Doxorubicin induces ferroptosis in endometrial cancer by suppressing the MKK6/p38/CEBPB axis

Abstract

Background

Doxorubicin is a cornerstone chemotherapeutic agent. However, its role in ferroptosis and the underlying molecular mechanisms in endometrial cancer (EC) remain inadequately explored.

Methods

We integrated transcriptomic profiling, machine learning, and protein–protein interaction network analyses to identify ferroptosis-related regulators in EC. Functional validation was performed using CCK-8 and EdU assays, transmission electron microscopy, biochemical assays of ferroptosis markers, and Western blotting. Mechanistic studies included RNA sequencing, molecular docking, cellular thermal shift assays, and surface plasmon resonance. In vivo effects were evaluated in nude mouse xenograft models.

Results

Bioinformatic analyses identified ferroptosis-related genes and highlighted a hub gene closely associated with drug sensitivity, with doxorubicin emerging as the agent most strongly linked to ferroptosis-related sensitivity. Doxorubicin markedly suppressed the proliferation of EC cells. This antiproliferative effect was partially reversed by ferrostatin-1. Morphological and biochemical analyses revealed features consistent with ferroptosis, including mitochondrial structural damage, increased lipid peroxidation, iron overload, and oxidative stress. Consistently, doxorubicin downregulated ferroptosis-associated proteins. Mechanistic studies demonstrated that doxorubicin binds directly to MKK6, thereby suppressing activation of the MKK6/P38/CEBPB signaling cascade. CEBPB positively regulates SLC7A11 expression and transactivates the SLC7A11 promoter in a motif-dependent reporter assay. CEBPB overexpression reverses both the anti-proliferative effects of doxorubicin and doxorubicin-induced ferroptosis in EC cells. In vivo, doxorubicin significantly reduced xenograft tumor growth while increasing ferroptosis-associated biochemical and molecular markers and inhibiting MKK6/P38 signaling.

Conclusion

Our findings uncover ferroptosis as a previously unrecognized mechanism of doxorubicin action in EC, establishing the MKK6/P38/CEBPB axis as a potential therapeutic target and opening new avenues for treatment optimization.

Graphical abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-026-08003-6.

Introduction

Endometrial cancer (EC) is one of the most common malignancies of the female reproductive system, and its incidence and mortality are increasing worldwide [1–3]. In China, EC was estimated to account for approximately 13,500 deaths in 2022, corresponding to an age-standardized mortality rate of 1.1 per 100,000 women [4]. While most patients diagnosed at an early stage achieve favorable outcomes after surgery with or without adjuvant therapy, advanced or recurrent EC remains a major clinical challenge and is associated with poor prognosis despite multimodal management [5–7]. Although systemic therapies including chemotherapy, hormonal therapy, and targeted agents are routinely used, clinical benefit is often modest and not durable, with platinum-based regimens (e.g., carboplatin–paclitaxel) still serving as the cornerstone for advanced and recurrent disease [8–11]. Therefore, elucidating additional molecular mechanisms that shape therapeutic response and resistance is urgently needed to improve outcomes for patients with advanced or metastatic EC [12, 13].

Chemotherapy remains a cornerstone for advanced or recurrent endometrial cancer, yet durable clinical benefit is often limited. Platinum–taxane regimens continue to serve as standard options, and agents such as doxorubicin or targeted therapies have been evaluated in selected settings. These unmet needs underscore the importance of defining mechanism-linked vulnerabilities that shape treatment response and resistance in EC [14–17].

Doxorubicin (DOX) is a classic anthracycline that induces cytotoxic stress through DNA intercalation, topoisomerase II inhibition, and reactive oxygen species generation. In endometrial cancer, however, its clinical utility has been constrained by cardiotoxicity and only modest activity relative to taxane–platinum regimens, and patient responses remain heterogeneous [15]. Formulation and combination strategies—including liposomal DOX and emerging co-delivery or combinatorial approaches—have been explored to improve tolerability and efficacy, but these advances have not fully clarified the key molecular determinants and signaling pathways governing DOX sensitivity in EC [18–20]. Therefore, defining mechanism-linked vulnerabilities that mediate DOX response in EC remains an important unmet need.

Ferroptosis is a regulated cell death modality driven by iron-dependent lipid peroxidation and compromised GPX4-centered antioxidant defense, which is mechanistically distinct from apoptosis and necrosis [21, 22]. Accumulating evidence links ferroptosis to cancer vulnerability and therapy response, as ferroptosis induction can potentiate antitumor efficacy and help overcome resistance to conventional therapies across multiple malignancies [23–25]. In endometrial cancer, although metabolic dysregulation suggests a permissive context for ferroptosis, its functional contribution and regulatory circuitry remain incompletely defined [21]. Emerging studies have begun to implicate ferroptosis-associated resistance or sensitivity programs in EC progression and treatment response, supporting ferroptosis as a therapeutically relevant pathway in this disease [26–30].

Although doxorubicin has long been used in chemotherapy regimens for endometrial cancer, the molecular determinants that govern its antitumor activity and patient-to-patient variability remain incompletely defined. Clinically, responses to doxorubicin-based therapy are heterogeneous, with intrinsic or acquired resistance observed in a subset of patients, suggesting that specific molecular programs modulate drug sensitivity [31, 32]. Ferroptosis has emerged as a therapy-relevant vulnerability that can influence tumor cell survival and treatment response across malignancies [33–36]. However, whether doxorubicin engages ferroptosis in endometrial cancer—and which signaling networks connect drug exposure to ferroptotic execution—has not been established. Defining this DOX–ferroptosis linkage may not only refine the mechanistic understanding of doxorubicin in EC but also inform strategies to improve therapeutic efficacy and address chemoresistance.

In this study, we integrated bioinformatic screening with experimental validation to interrogate ferroptosis-associated therapeutic vulnerabilities in endometrial cancer and to prioritize doxorubicin as a candidate agent linked to ferroptosis-related sensitivity. We then performed in vitro and in vivo experiments to test whether doxorubicin triggers ferroptosis in EC cells and to delineate the signaling events connecting drug exposure to ferroptotic regulation. Together, our findings clarify a previously underappreciated dimension of doxorubicin action in endometrial cancer and support ferroptosis-oriented strategies as a potential route for treatment optimization.

Materials and methods

Data collection and identification of differentially expressed genes (DEGs)

Gene expression datasets for EC were obtained from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/). The GSE17025 dataset includes 12 normal endometrium samples and 91 EC samples, derived from the GPL570 [HG-U133_Plus_2] Affymetrix Human Genome U133 Plus 2.0 Array. Ferroptosis-associated genes were retrieved from previously published studies [37, 38] (Supplementary Table 1). Using the R (4.3.1) ‘limma’ package, DEGs were identified from the merged datasets with cutoff criteria of |logFC| ≥ 0.2 and p-value < 0.05.

Gene ontology (GO) and kyoto encyclopedia of genes and genomes (KEGG) enrichment analysis

GO and KEGG pathway enrichment analyses were performed to investigate the functional relevance of the DEGs in endometrial cancer. GO analysis classified the DEGs into the three major domains of biological processes, molecular functions, and cellular components, while KEGG analysis identified signaling pathways potentially implicated in tumor progression. Both analyses were conducted using the “clusterProfiler” R package.

Least absolute shrinkage and selection operator (LASSO)

LASSO logistic regression was performed using the glmnet R package to prioritize ferroptosis-related differentially expressed genes for tumor–normal classification. Expression values of the candidate genes were converted into a predictor matrix (x) and the binary outcome (y) was encoded as 0 (Normal) and 1 (Tumor). We fitted a binomial model with an L1 penalty (family = “binomial”, alpha = 1). The optimal regularization parameter (λ) was selected by 10-fold cross-validation using cv.glmnet (nfold = 10; type.measure = “deviance”). Genes with non-zero coefficients at λmin were retained as LASSO-selected features for downstream intersection with other algorithms.

Support vector machine recursive feature elimination (SVM-RFE)

SVM-RFE was applied to further rank and select informative features using an SVM-based wrapper. Briefly, SVM models were trained and features were recursively eliminated based on their contribution to classification performance. To reduce overfitting and estimate feature stability, SVM-RFE was conducted under a cross-validation framework. Specifically, the dataset was randomly partitioned into 10 folds for outer resampling; within each iteration, feature selection was performed on the training portion using the SVM-RFE procedure with an internal k-fold resampling step (k = 5), and the cross-validation error was tracked as the number of retained features increased. The optimal feature set was defined as the number of features corresponding to the minimum cross-validation error, and the top-ranked genes at this point were retained as SVM-RFE-selected candidates. Parameters for feature reduction followed the script setting (halve.above = 100).

