We scoured articles in a number of databases, such as PubMed (https://pubmed.ncbi.nlm.nih.gov), Scopus (https://www.scopus.com), Web of Science (https://www.webofscience.com), ScienceDirect (https://www.sciencedirect.com), and the Cochrane Library (https://www.cochranelibrary.com). The time span for the literature we were looking for was from January 2000 to August 2024. The search was limited to peer-reviewed articles published in English. Zotero document management software was used to screen and remove duplicates, which were identified using a search strategy (Supplementary Table 1) that combined the terms “quercetin” and “hepatocellular carcinoma or HCC or liver cancer”.

(1) Animal Selection Criteria: When selecting animals for experimental models, the predominant choice was rodents such as mice, nude mice, and rats. (2) Model Selection Criteria: to facilitate the observation of whether the tumor model was successfully established and to make it convenient for the measurement of subsequent relevant results, only subcutaneous models were considered, whereas orthotopic models were excluded. (3) Intervention Measures: QT was used alone for treatment, and there was a placebo group or blank control group. (4) Outcomes: The impact of QT on animal models of HCC after tumor implantation, including changes in tumor volume, changes in tumor weight, mortality rate of mice during the treatment process, and changes in the animals’ own body weight before and after QT injection.

(1) Research Subjects: experimental animals for which basic information could not be extracted, or animal models with only orthotopic tumors; (2) Intervention Measures: Studies where QT was used in combination with other drugs; (3) Study design: non-randomized controlled trials (RCTs); (4) Data availability: studies with unavailable or missing relevant original experimental data; (5) Literature type: literature published in the forms of letters, editorials, abstracts, conference proceedings, experience summaries, or case reports; (6) Literature duplication: similar or duplicate studies; and (7) Literature timeliness/credibility: outdated articles lacking significance and credibility.

Authors Zhiguo Tan (ZT) and Yu Chen (YC) designed a data extraction form covering the first author, publication year, country, animal baseline data, detailed intervention information, and required specific outcome data. ZT and YC extracted the data independently. In case of any discrepancies that arise during the data extraction process, they will either resolve them through discussion or turn to a third author, Xu Chen (XC). If the relevant basic information and outcome indicators in an article are not clear, we will attempt to contact the first author of the article or reach out for consultation based on the email address of the corresponding author. Should the outcome indicators be unavailable and presented in the form of pictures, we will utilize the GetData Graph Digitizer (v2.24, GetData Pty Ltd., Brisbane, Queensland, Australia) software to reproduce the relevant data. Moreover, we will use Excel spreadsheets to independently extract the above data and materials from the articles that are finally included. ZT and YC will evaluate these preclinical animal studies using SYRCLE’s Risk of Bias tool, evaluating bias risk (e.g., publication bias, selective reporting) for each study. Results were categorized as “Yes” (low bias risk), “No” (high bias risk), or “Uncertain” (unclear bias risk). To evaluate the influence of high-risk (No) studies on our overall conclusions, we performed sensitivity analysis using two methods. First, we re-ran the primary analysis (low/unclear risk only) after including the high-risk studies, and compared the new pooled effect size to the original. Second, we systematically excluded each high-risk study and recalculated the pooled effect size to observe its impact. We independently evaluated the quality of each RCT, and any difference in opinion was settled through discussion until an agreement was reached or with the help of the third author (XC).

We conducted a statistical analysis using the Review Manager software (v5.4, Cochrane Collaboration, Copenhagen, Denmark). Standardized mean difference (SMD) with 95% confidence interval (CI) was employed to quantify effect sizes, while relative risk (RR) with 95% CI was utilized for the dichotomous mortality variable. First, we considered the heterogeneity of the included studies and conducted meta-synthesis after excluding clinical and methodological heterogeneity. The Q test and I² test were employed to determine the heterogeneity of the research results. In the Q test, p 0.1 was defined as the presence of heterogeneity among the studies. If the heterogeneity was relatively small (p $>$ 0.1, I² 50%), the fixed-effect model was chosen. If the I² test result was between 50% and 75%, the random-effect model was used for data synthesis. If I² exceeded 75%, we first identified potential sources of heterogeneity from two dimensions: clinical factors (e.g., QT intervention dose or administration route) and methodological factors (e.g., sample size or outcome measurement method). Based on these potential factors, we conducted exploratory subgroup analyses to investigate the impact of specific variables on heterogeneity. Statistical analyses were performed utilizing SPSS (v20.0, IBM Corporation, Armonk, NY, USA) or GraphPad Prism (v9.4.1, GraphPad Software, LLC, San Diego, CA, USA) software. Mean $\pm$ standard deviation was used to present the data, with all results derived from at least three independent experiments.

