Ginsenoside Rg3 inhibits melanoma progression by inducing ferroptosis via the p53/SLC7A11/GPX4 pathway

Graphical abstract

Highlights

• •Machine learning identified ferroptosis pathway in melanoma.
• •Screening multiple ginsenosides identified Rg3 for melanoma treatment.
• •In vitro and in vivo studies demonstrated Rg3′s efficacy against melanoma.
• •Integrated multi-omics revealed Rg3 modulated ferroptosis via p53/SLC7A11/GPX4 pathway.
• •Molecular docking and dynamics simulations confirmed stable Rg3 binding to pathway proteins.

Abstract

Introduction

Melanoma represents an aggressive cutaneous malignancy with limited treatment options. Ginsenoside Rg3, an active component extracted from the roots of Panax ginseng, has been extensively demonstrated to possess significant anti-tumor efficacy, showing promising application potential in the treatment of various malignancies. However, the role of Rg3 in melanoma treatment and its related mechanisms have not been reported in detail. Thus, exploring and elucidating reliable molecular mechanisms is critical.

Objectives

This study aimed to investigate the therapeutic efficacy of ginsenoside Rg3 in melanoma and its underlying mechanisms.

Methods

This research investigated the mechanism of melanoma pathogenesis using a clinical cohort database and established a corresponding disease model. By integrating transcriptomic and metabolomic approaches, we systematically explored the intrinsic molecular mechanisms by which Rg3 regulated ferroptosis in melanoma. Then, immunohistochemistry (IHC), reverse transcription quantitative polymerase chain reaction (RT-qPCR), western blotting (WB), molecular docking (MD), and molecular dynamics simulation (MDS) were used for validation.

Results

The database analysis revealed that melanoma was associated with the ferroptosis pathway. In vitro experiments showed that Rg3 inhibited the proliferation of melanoma cells. The functional annotation and enrichment analysis of differentially expressed genes and metabolites based on animal transcriptome and metabolome experiments showed that Rg3 exerted therapeutic effects by regulating the glutathione metabolism pathway and ferroptosis pathway. The detection based on the reagent kit revealed significant changes in ferroptosis-related biomarkers. IHC, RT-qPCR, and WB analyses confirmed the expression patterns of ferroptosis-associated mRNAs and proteins. MD and MDS confirmed stable binding between Rg3 and p53, SLC7A11, GPX4, and FTH1 proteins.

Conclusion

Rg3 induced melanoma ferroptosis via the p53/SLC7A11/GPX4 pathway, offering therapeutic potential for melanoma.

Introduction

Melanoma is a form of skin cancer. It results from the malignant transformation of melanocytes, which can develop from normal melanocytes or a primitive Epidermal nevus [1]. This type of cancer is marked by a high metastatic potential and a low cure ratio [2]. At present, the available treatment approaches for melanoma are quite restricted. Current therapeutic options remain limited to conventional approaches, including surgery, radiotherapy, and chemotherapy. However, surgery is only curative for localized primary tumors and is ineffective against metastatic melanoma, which is highly invasive [3]. Radiotherapy is often limited by radioresistance and cannot be employed as a systemic treatment [4], whereas chemotherapy is challenged by significant systemic toxicity and the rapid emergence of resistance [5,6]. Consequently, the efficacy of these conventional modalities remains unsatisfactory for melanoma. The prognosis is especially poor for metastatic cases, with only 32 % of patients surviving beyond five years, accounting for approximately 57,000 deaths annually worldwide [7]. These clinical challenges underscore the urgent need to identify key regulatory pathways in melanoma progression and develop more effective treatment strategies [8,9].

Iron is an essential micronutrient that plays vital roles in biological systems, participating in fundamental cellular processes including growth regulation, differentiation, and metabolic pathways [10]. Ferroptosis, an iron-mediated cell death process characterized by lethal lipid peroxidation, has gained attention as a potential cancer treatment target [11,12]. Emerging research has established crucial evidence demonstrating the significant involvement of ferroptosis in diverse pathological conditions, including tumor metastasis, multiple malignancies, tissue/organ injuries, and inflammatory or infectious diseases [13,14]. Studies designate glutathione peroxidase 4 (GPX4) and ferritin heavy chain 1 (FTH1) as central ferroptosis modulators [15]. The cystine/glutamate transporter SLC7A11 plays a critical role in cellular redox homeostasis by mediating cystine import for glutathione production, with its elevated expression observed in various malignancies, establishing it as a key ferroptosis regulator. Furthermore, the tumor suppressor p53 promotes ferroptosis susceptibility through transcriptional repression of SLC7A11, thereby limiting cellular cystine acquisition [16]. Consequently, developing therapeutic approaches to augment ferroptosis represents a promising avenue for melanoma intervention [17].

Ginsenosides, the primary bioactive components of Panax ginseng, constitute a diverse group of approximately 150 identified compounds [18]. Of the nearly 150 identified ginsenoside variants, Rg3 exhibits particularly notable bioactivity, demonstrating diverse therapeutic potentials including antioxidative, antineoplastic, immunomodulatory, and metabolic regulatory effects, along with anti-allergic properties [19]. Extensive experimental and clinical evidence has established Rg3′s potent antitumor properties [20]. In gallbladder carcinoma models, Rg3 induced growth arrest and promoted mitochondrial-mediated apoptosis through p53 pathway activation, ultimately triggering cellular senescence [21]. Furthermore, in triple-negative breast cancer, Rg3 enhanced paclitaxel’s therapeutic efficacy by suppressing NF-κB signaling while modulating apoptotic regulators, as evidenced by an elevated Bax/Bcl-2 ratio [22]. Additionally, ginsenoside Rg3 has been demonstrated to suppress TGF-β1 secretion, effectively reducing fibrotic scar formation [23]. Earlier research also revealed its ability to impede proliferation and DNA replication in B16F10 melanoma cells [24]. Notably, emerging evidence has begun to establish a connection between ginsenosides and ferroptosis regulation. For instance, ginsenoside Rh2 could repress prostate cancer progression through a combination of mitophagy and ferroptosis that led to mitochondrial damage [25]. Similarly, ginsenoside Rg3 has been reported to promote ferroptosis in hepatic stellate cells by epigenetically regulating ACSL4 to suppress liver fibrosis progression [26].