Random forest (RF)

Random forest modeling was conducted using the randomForest R package to estimate variable importance. Samples were randomly split into a training set (70%) and a test set (30%) with a fixed random seed (set.seed = 123). A random forest classifier was trained on the training set with ntree = 500 and variable importance enabled (importance = TRUE). Variable importance was quantified by the IncNodePurity metric, and the top 10 genes were selected for downstream intersection analyses. Model behavior was inspected using the out-of-bag error trajectory and predictions were generated for both training and test sets.

Construction and topological analysis of the protein–protein interaction (PPI) network

The identified ferroptosis-related genes were imported into the STRING database to construct a PPI network, which was subsequently visualized and analyzed using Cytoscape software (version 3.10.1). Within Cytoscape, degree centrality was calculated to assess the connectivity of individual nodes, enabling the identification of highly interactive proteins that may serve as potential hub genes for further investigation.

Gene set enrichment analysis (GSEA)

GSEA was performed to identify significantly enriched pathways associated with DEGs in endometrial cancer. The analysis was conducted using the GSEA software (version 4.1.0), which evaluates whether predefined gene sets show statistically significant, concordant differences between two biological states. The hallmark gene sets and KEGG pathways were selected for analysis. Results from GSEA were visualized using enrichment plots, which represent the ranking of genes within a predefined gene set, highlighting whether the gene set is positively or negatively correlated with the phenotype.

Immune infiltration analysis

Immune cell infiltration was evaluated using two complementary approaches to characterize the tumor immune microenvironment. The single-sample gene set enrichment analysis (ssGSEA) algorithm was applied to quantify the relative enrichment of immune cell populations in each tumor sample based on established immune-related gene signatures. In parallel, the CIBERSORT algorithm was employed to deconvolute the bulk transcriptomic data and estimate the relative proportions of distinct immune cell subsets, providing a complementary view of the immune landscape.

Single-cell transcriptomic analysis

Single-cell transcriptomic analysis was conducted using the GSE173682 dataset to further characterize the cellular distribution of candidate genes within the tumor microenvironment. Data preprocessing, quality control, and normalization were performed in R, followed by dimensionality reduction and clustering using the Seurat package. Uniform manifold approximation and projection (UMAP) was applied to visualize transcriptional heterogeneity and define major cellular populations. The expression patterns of hub genes were then mapped across identified clusters to assess their lineage specificity.

Drug sensitivity analysis

Drug sensitivity analysis was performed to explore the association between gene expression and chemotherapeutic response in endometrial cancer. Using the TCGA-UCEC dataset, patients were stratified into two groups based on expression levels of the candidate hub gene, and the R package “pRRophetic” was applied to estimate the half-maximal inhibitory concentration (IC50) values for commonly used agents. Differences in predicted sensitivity between groups were statistically assessed, with a significance threshold of p < 0.05.

Molecular docking

Molecular docking was performed to evaluate the interaction between doxorubicin and MKK6. The three-dimensional structure of MKK6 was retrieved from the Protein Data Bank (PDB), and doxorubicin was prepared using ChemBio3D. Docking was carried out with AutoDock Vina (v1.5.7), focusing on the active site of MKK6. The binding poses were evaluated based on binding energy, and key interactions were visualized using PyMOL (v2.5.0a0) to identify important residues involved in the binding.

Molecular dynamics simulation

Molecular dynamics (MD) simulations were performed to evaluate the dynamic stability of the MKK6–doxorubicin (DOX) complex. The initial complex conformation was taken from the docking-derived binding pose and prepared for simulation using GROMACS 2020. The Amber99sb-ildn force field was applied to the protein, while DOX parameters were generated using the General Amber Force Field (GAFF) to obtain the ligand topology. The complex was solvated in an explicit SPC/E water box with periodic boundary conditions, and counterions were added to neutralize the system; additional ions were included to mimic physiological ionic strength. Long-range electrostatic interactions were treated using the particle mesh Ewald (PME) method. The Coulomb and van der Waals cutoffs were both set to 1.0 nm. Energy minimization was carried out using the steepest descent algorithm with a maximum of 500,000 steps. The system was subsequently equilibrated under an NVT ensemble for 100 ps followed by an NPT ensemble for 100 ps. Temperature was maintained at 298.15 K using the V-rescale thermostat, and pressure was maintained at 1 bar. Production MD was then performed for 100 ns. Trajectory analyses were conducted using built-in GROMACS utilities, including root mean square deviation (RMSD), root mean square fluctuation (RMSF), radius of gyration (Rg), hydrogen bond analysis, and solvent-accessible surface area (SASA). Binding free energy was further estimated using the gmx_MMPBSA workflow, and per-residue energy decomposition was performed to identify residues contributing to ligand binding.

Cell culture

The human endometrial cancer cell lines Ishikawa and RL-95-2 were purchased from Procell Life Science & Technology Co., Ltd. (Wuhan, China). Cells were maintained in minimum essential medium supplemented with 10% fetal bovine serum, 100 U/mL penicillin, and 100 µg/mL streptomycin, and incubated at 37 °C in a humidified atmosphere containing 5% CO₂.

RNA extraction and quantitative real-time PCR (qPCR)

Total RNA was extracted from cultured cells using Total RNA Extraction Kit (G3013, Servicebio, China) according to the manufacturer’s instructions, and cDNA was synthesized using HiScript Ⅲ All-in-one RT Super Mix (R333, Vazyme, China). qPCR was performed with 2 × SYBR Green qPCR Master Mix (G3320-15, Servicebio, China) on a StepOnePlus Real-Time PCR System (Applied Biosystems, USA). Relative gene expression levels were calculated using the 2^−ΔΔCt method, with GAPDH serving as the internal control.

Western blotting (WB)

WB was performed to assess the expression of target proteins in cell and tissue samples. Briefly, proteins were extracted using RIPA lysis buffer containing protease inhibitors, and protein concentrations were determined using the BCA assay. Equal amounts of protein were loaded onto a 10% SDS-PAGE gel and separated by electrophoresis. After transferring the separated proteins to a polyvinylidene fluoride (PVDF) membrane, the membrane was blocked with 5% non-fat dry milk in TBST for 1 h at room temperature to prevent nonspecific binding. The membrane was then incubated overnight at 4 °C with primary antibodies: Anti-CEBPB antibody (T55276, Abcam, UK), Anti-SLC7A11 antibody (GB115276, Servicebio, China), Anti-GPX4 antibody (ET1706-45, Huaan Biology, China), Anti-MKK6 antibody (GB113063-50, Servicebio, China), Anti-P-MKK6 antibody (PJP11266, Abmart, China), Anti-p38 antibody (GB114685, Servicebio, China), Anti-P-p38 antibody (TA4001, Abmart, China), and Anti-GAPDH antibody (ab9485, Abcam, UK). Following primary antibody incubation, the membrane was washed with TBST and incubated with an appropriate HRP-conjugated secondary antibody for 1 h at room temperature. Protein bands were visualized using enhanced chemiluminescence (ECL) detection reagent (SuperKine Ultra-sensitive ECL Chemiluminescence Reagent, BMU103-CN, Abkine, China). GAPDH was used as the loading control to ensure equal protein loading across samples. Densitometric analysis of the bands was performed using ImageJ software. All experiments were repeated at least three times as independent biological replicates.

Cell Counting Kit-8 (CCK8)

Cell viability and proliferation were assessed using the Cell Counting Kit-8 (CCK-8, Ultra-sensitive Cell Proliferation Kit, BMU106-CN, Abbkine, China). Briefly, Ishikawa and RL-95-2 cells were seeded into 96-well plates at an appropriate density and allowed to adhere overnight. Cells were then treated with doxorubicin, ferrostatin-1, or corresponding controls for the indicated times. Doxorubicin working concentrations for in vitro assays were chosen based on dose–response testing in each cell line and were set near the corresponding IC50 values to facilitate mechanistic analyses under cytotoxic yet interpretable conditions. At each time point, 10 µL of CCK-8 solution was added to each well, followed by incubation at 37 °C for 2 h. The absorbance at 450 nm was measured using a microplate reader to determine cell viability.

