Targeting MCM10 disrupts cancer stemness and counteracts sorafenib resistance in hepatocellular carcinoma

Abstract

Hepatocellular carcinoma (HCC) is the most prevalent primary malignant tumor, with sorafenib as the main treatment for advanced cases. However, the development of resistance to sorafenib, often driven by cancer stemness, significantly limits its therapeutic efficacy. Minichromosome maintenance complex component 10 (MCM10), a critical regulator of DNA replication and tumor progression, has been implicated in cancer stemness and therapeutic resistance. This study utilized datasets from TCGA and ICGC alongside in vitro and vivo experiments on clinical HCC tissues and sorafenib-resistant cell lines to evaluate MCM10’s role in HCC. The Connectivity Map (CMap) was employed to identify TW-37, a potential gene silencing agent targeting MCM10 transcription. The effects of TW-37 on MCM10 expression, cancer stemness, and sorafenib sensitivity were assessed. Elevated MCM10 expression was observed in sorafenib-resistant HCC cell lines and was associated with poor patient outcomes. MCM10 knockout diminished cancer stemness and restored sorafenib sensitivity in resistant cells. Furthermore, TW-37, identified via CMap, effectively downregulated MCM10, reduced cancer stemness, and enhanced sorafenib efficacy, offering a promising therapeutic approach. MCM10 plays a pivotal role in promoting cancer stemness and sorafenib resistance in HCC. Targeting MCM10 transcription with TW-37 represents a novel strategy to overcome sorafenib resistance and improve therapeutic outcomes in HCC patients.

Introduction

Hepatocellular carcinoma (HCC) is the most prevalent form of liver cancer globally, primarily due to its high incidence and mortality rates. In China alone, nearly half of all global HCC cases are reported annually, highlighting the significant burden of this disease. Due to late-stage diagnoses in the majority of patients, the prognosis remains poor, with a median survival of only 9 months and a 5-year survival rate of ~20% [1–6]. Sorafenib, a multitarget tyrosine kinase inhibitor (TKI), has been the first-line systemic therapy for advanced HCC. Its mechanisms include antiangiogenic and antiproliferative effects, which have demonstrated a modest extension of median survival in clinical trials [7]. In landmark studies such as the SHARP and Asia-Pacific trials, sorafenib significantly improved outcomes for late-stage HCC patients, sparking widespread clinical interest [8]. Despite these advances, sorafenib benefits only around 30% of patients, and resistance typically develops within 6 months of treatment [9]. While “sorafenib responders” fail to achieve a median recurrence-free survival (RFS) during therapy, “non-responders” exhibit a median RFS of merely 28 months [10]. This resistance severely limits the clinical utility of sorafenib. Mechanistically, the dysregulation of pathways such as phosphatidylinositol 3-kinase (PI3K)/AKT has been implicated in promoting acquired sorafenib resistance by enhancing cancer stemness in HCC cells [11].

Initially conceptualized by Prof. Mackillop in 1983, cancer stem cells represent a subset of tumor cells with high tumorigenicity and stem-like properties. These cells express specific signaling molecules and surface markers, including KLF4, SOX2, and CD133, which activate pathways such as Wnt/β-catenin, Hedgehog, and Notch. These pathways facilitate self-renewal, enhance drug resistance, and drive therapeutic failure [12–14]. Furthermore, tumor stem cells promote epithelial-mesenchymal transition (EMT), which enhances cellular mobility and invasiveness while increasing resistance to anti-cancer therapies, including sorafenib [15, 16]. Although these insights are promising, the precise mechanisms of tumor stem cell-mediated sorafenib resistance remain poorly understood, underscoring the need for further research.

Among the molecular regulators of cancer, minichromosome maintenance (MCM) proteins, particularly MCM10, have garnered attention for their critical role in DNA replication and tumor progression. MCM10 is essential for maintaining DNA replication fork integrity and promoting cellular proliferation. Overexpression of MCM10 has been observed in multiple malignancies, including ovarian, liver, and breast cancers, suggesting its role in tumor development and progression [17–22]. However, its specific involvement in maintaining cancer stemness and drug resistance in HCC remains underexplored.

This study leverages sequencing data from sorafenib-resistant HCC cell lines to identify resistance-associated genes. A prognostic model was constructed and verified using public databases such as TCGA and ICGC, correlating gene expression with patient outcomes, tumor grade, and stage. Among the genes analyzed, MCM10 demonstrated the highest correlation with the cancer stemness index in TCGA-HCC patients, establishing its relevance in this context. Additionally, we utilized the Connectivity Map (CMap) to identify TW-37, a promising small molecule, as a potential therapeutic agent. TW-37 was found to downregulate MCM10 expression, suppress stemness, and restore sorafenib sensitivity in resistant HCC cells. These findings provide novel insights into the molecular relationship between cancer stemness and drug resistance in HCC, paving the way for innovative therapeutic strategies targeting MCM10.

Materials and methods

Collection of public data

RNA-sequencing data and clinical follow-up details from 367 hepatocellular carcinoma (HCC) patients in the Cancer Genome Atlas (TCGA) (https://portal.gdc.cancer.gov/) were used to identify prognostic biomarkers. A training cohort comprising 367 HCC specimens was established to develop a prognostic model. Subsequently, 243 rigorously selected samples from the International Cancer Genome Consortium (ICGC) (https://dcc.icgc.org/projects/LIRI-JP) served as validation cohorts to confirm the efficacy of the prognostic risk model.