This study was designed to determine the role of Rg3-induced ferroptosis in the treatment of melanoma and to elucidate the underlying mechanism through a multi-omics strategy. Firstly, using a clinical cohort database, we employed a variety of machine learning methods to screen core genes and construct a survival prediction model. Pathway enrichment analysis was then performed on the screened core genes. We adopted an approach that combined transcriptomics and metabolomics with in vivo and in vitro models to investigate the therapeutic mechanisms targeting melanoma. Differential genes and their enriched pathways were screened based on transcriptome data, while metabolomics detection was used to identify characteristic metabolites related to pathway functions, such as GSH and lipid peroxides. This multi-omics integration uncovered the mechanistic relationship between gene expression patterns, metabolic phenotypes, and clinical outcomes in melanoma treatment. Furthermore, methods including immunohistochemistry (IHC), reverse transcription quantitative polymerase chain reaction (RT-qPCR), western blotting (WB), molecular docking (MD), and molecular dynamics simulation (MDS) were utilized to observe the effect of Rg3 on the expression of ferroptosis markers in mouse melanoma, aiming to further clarify the mechanism of Rg3 in treating melanoma.

Materials and methods

Reagents

Rg3 (purity > 98 %) was purchased from Chengdu Mansite Biotechnology Co., Ltd. Rg5 (purity > 98 %) was obtained from Chengdu Aifa Biotechnology Co., Ltd. Rh2 was purchased from Beijing Dongge Boye Biotechnology Co., Ltd. Rg1 (purity ≥ 98 %) was acquired from Shijiazhuang Ximo Technology Co., Ltd. Penicillin-streptomycin solution (biosharp, China), Roswell Park Memorial Institute 1640 (RPMI 1640) medium (Gibco, China), 0.25 % trypsin (Gibco, China), phosphate-buffered saline (PBS) (biosharp, China), and fetal bovine serum (FBS) were all purchased from Hangzhou Zhengbo Biotechnology Co., Ltd. Bicinchoninic Acid (BCA) protein assay kit (Beyotime, China), Radio-Immunoprecipitation Assay (RIPA) lysis buffer (biosharp, China), and Sodium Dodecyl Sulfate-Polyacrylamide Gel Electrophoresis (SDS-PAGE) protein loading buffer were purchased from Yamei Biotechnology Co., Ltd. (Shanghai, China). RNA extraction kit and Evo M−MLV reverse transcription premix kit were obtained from Hunan Aikerui Bioengineering Co., Ltd. 3-(4,5-Dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT), malondialdehyde (MDA) assay kit, oxidized glutathione (GSSG) detection kit, GSH content detection kit, and iron content detection kit were all purchased from Jiangsu Addison Biotechnology Co., Ltd. Polyclonal antibodies against p53, SLC7A11, GPX4, and FTH1 were purchased from Hangzhou Huanan Biological Co., Ltd.

Cell and animals

B16F10 cells were purchased from the Cell Resource Center (Shanghai Institute of Life Sciences, Chinese Academy of Sciences, Shanghai, China). Male C57BL/6 mice (6–7 weeks old, weighing 18–22 g) were obtained from the Experimental Animal Center of Zhejiang Chinese Medical University.

Dataset source

Transcriptomic profiles and corresponding clinical data were obtained for 472 melanoma patients and 1 normal skin sample from The Cancer Genome Atlas (TCGA) database. Two transcriptome datasets of melanoma patients were obtained from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/), including GSE54467 (n = 79) and GSE59455 (n = 141). Since there was only one normal skin sample in the TCGA database, we also retrieved transcriptome data of 812 normal skin tissues from the Genotype-Tissue Expression (GTEx) database and merged them with the data from the TCGA database.

Screening method for prognosis-related programmed cell death genes

Programmed cell death (PCD), alternatively referred to as regulated cell death, represented a key physiological process with a vital role in preserving tissue homeostasis and removing damaged or superfluous cells. PCD could be executed via multiple mechanisms such as apoptosis, anoikis, autophagy, and ferroptosis, among others. It operated through an intrinsic regulatory system that orchestrated the removal of compromised or superfluous cells, thereby preserving tissue integrity [27].

The limma software package was applied to conduct differential expression analysis of PCD-associated genes between normal and tumor samples in the TCGA and GTEx databases, with the screening thresholds set at p < 0.05 and log10 (FC) > 0.5. Univariate Cox regression was performed on the identified differential genes, and p < 0.05 was used as the criterion to evaluate their prognostic value for melanoma patients, thereby identifying prognosis-related PCD genes.

Multi-model construction and survival analysis design

In this study, a ten-fold cross-validation method was used to fit 114 prediction models, including Random Forest (RF), Lasso, glmBoost, Enet, plsRglm and other models, so as to realize the feature screening of transcriptome data. By setting the screening condition that the number of features was greater than 5, the models that met the requirements were retained, and the most valuable model was determined by taking the concordance index (C-index) as the indicator. In terms of dataset processing, the training set was composed of the merged data of TCGA data, 526 GTEx normal tissue data and GSE54467 dataset. The test set was the merged result of 285 GTEx normal tissue data and GSE59455 dataset. Based on the expression levels of core genes in tumor tissues, patients were stratified into high-and low-expression groups, and survival analysis was performed to compare outcomes between these cohorts.

Functional annotation and enrichment analysis of genes

The ‘ClusterProfiler’ R package was used to perform assessments based on the Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology (GO), aiming to identify the enriched gene functions and cellular pathways among the core genes in the network.

Immune microenvironment analysis

Ferroptosis in tumor cells involved a dynamic interplay between inflammatory and immune factors. On one hand, ferroptosis-derived lipid peroxidation products triggered immune responses, while anti-inflammatory factors like IL-10 were adaptively upregulated to prevent excessive inflammation. On the other hand, increased expression of immune factors such as IFN-γ might be enhanced due to immune cell activation, which could kill tumor cells by synergizing with the ferroptosis pathway. To characterize this immune microenvironment, we employed six analytical approaches: CIBERSORT, EPIC, MCP-counter, quanTIseq, single-sample gene set enrichment analysis (ssGSEA), and xCell for comparative analysis of normal and tumor tissues.

Cell cultures

Cell cultures were maintained at 37 °C with 5 % CO₂ humidified atmosphere in RPMI-1640 medium containing 10 % fetal bovine serum (FBS) and 1 % penicillin–streptomycin.