The TCMSP database (https://www.tcmsp-e.com/) was adopted to search for the PubChem Compound ID of “quercetin”. Subsequently, search for Canonical SMILES within the PubChem database (http://pubchem.ncbi.nlm.nih.gov/). Ultimately, employ the SwissTargetPrediction database (https://www.swisstargetprediction.ch/) to gather the targets of compounds that have a probability score higher than zero. Genes associated with HCC were identified by searching the GeneCards (http://www.genecards.org) and Online Mendelian Inheritance in Man (http://omim.org) databases using the keywords “hepatocellular carcinoma” and “liver cancer”. We selected genes that scored above 1 as potential targets from the GeneCards database. Genes from the two databases were de-integrated, with the Uniprot database (https://www.uniprot.org/) employed for the standardization of gene names.

The potential anti-HCC targets of QT were uploaded into the STRING 11.5 database (https://cn.string-db.org/) for construction of the PPI. “Homo sapiens” was designated as the target protein species, and a minimum interaction threshold of “high confidence $>$0.4” was adopted. Meanwhile, free targets were concealed, and the remaining parameters were left at their default settings. Subsequently, PPI data were imported into Cytoscape (v3.10.2, Cytoscape Consortium, San Diego, CA, USA) Software and analyzed via the “Network Analyzer” function [15].

The potential targets of QT for treating HCC were analyzed using R. Gene Ontology (GO) functional enrichment was analyzed using the R package org.Hs.eg.db. (3.20.0, Bioconductor Project, Seattle, WA, USA) As for the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment results, these were analyzed with the R Package clusterProfiler (4.18.1,Yu Lab, Guangdong, China). Those top-ranked results where the p-values were less than 0.05 and the false discovery rate (FDR) was less than 0.25 were visualized and analyzed. GO enrichment and KEGG pathway analyses were plotted via the tool available on https://www.bioinformatics.com.cn.

The two-dimensional structure of QT was retrieved as a Spatial Data File from the National Center for Biotechnology Information PubChem database. Target proteins were obtained in Protein Data Bank (PDB) format from the PDB [16]. Using PyMOL (v2.5.2, Schrödinger, LLC, New York, NY, USA), water molecules and ligands were removed from the crystal structure complex; then the target proteins were imported into AutoDock Tools (v1.5.6, Molecular Graphics Laboratory, La Jolla, CA, USA), where hydrogenation was performed, followed by saving in PDBQT format. The protein’s protonation states were calculated using the Add Gasteiger Charges module. The active site was defined using a grid box, centered on the protein’s co-crystallized ligand binding pocket for full coverage. Finally, AutoDock Vina [17] was used to perform molecular docking of the receptor protein with the ligand and calculate its binding energy score. Binding energy less than zero denotes spontaneous binding potential between the ligand and receptor. It is generally acknowledged that the lower the binding energy between the receptor and ligand, the more probable their binding will be [17]. Molecular dynamics (MD) simulations were conducted using GROMACS (v2018.1, GROMACS Development Team, Stockholm, Sweden) [18]. The protein’s topology was generated with the GROMACS 54a7 force field, whereas the ligand’s parameters were obtained from the ATB website (https://atb.uq.edu.au/). The system was placed in a cubic box with simple point charge water and neutralized with Na⁺/Cl^– ions [19, 20]. Prior to the production runs, the system was prepared through a series of steps: first, energy minimization was performed via 50,000 steepest descent steps, with 100 ps NVT (constant atomic number, volume, and temperature) and 100 ps NPT (constant atomic number, pressure, and temperature) equilibration carried out sequentially [21, 22]. The production simulation was carried out for 100 ns under equilibrated conditions (310 K, 1.0 bar) [23]. Long-range electrostatic interactions were evaluated using the Particle Mesh Ewald method. Trajectory analyses (root mean square deviation [RMSD], root mean square fluctuation [RMSF], radius of gyration, solvent accessible surface area, and hydrogen bonds [HBs]) were performed using GROMACS modules, with data visualized using Xmgrace (v5.1.25, The Grace Development Team, Göttingen, Lower Saxony, Germany). To ensure the reliability of results, MD simulations were independently repeated three times with different initial velocity distributions.