Cell culture

The HCC cell lines HepG2 (ATCC: CRL-3533) and Hep3B (ATCC: HB-8064) were obtained from the American Type Culture Collection (ATCC). Both cell lines were cultured in DMEM (Gibco, Carlsbad, CA, United States) supplemented with 10% (v/v) fetal bovine serum (FBS, Biological Industries, Israel) and antibiotics. The cells were maintained at 37 °C in a 5% CO₂ atmosphere. Sorafenib-resistant cell lines (HepG2-SR and Hep3B-SR) were developed by gradually increasing sorafenib concentrations in continuous cultures of HepG2 and Hep3B cells over approximately one and a half years. These cell lines were deemed resistant when the parental cells could proliferate at a concentration of 5 µM. These cell lines were passaged < 10 times after the initial recovery of the frozen stocks.

Plasmids

A lentiviral vector overexpressing MCM10 was generated by cloning MCM10 cDNA into a pLV3-CMV-3×Myc-Neo vector (Miaoling Biology, Hubei, China). Unique single guide RNA (sgRNA) sequences were cloned into a puro-resistant lentiCRISPRv2 construct containing puro resistance for generation of lentiviral vectors for knockout of MCM10 (MCM10 KO#1: 5′-CACCGAACGACTCAACCCATCTGTG-3′; MCM10 KO#2: 5′-CACCGCGGTGAATCTTATACAGAAG-3′). Plasmid transfection was performed using packaging plasmid and ExFectTransfection Reagent (Vazyme, Nanjing, China), following the manufacturer’s instructions. The supernatant-containing virus was collected to infect HepG2 and Hep3B, and the colonies with stable overexpressing or knockout of MCM10 were screened on Geneticin or Puromycin. lentiCRISPR v2 was a gift from Feng Zhang (Addgene plasmid # 52961; http://n2t.net/addgene:52961; RRID:Addgene_52961) [23]. All plasmid vectors were verified via sequencing.

Weighted gene correlation network analysis (WGCNA)

WGCNA was employed to identify highly synergistic genomes and potential biomarker genes, leveraging the endogeneity of the genome and its association with phenotype [24]. The TCGA-HCC cohort was divided into two groups and differentially expressed genes (DEGs) were identified using limma analysis. Subsequently, utilizing the ‘WGCNA’ R package, upregulated genes were partitioned into distinct modules based on their correlations. We correlated the modules with various sample features, including grade, stage, Overall Survival (OS) time, etc. Genes within the modules most correlated with sample traits were then subjected to correlation analysis with candidate genes.

Protein-protein interactions (PPI) networks

Moledule genes from WGCNA were imported into the Retrieval of Interacting Genes (STRING, https://cn.string-db.org/) database, while the PPI network was visualized by the plugin cytoNCA in Cytoscape software (version 3.10).

RNA sequencing

Total RNA from sorafenib-resistant HepG2, Hep3B (HepG2-SR, Hep3B-SR), and indicated parental cells were collected and subjected to RNA sequencing performed by Majorbio Biopharmaceutical Technology (China).

Construction and validation of the sorafenib resistant risk model

Using LASSO regression and the R package “glmnet” [25] candidate genes for the sorafenib resistance (SR) related model were determined through maximum likelihood fitting penalties, thus obtaining SR-related mRNA. Based on the results of LASSO, variables were selected and a predictive resistance model was established using the glmnet R package. ROC curve analysis was performed using the “timeroc” package in R, and visualized using “ggplot2”. The risk score for each patient was calculated as follows: Risk score = Sum (expression level of each candidate gene × corresponding LASSO regression coefficient). Subsequently, TCGA-HCC samples were used as the training set and divided into high-risk and low-risk groups based on median risk scores. Partial least squares discriminant analysis (PLS-DA) was conducted using the R package “mixOmics” to identify intergroup differences. Kaplan-Meier (KM) survival curves were plotted using the “survival” package in R, and risk charts were generated using the “ggrisk” package. Risk factor plots and KM curves were utilized to evaluate the survival status of high-risk and low-risk groups.

Functional enrichment analysis

The “limma” R package was used to identify DEGs between the high- and low-risk groups in TCGA-HCC cohort (|log₂FC| > 1.2, P < 0.05). Risk genes were subjected to gene set enrichment analysis (GSEA) and gene set variation analysis (GSVA) using the “gseaplot2”, “GSVA” package and GSEA software to analyze their biological functions and signal transduction pathways [26].

mRNA expression-based stemness index calculation

The mRNA expression-based stemness index (mRNAsi) was calculated to assess the degree of stemness in hepatocellular carcinoma (HCC) samples [27]. The stemness index was derived using the one-class logistic regression (OCLR) algorithm. The OCLR model was trained using stem cell gene expression profiles to distinguish between stem-like and non-stem-like features in tumor samples. The mRNAsi score for each tumor sample was derived by applying the trained model to its mRNA expression data, yielding a continuous score that reflects the degree of stemness. Higher mRNAsi values indicate greater similarity to stem cell-like gene expression.

Western blot

After digestion, the cells were collected and washed twice in cold phosphate-buffered saline (PBS). Subsequently, they were centrifuged at 15,000 × g for 15 min at 4 °C following lysis on ice for 30 min using lysate buffer (#87787, Thermo Fisher) and a protease inhibitor (#P1010, Beyotime). The BCA assay (#E112-01, Vazyme) determined the protein concentration in the supernatant. The western blot analysis utilized the following antibodies: rabbit anti-human MCM10 antibody (#DF12162, Affinity), rabbit anti-human KLF4 antibody (#11880-1-AP, Proteintech), rabbit anti-human SOX2 antibody (#A0561, Abclonal), mouse monoclonal anti-human AKT antibody (#YM3618, Immunoway), rabbit anti-human p-AKT antibody (#YM8304, Immunoway), mouse monoclonal anti-human PI3K antibody (#60225, Proteintech), rabbit anti-human p-PI3K antibody (#AF3242, Affinity), rabbit anti-human S6K1 antibody (#380469, Immunoway), rabbit anti-human p-S6K1 antibody (#310310, Immunoway), rabbit anti-human 4EBP1 antibody (#R24197, Immunoway), rabbit anti-human p-4EBP1 antibody (#R22929, Immunoway) and mouse monoclonal anti-human β-actin antibody (#AF7018, Affinity).