Cell viability assays

B16F10 cells were seeded into 96-well plates at a density of 3.5 × 10⁴ cells per well and incubated overnight at 37 °C. After cell attachment, the cells were treated with increasing doses of Rg3, Rg5, Rg1, Rh2 (7.8125, 15.625, 31.25, 62.5, 125, 250, 500, and 1000 μM) for 24 h. Then, 20 µl of the MTT solution (5 mg/ml) was added to each well and incubated for 4 h at 37 °C. Subsequently, the cytotoxicity of B16F10 cells were determined by MTT assay. The absorbance was measured at 490 nm using a microplate reader (Bio-Tek, American).

Colony formation assay

The proliferative ability of B16F10 cells was assessed via colony formation assay under ultra-low-density conditions (n = 3). The proliferative ability of B16F10 cells was assessed via colony formation assay under ultra-low-density conditions. Briefly, B16F10 cells were plated into 6-well plates at 1000 cells/well. After the cells adhered to the wall, they were treated with 0, 10 μM, 20 μM, and 40 μM Rg3 for 48 h [28]. Cells were routinely cultured at 37 °C. After 7 days, the medium was aspirated, followed by two gentle PBS washes. After removing PBS, the cells were fixed with formaldehyde for 30 min and then stained with 0.1 % crystal violet at a volume of 1 ml per well for 30 min. Colonies were then imaged for proliferation analysis.

Wound healing assay

The wound healing assay was employed to evaluate cell migration (n = 3). In brief, 5 × 10⁵ cells were plated into each well of a 6-well plate and cultured overnight to form a confluent cell monolayer. Subsequently, a sterile scratch was made through the middle of each well using a 200 μL pipette tip [29]. After removing non-adherent cells, fresh serum-free RPMI 1640 medium containing Rg3 (0, 10, 20, and 40 μM) was added to each well. The plates were then incubated under standard conditions to assess cell migration. Images were captured at 0 and 24 h using an inverted microscope (Nikon, Japan). The migration distances were quantified with ImageJ software.

$$\mathrm{M}\mathrm{i}gration\left(\%\right)=\frac{\mathit{Initialscratcharea}-Finalscratcharea}{Ini\mathrm{t}\mathrm{i}\mathrm{a}\mathrm{l}\mathrm{s}\mathrm{c}\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{c}\mathrm{h}\mathrm{a}\mathrm{r}\mathrm{e}\mathrm{a}}\times 100\%$$

Xenograft model and treatment

B16F10 tumor cells were injected subcutaneously into the right axillary region of C57BL/6 mice at a concentration of 1 × 10⁶ cells per milliliter. When tumors grew to ∼ 100 mm3, mice were randomized into three groups (n = 6 per group): the model group (0.2 ml/10 g normal saline daily by gavage), the low-dose group (20 mg/kg Rg3 daily by gavage), and the high-dose group (40 mg/kg Rg3 daily by gavage). Tumor volume and body weight were measured daily. After 7 days of treatment, mice were euthanized for tumor collection and subsequent analysis.

Hematoxylin-Eosin staining (H&E)

Tumor tissues were fixed in 4 % paraformaldehyde, then subjected to dehydration and paraffin embedding. Following dewaxing, tissue sections were stained with H&E. Histopathological images were acquired using a digital slide scanner (KFBIO, China).

Immunohistochemical analyses

Tumor tissue sections were dewaxed using xylene and ethanol. Following antigen retrieval, they were blocked with 3 % bovine serum albumin (BSA) and then incubated with antibodies against p53, SLC7A11, GPX4, and FTH1 [30]. Under the microscope, positive tissue areas showed brown or tan staining, and tissue images were captured with an inverted microscope. Three random visual fields were selected from each section. The average optical density (AOD) of the tissue sections was measured using ImageJ software, and this AOD value was taken as the relative expression level of the corresponding protein.

Metabolomics analysis

To 10 mg of tumor tissue samples, 1300 μL of pre-chilled 80 % methanol was added. The cells were then lysed through two rounds of rapid freezing and thawing using liquid nitrogen. After storage at −20 °C for 1 h, supernatants were collected by centrifugation (12,000 × g, 10 min, 4 °C) and concentrated using nitrogen blow-down. Samples were reconstituted in 100 μL of 50 % aqueous acetonitrile before UHPLC/Q-TOF-MS analysis.

Transcriptome sequencing

Total RNA was extracted using the RNeasy Mini Kit (Qiagen, Germany). Using the TruSeq Stranded Total RNA Kit (Illumina, USA), strand-specific libraries were constructed per manufacturer’s instructions. mRNA was purified by oligo (dT) selection, thermally fragmented (86 °C, 6 min) in the presence of divalent cations, and converted to first-strand cDNA using random priming. Second-strand synthesis was performed with DNA Polymerase I and RNase H. After end repair, adenylation, and adapter ligation, libraries were PCR-amplified, quantified by Qubit 2.0 Fluorometer (Life Technologies), and validated on an Agilent 2100 bioanalyzer for size distribution and concentration. Sequencing was performed on an Illumina NovaSeq 6000 at 10 pM library concentration.

Integration of metabolomic and transcriptomic analysis

R language was used for multivariate statistical analysis to investigate associations metabolome-transcriptome associations. Firstly, Spearman correlation analysis systematically evaluated relationships between all detected metabolites and mRNA expression levels in key pathways. To further dissect the association patterns of the two omics data, two multivariate statistical methods, namely Canonical Correlation Analysis (CCA) and Two-way Orthogonal Partial Least Squares (O2PLS) regression, were further applied to conduct in-depth correlation mining on the same dataset.

Detection of IL-10, IFN-γ, TGF-β, IL-6, CXCL8, TNF-α, GSSG, GSH, GSSG, MDA and Fe²⁺by using ELISA

Tumor samples were weighed, homogenized in chilled PBS, and centrifuged (12,000g, 10 min) to obtain supernatant. The concentrations of IL-10, IFN-γ, TGF-β, IL-6, CXCL8, TNF-α, MDA, GSSG, GSH, and iron ions in mouse tumor tissues were detected using corresponding ELISA kits. The absorbance at 450 nm (OD value) was determined with a microplate reader, and the concentrations were computed from standard curves in accordance with the manufacturer’s instructions.

RT-qPCR

Total RNA was extracted with Trizol reagent following the manufacturer’s guidelines. cDNA was synthesized via reverse transcription using the Evo M−MLV Reverse Transcription Premix Kit. mRNA levels were quantified using SYBR Green-based RT-qPCR (Thermo Fisher system). GAPDH served as the reference gene for normalization, and data analysis was conducted using the comparative 2-ΔΔCT method [31].