For the Cell Counting Kit-8 (CCK-8) assay, 1 $\times$ 10³ MHCC97-H HCC cells, supplied by the Fudan Institutes of Biomedical Sciences Cell Center (Shanghai, China), were seeded into 96-well plates. Cells were cultured in DMEM medium with 10% fetal bovine serum. Following cell adhesion, cells were treated with medium containing QT (SQ8030, Solarbio, Beijing, China) at a series of concentrations (0, 5, 10, 20, 40, 80, 160 µM) for 24 and 48 h of treatment, respectively. Subsequently, 10 µL CCK-8 reagent (C0037, Beyotime, Beijing, China) was added to each well. We measured absorbance at 450 nm via a microplate spectrophotometer (ELx800, Biotech Inc., Winooski, VT, USA) after 2 h of incubation. All cell lines were validated by STR profiling and tested negative for mycoplasma.

For the colony formation assay, MHCC97-H cells were seeded into 6-well plate wells. After adherence, medium was replaced with fresh medium containing QT (40 and 80 µM) for 24 h. Then cells were cultured in standard growth medium for 14 days, fixed in 4% paraformaldehyde at room temperature for 10 min, and washed three times with phosphate-buffered saline. Finally, cells were stained with crystal violet for 15 min, rinsed, and then analyzed directly. Colonies were quantified using ImageJ software (v1.54, National Institutes of Health, Bethesda, MD, USA). The software was set to automatically recognize stained colonies based on color and size thresholds (area $\ge$500 pixels), followed by manual verification to exclude debris or overlapping colonies for accurate counting.

The EdU (C0078S, Beyotime, Beijing, China) proliferation test was performed according to the manufacturer’s protocol. After incubation at 37 °C, cells were added with 50 µM EdU and incubated for 2 h. Subsequently, the corresponding reagents were added sequentially in the specified order. The images were taken using Leica Application Suite (version 3.0.0, Leica Microsystems, Wetzlar, Germany).

For cell cycle assessment, MHCC97-H cells treated with QT were trypsinized and processed using a flow cytometry cell cycle kit (C1052, Beyotime, Beijing, China), with propidium iodide (PI) staining. We collected cell cycle data via a CytoFLEX flow cytometer (Beckman Coulter, Brea, CA, USA) and analyzed with CytoFLEX analysis software (v2.4, Beckman Coulter Life Sciences, Brea, CA, USA). The gating strategy began by excluding debris and aggregates using a forward scatter-side scatter plot to select only single cells. Subsequently, G0/G1, S, and G2/M cell cycle distribution was identified by analyzing PI fluorescence. The controls included unstained cells (to set the autofluorescence baseline) and untreated PI-stained cells (normal cycle reference). Samples were analyzed on the CytoFLEX cytometer, with $\ge$20,000 events recorded per sample.

All in vivo experiments used 6-week-old male BALB/c nude mice (Slack Jingda Experimental Animal Co., Ltd, Hunan, China). A 5 $\times$ 10⁶ MHCC97-H cell in suspension were injected into the right axillary fossa of mice, and subcutaneous tumors formed after about 1 week. We next randomly assigned the mice to two groups (n = 5/group): 50 mg/kg QT was delivered to the QT group via daily gavage, with the control group was administered an equivalent volume of normal saline. Tumor volume was measured with a vernier caliper every 3 days. After 3 weeks, the mice were sacrificed by cervical dislocation. Their body weights were measured, and subcutaneous tumors were excised to record size and weight. Tumor volume (mm³) was calculated as (L $\times$ W²)/2 (L = long axis, W = short axis). Blinding was applied during the tumor volume measurement. All experiments were approved by the Ethics Committee of Hunan Provincial People’s Hospital (Ethical Number: [2023]-151).