H&E staining

Paraffin tissue sections were deparaffinized with xylene (twice for 10 min each) and then placed in anhydrous ethanol (twice for 10 min each), 90% ethanol (5 min), 80% ethanol (5 min), 70% ethanol (3 min) and the tissues were washed with primary water. Subsequently, the tissues were sequentially stained with hematoxylin staining solution (3 min), differentiation solution (1 s), eosin solution (15 s), 95% ethanol (5 s), anhydrous ethanol (twice for 1 min each) and xylene (twice for 2 min each). Finally, the slices were sealed with neutral resin and observed by microscope (Carl Zeiss, Suzhou, China).

Immunohistochemistry and immunofluorescence

Immunohistochemical staining was conducted and analyzed according to described previously [26]. For immunofluorescence staining, cells were seeded on glass coverslips in 12-well plates, washed with PBS, and then fixed by 4% paraformaldehyde fixation. Cells were subsequently incubated with PBS containing 0.25% Triton X-100 and 5% BSA for 1 h. Then incubated them with MCM10 antibody (#DF12162, Affinity), CD133 antibody (#A25678, ABclonal, 1:100), and CD44 antibody (#15675-1-AP, Proteintech) overnight at 4 °C in a humidified chamber. Slides were then incubated with fluorescein-conjugated secondary antibodies for 1 h at room temperature in the dark. Nuclei were counterstained with 4’, 6-diamidino-2-phenylindole (DAPI). We used confocal microscopy (Leica Microsystems CMS GmbH, Mannheim, Germany) for imaging.

Tumorsphere formation

Each well of the ultra-low-adhesion 12-well plate was seeded with 2000 cells and then incubated at 37 °C with 5% CO₂ for 7–14 days. The number of tumorspheres with a diameter larger than 50 µm was observed under a microscope for statistical analysis.

Flow cytometry

Cells in the logarithmic growth phase were processed into a single-cell suspension centrifuged for 5 min, and this procedure was repeated twice to discard the supernatant. Subsequently, 1 × 10⁶ cells were resuspended in 100 μL of PBS supplemented with 5% BSA and CD133 primary antibody (#A25678, ABclonal, 1:100), and incubated for 30 min at 4 °C in the dark. Following two rounds of centrifugation in PBS and removal of the supernatant, the cells were resuspended in 100 μL of BSA-diluted fluorescent marker. The cells were then incubated with 100 μL of Fluor488-conjugated Goat Anti-Mouse IgG(H + L) secondary antibody (#S0017, Affinity, 1:100) for 20 min at 4 °C in the dark, centrifuged twice with PBS, and the supernatant was discarded. The cells were finally resuspended in 100 μL of 1 × PBS with 5 μL of PI stain added and incubated for 2 min at 4 °C in the dark. An additional 400 μL of 1 × PBS was added before analysis via BD FACS Lyrics flow cytometer. Data were analyzed using FlowJo software (v10.8.1, FlowJo).

Identification of compounds negatively associated with sorafenib resistance

Differentiated gene expression analysis was conducted using the “edgeR” R package on sequencing data from sorafenib-resistant HCC and parental cell lines. to identify. Subsequently, the top 150 upregulated and downregulated genes were imported into the Connectivity Map (CMap) database (https://clue.io/) to identify compounds that inversely correlated with sorafenib resistance. In the CMap analytical framework, compounds exhibiting lower (more negative) scores demonstrate stronger inverse correlations with sorafenib resistance phenotypes. These agents with pronounced negative associations hold greater potential as candidate therapeutics for reversing sorafenib resistance in HCC.

Cell counting Kit-8 (CCK8) assay and combination index (CI)

Cells were seeded into 96-well plates at 2000 cells per-well density. After 24 h of seeding, the cells were treated with escalating concentrations of sorafenib for 48 h. Ten microliters of CCK-8 reagent (Vazyme, China) were added to each well of the cells and then incubated for 2 h at 37 °C. Absorbance at 450 nm was measured with a Synergy 2 microplate reader (BioTek, USA). Subsequently, the IC50 values for HepG2, Hep3B, HepG2-SR, and Hep3B-SR were visualized using GraphPad Prism (v 9.1.0) (GraphPad Software, San Diego, CA, USA).

To conduct synergy analysis, HepG2-SR/Hep3B-SR HCC cells were seeded at 1,500 cells per well in a 96-well plate and incubated overnight. The cells were treated with TW-37 (HY-12020, Medchemexpress), sorafenib, or a control vehicle for 48 h, both as single agents and in combination. Cell viability was assessed using a CCK-8 assay, and CI was generated with CompuSyn software (Cambridge, UK) to quantify the effects of drug combinations. The drugs’ synergistic, additive, and antagonistic effects were defined by CI values of less than 1, equal to 1, and greater than 1, respectively [28].