Western blotting

Protein expression levels were assessed via WB. Briefly, ∼40 mg of tumor tissue was homogenized in RIPA protein lysis buffer containing protease inhibitors at a ratio of 1:9 (g/ml) [32]. Total proteins were extracted, and their concentrations were determined using a BCA protein assay kit (Beyotime, China). Samples in 5 × loading buffer were heat-denatured at 95 °C for 15 min followed by cooling. A 10 % separating gel was prepared, and 5 μL of each protein sample was loaded to ensure a consistent 40 μg protein per lane. Following SDS-PAGE, proteins were wet-transferred to PVDF membranes, blocked with 5 % skim milk (1 h, RT), and probed overnight (4 °C) with primary antibodies: anti-p53 (1:1000), anti-SLC7A11 (1:1000), anti-GPX4 (1:10000), and anti-FTH1 (1:5000). Following three 5-minute washes with Tris-Buffered Saline with Tween-20 (TBST), the membranes were incubated with secondary antibodies (1:5000) for 60 min. After another three washes with TBST, bands were visualized using a chemiluminescence imager (Monad, China). Protein expression was quantified using ImageJ software.

Molecular docking

To explore the interaction between the active component Rg3 and core targets, molecular docking simulations were conducted using the AutoDock Vina. Rg3’s molecular structure was obtained from PubChem Compound, with key target protein structures downloaded from the Protein Data Bank. Following the removal of crystal water and heteroatoms via PyMOL software, these proteins were prepared and loaded as molecular receptors. AutoDock Vina software was used for molecular docking with p53 (PDB ID: 4cz7), SLC7A11 (PDB ID: 7p9u), GPX4 (PDB ID: 7u4j), and FTH1 (PDB ID: 8wie), and the molecular docking data were visualized using the PyMOL molecular graphics system [33]. A lower value indicated a higher binding affinity. In this study, a binding energy of ≤ -5 kcal/mol was used as the screening threshold. LigPlot + was used to display and evaluate the interactions between ligands and proteins.

Molecular dynamics simulation

The binding stability of Rg3 with p53/SLC7A11/GPX4/FTH1 was evaluated through 200 ns MD simulations (GROMACS 2021.6) using Amber_14SB and TIP3P force fields. Starting from the lowest-energy docking conformation, systems were prepared with Sobtop (ligand) and Gromacs (protein) topologies, solvated in a charge-neutralized box (Na⁺ added). After energy minimization and 10 ns NPT equilibration (298.15 K, 1 bar), production simulations (200 ns, leapfrog algorithm) were conducted with V-rescale thermostat and Parrinello-Rahman barostat, saving trajectories every 10 ps. Based on the trajectories of the production simulation, the root mean square deviation (RMSD), radius of gyration (Rg), and solvent-accessible surface area (SASA) were calculated using GROMACS trajectory analysis tools, and hydrogen bond (HBOND) analysis and principal component analysis (PCA) were performed. The Molecular Mechanics Poisson-Boltzmann Surface Area (MMPBSA) method was employed to estimate binding free energies. The binding free energy was calculated using the tool. Visualization was performed using Visual Molecular Dynamics (VMD) software.

Ethics statement

All animal experiments were performed in accordance with the National Research Council Guidelines for Laboratory Care and Use and approved by the Animal Experimental Ethics Committee of Zhejiang Chinese Medical university, with the ethical approval number IACUC-20240923–10.

Statistical analysis

All data were expressed as mean ± standard deviation. Statistical analyses were performed using GraphPad Prism 8.0.2, with two-group comparisons assessed by two-tailed non-parametric t-tests and multi-group comparisons by one-way ANOVA. p-value < 0.05 was considered statistically significant.

Result

Screening results of prognosis-related programmed cell death genes

Through analysis with the limma package, a total of 344 PCD-related genes with differential expression between normal and tumor samples were screened out (Fig. 1A). After univariate Cox regression analysis, 116 genes with p < 0.05 were identified (Fig. 1B). These genes were members of the PCD gene family and were linked to the prognostic results of melanoma patients. The expression levels of these genes in the samples were illustrated in Fig. 1C.

Fig. 1.

Differential genes between normal samples and tumor samples (A), univariate Cox analysis to identify potential prognostic genes (B), expression levels of prognosis-related genes (C), C-index of 67 prediction models (D), AUC of the training set (E) and test set (GSE59455) (F).

Comparison of multi-model performance and screening results of core genes

The model analysis results showed that after ten-fold cross-validation, the Enet [alpha = 0.2] models had the highest C-index, and 65 core genes were identified (Fig. 1D). The AUC index analysis of the Enet [alpha = 0.2] model indicated that the accuracy rate of the training set model was 1, and that of the test set model reached 0.999 (Fig. 1E-F). The model confusion matrix showed that the training set data were successfully identified. In the validation set, 6 normal samples were misjudged as tumor samples, and all tumor samples were correctly identified (Fig. 2A). Survival analysis revealed that patients with lower expression levels of the 65 core genes exhibited significantly better prognosis and longer survival compared to high-expression groups (Fig. 2B).

Fig. 2.

Confusion matrix of Enet [alpha = 0.2] (A), patient survival time in the training set and test set (GSE59455) (B), GO analysis (C), KEGG analysis (D), sample immune microenvironment map (E), comparison map of immune cells between normal and tumor samples (F).

Core gene pathway enrichment analysis

GO enrichment analysis was conducted from the perspective of gene function classification, such as biological processes, molecular functions, and cellular components, while KEGG enrichment analysis was performed through signal pathway mapping.

The KEGG enrichment analysis results showed that the incidence of melanoma was related to cytokine-cytokine receptor interaction, JAK-STAT signaling pathway, ferroptosis, and lysosome (Fig. 2C), while the GO enrichment analysis results showed that it was related to ferrous iron binding, plasma membrane raft and glutamate metabolic process (Fig. 2D). These findings strongly implicated ferroptosis in melanoma development.

Comparison of immune cells between normal and tumor samples

Analysis using the above 6 methods demonstrated that, at the overall level of immune cells, the number of immune cells in tissue samples from tumor patients was significantly higher than that in normal tissue samples (Fig. 2F). These findings indicated distinct immune microenvironment profiles between tumor and normal tissues, with immune cell infiltration potentially contributing to tumorigenesis and progression.

