2013) Obesity-induced gut microbial metabolite promotes liver cancer through senescence secretome. Nature 499:97–101 [DOI] [PubMed] [Google Scholar]
  69. Zhang J, Huang D, Saw PE, Song E (2022) Turning cold tumors hot: from molecular mechanisms to clinical applications. Trends Immunol 43:523–545 [DOI] [PubMed] [Google Scholar]
  70. Zheng H, Liu H, Li H, Dou W, Wang J, Zhang J et al (2022) Characterization of stem cell landscape and identification of stemness-relevant prognostic gene signature to aid immunotherapy in colorectal cancer. Stem Cell Res Ther 13:244 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary file1 (DOCX 2340 KB) (2.3MB, docx)
Supplementary file2 (XLSX 51 KB) (51.5KB, xlsx)

Data Availability Statement

The data are available from the corresponding author for reasonable requests.

Articles from Journal of Cancer Research and Clinical Oncology are provided here courtesy of Springer

RESOURCES