The tumor tissue was fixed, embedded, sectioned, stained, dehydrated, and sealed. Subsequently, 5% goat serum (31872, Invitrogen, Carlsbad, CA, USA) was used to block sections for 30 min, followed by the addition of primary antibody (p-AKT, AP0637, ABclonal, Wuhan, China, 1:400) and incubation at 4 °C for 12–16 h. On the subsequent day, the secondary antibody (G1213, Servicebio, Wuhan, China, 1:200) was incubated for 2 hours at ambient temperature. To ensure the specificity and reliability of the immunohistochemical staining results, a negative control group was included. Staining intensity multiplied by positive cell percentage yielded the immunoreactive score [24]. For H&E staining, the reagent we used was the HE staining kit (G1076, Servicebio, Wuhan, China).

The search results yielded a total of 1079 potential articles, with 276 from the Web of Science, 152 from PubMed, 356 from Science Direct, 253 from Wiley Online Library, and 42 from Scopus. Subsequently, after the Zotero document management software identified and removed duplicate documents, leaving 752 articles whose titles and abstracts were then screened. According to the research selection criteria, 56 articles were filtered out. Among these, 47 articles were excluded, and ultimately, 9 articles were included in this study [25, 26, 27, 28, 29, 30, 31, 32, 33] (Fig. 1).

Fig. 1.

Flow diagram of the identified, included, and excluded studies. QT, quercetin.

The basic data of the included relevant literature are shown in Table 1 (Ref. [25, 26, 27, 28, 29, 30, 31, 32, 33]) and Table 2. A total of 9 papers met the final inclusion criteria. The experimental animals used were nude mice or regular mice, and subcutaneous tumor models were established using liver cancer cells (e.g., HepG2, HH-7, and H22). Compared with the control group, QT was mainly administered in the form of QT solution or QT nanoparticles. The dosage of QT ranged from 10 to 150 mg/kg, and was administered intermittently or continuously through intraperitoneal injection, gavage, or intravenous injection. The study duration varied from 1 to 7 weeks. The study quality was assessed using the SYRCLE animal experiment risk assessment tool (Fig. 2). Generally, the overall quality of the included studies was moderate.

Fig. 2.

Risk of Bias Assessment. (A) Classification of bias risk among included articles. (B) Per-article bias characteristics. ?: Unclear risk; +: Low risk; -: High risk.

Nine included articles reported tumor volume changes in the experimental (QT-treated) and control groups. Among them, Wu et al. [27] tested two QT concentrations (25 and 50 mg/kg); their data were treated as independent sets (Wu Ruoxiao 1, Wu Ruoxiao 2). Wang et al. [26] used two administration forms (ordinary QT solution and QT nanoparticles); their data were also split into independent sets (Wang Can 2016a, Wang Can 2016b). In total, 11 datasets from the 9 articles were included. Sensitivity analyses of these 11 sets showed no significant changes in meta-analysis results or consistency after excluding any single dataset. Pooled analysis of the 11 sets (Fig. 3A) yielded the SMD and 95% CI. Due to substantial heterogeneity (I² = 69%), a random-effects model was used. Compared with the control group, the QT group inhibited liver cancer volume growth (SMD = 4.63, 95% CI: 3.44–5.82, Z = 7.63, I² = 69%, p 0.00001). The dosage cut-off for subgroup analysis was determined based on the classification of QT dosages into preventive and therapeutic categories. A previous study on QT for the treatment of prostate xenografts adopted 75 mg/kg QT as the therapeutic dose [34]. Therefore, considering the administered doses in all of the RCTs included in this study, we finally selected 60 mg/kg (slightly lower than 75 mg/kg) as the threshold to distinguish between preventive and therapeutic doses. We further performed subgroup analysis to explore the heterogeneity sources. Based on dosage, we divided the data into two subgroups: QT $\le$60 mg/kg (preventive) and QT $>$60 mg/kg (therapeutic). In the $>$60 mg/kg (therapeutic) subgroup, heterogeneity (I²) was reduced to 16% (SMD = 2.90, 95% CI: 1.68–4.12, Z = 4.65; p 0.00001). However, the $\le$60 mg/kg subgroup still had significant heterogeneity, indicating that dosage difference was not the root cause of heterogeneity (Table 3).