Screening the ginsenoside component that exhibited the strongest inhibitory effect on B16F12 cells

To screen out ginsenoside components with significant inhibitory effects on melanoma, this study selected ginsenosides Rg3, Rg5, Rh2, and Rg1 for in vitro pharmacodynamic evaluation. The half-maximal inhibitory concentration (IC₅₀) was calculated by detecting the inhibition rate of each component on melanoma cell proliferation at different concentrations. The IC₅₀ value reflected drug potency, with lower values indicating stronger anti-tumor effects at reduced concentrations. Ginsenoside Rg3 showed the strongest inhibitory ability on melanoma cells among the four components, with its IC₅₀ value significantly lower than those of the other three saponins (Fig. 3A), suggesting that it had higher application potential in melanoma treatment. Therefore, in subsequent studies, ginsenoside Rg3 was taken as the core research object to further explore its in vivo pharmacodynamic effects and molecular mechanism against melanoma.

Fig. 3.

IC₅₀ values of Rg3 (A), Rg5 (B), Rh2 (C) and Rg1 (D) on B16F12 cells, wound healing assay of B16F10 cells (E), colony formation assay of B16F10 cells (F), body weight change of mice from day 7 to 14 (n = 6) (G), tumor volume change of mice from day 7 to 14 (n = 6) (H), H&E staining images of tumor tissues (I). Data are shown as mean ± SD (n = 6). * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001 compared to the model.

Rg3 significantly inhibited proliferation and migration of B16F10 cells

To verify the anti-melanoma effect of Rg3, wound healing assay and cell clone formation assay were used to detect the inhibitory effect of Rg3 on the migration and proliferation abilities of B16F10 cells. In the wound healing assay, the migration rates of B16F10 cells in the control group and those treated with 10 μM, 20 μM, and 40 μM Rg3 were 77.22 %, 52.55 %, 25.10 % and 3.61 % respectively, indicating that Rg3 significantly inhibited the migration ability of Rg3 cells in a dose-dependent manner (Fig. 3E). In colony formation assay, compared with the control group, after 7 days of culture with Rg3, the number of colony formations decreased significantly with the increase of dose (Fig. 3F). These results demonstrated that Rg3 exhibited potent anti-tumor activity against melanoma, significantly inhibiting both proliferation and metastasis of B16F10 cells.

Evaluation of tumor development and pathological alterations

B16F10 cells were injected into the left axilla of mice to successfully establish a melanoma mouse model, which was used to evaluate the in vivo anti-tumor efficacy of Rg3. Body weight measurements revealed no significant differences between the model and treatment groups (Fig. 3G). The tumor growth curve showed that high-dose Rg3 treatment significantly slowed tumor progression (Fig. 3H). At day 7 post-treatment, a statistically significant reduction in tumor volume was observed in the high-dose group (p < 0.05), whereas the low-dose regimen showed negligible inhibitory effects. In addition, under Rg3 treatment, H&E staining showed that in the model control group, tumor cell nuclei were deeply stained, cells were hypertrophic, arranged closely, and the boundaries between cells were clear. In the high-dose Rg3 group, scattered reduced density, cell rupture, a large number of vacuolar structures, and obvious nuclear condensation, fragmentation and dissolution were observed, indicating cell necrosis, while the effect in the low-dose group was not obvious (Fig. 3I).

Results of Rg3 regulating metabolomics in melanoma mice

Studies have shown that Rg3 can inhibit the proliferation of melanoma, but its underlying mechanism remains unclear. To further investigate the changes in metabolites of melanoma mice induced by Rg3, we conducted a metabolomic study. The results in positive ion mode were presented in Fig. 4A, C, E, while those in negative ion mode were shown in Fig. 4B, D, F. As shown in the PCA (Fig. 4A-B), there were significant differences in the global metabolic profiles between the Model group and the Rg3-H group. To further distinguish metabolic differences, OPLS-DA was performed on all analyzed metabolites to evaluate the changes in metabolites in mice. The metabolic profiles of tumor tissues in the Model group and the Rg3-H group showed complete differentiation, which suggested marked differences between the two groups (Fig. 4C-D). Differential analysis was performed on the identified metabolites (Fig. 4E-F). The differential metabolites simultaneously met the criteria: FC >= 1.2 or FC <= 1/1.2, and p value < 0.05. By combining the results of positive and negative spectra, a total of 53 differential metabolites were identified, and their expression levels in the samples were shown in Fig. 4G. Through KEGG analysis of differential metabolites (Fig. 4H), these metabolites with differential abundances were considered to be involved in amino acid metabolic pathways, such as histidine metabolism, tyrosine metabolism, and glutathione metabolism. GSH, as the central antioxidant in ferroptosis, eliminated lipid peroxides via GPX4. GSSG indicated oxidative stress, and an imbalance in the ratio promotes ferroptosis. The degradation intermediate product cysteinylglycine impaired GSH regeneration. These three molecules regulated ferroptosis through metabolic balance and were therapeutic targets for related diseases. Among the differential metabolites, cysteinylglycine, GSH and GSSG were identified, all of which were key molecules in the glutathione metabolic pathway. The GSEA of the glutathione metabolic pathway was shown in Fig. 4I.

Fig. 4.

PCA analysis in positive ion mode (A), OPLS-DA analysis in negative ion mode (B), OPLS-DA analysis in positive ion mode (C), PCA analysis in negative ion mode (D), volcano plot of metabolite changes between Model group and Rg3-H group in negative ion mode (E), metabolite changes between Model group and Rg3-H group in negative ion mode (F), metabolite changes between Model group and Rg3-H group (G), KEGG pathway enrichment analysis between Model group and Rg3-H group (H), GSEA enrichment analysis of glutathione metabolism (I).

Rna-seq analysis of Rg3-regulated gene expression in melanoma tissues

The transcriptomic analysis evaluated the effect of Rg3 on gene expression in melanoma mice. PCA showed clustering of mice in the same phenotypic group (Fig. 5A). To further distinguish gene differences, OPLS-DA was performed on all analyzed genes to assess transcriptional differences in mice (Fig. 5B). As shown in the volcano plot (Fig. 3C), a total of 1,325 differentially expressed genes (DEGs) were identified between the model group mice and the high-dose Rg3 group mice, among which 1,182 were upregulated and 143 were downregulated. The relative expression levels of these DEGs were visualized in the heatmap (Fig. 5D). To investigate the functional roles of these DEGs, KEGG pathway analysis was performed (Fig. 5E). The 7 enriched pathways included ECM-receptor interaction, Superoxide radical degradation, p53 signaling pathway, etc., which were related to the ferroptosis pathway. GSEA showed that the p53 pathway was positively enriched, indicating that after Rg3 treatment, the p53 pathway in melanoma cells was activated (Fig. 5F). The p53 pathway was an upstream regulatory factor of abnormal glutathione metabolism, and the genes in the glutathione metabolic pathway showed positive enrichment. The depletion of glutathione in the metabolome was the core metabolic hub connecting p53 and ferroptosis. Therefore, based on metabolomics and transcriptomics analyses, we confirmed that Rg3 induced ferroptosis in melanoma through the p53 pathway.