EdU fluorescence staining assay

To evaluate DNA synthesis, an EdU incorporation assay was performed using the Click-iT EdU-488 Cell Proliferation Kit (G1601, Servicebio, China). Briefly, Ishikawa and RL-95-2 cells were seeded in 6-well plates and cultured overnight to allow cell attachment. The cells were then treated with doxorubicin, ferrostatin-1, or appropriate controls for 24 h. Following treatment, EdU was added to the culture medium at a final concentration of 10 µM, and the cells were incubated for an additional 2 h. Cells were then fixed with 4% paraformaldehyde and permeabilized with 0.5% Triton X-100. EdU incorporation was detected using a fluorescent dye, and images were captured using a fluorescence microscope. The percentage of EdU-positive cells was quantified to assess DNA synthesis.

BODIPY 581/589-C11 staining and fluorescence imaging

Lipid peroxidation was measured using BODIPY 581/589-C11. After the indicated treatments, cells were washed with PBS and incubated with BODIPY 581/589-C11 (2 µM, serum-free medium) at 37 °C for 30 min in the dark. Cells were then washed twice with PBS and immediately imaged under identical microscope settings. Fluorescence images were analyzed in ImageJ to quantify the oxidized (green)/reduced (red) fluorescence intensity ratio for each field, and mean ratios were calculated per group.

Transmission electron microscopy (TEM)

Transmission electron microscopy (TEM) was employed to examine ultrastructural changes in Ishikawa and RL-95-2 cells following 24 h of doxorubicin treatment. Cells were collected, fixed with electron microscopy–grade fixatives, and subjected to standard procedures including osmium tetroxide post-fixation, graded ethanol dehydration, resin embedding, and ultrathin sectioning. Sections were stained with uranyl acetate and lead citrate to enhance contrast, and images were acquired under a transmission electron microscope to assess characteristic ferroptotic features.

Determination of lipid peroxidation

For cell samples, harvested pellets were lysed and homogenized before centrifugation, with supernatants collected for analysis; tissue samples were similarly homogenized and processed under cold conditions. Reaction mixtures were prepared according to the manufacturer’s protocol, followed by heating, cooling, and centrifugation to separate the reaction products. The absorbance of the supernatants was then measured at 532 nm using a microplate reader, and Malondialdehyde (MDA)concentrations were normalized to protein content. All assays were performed in triplicate to ensure reproducibility.

Assessment of cellular redox status (GSH/GSSG ratio)

Tissue samples were homogenized in extraction buffer at a defined weight-to-volume ratio, while cultured cells were lysed and subjected to ultrasonic disruption to obtain clear suspensions. After centrifugation, supernatants were collected and kept on ice for further analysis. Standard solutions were freshly prepared, and working reagents were mixed immediately prior to use to ensure stability. For selective quantification, GSH was derivatized and removed before total glutathione and oxidized glutathione were measured, allowing the GSH/GSSG ratio to be calculated. All assays were performed in triplicate to ensure reproducibility.

Quantification of cellular iron content

For cell samples, approximately 1 × 10⁶ cells were lysed in extraction buffer, incubated on ice, and centrifuged to obtain clear supernatants. Tissue samples were homogenized in buffer under cold conditions, and the resulting lysates were similarly clarified by centrifugation. Both total iron and ferrous iron (Fe²⁺) levels were quantified in separate reactions, with samples incubated at 37 °C for defined periods to allow chromogenic development. Absorbance was measured at 593 nm using a microplate reader, and values were normalized to protein concentration. All measurements were performed in triplicate to ensure reproducibility.

Measurement of intracellular ROS production

Cells were washed with serum-free medium and incubated with the working solution of DCFH-DA at 37 °C for 1 h in the dark. After incubation, excess dye was removed by repeated washing, and cells were collected by trypsinization and centrifugation before being resuspended in serum-free medium. Fluorescence intensity of DCF was measured by flow cytometry, with 20,000 cells acquired per sample. All assays were performed in triplicate to ensure reproducibility.

RNA sequencing (RNA-Seq)

Ishikawa cells were seeded in 6-cm culture dishes and treated with doxorubicin for 24 h after reaching approximately 85% confluence. Total RNA was extracted using a commercial RNA isolation reagent under RNase-free conditions to prevent degradation, and samples were submitted to a sequencing platform (Shanghai Bioengineering Co., Ltd.) for high-throughput RNA sequencing. Libraries were prepared following standard protocols and sequenced on the Illumina platform. Raw data quality was assessed using FastQC, and clean reads were aligned to the human reference genome with STAR. Gene expression levels were quantified as FPKM and TPM for downstream analyses.

Dual-luciferase reporter assay

Cells were seeded at 5 × 10⁵ per well in 6-well plates and cultured for 18–24 h at 37 °C. For plasmid transfection, the following ratios of plasmids and Lipofectamine 8000 were used: GPL3-Basic, pcDNA3.1, Renilla, and Lipofectamine 8000 (100 ng:400 ng:10 ng:4 µL) for control and experimental conditions, including CEBPB and SLC7A11-WT or SLC7A11-MT promoter constructs. After transfection, cells were incubated for 24–48 h before the culture medium was removed and the cells were washed twice with PBS. Lysis buffer was added, and cells were lysed at room temperature with shaking for 15 min. The cell lysate was centrifuged at 10,000–15,000 rpm for 3–5 min, and the supernatant was collected for subsequent luciferase assays. Firefly luciferase activity was measured using Firefly Luciferase Assay Buffer with D-Luciferin (0.2 mg/mL), and Renilla luciferase activity was assessed using Coelenterazine in Renilla Luciferase Assay Buffer. Both luminescence signals were measured using a chemiluminescence reader at 2-second intervals for 10 s. Relative enzyme activity was calculated by normalizing firefly RLU to Renilla RLU.

Cell thermal shift assay (CETSA)

Cells were cultured to approximately 90% confluence and treated with doxorubicin or DMSO as a control for 4 h at 37 °C. After treatment, cells were washed, scraped, and collected by centrifugation at 3000 rpm for 5 min. The cell pellet was resuspended in PBS with 1% PMSF. For thermal treatment, cell suspensions were heated at different temperatures (37 °C, 42 °C, 47 °C, 52 °C, 57 °C, 62 °C) for 3 min each, followed by cooling on ice. Cell lysis was performed using a freeze-thaw cycle, and protein extracts were obtained by centrifugation at 12,000 rpm for 15 min. Protein samples were mixed with SDS loading buffer, heated at 100 °C for 5 min, and analyzed by SDS-PAGE. A temperature-protein stability curve was constructed based on the protein stability at each temperature.

Surface plasmon resonance (SPR)

SPR experiments were conducted to assess the interaction between MKK6 and doxorubicin using a Biacore system. The CM5 chip was activated with a mixture of EDC and NHS, followed by immobilization of MKK6 protein on the second flow cellat a flow rate of 10 µL/min. The chip was blocked with ethanolamine to prevent non-specific binding, and a reference channel was treated similarly but without protein. For the interaction analysis, compounds were prepared in PBS with 5% DMSO, and their concentrations were injected into the flow cell at a flow rate of 30 µL/min for 150 s. After each injection, the chip was regenerated with 10 mM glycine-HCl (pH 2.0) for 5 min. Data were analyzed using Biacore Insight software (Cytiva Marlborough, MA, USA) and fitted to a 1:1 Langmuir binding model to determine the binding and dissociation constants.

Animal experiments

For the establishment of the subcutaneous xenograft tumor model, 4–5 week-old female BALB/C nude mice (body weight: 15–20 g) were used. The mice were housed under specific pathogen-free (SPF) conditions with sterile food and water. Four mice were kept per cage, and they were allowed to acclimatize for 5 days prior to cell implantation. The animals were monitored daily for general health, including diet, activity, and overall condition. During the acclimatization phase, the mice were housed under standard laboratory conditions, with a 12-hour light/dark cycle, at a temperature of 22–24 °C, and humidity levels of 50–60%. The cages were cleaned regularly, and the animals were provided with ad libitum access to sterile food and water. Following the acclimatization period, RL-95-2 cells (1 × 10^7 cells/mL) in a sterile PBS solution were subcutaneously injected into the right axillary region of the mice after iodine disinfection. After injection, the area was disinfected again, and the tumor site was lightly pressed to avoid leakage of the injection fluid. The mice were monitored closely for any abnormal signs after the injection. After one week, tumor growth was evaluated.