Fig. 3.

Forest plots of tumor growth, body weight, and mortality in mice across different interventions. (A) Tumor volume, (B) Tumor weight, (C) Mice own weight and (D) Mice mortality rate.

Notably, some experiments used QT nanoparticles (a potential heterogeneity source), so we conducted subgroup analysis by administration form. The results showed reduced within-group heterogeneity: the QT nanoparticle group had I² = 0%, whereas the common QT group still had relatively high heterogeneity (I² = 59%) despite a decrease. Notably, the interaction between dosage and administration can explain the observed results. At high doses (therapeutic, $>$60 mg/kg), QT nanoparticles improved solubility and targeting to overcome the absorption heterogeneity of the standard drug form and boost efficacy. However, at low doses (preventive, $\le$60 mg/kg), this effect is masked by a “threshold effect”, where both forms achieve effective concentrations. This dosage-dependent interaction could explain why the nanoparticle group showed less variability than the standard group. Future cross-subgroup analyses may confirm this theory. Nevertheless, QT nanoparticles significantly inhibited subcutaneous tumor volume increase (SMD = 7.87, 95% CI: 5.71–10.04, Z = 7.13; p 0.00001). No significant heterogeneity sources were identified in each subgroup analysis (Table 3).

Among the 9 articles included in our study, 4 articles [27, 28, 29, 30], which contained five groups of data in total, reported on the changes in tumor weight within both the experimental group and control group following QT intervention. Subsequently, we conducted a sensitivity analysis. It was found that after excluding any group of data from any of these articles, the remaining literature’s meta-analysis results remained unchanged and were thus stable. When conducting a meta-analysis on the five groups of data from these four articles (Fig. 3B), we opted for the fixed-effects model for data synthesis. The combined results revealed that the experimental group (with QT intervention) significantly suppressed liver cancer weight gain relative to the control group (SMD = 3.00, 95% CI: 2.25–3.7, Z = 7.87, I² = 32%, p 0.05).

Among the 9 articles incorporated into our study, 3 articles [29, 30, 31], which contained three groups of data in total, described the weight variations of experimental mice following treatment. Changes in the body weights of mice were used to assess adverse reactions associated with QT intervention. Subsequently, a meta-analysis was conducted on the group data from these three articles (Fig. 3C). It was found that no obvious effect on body weights was observed in the QT group (SMD = 0.45, 95% CI: 0.19–1.08, I² = 22%, Z = 1.38; p = 0.17).

Three of the nine articles included in our study provided three groups of data that described the changes in mortality [25, 32, 33]. In total, 52 mice were involved in these studies, with 26 mice in the QT group and another 26 in the control group. A meta-analysis was performed on the above-mentioned three groups of data (Fig. 3D). No significant heterogeneity was observed among the included studies, as shown by the results (p = 0.90, I² = 0%). Subsequently, we selected a fixed model for data synthesis. The combined results revealed that the mortality rate of the QT group was lower than that of the control group (RR = 0.56, 95% CI: 0.40–0.80; p = 0.001). These findings suggest that QT is capable of reducing the mortality rate of mice with transplanted tumors.