In vivo tumorigenesis assay

20 female BALB/c mice, about 4 weeks old, were purchased from Beijing Vital River Laboratory Animal Technology Ltd (China). The animal study protocol was approved by the Ethics Committee of The Second Xiangya Hospital of Central South University (Animal Permit No. 2022027, Human samples Permit No.2023-0315). MCM10-overexpressing HepG2 or MCM10-knockout HepG2-SR cells, along with their corresponding control cell suspensions, were subcutaneously injected 100 μL into the axilla of each nude mouse (2 × 10⁶ cells/mouse). Thereafter, the body weight of the mice was monitored every 5 days. The short and long diameters of the tumors were measured every 5 days starting on day 10 of inoculation, with the volume calculated by the formula: tumor volume (mm³) = 0.5 × (short diameter)² × (long diameter) until day 40 of inoculation when the nude mice were euthanized, and the tumors were exfoliated, photographed, and weighed. Partial tumors were formalin-fixed, embedded in paraffin blocks, and sectioned for subsequent IHC assays.

Statistical analysis

Each experiment was conducted a minimum of three times. Statistical data analysis was performed using the R package (version 4.2.0) or GraphPad Prism 9.0. The Student’s t-test was utilized to compare the two groups. Survival analysis was carried out using Cox hazards regression and the Kaplan-Meier method in this study. Correlation analysis was performed using Spearman’s rank correlation test. The statistical significance of the results was confirmed when P < 0.05.

Results

Identifying and screening of 18 sorafenib-resistance feature genes in hepatocellular carcinoma

Differential expression analysis was conducted using RNA sequencing data from sorafenib-resistant and parental HepG2 and Hep3B cells. In HepG2 cells, 5933 upregulated and 2046 downregulated differentially expressed genes (DEGs) were identified, while Hep3B cells showed 3799 upregulated and 1730 downregulated DEGs (|log₂FC| > 1.5, P < 0.05; Fig. 1A, B). The intersection of DEGs from both cell lines revealed 2110 upregulated and 489 downregulated sorafenib resistance (SR)-related DEGs, which were considered potential candidates for further analysis (Fig. 1C, D).

Fig. 1. Screening of sorafenib resistance feature genes.

A Volcano plot of 7979 DEGs (5933 upregulated genes and 2046 down-regulated genes) associated with sorafenib-resistance (sorafenib-resistant cells versus parental cells) with |log2FC| > 1.5 and P < 0.05 from sequencing data for the cell line HepG2. B Volcano plot of 5529 DEGs (3799 upregulated genes and 1730 down-regulated genes) from sequencing data for Hep3B (|log2FC| > 1.5 and P < 0.05). C, D The Venn diagram illustrates 2110 DEGs commonly upregulated and 489 DEGs commonly downregulated in HepG2 and Hep3B cells. E Heatmap in WGCNA of the correlation between the characteristic genes of the module and the clinical characteristics of HCC. F Correlation analysis of the brown module and clinical traits with grade. The hub genes were elaborated with MM > 0.8 and GS > 0.1. G Protein-protein interaction (PPI) network of 128 genes from the brown module was constructed using Cytoscape 3.10 software on STRING database. H The Forest plot shows univariate regression analysis of the top 20 of 312 risk genes from 2110 intersecting upregulated genes. HR > 1, P < 0.0001. I Volcano plot showing differential gene expression (HCC group versus control group) from TCGA. J The Venn diagram shows four comparisons’ intersections: the hub genes in the brown module from WGCNA, hub protein from PPI network, SR risk genes, and HCC DEGs.

To refine these candidates, a multi-step approach was employed, including weighted gene co-expression network analysis (WGCNA), protein-protein interaction analysis (STRING), univariate Cox regression, and comparison of tumor and control samples from the TCGA database. Using WGCNA, the 2110 upregulated genes were grouped into five co-expression modules: turquoise, brown, blue, yellow, and gray. The brown module exhibited the strongest positively correlation with tumor grade (Fig. 1E). Within this module, 128 genes were found to positively correlate with tumor grade and stage while negatively correlating with overall survival. Based on module membership (MM > 0.8) and gene significance (GS > 0.1), 29 hub genes were identified (Fig. 1F).

Next, the 128 brown module genes were analyzed using the STRING database to assess protein-protein interactions (PPI). The top 30 proteins, ranked by network degree, were selected as key hub proteins (Fig. 1G). These genes were further examined through univariate Cox regression analysis of TCGA clinical data, which identified the first 20 risk genes significantly associated with poor survival (HR > 1, P < 0.001; Fig. 1H).

Finally, differential expression analysis of HCC tumor and control samples from TCGA revealed 1610 upregulated and 2067 downregulated DEGs (Fig. 1I). Integration of results from WGCNA, STRING, univariate regression, and TCGA DEG analysis identified 18 overlapping genes as sorafenib-resistance feature genes (Fig. 1J). These 18 genes demonstrated high expression in HCC, were strongly associated with tumor progression, and served as candidates for prognostic risk modeling. This streamlined approach highlights the significance of these 18 genes in sorafenib resistance and their potential utility in predicting prognosis and therapeutic outcomes in HCC.

Construction and validation of prognostic risk model using machine learning

The logistic least absolute shrinkage and selection operator (LASSO) regression method was employed to develop a robust prognostic model based on the above 18 genes. This machine learning approach, which incorporates L1 regularization for feature selection, identified nine candidate genes strongly associated with HCC prognosis: SASS6, MCM10, WDHD1, ECT2, FBXO5, PRR11, CENPI, ATAD5, and KIF20B (Fig. 2A). A sorafenib-resistant prognostic risk score was formulated based on these genes using the following equation: RiskScore = 0.1627*MCM10 + 0.0833*ECT2 + 0.0374*CENPI + 0.0741*WDHD1 + 0.0717*FBXO5 + 0.1619*SASS6 + 0.0155*PRR11-0.1850*ATAD5-0.1407*KIF20B.

Fig. 2. Constructing a prognostic model by candidate genes.