Fig. 5.

PCA analysis (A), OPLS-DA analysis (B), volcano plot of gene changes between the Model group and Rg3-H group (C), heatmap of gene changes between the Model group and Rg3-H group (D), KEGG pathway enrichment analysis between the Model group and Rg3-H group (E), GSEA enrichment analysis of the p53 signaling pathway (F).

Rg3 inhibited melanoma ferroptosis through glutathione-related metabolites and key genes

To investigate the mechanism underlying Rg3’s therapeutic effects in melanoma, we integrated metabolomic and transcriptomic data for correlation analysis. In multi-omics association studies, we first performed a comprehensive Spearman analysis to explore relationships between metabolomic and transcriptomic datasets (Fig. 6A). Based on findings from 3.7 metabolomic analyses, we focused on three key metabolites (cysteinylglycine, glutathione, and oxidized glutathione) and identified the top 10 mRNAs most strongly correlated with each metabolite (Fig. 6B). Results showed significant correlations between these metabolites and genes such as complement factor I (Cfi), myosin light chain kinase 2 (Mylk2), and adenylate cyclase 1 (Adcy1), suggesting their potential involvement in regulating redox homeostasis.

Fig. 6.

Spearman correlation heatmap (A), genes most strongly correlated with cysteinylglycine, glutathione, and oxidized glutathione (B), canonical correlation analysis (CCA) of cysteinylglycine, glutathione, oxidized glutathione and related genes (C), O2PLS analysis of cysteinylglycine, glutathione, oxidized glutathione (D) and related genes (E), correlation between oxidized glutathione and Adcy1 (F), correlation between cysteinylglycine and Cfi, correlations between glutathione and Gsk3b, Mylk2, Sv2b (G).

Further canonical correlation analysis (CCA) revealed synergistic relationships among multivariables (Fig. 6C), with genes including Vav1, Plin1, and Mylk2 contributing prominently to metabolite-gene co-expression patterns. Orthogonal partial least squares (O2PLS) analysis (Fig. 6D-E) identified genes such as Nkd1, Adcy1, and glycogen synthase kinase 3β (Gsk3b) as potential regulatory nodes. Integrating results from all three analytical methods, we observed that oxidized glutathione exhibited a significant correlation with Adcy1 (R² = 0.49, p = 0.011; Fig. 6F), cysteinylglycine showed a strong positive correlation with Cfi (R² = 0.63, p = 0.002; Fig. 6G), and glutathione was highly correlated with Gsk3b (R² = 0.83, p < 0.001), Mylk2 (R² = 0.66, p = 0.0013), and synaptosomal-associated protein 2B (Sv2b, R² = 0.76, p < 0.001; Fig. 6H). These high-confidence associations suggested that genes such as Cfi, Gsk3b, and Mylk2 were involved in Rg3-mediated inhibition of melanoma cell ferroptosis by regulating glutathione, oxidized glutathione, and cysteinylglycine levels.

Rg3 systemically changed the inflammatory and immune microenvironment in murine melanoma

To determine whether ferroptosis influenced the inflammatory and immune microenvironments, our aforementioned machine learning analysis revealed a significantly higher abundance of immune cells in tumor samples compared to normal tissues. We further analyzed the immune and metabolic microenvironments using murine transcriptomic data, with results shown in Fig. 7A and 7E, following Rg3 administration, immune cell expression was upregulated, while the expression of inflammation-related pathways was downregulated. Kit-based validation assays revealed that Rg3-treated mice showed significantly higher levels of IL-10, IFN-γ, and TGF-β in tumor tissues compared to the model group (Fig. 7B-D), along with markedly lower expression of pro-inflammatory cytokines IL-6, TNF-α, and CXCL8 (Fig. 7F-H). These findings demonstrated that Rg3 also exerted therapeutic effects on melanoma by modulating and improving the inflammatory and immune microenvironments.

Fig. 7.

Heatmap of the immune microenvironment (A), levels of IL-10 in tumor tissues (B), levels of IFN-γ in tumor tissues (C), levels of TGF-β in tumor tissues(D), heatmap of the inflammatory microenvironment (E), levels of IL-6 in tumor tissues (F), levels of CXCL8 in tumor tissues (G), levels of TNF-α in tumor tissues (H).

Rg3 induced ferroptosis to suppress tumor growth

To validate the metabolomic and transcriptomic findings, we performed biochemical analyses using assay kits. These experiments revealed a significant increase in intracellular Fe²⁺ levels (Fig. 8A), a reduction in GSH (Fig. 8B), elevated GSSG (Fig. 8C), a decreased GSH/GSSG ratio (Fig. 8D), and heightened MDA concentrations (Fig. 8E).

Fig. 8.

Levels of Fe²⁺ in tumor tissues (A), levels of GSH in tumor tissues (B), levels of GSSG in tumor tissues (C), GSH/GSSG ratio (D), levels of MDA in tumor tissues (E), IHC staining images showing protein expression of p53, SLC7A11, GPX4, and FTH1 (Scale bar = 100 μm) (F), mean optical density values (G), relative mRNA expression of p53, SLC7A11, GPX4, and FTH1 (H), protein expression levels of p53, SLC7A11, GPX4, and FTH1 (I,J).

Ferroptosis, an iron-dependent mode of cell death triggered by lipid peroxidation, was mechanistically associated with disturbances in glutathione metabolism, the buildup of lipid peroxides, and imbalances in iron homeostasis. As a central intracellular antioxidant, the reduced GSH levels and the diminished GSH/GSSG ratio indicated impaired antioxidant capacity, fostering conditions permissive for lipid peroxidation. Increased MDA levels, a characteristic marker of lipid peroxidation, served to further verify the presence of oxidative stress. Moreover, increased intracellular Fe²⁺, a key driver of ferroptosis, exacerbated lipid peroxidation cascades. These observations aligned with our metabolomic and transcriptomic data, collectively demonstrating that Rg3 induced ferroptosis in melanoma cells.