The meta-analysis results for tumor volume, tumor weight, variations in mouse weight, and mouse mortality showed the symmetrical distribution of studies in the upper region of the funnel plot (Fig. 4). Egger’s test was used to assess publication bias for the four above-mentioned outcome indicators (Supplementary Fig. 1). While the results showed no significant publication bias for tumor weight, mouse body weight, and mortality (p $>$ 0.05), potential publication bias was detected for tumor volume (p 0.05) (Supplementary Fig. 2). Therefore, we thus employed the trim-and-fill method for correcting publication bias. The results revealed no “missing” studies and no significant change in the pooled effect size, suggesting that the results were robust and not substantially impacted by this form of bias. Furthermore, we performed sensitivity analysis to assess the stability of our meta-analysis results for tumor volume, tumor weight, mouse body weight changes, and mouse mortality rate. The one-by-one exclusion approach was employed, where we removed one literature item at a time and subsequently conducted a meta-analysis on the remaining articles (n-1). Sensitivity was evaluated by observing the variations in the combined results. The results demonstrated that none of the included articles significantly influenced the overall SMD or RR value, confirming the reliability of the pooled results.

Fig. 4.

Funnel plots for assessing publication bias. (A) Tumor volume, (B) tumor weight, (C) mouse body weight, and (D) mouse mortality rate.

First, 89 intersection targets of QT and HCC were obtained through a Venn diagram (Fig. 5A). Subsequently, the PPI network of these intersection targets was acquired using the STRING database (Supplementary Fig. 3). Finally, using Cytoscape software, the top 10 core targets of QT’s action on HCC were screened based on the degree values of the targets, including AKT1, epidermal growth factor receptor (EGFR), Proto-oncogene tyrosine-protein kinase Src (SRC), matrix metalloproteinase 9 (MMP9), glycogen synthase kinase 3 beta (GSK3B), poly (ADP-ribose) polymerase 1 (PARP1), cyclin-dependent kinase 2 (CDK2), MMP2, KDR (also known as vascular endothelial growth factor receptor 2) and insulin-like growth factor 1 receptor (IGF1R), with AKT1 being ranked first (Fig. 5B).

Fig. 5.

Network pharmacology analysis. (A) Venn diagram illustrating overlapping gene targets between QT and HCC. (B) Screening of key targets. (C–E) GO analysis of MF, CC and BP. (F) KEGG analysis. (G) Network diagram of the PI3K/AKT signaling pathway. HCC, hepatocellular carcinoma; GO, Gene Ontology; MF, molecular function; CC, cellular composition; BP, biological process; KEGG, Kyoto Encyclopedia of Genes and Genomes; PI3K, phosphoinositide 3-kinase; AKT, protein kinase B.

To elucidate the molecular function (MF), cellular composition (CC), and biological process (BP) of the predicted target proteins, 89 potential targets of QT for HCC treatment were subjected to GO analysis, with the top 10 items selected by p-value for visualization (Fig. 5C–E). At the BP level, enrichment was primarily observed in the response to oxidative stress, positive regulation of the mitogen-activated protein kinase (MAPK) cascade, phosphoinositide 3-kinase (PI3K) signaling, and the cell cycle G2/M phase transition. At the CC level, enrichment was mainly in the protein kinase complex, membrane raft, and membrane microdomain. At the MF level, enrichment was in protein serine/threonine kinase activity and kinase regulator activity. KEGG enrichment analysis showed (Fig. 5F) that the therapeutic effect of QT on HCC may be regulated through signaling pathways such as PI3K/AKT (Fig. 5G), MAPK, and mammalian target of rapamycin.