A Establishment of the sorafenib-resistant-related model: The selection process of the optimum value of the parameter λ in the Lasso regression model by cross-validation method. B Diagnostic ROC curves showed the prediction of patients’ survival status in this model using the training group from TCGA and the validation group from ICGC. C PLS-DA of OS from the low-score group against the high-score group. D Kaplan−Meier survival curves illustrate the prognostic value of risk score in the training cohort and validation cohort. E Distribution of the risk score, survival time, and survival status of the patients with HCC in the TCGA and ICGC cohorts.

The model was trained on 367 HCC cases from the TCGA cohort after excluding samples with incomplete clinical data. Validation was performed using an independent cohort of 243 samples from the ICGC database. Patients in both cohorts were stratified into high- and low-risk groups based on the median risk score. The model’s predictive performance was evaluated using time-dependent receiver operating characteristic (ROC) curves. In the training cohort, the area under the curve (AUC) for the risk score combined with clinical characteristics was 0.72, 0.72, and 0.74 at 1, 3, and 5 years, respectively. Validation in the ICGC cohort yielded AUCs of 0.74, 0.74, and 0.83 at 1, 3, and 4 years, respectively, confirming the model’s reliability (Fig. 2B). Partial least squares discriminant analysis (PLS-DA) further validated the model by projecting patients into distinct low-dimensional regions based on risk group, revealing clear differences in clinical characteristics between high- and low-risk groups (Fig. 2C). Kaplan–Meier survival analysis (Fig. 2D) showed significantly worse overall survival in high-risk patients compared to their low-risk counterparts. A risk factor graph (Fig. 2E) visually demonstrated the progressive increase in mortality with higher risk scores. These findings highlight the risk score’s high accuracy and clinical utility in predicting HCC patient prognosis, offering a valuable tool for stratifying patients and guiding personalized treatment strategies.

Association of risk score with clinical features of HCC

The risk score demonstrated a significant correlation with the clinical features of HCC. Patients with higher clinical grades and advanced stages had elevated risk scores, indicating the model’s association with disease progression (Fig. 3A, B). The cBioPortal database was used to analyze genomic alterations and serum alpha-fetoprotein (AFP) levels to explore further the clinical differences between high- and low-risk groups. The high-risk group exhibited significantly higher AFP levels (Fig. 3C), a well-established biomarker for HCC detection since the 1970s [29]. Additionally, patients in the high-risk group displayed a more significant fraction of genome alterations compared to the low-risk group, suggesting increased genetic instability (Fig. 3D). Detailed mutation analysis revealed a higher prevalence of gene mutations in the high-risk group, with TP53 being the most frequently mutated gene (Fig. 3E). As a critical tumor suppressor and transcription factor, TP53 regulates vital cellular processes such as DNA repair, apoptosis, and cell cycle arrest, underscoring its role in tumor progression [30]. These findings underscore the biological and clinical distinctions between the risk groups and highlight the relevance of the risk score in stratifying HCC patients based on tumor aggressiveness and genetic profile.

Fig. 3. Relationship between the risk score and clinical features.

Relationship between the risk score and grade (A) or stage (B) of HCC. Differences between the high-risk and low-risk groups in AFP at procurement (C) and Fraction genome altered (D). E Somatic mutations of the high-risk group and low-risk group are investigated. The waterfall map showed the top twenty most often altered genes in the high-risk and low-risk groups. The asterisks represented the statistical p-value (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001).

MCM10 is associated with cancer stemness and drug resistance of HCC cells

GSVA was performed to explore the molecular mechanisms underlying the risk score genes. The high-risk group exhibited enrichment in gene sets associated with stemness, such as the HCC stem cell subclass gene set, E2F binding gene set, and MYC downstream targets; Conversely, the low-risk group showed enrichment in metabolic pathway gene sets that inhibit tumor stemness, such as tyrosine catabolism. Moreover, the high-risk group demonstrated activation of TP53-related pathways, while the low-risk group was associated with the G3 subtype gene set, characterized by TP53 mutation and overexpression of cell cycle-regulating genes [31] (Fig. 4A). These results suggest that genes in the model may influence HCC stemness and contribute to sorafenib resistance via metabolic and other pathways.

Fig. 4. Identification of biological pathways associated with the risk score.

A GSVA of the risk score cohorts. Differentially expressed pathways were identified using GSVA through hallmark gene sets. B Scatterplot showing the top eight Pearson correlations between mRNAsi scores and the model genes in TCGA-HCC patients. C Gene set enrichment analysis (GSEA) with high and low MCM10 expression. The stemness-associated and drug-resistance-related sets enriched high MCM10 expression, including the E2F-targets, PI3K-AKT-MTOR signaling, are shown. D, E Western blots showing expression levels of total and phosphorylated AKT, PI3K, S6K1, and 4EBP1 proteins in MCM10‐overexpressed HepG2(D), MCM10‐knockout HepG2-SR(E), and control cells. P < 0.05, FDR < 0.25.

To investigate the relationship between risk genes and cancer stemness, we utilized the mRNAsi scores from TCGA-HCC. A scatter plot analysis revealed that MCM10 exhibited the highest correlation with cancer stemness (r = 0.47, P < 0.05; Fig. 4B). GSEA was performed to examine further the biological processes and pathways associated with MCM10. High MCM10 expression was significantly associated with stemness and drug resistance pathways, including the E2F and PI3K-AKT signaling pathways (Fig. 4C). Analysis in HepG2 cells also indicated significant enrichment of the PI3K-AKT-mTOR pathway in HCC samples with high MCM10 expression. Activation of this pathway is pivotal in maintaining cancer stem cell phenotypes and has been previously linked to sorafenib resistance in HCC. To validate whether MCM10 regulates stemness through the PI3K-AKT pathway and influences sorafenib sensitivity, we examined the phosphorylation status of pathway components. MCM10 overexpression in HepG2 cells increased the phosphorylation levels of PI3K, AKT, S6K1, and 4EBP1 (Fig. 4D and Fig. S1E). In contrast, depletion of MCM10 in sorafenib-resistant HepG2-SR cells reduced the phosphorylation of these proteins (Fig. 4E and Fig. S1F).