Once the tumors reached a size of 100 mm³, the mice were randomly assigned into two groups: a control group (treated with DMSO solution) and a doxorubicin treatment group (DOX group). The control group received the same volume of DMSO solution as the DOX group, while the DOX group was treated with doxorubicin (5 mg/kg body weight) dissolved in PBS. Doxorubicin was administered via intraperitoneal injection every 4 days for a total of 4 treatments. The mice’s general health, including food intake, activity levels, and tumor growth, were monitored throughout the treatment period. Tumor dimensions were measured every 4 days.

After the fourth dose of doxorubicin, the tumors were excised, weighed, and measured. Tumor tissues were washed with physiological saline and then used for protein extraction, RNA analysis, and biochemical assays to assess ferroptosis markers. Tumor tissues were preserved in liquid nitrogen for subsequent analysis.

Statistical analyses

Statistical analyses were performed using GraphPad Prism 9.0. Unless otherwise stated, experiments were performed with three independent biological replicates (n = 3), and data are presented as mean ± SEM. For comparisons between two groups, we used a two-sided Student’s t-test when data were approximately normally distributed; Welch’s t-test was applied when variances were unequal. Given the small sample sizes in several assays (typically n = 3), normality tests (Shapiro–Wilk) have limited power; therefore, when distributional assumptions were uncertain, we prioritized conservative non-parametric tests (two-sided Mann–Whitney U test). For multiple-group comparisons, one-way ANOVA followed by Dunnett’s post hoc test was used for approximately normal data, whereas the Kruskal–Wallis test with appropriate multiple-comparison procedures was used otherwise. Wherever possible, individual data points are shown to reflect biological replicate variability. A two-sided P value < 0.05 was considered statistically significant.

Results

Dataset processing and analysis

For transcriptomic analysis, the endometrial cancer dataset GSE17025 was obtained from the GEO database and subjected to standard preprocessing. After normalization, the distribution of expression values was assessed using box plots, which confirmed consistent medians across all samples and indicated that data quality and comparability were preserved (Fig. 1A). To further evaluate the dataset structure, principal component analysis (PCA) was performed, which revealed a clear separation between tumor and normal endometrial tissues, with the first two principal components (PC1 and PC2) explaining 10.8% and 5.9% of the overall variance, respectively (Fig. 1B). This clustering pattern underscored the distinct transcriptional profiles of malignant and non-malignant samples. To highlight the most significant transcriptional alterations, a heatmap was generated, displaying the top ten most strongly upregulated and downregulated genes across the dataset (Fig. 1C). Differential expression analysis subsequently identified 3,846 upregulated and 3,759 downregulated genes, which were visualized through a volcano plot using thresholds of |logFC| > 0.2 and P < 0.05 (Fig. 1D).

Fig. 1.

Transcriptomic analysis and identification of ferroptosis-related genes in EC. (A)Box plots of expression values for the endometrial cancer dataset GSE17025 after standard preprocessing and normalization; (B) PCA of the GSE17025 dataset, with clear separation between tumor and normal tissues, explaining 10.8% and 5.9% of the variance with PC1 and PC2, respectively; (C) Heatmap of the top ten most significantly upregulated and downregulated genes in the dataset; (D) Volcano plot displaying 3,846 upregulated and 3,759 downregulated genes, with thresholds of |logFC| > 0.2 and P < 0.05; (E) Venn diagram showing 24 overlapping ferroptosis-related DEGs from the GSE17025 dataset; (F) Chromosomal distribution of the 24 ferroptosis-related DEGs across multiple chromosomes; (G) LASSO regression identified a subset of genes with the most robust predictive power; (H) SVM-RFE analysis with 14 genes showing the strongest classification performance; (I) RF analysis identified the top 10 genes based on their importance in sample classification

Identification of ferroptosis-related DEGs

Following the identification of DEGs from the GSE17025 dataset, we next focused on ferroptosis-related factors to determine their relevance in endometrial cancer. A total of 41 ferroptosis-associated genes were retrieved from previously published studies, and an intersection analysis was performed between these genes and the DEGs identified in our dataset. This comparison yielded 24 overlapping genes, hereafter defined as ferroptosis-related DEGs (Fig. 1E) (Supplementary Table 1). To further characterize their genomic distribution, we mapped the chromosomal localization of these 24 genes, which demonstrated a relatively dispersed distribution across multiple chromosomes (Fig. 1F).

Machine learning–based screening of hub genes

To further refine the ferroptosis-related gene set, three complementary machine learning approaches were applied to the 24 candidate genes. LASSO regression was first performed, and the optimal penalty parameter (λ) was identified through cross-validation, resulting in the retention of a subset of genes with the most robust predictive power (Fig. 1G). In parallel, SVM-RFE analysis was carried out, in which the cross-validation error progressively decreased with the number of included features, and the optimal model was achieved with 14 genes, reflecting the strongest classification performance (Fig. 1H). RF analysis was then employed to rank variable importance, with the top 10 genes selected according to their IncNodePurity scores, indicating their significant contribution to sample classification (Fig. 1I). Finally, an intersection of the gene sets derived from the three algorithms yielded three overlapping candidates, which were defined as hub genes for subsequent analyses (Fig. 2A). They were acetyl-CoA carboxylase alpha(ACACA), glutamic-oxaloacetic transaminase 1(GOT1) and solute carrier family 1 member 5 (SLC1A5). To mitigate overfitting, the feature space was restricted to ferroptosis-related DEGs (n = 24), cross-validation was applied for LASSO and SVM-RFE, and hub candidates were prioritized based on the intersection of three independent feature-selection methods rather than a single model.

Fig. 2.

Identification and functional characterization of ferroptosis-related hub genes in EC. (A) Venn diagram showing the intersection of candidate genes derived from three machine learning algorithms; (B) Expression levels of the three hub genes (ACACA, GOT1, and SLC1A5) in both the GSE17025 and TCGA-UCEC datasets; (C) and (D) PPI network of the 24 ferroptosis-related differentially expressed genes, generated using the STRING database and visualized in Cytoscape; (E) GO enrichment analysis of DEGs associated with high SLC1A5 expression in the GSE17025 dataset; (F) GSEA results identifying significantly enriched signaling pathways associated with oncogenesis; (G) and (H) ssGSEA correlation analysis of immune cell infiltration in the TCGA-UCEC cohort; (I) Immune cell infiltration comparison between high and low SLC1A5 expression groups; (J) CIBERSORT analysis showing differences in immune cell proportions between high and low SLC1A5 expression groups. ns>0.05, *P < 0.05, **P < 0.01, ***P < 0.001

Validation of hub gene expression

The expression levels of the three hub genes were first assessed in the GSE17025 dataset. All three showed a consistent and significant upregulation in tumor samples compared with normal tissues (Fig. 2B). Specifically, ACACA expression increased from 8.593 ± 0.142 in the normal group to 9.896 ± 0.050 in the tumor group (U = 66, p < 0.001), GOT1 expression rose from 8.064 ± 0.1477 in normal tissues to 9.374 ± 0.086 in tumors (U = 54, p < 0.001), and SLC1A5 expression similarly shifted from 7.230 ± 0.216 in the normal group to 8.313 ± 0.087 in the tumor group (U = 105, p < 0.001). Here, TCGA-UCEC was used to independently validate the differential expression patterns of the candidate hub genes rather than to externally validate a finalized predictive classifier; further model-level validation in additional independent cohorts will be required to assess generalizability. To validate these findings, the TCGA-UCEC cohort was analyzed as an independent dataset (Fig. 2B). Consistent with the GEO results, ACACA again displayed elevated expression in tumor samples (normal 3.766 ± 0.064 vs. tumor 4.250 ± 0.035, U = 5346, p < 0.001), as did GOT1 (normal 5.186 ± 0.049 vs. tumor 5.993 ± 0.028, U = 2846, p < 0.001) and SLC1A5 (normal 6.047 ± 0.093 vs. tumor 6.481 ± 0.057, U = 7447, p < 0.001). The concordance of these expression patterns across two independent datasets supports the robustness of the differential-expression findings; however, additional independent cohorts will be needed for external validation of model generalizability.