To explore AKT1’s binding capacity to QT, we carried out molecular docking of the two to validate their binding (Fig. 6A,B). The findings demonstrated a binding energy of –9.6 kcal/mol, implying that AKT1 and QT underwent spontaneous and effective docking. The molecular docking results of EGFR, GSK3$\beta$, MMP9, SRC, and QT are presented in Supplementary Fig. 4. The RMSD is regarded as a criterion for measuring the stability of a system. During the simulation, the complex’s RMSD (represented by the red curve) initially showed an upward trend for the first 20,000 ps before stabilizing within a range of 0.2–0.4 nm. This stabilization of fluctuations indicated that the complex achieved a stable conformational state (Fig. 6C), revealing that after free AKT1 binds to the ligand, the stability of the complex is better than that in the free state, and binding to the small molecule ligand contributes to stabilization of the protein structure. The overall RMSF for the protein’s residues ranged from 0 to 0.5 nm, which is a measure of its flexibility. Analysis of residue fluctuations revealed a high degree of structural flexibility in specific regions, such as residues numbers 49 and 80, reaching nearly 0.5 nm. By contrast, the majority of residues displayed low to moderate fluctuations within the range of 0–0.3 nm, indicating a more rigid structure (Fig. 6D). HBs contribute to the stable binding of the complex. As observed in the figure, the number of HBs was frequently maintained within a range of 1–3. The HB interactions between the complexes showed favorable stability, evidenced by the large number of HBs that persisted for hundreds to thousands of ps. For example, the sustained presence of these HBs was observed over the 20,000–40,000 ps time window (Fig. 6E).

Fig. 6.

Molecular docking and MD simulation. (A) Molecular docking of AKT1 with QT (binding energies: –9.6 kcal/mol). (B) Molecular Docking 2D Structure. (C) RMSD curve of protein-ligand complexes. (D) RMSF curve of protein-ligand complexes. (E) Number of hydrogen bonds formed between AKT1 and QT. MD, molecular dynamics; RMSD, root mean square deviation; RMSF, root mean square fluctuation.

To validate the meta-analysis and network pharmacology findings, we performed in vitro experiments. First, the CCK-8 assays showed that QT inhibited the HCC cell line MHCC-97H in time- and dose-dependent manners. Treating MHCC-97H cells with 40 or 80 µM QT for 24 h resulted in ~10% and ~30% inhibition rates, respectively (Fig. 7A). Thus, these two concentrations were selected for subsequent experiments. Colony formation assays revealed that QT significantly reduced MHCC-97H colony numbers (Fig. 7B). A flow cytometry-based cell cycle analysis showed that QT induced G0/G1 phase arrest and prolonged the G2/M phase in MHCC-97H cells, indicating suppressed cell proliferation (Fig. 7C). Additionally, 5-ethynyl-2^′-deoxyuridine (EdU) incorporation assays demonstrated a significant decrease in the EdU-positive rate of QT-treated MHCC-97H cells (Fig. 7D). Collectively, these results confirmed that QT inhibited MHCC-97H proliferation. To further validate QT’s anti-HCC effects, we conducted in vivo experiments using tumor-bearing mice. The results showed that both the volume and weight of subcutaneous tumors in the QT group were significantly lower than those in the control group (Fig. 8A–C). Meanwhile, hematoxylin and eosin (H&E) staining indicated no obvious liver tissue damage in the QT group (Fig. 8D), suggesting that the therapeutic QT dose exerted no significant hepatic toxicity in mice. To explore the underlying mechanism of QT’s anti-HCC action, we performed immunohistochemistry. The results revealed significantly lower expression of phosphorylated AKT in the QT group (Fig. 8D), verifying the network pharmacology findings. Thus, QT likely inhibits HCC progression by suppressing the PI3K/AKT pathway.

Fig. 7.

Quercetin (QT) inhibits the proliferation of MHCC-97H cells in vitro. (A) CCK-8 shows time-and dose-dependent inhibition by QT. (B) Colony formation shows QT reduced colony numbers. (C) Flow cytometry shows QT induced G0/G1 arrest and prolonged G2/M phase. (D) EdU shows QT decreased EdU-positive rate. CCK-8, Cell Counting Kit-8; EdU, 5-ethynyl-2^′-deoxyuridine. Scale bar: 100 µm. ***: p 0.001.

Fig. 8.

QT inhibits HCC tumors by suppressing the PI3K/AKT signaling pathway in vivo. (A–C) Subcutaneous tumors in the QT group were remarkably lower than those in the control group. (D) Results of H&E staining and p-AKT IHC in the control group versus the QT group. H&E, Hematoxylin and Eosin Staining; p-AKT, Phosphorylated Protein Kinase B; IHC, Immunohistochemistry. Scale bar: 100 µm (main figure), 20 µm (enlarged view). ***: p 0.001.