These findings underscore the role of MCM10 as a critical regulator of cancer stemness and sorafenib resistance in HCC. The correlation between MCM10 expression and activation of the PI3K-AKT pathway further supports its involvement in driving stemness and resistance. Additionally, MCM10 emerges as a significant risk factor for poor prognosis in HCC patients, with its influence on resistance likely mediated through stemness-related mechanisms.

MCM10 promoted stemness in HCC cells

Immunohistochemical analysis of tissues from HCC patients revealed a strong positive correlation between MCM10 expression and key cancer stemness markers KLF4 and SOX2; Pearson correlation coefficients of 0.740 and 0.767, respectively, underscore the significant association between MCM10 optical density and these markers, highlighting MCM10’s role in maintaining stemness characteristics in HCC tissues(Fig. 5A). These findings were further corroborated by Western blot analysis, which showed elevated MCM10 levels in sorafenib-resistant HepG2 and Hep3B cell lines compared to their parental counterparts. Moreover, a corresponding increase in stemness-related proteins in these drug-resistant cell lines suggests that MCM10 is critical in conferring sorafenib resistance by sustaining the cancer stem cell phenotype(Fig. 5B and S1A).

Fig. 5. MCM10 promoted stemness in HCC cells.

A Immunohistochemical analysis illustrating the positive correlation between MCM10 expression and key stemness markers (KLF4, SOX2) in hepatocellular carcinoma cells. B Western blots results showed that KLF4 and SOX2 expression levels were consistent with MCM10 expression in HepG2 cells. C Tumor sphere formation results showed stable overexpression or knockout of MCM10 affects HepG2 cell stemness. D Immunofluorescence was examined for fluorescence intensity and CD44 and CD133 colocalization in HepG2 cells. E Flow cytometry results illustrate that the expression level of MCM10 affects the number of CD133⁺ cells. The asterisks represented the statistical p-value (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001).

Tumor sphere-formation assays (Fig. 5C and S1B) confirmed that overexpression of MCM10 significantly enhances oncogenic properties.

Functional assays further supported these findings. Tumor sphere formation assays demonstrated that MCM10 overexpression significantly enhanced the oncogenic capacity of HCC cells, as evidenced by robust tumor spheroid formation. Conversely, MCM10 knockout in sorafenib-resistant cells markedly reduced tumor sphere-forming ability, indicating its essential role in promoting stemness-associated properties(Fig. 5C and S1B). Immunofluorescence analysis revealed differential expression of stemness markers CD44 and CD133 in HCC cells with varying MCM10 levels. Elevated MCM10 expression correlated with increased levels of these markers, further underscoring its role in sustaining the cancer stem cell phenotype and promoting drug resistance (Fig. 5D and S1C). Flow cytometry analysis provided additional evidence of MCM10’s involvement in stemness and clinical outcomes. Cells with high MCM10 over-expression exhibited a significantly larger population of CD133⁺ cancer stem cells, while MCM10 knockoutwas associated with a reduction in this population (Fig. 5E and S1D).

MCM10 augments tumor growth in xenograft

Kaplan-Meier survival analysis, conducted using the Kaplan-Meier Plotter database(https://kmplot.com/analysis/), revealed that high MCM10 expression is significantly associated with reduced survival in HCC patients. The hazard ratios (HR) across Kaplan-Meier Plotter database ranged from 1.97 to 3.39, indicating a robust correlation between elevated MCM10 expression and poor prognosis (Fig. 6A, P < 0.001). An in vivo study was performed using xenograft models to validate these clinical observations. Nude mice were subcutaneously injected with MCM10-overexpressing HepG2 cells, MCM10-knockout HepG2-SR cells, or control cells. Body weight measurements taken post-inoculation showed no significant differences among the groups (Fig. 6B). However, tumor growth was markedly influenced by MCM10 expression. Mice injected with MCM10-overexpressing cells developed significantly larger tumors than the control group, whereas those injected with MCM10-knockout cells displayed substantially smaller tumors (Fig. 6C, D).

Fig. 6. MCM10 promotes HCC xenograft tumor growth by enhancing stemness.

A Kaplan-Meier survival curves displaying the relationship between high MCM10 expression and decreased survival rates in hepatocellular carcinoma patients. B Weight changes of nude mice after subcutaneous injections of HepG2-Vector/oeMCM10 and HepG2-SR-Ctrl/MCM10 KO cells in nude mice. C Growth curves of transplanted tumors after HCC cells injection. The transplanted tumors (D) and their weights (E) were obtained 40 days after HCC cells subcutaneous injection. F Paraffin-embedded sections of xenograft tumors were stained by H&E and immunohistochemistry (IHC) using MCM10, KLF4, SOX2, and Ki67 antibodies. The asterisks represented the statistical p-value (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001).

Histological and immunohistochemical analyses of the xenograft tumors provided additional insights into the mechanism of MCM10-mediated tumor progression. Hematoxylin and eosin (H&E) staining and immunohistochemistry (IHC) confirmed elevated MCM10 expression in the MCM10-overexpression group and minimal or absent expression in the knockout group. Furthermore, the overexpression group exhibited significantly increased levels of stemness markers, including KLF4, SOX2, and the proliferation marker Ki67, while their expression was markedly reduced in the knockout group (Fig. 6F).