PPI network analysis and selection of the core hub gene

To further refine the candidate hub genes, a PPI network was constructed using the 24 ferroptosis-related differentially expressed genes. The network was first generated through the STRING database and subsequently imported into Cytoscape for topological analysis (Fig. 2C and D). Within this framework, the degree value, which reflects the number of direct interactions a given node has with other proteins, was used as the primary ranking metric (Supplementary Table 2). Degree centrality is widely recognized as a measure of biological importance in network biology, as highly connected nodes often represent core regulators with essential roles in cellular processes and disease pathogenesis. The visualization of the interaction network highlighted several key nodes, and when cross-referenced with the three hub genes identified by machine learning approaches, SLC1A5 consistently exhibited the highest degree among them. This convergence of evidence from both computational screening and network centrality analysis supported the prioritization of SLC1A5 as the most functionally relevant hub gene, which was therefore selected for subsequent experimental investigation.

Enrichment analysis of SLC1A5

To explore the biological processes associated with SLC1A5 expression, tumor samples from the GSE17025 dataset were stratified into high- and low-expression groups, followed by differential expression analysis using thresholds of |logFC| > 0.2 and P < 0.05. GO enrichment analysis of the resulting DEGs revealed that the top biological process (BP) terms were predominantly related to extracellular structure remodeling, including extracellular matrix organization, extracellular structure organization, and external encapsulating structure organization. In the cellular component (CC) category, enriched terms included ion channel complex, transmembrane transporter complex, and transporter complex, while the molecular function (MF) category was characterized by enrichment in transmembrane receptor protein tyrosine kinase activity, transmembrane receptor protein kinase activity, and frizzled binding (Fig. 2E). To complement these findings, GSEA was performed, which identified significant enrichment of signaling pathways associated with oncogenic processes, particularly the WNT signaling pathway, TGF-β receptor signaling in skeletal dysplasias, and lncRNA regulation in canonical WNT signaling and colorectal cancer. Additional enriched pathways included basal cell carcinoma, ncRNAs involved in WNT signaling in hepatocellular carcinoma, and formation of the cornified envelope (Fig. 2F) (Supplementary Table 2).

Immune infiltration analysis

To further explore the potential immunological role of SLC1A5 in endometrial cancer, immune cell infiltration was analyzed in the TCGA-UCEC cohort using the ssGSEA algorithm. Correlation analysis demonstrated that SLC1A5 expression was significantly associated with multiple immune cell populations, with the strongest positive correlations observed for activated dendritic cells (aDC, R = 0.217, p < 0.001), T follicular helper cells (TFH, R = 0.213, p < 0.001), neutrophils (R = 0.200, p < 0.001), and Th1 cells (R = 0.193, p < 0.001), while weaker or negative correlations were noted with T helper cells (R = -0.119, p = 0.005) and central memory T cells (Tcm, R = -0.096, p = 0.023) (Fig. 2G and H). To further support these associations, samples were stratifiedinto high- and low- SLC1A5 expression groups, and enrichment scores for immune cell subsets were compared. Comparative analysis revealed that a broad range of immune cell infiltration levels exhibited significant alterations between the high and low SLC1A5 expression groups, indicating that SLC1A5 expression is associated with widespread remodeling of the tumor immune microenvironment (Fig. 2I). Complementary analysis using the CIBERSORT algorithm further highlighted distinct immune landscape patterns between the two groups, as reflected in stacked bar plots of immune cell proportions. In particular, high SLC1A5 expression was accompanied by elevated proportions of macrophage and dendritic cell populations, along with shifts in regulatory T cells, indicating that SLC1A5 expression may be closely linked to remodeling of the immune microenvironment in endometrial cancer (Fig. 2J).

Single-cell RNA-seq analysis

To further delineate the cellular context of the identified hub genes, we performed single-cell transcriptomic analysis using the GSE173682 dataset. UMAP clustering identified 14 distinct transcriptional clusters, which were subsequently consolidated into seven major cellular lineages, including malignant epithelial cells, fibroblasts, endothelial cells, macrophages, mast cells, lymphoid cells, and T cells, reflecting the cellular heterogeneity of the tumor microenvironment (Fig. 3A and B). Visualization of hub gene expression across these 14 clusters revealed that all three candidates—ACACA, GOT1, and SLC1A5—were differentially distributed among the clusters, with SLC1A5 demonstrating the most distinct enrichment in malignant epithelial populations (Fig. 3C). Mapping the expression of these genes onto cell types highlighted that SLC1A5 was particularly abundant in tumor epithelial cells and subsets of macrophages, suggesting dual roles in metabolic reprogramming and immune regulation. In contrast, ACACA and GOT1 exhibited broader yet less specific expression patterns across stromal and immune subsets. Spatial feature plots further corroborated these findings, while boxplot-based comparisons confirmed that SLC1A5 expression was significantly higher in malignant clusters relative to other cell types (Fig. 3D and E). Collectively, these results indicate that among the three hub genes, SLC1A5 exhibits the strongest tumor-associated expression pattern, reinforcing its prioritization as the core candidate for downstream functional validation.

Fig. 3.

Single-cell transcriptomic analysis of hub gene expression in the tumor microenvironment. (A) and (B) UMAP clustering of single-cell RNA sequencing data from the GSE173682 dataset identified 14 distinct transcriptional clusters, which were grouped into seven major cellular lineages, including malignant epithelial cells, fibroblasts, endothelial cells, macrophages, mast cells, lymphoid cells, and T cells; (C) Expression of the hub genes ACACA, GOT1, and SLC1A5 across the 14 identified clusters; (D) and (E) Spatial feature plots and boxplot-based comparisons of ACACA, GOT1, and SLC1A5 expression in various cell types

Chemotherapy sensitivity analysis

To investigate whether SLC1A5 expression is associated with differential chemotherapeutic response in endometrial cancer, drug sensitivity analysis was conducted using the TCGA-UCEC cohort. Patients were stratified into high- and low-expression groups according to SLC1A5 levels, and the predicted IC50 values for five commonly used agents were compared. Among cisplatin, docetaxel, paclitaxel, temsirolimus, and doxorubicin, only doxorubicin exhibited a statistically significant difference between the two groups, with tumors in the low-SLC1A5 cohort showing increased sensitivity as reflected by lower predicted IC50 values (p < 0.05, Fig. 4A). No meaningful differences were observed for the other agents, suggesting that SLC1A5 expression specifically modulates responsiveness to doxorubicin in endometrial cancer. These findings support the hypothesis that doxorubicin sensitivity may be mechanistically linked to ferroptosis pathways regulated by SLC1A5 expression.

Fig. 4.

Doxorubicin-induced inhibition of EC cell proliferation and the role of ferroptosis. (A) Drug sensitivity analysis in the TCGA-UCEC cohort stratified patients into high- and low-SLC1A5 expression groups. Among cisplatin, docetaxel, paclitaxel, temsirolimus, and doxorubicin, only doxorubicin showed a statistically significant difference in IC50 values, with tumors in the low-SLC1A5 group exhibiting increased sensitivity to doxorubicin (p < 0.05); (B) CCK-8 assay for assessing doxorubicin-induced cytotoxicity in Ishikawa and RL-95-2 cells. Dose–response curves demonstrate a concentration-dependent reduction in cell viability; (C) Time-course proliferation assays using CCK-8 further confirmed the sustained inhibition of cell growth by doxorubicin in both Ishikawa and RL-95-2 cells. Significant suppression of proliferation was observed at 24, 48, 72, and 96 h in both cell lines; (D) EdU incorporation assay showing significantly reduced DNA synthesis in doxorubicin-treated Ishikawa and RL-95-2 cells compared to controls, with marked decreases in the proportion of EdU-positive cells; (E) Co-treatment with Fer-1 partially reversed doxorubicin-induced inhibition of cell viability in both Ishikawa and RL-95-2 cells, with OD450 values increasing at 24, 48, 72, and 96 h; (F) EdU incorporation assay demonstrating that Fer-1 co-treatment also reversed the reduction in EdU-positive cells induced by doxorubicin. ns>0.05, *P < 0.05, **P < 0.01, ***P < 0.001

The machine-learning pipeline was designed to prioritize ferroptosis-associated hub genes in EC and to identify therapeutic vulnerabilities linked to ferroptosis-related metabolic programs. Among the three hub genes, SLC1A5 showed the highest network centrality and tumor-enriched expression, and its expression stratified predicted chemosensitivity in TCGA-UCEC. Notably, doxorubicin was the only agent showing a significant IC50 difference between the SLC1A5-high and -low groups, prompting us to prioritize doxorubicin for subsequent mechanistic interrogation. To identify the signaling events connecting doxorubicin exposure to ferroptosis, we performed unbiased transcriptomic profiling in doxorubicin-treated EC cells, which highlighted MKK6 as a doxorubicin-responsive node within the stress kinase (p38) pathway. Notably, MKK6 was not derived from the initial tumor–normal ferroptosis-DEG feature-selection workflow; rather, it was identified in an orthogonal drug-perturbation transcriptomic context as a doxorubicin-responsive signaling node.