Rg3 induced ferroptosis by modulating p53, SLC7A11, GPX4, and FTH1 expression in murine melanoma tissues

Given that SLC7A11, GPX4, and FTH1 were key regulators of ferroptosis, we examined their expression alongside p53. WB and RT-qPCR analyses assessed the expression levels of p53, SLC7A11, GPX4, and FTH1 in mouse tumor tissues. In Rg3-treated tumors, p53 levels were elevated, whereas SLC7A11, GPX4, and FTH1 levels were significantly reduced (p < 0.05).

To further validate the expression trends and localization of these differentially expressed proteins, IHC staining was performed on tumor sections. Compared with the model group, Rg3 treatment significantly upregulated the ferroptosis-related protein p53 and downregulated SLC7A11, GPX4, and FTH1 (Fig. 8F-G). These findings were consistent with RT-qPCR (Fig. 8H) and WB results (Fig. 8I-J), collectively demonstrating that Rg3 modulated the expression of p53, SLC7A11, GPX4, and FTH1 to induce ferroptosis.

Molecular docking and molecular dynamics simulations revealed high affinity of Rg3 for ferroptosis-related proteins

To further investigate the binding potential of Rg3 with key proteins in the p53/SLC7A11/GPX4 signaling pathway, molecular docking was performed using AutoDock (Fig. 9A). The binding energies of Rg3 with p53, SLC7A11, GPX4, and FTH1 were calculated as −8.2, −7.8, −9.1, and −13.2 kcal/mol, respectively (Fig. 9A). MDS were employed to analyze the stability of protein–ligand interactions and protein structural changes upon ligand binding. RMSD assessed structural fluctuations of the receptor-ligand complexes, with lower values indicating greater conformational stability. The Rg3-FTH1 complex exhibited the most stable conformation (Fig. 9B). The Rg reflected spatial folding during dynamics, with lower values indicating a more compact structure, indicating that the Rg3-GPX4 complex was the most compact. SASA represented the area of protein surface in contact with solvent, with larger SASA values indicating greater surface exposure of the protein (Fig. 9F). This revealed that SLC7A11 had the highest surface exposure. Analysis showed that protein–ligand complexes reached equilibrium after 200 ns of production simulation. Hydrogen bond analysis revealed a consistent number of hydrogen bonds between the proteins and Rg3, with an increase in hydrogen bond count corresponding to larger protein molecular weights, suggesting that larger protein receptors possessed more hydrogen-bonding sites. Hydrogen bonds between FTH1, GPX4, SLC7A11, and Rg3 exhibited slow decay, indicating stable hydrogen-bonding interactions (Fig. 9D). Further MMPBSA analysis of interaction modes revealed that the binding free energy trends of protein–ligand complexes correlated with the number of hydrogen bonds (Fig. 9E), suggesting that binding was primarily hydrogen bond-driven. Decomposition of binding free energies revealed that these energies stemmed from a combination of gas-phase free energy and polar solvation free energy. Electrostatic and hydrophobic interactions between the protein and ligand collectively drove hydrogen bond formation and stabilized binding, which was counterbalanced by polar solute–solvent interactions, while nonpolar solvation energy contributed negligibly. PCA of binding modes showed distinct conformational dynamics in all four proteins upon Rg3 binding (Fig. 9G), suggesting self-assembly optimized ligand engagement. These results demonstrated Rg3′s high affinity for p53, SLC7A11, GPX4 and FTH1.

Fig. 9.

Molecular docking results and Ligplot + diagrams of Rg3 with FTH1, GPX4, p53, and SLC7A11 (A), RMSD plots of FTH1, GPX4, p53, and SLC7A11 (B), Rg plots of FTH1, GPX4, p53, and SLC7A11 (C), plots of the number of hydrogen bonds in FTH1, GPX4, p53, and SLC7A11 (D), binding energy plots of FTH1, GPX4, p53, and SLC7A11 with Rg3 (E), SASA plots of FTH1, GPX4, P53 and SLC7A11 (F), free energy landscape plots of FTH1, GPX4, p53and SLC7A11 (G).

Discussion

Malignant melanoma mainly originates from skin tissue and is a highly aggressive tumor [34]. The existing treatment regimens have limited efficacy and often cause a large number of toxic and adverse side effects [35]. Therefore, there is an urgent need to develop safe and effective therapeutic agents that can effectively inhibit the progression and metastasis of melanoma. Notably, ginseng’s potential anti-tumor properties were identified as early as the 1960 s-1970 s, with studies demonstrating its ability to inhibit the growth of certain aggressive cancers [36]. Numerous studies have identified ginsenosides as the principal bioactive compounds contributing to ginseng’s antitumor efficacy [37]. Among these, the steroidal saponin Rg3 has demonstrated diverse pharmacological effects, with particularly prominent anticancer potential [38]. When compared to other recent similar approaches, ginsenoside Rg3 demonstrates superior potency. Ginsenoside Rg3 is the active ingredient in Shenyi Capsule, an approved anticancer drug which is clinically used to treat lung and liver cancer. However, its potential in melanoma therapy has not been extensively investigated. Although some preliminary in vitro studies existed, evidence supporting its use for melanoma remained scarce. However, our study found that ginsenoside Rg3 could treat melanoma and broaden its therapeutic scope, which provided a reference for clinical medication. For the treatment mechanism of melanoma, previous studies have focused on GSK3 β/β-Catenin/MITF axis [39], AhR-ALKAL1 protein complex [40], ATAD2 protein [41], etc. Our study revealed that ginsenoside Rg3 exerted its antitumor effect in melanoma by inducing ferroptosis through the p53/SLC7A11/GPX4 signaling pathway. Our findings elucidated a novel mechanism of action for this compound. Our study also used a new method of integrating machine learning with multi omics to discover and validate the mechanism of action of ginsenoside Rg3 in the treatment of melanoma. This method was a universal approach that was also applicable in the treatment of other drugs and diseases.