These results highlight MCM10’s ability to accelerate tumor growth in vivo by promoting stemness and cellular proliferation. The enhanced expression of stemness-associated markers in the presence of MCM10 suggests its pivotal role in driving HCC progression and poor clinical outcomes.

TW-37 suppresses MCM10 expression and enhances the efficacy of sorafenib in sorafenib-resistant HCC cells

TW-37, identified as a potential gene silencing agent of MCM10 through the CMap database, showed significant negative connectivity scores, suggesting its ability to suppress MCM10 expression in HepG2 and Hep3B sorafenib-resistant (SR) cell lines (Fig. 7A). The half-maximal inhibitory concentration (IC50) values of TW-37 in HepG2-SR and Hep3B-SR cells were determined using CCK-8 assays and calculated as 43.477 µM and 57.788 µM, respectively (Fig. 7B). TW-37 treatment significantly downregulated the expression of MCM10 and key stemness markers KLF4 and SOX2 in a time-dependent manner (Fig. 7C). A dose-response analysis revealed that increasing concentrations of TW-37 over 48 h effectively reduced the levels of these proteins, reinforcing its potential to mitigate cancer stemness (Fig. 7D). In addition, The mRNA level of MCM10 was significantly down-regulated following TW-37 treatment in a time- and dose-dependent manner (Figs. S2A, S2B).

Fig. 7. TW-37 suppresses MCM10 expression and improves sorafenib efficacy in resistant HCC cells.

A Connectivity Map (CMap) analysis identified TW-37 as a potential inhibitor impacting the expression profiles of HepG2 and Hep3B sorafenib-resistant (SR) cell lines. B CCK8 was conducted to determine the half maximal inhibitory concentration (IC50) of TW-37 in HepG2-SR and Hep3B-SR cell lines. C Western blot analysis assessed the time-dependent effects of 50 µM TW-37 on MCM10, KLF4, and SOX2 protein levels in HepG2-SR and Hep3B-SR cells over 48 h. D Dose-dependent effects of TW-37 on the expression levels of the stemness markers were analyzed after 48 h of treatment in HepG2-SR and Hep3B-SR cell lines. E Effects of TW-37 and sorafenib on HepG2-SR and Hep3B-SR cell viability were quantified with CCK8 assays. F Cell viability was assessed in CCK8 assays after treatment of HepG2-SR and Hep3B-SR with Sorafenib alone, TW-37, or a two-drug combination time gradient. No significance (NS) was observed in specific pairings. The asterisks represented the statistical p-value (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001).

To evaluate the potential effects of TW-37 in combination with sorafenib, HepG2-SR, and Hep3B-SR cells were treated with various concentrations of TW-37 (0, 10, 15, 20, 30, 35, 40, 50, 60 µM) and sorafenib (5, 10, or 15 µM) for 48 h. Combination index (CI) values calculated using CompuSyn software consistently fell below 1, suggesting TW-37 enhances sorafenib’s inhibitory effects on resistant cell lines (Fig. 7E and Tables S1, S2). Time-course CCK-8 assays further demonstrated that the combination of TW-37 and sorafenib produced a more pronounced antiproliferative effect than either agent alone under multiple treatment conditions (Fig. 7F). These findings demonstrate that TW-37 achieves dual therapeutic effects: significant suppression of MCM10 expression, and potentiation of sorafenib’s anti-proliferative and stemness-inhibiting activities in sorafenib-refractory HCC models. This combinatorial therapeutic profile nominates TW-37 as a promising adjunctive agent for advanced HCC treatment regimens.

Discussion

Sorafenib remains the cornerstone of systemic therapy for advanced HCC. However, its therapeutic effectiveness is limited, with only about 30% of patients deriving significant benefit due to the development of drug resistance [10]. This underscores the critical need for strategies to overcome sorafenib resistance and improve patient outcomes. Recent evidence highlights a strong connection between sorafenib resistance and the stemness phenotype in cancer cells, offering a potential target for novel therapeutic interventions.

Sorafenib-resistant HCC cells often exhibit enhanced stemness characteristics, such as elevated expression of markers like CD44 and CD133. For instance, downregulation of CD44 has been shown to protect HCC cells from sorafenib-induced apoptosis [32], while high CD133 expression correlates with poor therapeutic response to sorafenib [33]. The PI3K/AKT signaling pathway is central to maintaining stemness in cancer cells, promoting the upregulation of transcription factors like OCT4 and SOX2 [34], which are closely linked to tumor progression and therapeutic resistance. These findings underscore the pivotal role of cancer stemness in the failure of sorafenib therapy and highlight the potential of targeting stemness-associated pathways to counter resistance.

Our study developed a prognostic risk model based on nine sorafenib-resistance-related genes, identifying MCM10 as the most significant gene associated with the cancer stemness index. Elevated MCM10 expression strongly correlated with advanced tumor grade, poor prognosis, and high TP53 mutation rates in HCC patients. TP53 mutations, which are enriched in high-risk patients, can activate the PI3K/AKT signaling pathway by deregulating PTEN, thus amplifying tumor stemness and poor survival [35]. This mechanistic link validates the clinical relevance of our model in stratifying patients based on risk and guiding therapeutic decision-making.

While existing LASSO regression-based prognostic models in HCC provide valuable prognostic information, they often focus primarily on clinical features or single gene expression patterns, failing to adequately integrate multidimensional biological features, such as drug resistance-related genes and cancer stemness characteristics. In contrast, our newly developed model combines various biological features, particularly resistance-related genes, to offer a more comprehensive and accurate prognosis assessment. Further, we enhanced the robustness of the model through cross-validation across different datasets (TCGA and ICGC), ensuring its applicability and generalizability across diverse clinical settings. Interestingly, the risk score model performed better in the validation cohort (ICGC) than in the training cohort (TCGA). Several factors may contribute to this difference, including variations in the data sources and sample populations. Additionally, the differences in patient demographics, clinical characteristics between the two datasets may also have influenced the model’s performance.