Doxorubicin inhibits proliferation of endometrial cancer cells

The cytotoxic effect of doxorubicin on endometrial cancer cells was first evaluated by determining the IC50. Dose–response curves demonstrated a concentration-dependent reduction in cell viability in both Ishikawa and RL-95-2 cells following 24 h of treatment. The calculated IC50 values were 1.016 µM for Ishikawa cells and 2.003 µM for RL-95-2 cells, respectively. Consistently, significant decreases in OD450 values were observed at increasing drug concentrations. In Ishikawa cells, treatment with 0.2 µM (t = 3.864, P = 0.003), 0.4 µM (t = 3.730, P = 0.004), 0.5 µM (t = 10.30, P < 0.001), 1.0 µM (t = 8.668, P < 0.001), 2.0 µM (t = 25.22, P < 0.001), and 2.5 µM (t = 34.96, P < 0.001) resulted in progressive inhibition of viability. Similarly, RL-95-2 cells showed a marked reduction in viability at concentrations of 1.0 µM (U = 0, P = 0.008), 1.2 µM (U = 0, P = 0.008), 1.4 µM (U = 0, P = 0.008), 1.6 µM (U = 0, P = 0.008), and 2.2 µM (U = 0, P = 0.008), whereas differences at 0.5 µM (U = 6, P = 0.22) and 0.8 µM (U = 9, P = 0.55) were not statistically significant (Fig. 4B).

Time-course proliferation assays using CCK-8 further confirmed that doxorubicin substantially impaired cell growth. In Ishikawa cells treated with 1 µM doxorubicin, cell proliferation was markedly suppressed compared with the DMSO control. At 0 h, OD450 values were comparable between the two groups (DMSO: 0.563 ± 0.003 vs. DOX: 0.548 ± 0.003; t = 2.221, P = 0.1394). However, at 24 h, DOX-treated cells exhibited significantly lower OD450 values (0.533 ± 0.012) compared with controls (0.883 ± 0.013; t = 20.07, P < 0.001). This trend persisted at 48 h (DOX: 0.540 ± 0.007 vs. DMSO: 1.470 ± 0.029; t = 31.25, P < 0.001), 72 h (DOX: 0.561 ± 0.020 vs. DMSO: 2.320 ± 0.095; t = 0, P < 0.001), and 96 h (DOX: 0.383 ± 0.011 vs. DMSO: 2.300 ± 0.061; t = 0, P < 0.001), confirming sustained inhibition of proliferation. A similar inhibitory effect was observed in RL-95-2 cells treated with 2 µM doxorubicin. At baseline, OD450 values were not significantly different (DMSO: 0.320 ± 0.004 vs. DOX: 0.322 ± 0.003; U = 47, P = 0.85). By 24 h, however, proliferation was already significantly impaired in the DOX group (0.489 ± 0.004) compared with controls (0.968 ± 0.007; t = 58.6, P < 0.001). The differences remained significant at 48 h (DOX: 0.418 ± 0.005 vs. DMSO: 1.048 ± 0.017; U = 0, P < 0.001), 72 h (DOX: 0.489 ± 0.004 vs. DMSO: 2.287 ± 0.018; t = 95.33, P < 0.001), and 96 h (DOX: 0.603 ± 0.014 vs. DMSO: 3.305 ± 0.024; t = 97.39, P < 0.001), demonstrating a profound time-dependent inhibition of cell growth (Fig. 4C).

To corroborate these findings, EdU incorporation assays were performed to directly assess DNA synthesis. In Ishikawa cells, the proportion of EdU-positive cells was significantly reduced in the doxorubicin-treated group (15.120% ± 0.310%) compared with the control group (23.770% ± 0.800%; U = 0, P < 0.001). Similarly, in RL-95-2 cells, doxorubicin treatment lowered the EdU-positive fraction to 7.799% ± 0.139%, compared with 14.180% ± 0.425% in controls (U = 0, P < 0.001) (Fig. 4D).

Taken together, these results demonstrate that doxorubicin significantly suppresses the proliferation of endometrial cancer cells in a dose- and time-dependent manner, as evidenced by reduced viability and impaired DNA synthesis in both Ishikawa and RL-95-2 cell lines.

The antiproliferative effect of doxorubicin on endometrial cancer cells is partially reversed by ferrostatin-1

To determine whether the inhibitory effect of doxorubicin on cell proliferation was associated with ferroptosis, ferrostatin-1 (Fer-1) was administered in combination with doxorubicin, and cell viability was assessed using CCK-8 assays. In Ishikawa cells, OD450 values did not differ significantly between the DOX group (0.548 ± 0.003) and the DOX + Fer-1 group (0.515 ± 0.027) at 0 h (U = 50, P > 0.99). At 24 h, however, Fer-1 co-treatment partially reversed the inhibitory effect of doxorubicin, with OD450 values of 0.664 ± 0.020 compared with 0.533 ± 0.012 in the DOX group (t = 5.554, P < 0.001). This reversal persisted at 48 h (DOX + Fer-1: 0.696 ± 0.010 vs. DOX: 0.540 ± 0.007; t = 12.47, P < 0.001), 72 h (0.695 ± 0.020 vs. 0.561 ± 0.020; U = 5, P < 0.001), and 96 h (0.536 ± 0.021 vs. 0.383 ± 0.011; U = 0, P < 0.001). A similar trend was observed in RL-95-2 cells. At baseline (0 h), OD450 values were comparable between the DOX group (0.322 ± 0.003) and the DOX + Fer-1 group (0.328 ± 0.006; U = 40, P = 0.48). At 24 h, cells treated with DOX + Fer-1 showed significantly higher viability (0.505 ± 0.003) than those treated with DOX alone (0.489 ± 0.004; t = 3.347, P = 0.004). The difference became more pronounced at 48 h (0.510 ± 0.010 vs. 0.418 ± 0.005; t = 8.521, P < 0.001), 72 h (0.701 ± 0.004 vs. 0.489 ± 0.004; t = 36.48, P < 0.001), and 96 h (0.747 ± 0.014 vs. 0.603 ± 0.014; t = 7.193, P < 0.001) (Fig. 4E).

The reversal effect of Fer-1 on proliferation was further confirmed by EdU incorporation assays. In Ishikawa cells, the proportion of EdU-positive cells increased from 15.120% ± 0.310% in the doxorubicin group to 16.420% ± 0.389% in the DOX + Fer-1 group (t = 2.632, P < 0.05). In RL-95-2 cells, EdU positivity increased from 7.799% ± 0.139% with doxorubicin alone to 9.045% ± 0.276% with DOX + Fer-1 (t = 4.036, P < 0.001) (Fig. 4F). Collectively, these results indicate that ferroptosis contributes to the antiproliferative effects of doxorubicin in endometrial cancer cells, and inhibition of ferroptosis by Fer-1 partially rescues the proliferative capacity of both Ishikawa and RL-95-2 cells. The incomplete reversal further suggests that ferroptosis contributes to, but is unlikely to be the sole determinant of, doxorubicin-mediated antiproliferative activity in EC cells.

Doxorubicin induces ferroptosis in endometrial cancer cells

To evaluate whether doxorubicin induces ferroptosis in endometrial cancer cells, we first examined ultrastructural changes by transmission electron microscopy. In both Ishikawa and RL-95-2 cells, mitochondria in the DMSO control group displayed intact morphology with well-preserved cristae and membranes. By contrast, doxorubicin treatment for 24 h induced typical ferroptosis features, including mitochondrial shrinkage, increased membrane density, disappearance of cristae, and overall structural disruption (Fig. 5A).