In this study, we first systematically retrieved the transcriptomic data and corresponding clinical information of melanoma patients from TCGA and GEO databases, laying a solid data foundation for subsequent in-depth analyses. Using the limma package, we focused on the genes related to PCD and conducted a meticulous differential expression analysis between normal and tumor tissue samples from the aforementioned databases. Through this analysis process, we successfully screened out the PCD genes closely associated with the patients’ prognosis. To achieve feature selection of the transcriptomic data and sample classification, we adopted the ten-fold cross-validation method to fit and optimize 114 predictive models. Results from the survival analysis indicated that, according to the expression levels of the 65 screened core genes, patients could be distinctly categorized into a high-expression group and a low-expression group. Notably, the patients in the low-expression group exhibited significantly better prognosis and longer survival times. Additionally, the R package ‘ClusterProfiler’ was employed to carry out KEGG pathway enrichment analysis as well as GO functional enrichment analysis. The results indicated that the pathways enriched by these core genes were closely related to ferroptosis. Metabolomic pathway analysis detected glutathione metabolism as a prominently enriched pathway. In the transcriptomic analysis, the enrichment analysis of differentially expressed genes identified the p53 pathway. The depletion of glutathione in the metabolome constituted a core metabolic hub linking p53 and ferroptosis. Our data demonstrated that Rg3 acted as a potent inducer of ferroptosis, a conclusion supported by characteristic marker changes including depleted GSH, elevated GSSG, a lowered GSH/GSSG ratio, increased MDA, and Fe²⁺ accumulation. Moreover, Rg3 exerted potent antitumor effects in vivo by up-regulating p53 and concurrently down-regulating SLC7A11, GPX4, and FTH1 in tumor tissues, thereby disrupting the redox balance. As was shown in Fig. 10, these findings provided compelling evidence that Rg3 achieved melanoma suppression by triggering ferroptosis, highlighting its promising therapeutic value. In our in vitro study, Rg3 exhibited significant inhibitory effects on B16F10 cells at very low concentrations, with an IC₅₀ of 17.04 μM. In in vivo studies, Rg3 exhibited an approximate 40 % inhibition rate on tumor growth after oral administration at a dose of 40 mg/kg for 7 days. In traditional melanoma treatment regimens, Dacarbazine, the most commonly used single-agent chemotherapy for melanoma over the past decades, demonstrates a slow onset of action and modest efficacy. In one study, a regimen of 10 mg/kg via tail vein injection in mice achieved a tumor inhibition rate of merely 40 % only after 24 days of administration [42]. Additionally, several studies have reported an increased risk of hepatotoxicity with its clinical use [43]. In previous therapeutic strategies, immune checkpoint inhibitors (ICIs), such as Pembrolizumab and Tislelizumab, have been widely used. However, they are associated with significant adverse effects, including pruritus, diarrhea, and skin reactions [44,45]. Moreover, long-term use of these targeted agents often leads to drug resistance. Our study demonstrated that Rg3 exerted its therapeutic effects by activating ferroptosis and remodeling the immune microenvironment, suggesting that its combination with ICIs could allow for dose reduction, thereby minimizing side effects and potentially countering or delaying the onset of resistance. One study on therapy-resistant breast cancer reported that a liposomal formulation co-delivering paclitaxel and ginsenoside Rg3 increased the tumor inhibition rate by 5-fold compared to paclitaxel alone, achieving an inhibition rate of 51 % [20]. Previous studies have indicated that ferroptosis inducers could overcome drug resistance. Studies in advanced intrahepatic cholangiocarcinoma models have demonstrated that the small-molecule GPX4 inhibitor JKE-1674 overcame gemcitabine and cisplatin resistance by inducing ferroptosis via suppression of the PAX8-AS1/GPX4 axis [46]. In the present study, as Rg3 induced ferroptosis in melanoma, it may similarly serve as a promising strategy to overcome drug resistance in this tumor type. Although clinical data on Rg3 specifically for melanoma remain scarce, these collective findings provide a valuable rationale for its potential to reverse chemoresistance, supporting further validation in targeted clinical trials. Among reported ferroptosis inducers for melanoma, one study showed that Tanshinone IIA treated cutaneous melanoma by activating the ferroptosis pathway [47]. After 28 days of oral administration at a dose of 50 mg/kg, the tumor inhibition rate reached nearly 40 %. Moreover, previous studies have established that Mefloquine promoted ferroptosis in melanoma cells by upregulating LPCAT3 via the IFN-γ-STAT1-IRF1 axis. However, achieving a 40 % tumor suppression rate in vivo required a high oral dose of 50 mg/kg daily for nine days. It is well-established that Mefloquine is associated with severe neuropsychiatric adverse effects [48]. Compared with the above studies, Rg3 had higher efficacy with fewer doses and shorter dosing cycles. Moreover, Rg3 is a naturally occurring drug, and our study and previous research have not found any side effects. Therefore, Rg3 may be a better choice for clinical treatment.

Fig. 10.

Schematic diagram of the mechanism by which Rg3 promotes ferroptosis in melanoma via the p53/SLC7A11/GPX4 pathway.

Furthermore, we focused on the tumor immune and inflammatory microenvironment during ferroptosis to explore additional therapeutic directions. To determine whether ferroptosis affects the inflammatory and immune microenvironment, the analysis of the immune microenvironment in normal and tumor tissue samples revealed that the number of immune cells in tumor patients was significantly higher than that in normal samples. Within the tumor immune microenvironment of mice in the treatment group, the expression levels of IL-10, IFN-γ, and TGF-β were found to be notably increased, whereas the expression of pro-inflammatory cytokines including IL-6, TNF-α, and CXCL8 was significantly reduced. These data suggested that Rg3-induced ferroptosis might reverse immunosuppression and remodel the inflammatory microenvironment. An earlier study demonstrated that ginsenoside Rg3 alleviated cisplatin resistance in lung cancer by downregulating PD-L1 and restoring immune function. Treatment with Rg3 resulted in a fourfold reduction in PD-L1 expression compared to the control. Additionally, Rg3 has been reported to modulate the circFXOP1-miR-4477a-PD-L1 signaling axis, thereby restoring CD8⁺T cell function and reshaping the immune microenvironment [49]. Therefore, combining Rg3 with immune checkpoint inhibitors or conventional anticancer drugs represents a promising synergistic strategy.

Despite these novel findings, this study was subject to several limitations that warrant consideration. While the in vivo efficacy and functional role of ferroptosis were validated in a mouse model, future investigations employing patient-derived xenograft models would provide greater clinical relevance by better recapitulating the tumor microenvironment and heterogeneity of human melanoma. The clinical efficacy of Rg3 in melanoma patients remains to be established.

Conclusion

In summary, in this study, we comprehensively employed multiple strategies including machine learning, transcriptomics, and metabolomics for combined analysis, and revealed that Rg3 activated the p53/SLC7A11/GPX4 signaling pathway to induce ferroptosis in murine melanoma, which has not been fully explored in previous studies. This provides a new direction and potential guidance for future treatment strategies for melanoma patients.

Compliance with ethics requirements

All Institutional and National Guidelines for the care and use of animals (fisheries) were followed.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.