Building on these findings, our study identified MCM10 as a key driver of HCC progression and sorafenib resistance. Historically, MCM10 has been recognized for its role in DNA replication and cell cycle regulation [36]. Our findings expand on this understanding, showing that MCM10 not only drives HCC progression but also enhances the PI3K/AKT pathway, thereby maintaining stem-like properties and contributing to sorafenib resistance. MCM10’s preferential overexpression in HCC tissues, compared to normal liver tissues, further highlights its potential as a biomarker for malignant behavior. Notably, high MCM10 expression exacerbates DNA replication stress in cancer cells, a vulnerability that could be exploited for targeted therapy.

Through Connectivity Map (CMap) analysis, TW-37 was identified as a promising small-molecule agent capable of downregulating MCM10 expression. TW-37, originally developed as a BCL-2 family inhibitor, disrupts anti-apoptotic pathways and induces cancer cell apoptosis [37–40]. In our study, TW-37 significantly downregulated MCM10, KLF4, and SOX2 expression, thereby reducing stemness-associated features in sorafenib-resistant HCC cells. Further analysis revealed that TW-37 treatment was associated with alterations in the PI3K-AKT signaling pathway, a pathway known to be critical for stemness maintenance and drug resistance in HCC. This observation suggests that the observed tumor-suppressing effects of TW-37 may, at least in part, be mediated by its influence on PI3K-AKT signaling.

TW-37 demonstrated additive effects with sorafenib, as evidenced by combination index (CI) values below 1 and superior anti-proliferative activity compared to monotherapy. Mechanistically, TW-37 may impair mitochondrial integrity, increasing oxidative stress and apoptosis in stem-like cancer cells [41]. Additionally, it likely interferes with pathways critical for stem cell survival, such as Wnt/β-catenin and Notch signaling, although further research is needed to elucidate these effects [42, 43].

The dual functionality of TW-37, inducing apoptosis and inhibiting cancer stemness, makes it an attractive candidate for advanced HCC therapy. By targeting the heterogeneous populations within HCC tumors, TW-37 could address both stem-like and non-stem-like cancer cells, thereby improving therapeutic efficacy and reducing relapse risk.

Our findings suggest that MCM10 serves as a robust prognostic marker and therapeutic target for HCC. Its high expression in malignant tissues and its critical role in maintaining cancer stemness make it a compelling focus for future drug development. Moreover, the ability of TW-37 to attenuate MCM10 expression and additive with sorafenib underscores its potential as an adjunct therapy, capable of overcoming drug resistance and enhancing patient outcomes.

Several limitations should be acknowledged. While our results showed that TW-37 downregulates MCM10 expression and alters PI3K-AKT signaling, its tumor-suppressing effects cannot be exclusively attributed to MCM10 inhibition. Given that TW-37 was originally developed as a BCL-2 inhibitor, TW-37 directly binds to Bcl-2 [44], preventing it from activating the NF-κB pathway. Since Bcl-2 normally interacts with IKK (IκB kinase) to promote NF-κB activation [45], TW-37 will suppresses NF-κB activity. Studies have shown that NF-κB activation upregulates multiple DNA replication factors, including minichromosome maintenance proteins (MCMs), to promote S-phase progression [46]. These reports suggest that TW-37 may regulate MCM10 expression via NF-κB pathway. In future studies, we will focus on investigating whether TW-37 affects tumor cell stemness and sorafenib resistance, as well as the molecular mechanisms through which it regulates MCM10 expression. Additionally, we aim to determine whether TW-37 exerts its effects on inhibiting tumor stemness and reversing sorafenib resistance through MCM10.

Conclusion

This study establishes MCM10 as a key regulator of cancer stemness and sorafenib resistance in HCC, linking its overexpression to poor clinical outcomes and aggressive tumor behavior. TW-37 emerges as a promising therapeutic candidate, effectively targeting stemness and reversing resistance to sorafenib. Together, these findings highlight the potential of integrating MCM10-targeted strategies into clinical practice, paving the way for improved management of advanced HCC.

Author contributions

TW conceptualized and designed the research. ZYZ conducted the experiments. LL, YL, YF, YL, ZFH, and XMW compiled the data and performed the biostatistical analyses. ZYZ drafted the manuscript, which was then critically revised by TW, YQJ, and LC. All authors contributed significantly to the work and approved the final version for submission.

Funding

This work is supported by National Natural Science Foundation of China(82203023), the Scientific Research Launch Project for new employees of the Second Xiangya Hospital of Central South University(to TW).

Data availability

The data presented in this study are openly available in TCGA database (https://portal.gdc.cancer.gov/), ICGC database (https://dcc.icgc.org/projects/LIRI-JP), Kaplan-Meier Plotter database (https://kmplot.com/analysis/).

Competing interests

The authors declare no competing interests.

Ethical approval and informed consent

All human tissue samples used in this study were obtained with written informed consent from participants, in accordance with the Declaration of Helsinki. The animal study protocol was approved by the Ethics Committee of The Second Xiangya Hospital of Central South University (Animal Permit No. 2022027, Human samples Permit No.2023-0315). The study adhered to the guidelines set by the committee. The term “INFORMED CONSENT” is hereby acknowledged as a formal requirement fulfilled in this research.

Supplementary information

The online version contains supplementary material available at 10.1038/s41417-025-00946-0.