Fig. 5.

Doxorubicin induces ferroptosis in EC cells. (A) Transmission electron microscopy images of Ishikawa and RL-95-2 cells treated with DMSO or doxorubicin for 24 h. Control cells exhibited intact mitochondria with preserved cristae, while doxorubicin-treated cells displayed typical ferroptosis features, including mitochondrial shrinkage, increased membrane density, and disappearance of cristae; (B) MDA levels were significantly increased in both Ishikawa and RL-95-2 cells following doxorubicin treatment. Co-treatment with Fer-1 significantly reduced MDA levels in both cell lines; (C) Doxorubicin treatment elevated the Fe²⁺/Fe ratio in both Ishikawa and RL-95-2 cells. Fer-1 co-treatment reduced this elevation in both cell types; (D) The GSH/GSSG ratio was markedly reduced by doxorubicin in both Ishikawa and RL-95-2 cells, while Fer-1 co-treatment restored the redox balance in both cell lines; (E) ROS accumulation increased in both Ishikawa and RL-95-2 cells after doxorubicin exposure, and this was significantly reduced by Fer-1 co-treatment; (F) Western blot analysis showed that doxorubicin treatment decreased the expression of ferroptosis-associated proteins SLC7A11 and GPX4 in both Ishikawa and RL-95-2 cells. The downregulation of these proteins was significantly reversed by Fer-1 co-treatment. *P < 0.05, **P < 0.01, ***P < 0.001

Biochemical analyses further confirmed ferroptosis induction. In Ishikawa cells, doxorubicin treatment significantly increased MDA levels compared with DMSO controls (F = 49.55, P < 0.001), while co-treatment with Fer-1 markedly reduced MDA relative to doxorubicin alone (F = 49.55, P < 0.001). A similar trend was observed in RL-95-2 cells, where MDA was elevated following doxorubicin exposure (F = 155.2, P < 0.001) and suppressed by Fer-1 (F = 155.2, P = 0.001) (Fig. 5B). The Fe²⁺/Fe ratio was also significantly increased in both Ishikawa (F = 42.5, P = 0.007) and RL-95-2 (F = 55.45, P = 0.002) cells after doxorubicin treatment, while Fer-1 reduced this elevation in both cell types (F = 42.5 and 55.45, P < 0.001) (Fig. 5C). In parallel, doxorubicin markedly reduced the GSH/GSSG ratio in Ishikawa (F = 504.6, P < 0.001) and RL-95-2 (F = 13185, P < 0.001) cells, whereas Fer-1 co-treatment significantly restored the balance (P < 0.001 for both) (Fig. 5D). Consistent with these results, ROS accumulation was increased in Ishikawa (F = 22.47, P = 0.001) and RL-95-2 (F = 10.19, P = 0.009) cells following doxorubicin exposure, and was significantly reduced upon Fer-1 addition (P = 0.002 and P = 0.03, respectively) (Fig. 5E). Collectively, these findings demonstrate that doxorubicin triggers ferroptosis in both endometrial cancer cell lines, and this process can be effectively suppressed by the ferroptosis inhibitor Fer-1.

We next evaluated the expression of ferroptosis-associated proteins SLC7A11 and GPX4. Western blotting revealed that in Ishikawa cells, SLC7A11 expression decreased from 1.000 ± 0.034 in controls to 0.841 ± 0.044 after doxorubicin treatment (t = 2.850, P = 0.046), while GPX4 decreased from 1.000 ± 0.055 to 0.820 ± 0.020 (t = 3.056, P = 0.020). Similar changes were observed in RL-95-2 cells, where SLC7A11 expression declined from 1.000 ± 0.060 to 0.791 ± 0.040 (t = 2.902, P = 0.040), and GPX4 from 0.997 ± 0.055 to 0.671 ± 0.086 (t = 3.176, P = 0.010) (Fig. 5F). Together, these data provide strong evidence that doxorubicin promotes ferroptosis in endometrial cancer cells, in part by suppressing the SLC7A11–GPX4 axis, thereby disrupting redox homeostasis and enhancing lipid peroxidation. While ROS accumulation and mitochondrial injury are not exclusive to ferroptosis, the Fer-1–sensitive lipid peroxidation phenotype (BODIPY 581/589-C11 oxidation and MDA elevation), together with iron overload/redox imbalance and coordinated downregulation of SLC7A11 and GPX4, supports ferroptosis involvement in the doxorubicin response in EC cells.

Transcriptomic profiling suggests doxorubicin suppresses MKK6 to induce ferroptosis

To further elucidate the molecular mechanisms underlying doxorubicin-induced ferroptosis, transcriptomic profiling was performed in Ishikawa cells following doxorubicin treatment. Comparative RNA-seq analysis identified a total of 4,029 DEGs relative to the DMSO control group, of which 2,961 were significantly upregulated and 1,068were downregulated (Fig. 6A). Notably, among the downregulated transcripts were the ferroptosis regulator SLC7A11 and the signaling kinase MKK6 (Mitogen-Activated Protein Kinase Kinase 6), suggesting that doxorubicin may concurrently modulate both ferroptotic and inflammatory signaling networks. KEGG enrichment analysis of the downregulated genes revealed significant perturbations in the ferroptosis pathway and the TNF signaling pathway. To further link the transcriptomic response to ferroptosis, ferroptosis-focused GSEA was performed using the DOX-versus-control RNA-seq dataset, revealing significant enrichment of gene sets related to ferroptosis, redox regulation, and lipid metabolism (Supplementary Fig. 1 and Supplementary Table 2). Within these enriched pathways, SLC7A11 was mapped as a critical ferroptosis effector, whereas MKK6 emerged as a key component of the TNF pathway, highlighting its potential role as a doxorubicin-responsive signaling mediator. Together, these transcriptomic findings point to MKK6 as a candidate regulatory node through which doxorubicin may suppress pro-survival signaling and promote ferroptotic cell death in endometrial cancer. Among the DOX-responsive signaling candidates highlighted by transcriptomic profiling, MKK6 was prioritized because it connects the stress-kinase (p38) axis to a downstream transcriptional program with plausible redox relevance. Subsequent biochemical, target-engagement, and rescue experiments (p38 pathway inhibition, CETSA/SPR, and MKK6 restoration) were therefore designed to test the specificity and functional consequence of this candidate node.

Fig. 6.

Doxorubicin engages MKK6 and suppresses the MKK6/p38/CEBPB axis. (A) RNA-seq volcano plot showing DEGs in DOX-treated Ishikawa cells, including decreased SLC7A11 and MKK6; (B–C) WB showing reduced p-MKK6/p-p38 and decreased CEBPB in Ishikawa and RL-95-2 cells after DOX; (D–E) Docking model of DOX–MKK6 interaction and predicted contact residues; (F) CETSA showing DOX-dependent thermal stabilization of MKK6 in cells; (G) SPR sensorgrams and KD indicating direct DOX–MKK6 binding. *P < 0.05, **P < 0.01, ***P < 0.001

Doxorubicin suppresses the MKK6/P38/CEBPB signaling axis

To validate the transcriptomic findings suggesting the involvement of the TNF signaling cascade, Western blot analyses were performed to examine the phosphorylation status of MKK6 and p38 MAPK, as well as the expression of the downstream transcription factor CCAAT enhancer binding protein beta (CEBPB). In Ishikawa cells, doxorubicin treatment significantly reduced the phosphorylation of MKK6 and p38. Specifically, the ratio of P-MKK6/MKK6 decreased from 1.000 ± 0.014 in the DMSO group to 0.839 ± 0.019 in the DOX group (t = 6.874, P = 0.002), while P-P38/P38 levels declined from 1.000 ± 0.058 to 0.731 ± 0.051 (t = 3.489, P = 0.006), indicating marked inhibition of the MKK6–P38 axis. A similar pattern was observed in RL-95-2 cells, where P-MKK6/MKK6 ratios were reduced from 1.000 ± 0.029 to 0.882 ± 0.016 (t = 3.546, P = 0.020), and P-P38/P38 levels decreased from 0.977 ± 0.068 to 0.658 ± 0.058 (t = 3.573, P = 0.003), further confirming pathway suppression (Fig. 6B).

In addition, the downstream transcription factor CEBPB was markedly downregulated following doxorubicin exposure. In Ishikawa cells, CEBPB protein levels decreased from 1.000 ± 0.040 in controls to 0.306 ± 0.036 in the DOX group. Consistent findings were observed in RL-95-2 cells, where CEBPB expression declined from 1.000 ± 0.139 to 0.596 ± 0.026 (t = 2.854, P = 0.046) (Fig. 6C).

Taken together, these results demonstrate that doxorubicin significantly suppresses the phosphorylation of MKK6 and p38 and reduces the expression of CEBPB, thereby inhibiting the MKK6/P38/CEBPB signaling axis in endometrial cancer cells. Conceptually, these data support a cascade in which doxorubicin first suppresses MKK6–p38 activation at the kinase level, followed by reduced CEBPB abundance and subsequent attenuation of the SLC7A11/GPX4 antioxidant defense program, thereby creating a permissive redox/iron milieu for lipid peroxidation and ferroptotic execution.

Doxorubicin induces ferroptosis via direct binding to MKK6

To clarify the molecular mechanism by which doxorubicin modulates the MKK6/P38/CEBPB axis, molecular docking was first performed to predict the interaction between doxorubicin and MKK6. Analysis revealed that DOX stably binds within the active pocket of MKK6, with a predicted binding energy of -9.8 kcal/mol, indicating strong binding affinity. Molecular interaction analysis of the binding was performed using PLIP, and the results were visualized in PyMOL. In the vicinity of MKK6’s active pocket, DOX forms hydrogen bonds with residues S201, D197, K181, K82, G65, R61, D137, K138, and T134, while hydrophobic interactions are observed with M132, L186, N184, and V67 (Fig. 6D and E).

Molecular dynamics simulation supports a stable interaction between doxorubicin and MKK6

To further assess the dynamic stability of the doxorubicin (DOX)–MKK6 complex and characterize the interaction features in an explicit-solvent environment, a 100 ns molecular dynamics (MD) simulation was performed. Structural convergence was first evaluated by root-mean-square deviation (RMSD). As shown in Supplementary Fig. 2A, the complex exhibited noticeable fluctuations during the initial ~ 10 ns, consistent with early-stage relaxation from the starting conformation. Thereafter, RMSD variations became limited and progressively stabilized over time, indicating that the system approached equilibrium. At the end of the 100 ns simulation, the RMSD of the complex reached 0.35176 nm, supporting an overall stable binding conformation throughout the production phase.

We next examined residue-level flexibility of MKK6 in the bound state using root-mean-square fluctuation (RMSF). As depicted in Supplementary Fig. 2B, the majority of residues displayed relatively restrained fluctuations during the simulation, while a localized increase in flexibility was observed around residues 44–45, suggesting a more mobile region within the protein structure under the simulated conditions. To evaluate global compactness of the protein–ligand system, the radius of gyration (Rg) was analyzed. As shown in Supplementary Fig. 2C, Rg values remained within a narrow range (approximately 1.9–2.0 nm) during the entire simulation, and the terminal Rg value at 100 ns was 1.9791 nm, indicating that the overall folding state and compactness of the complex were maintained without evidence of global unfolding or structural destabilization. Hydrogen bonding is an important contributor to protein–ligand stability. Consistent with sustained binding, analysis of hydrogen bonds between DOX and MKK6 revealed persistent polar interactions throughout the 100 ns trajectory. As shown in Supplementary Fig. 2D, the number of intermolecular hydrogen bonds fluctuated between 1 and 3, supporting continuous contact between the ligand and the binding pocket during simulation.

In addition, solvent-accessible surface area (SASA) analysis was conducted to assess solvent exposure and potential changes in surface properties during the trajectory. As shown in Supplementary Fig. 2E, SASA (reported as buried/exposed surface-related variation in the simulation output) showed higher fluctuations during the early-to-mid simulation stage and became more stable toward the later stage, with the surface-related metric fluctuating around ~ 160 nm² near the end of the simulation. Collectively, these results indicate that the DOX–MKK6 complex remained structurally stable over the 100 ns timescale, with maintained compactness and persistent intermolecular interactions.

To further quantify the energetic favorability of DOX binding, binding free energy was estimated using the gmx_MMPBSA approach. As summarized in Fig. 2, the total gas-phase contribution (GGAS) was negative, reflecting favorable nonbonded interactions. Specifically, both van der Waals energy (VDWAALS < 0) and electrostatic energy (EEL < 0) contributed favorably to binding. In contrast, the polar solvation term (EGB > 0) opposed binding, while the nonpolar solvation term (ESURF < 0) provided a modest favorable contribution. The resulting total binding free energy was − 26.74 ± 6.48 kcal/mol, indicating an overall favorable interaction between DOX and MKK6 under the simulated conditions (Supplementary Fig. 2F). Finally, per-residue energy decomposition was performed to identify residues contributing prominently to ligand binding. As shown in Supplementary Fig. 2G, Leu59, Gly62, and Val67 exhibited relatively strong energetic contributions, suggesting that these residues may represent key interaction determinants that stabilize the DOX–MKK6 complex.

Taken together, trajectory-based structural analyses (RMSD, RMSF, Rg, hydrogen bonds, and SASA) and MM-PBSA energy calculations consistently support that DOX can form a dynamically stable complex with MKK6 and maintain sustained interactions over the 100 ns simulation period.

CETSA confirms DOX binding to MKK6 by stabilizing MKK6 thermal denaturation

To confirm this interaction in cells, CETSA was carried out in Ishikawa and RL-95-2 cells exposed to doxorubicin for 4 h. In the DMSO control group, MKK6 protein levels declined sharply with increasing temperatures, whereas in the doxorubicin-treated group, thermal degradation was markedly attenuated, suggesting that doxorubicin stabilized MKK6 against heat-induced denaturation. In Ishikawa cells, CETSA analysis showed that MKK6 protein levels declined progressively with increasing temperatures in the control group, whereas doxorubicin markedly preserved protein stability. At 42 °C, MKK6 levels were comparable between the DMSO group (105.0 ± 7.000%) and the DOX group (108.7 ± 2.910%; t = 0.484, P = 0.65). However, significant differences emerged at higher temperatures: at 47 °C, protein levels were 70.67 ± 7.513% in controls compared with 101.7 ± 5.239% with DOX (t = 3.385, P = 0.03); at 52 °C, 28.67 ± 4.667% versus 99.33 ± 6.333% (t = 8.983, P < 0.001); at 57 °C, 8.000 ± 2.309% versus 92.67 ± 8.838% (t = 9.269, P < 0.001); and at 62 °C, 6.333 ± 0.882% versus 29.33 ± 4.631% (t = 4.879, P = 0.008) (Fig. 6F). A similar pattern was observed in RL-95-2 cells. At 42 °C, MKK6 protein retention was significantly higher in the DOX group (104.0 ± 2.517%) compared with controls (86.67 ± 1.202%; t = 6.215, P = 0.003). This protective effect persisted across higher temperatures: at 47 °C, 97.00 ± 6.110% versus 51.33 ± 1.856% (t = 7.151, P = 0.002); at 52 °C, 84.67 ± 10.48% versus 36.67 ± 2.404% (t = 4.465, P = 0.01); at 57 °C, 54.00 ± 3.786% versus 31.33 ± 4.333% (t = 3.939, P = 0.02); and at 62 °C, 54.00 ± 4.359% versus 20.67 ± 1.667% (t = 7.143, P = 0.002). Collectively, these results demonstrate that doxorubicin markedly stabilizes MKK6 against heat-induced denaturation in both Ishikawa and RL-95-2 cells (Fig. 6F).

To further validate the direct binding of doxorubicin to MKK6, SPR analysis was performed, yielding a dissociation constant (KD) of 2.41 × 10^−5 M, supporting a direct interaction between doxorubicin and MKK6 (Fig. 6G). Collectively, the molecular docking, CETSA, and SPR results consistently demonstrate that doxorubicin binds to MKK6, with predicted involvement of residues Q93, L204, R178, R224, and K232, thereby suppressing activation of the MKK6/P38/CEBPB signaling pathway and promoting ferroptotic cell death in endometrial cancer. Together, molecular docking, CETSA, and SPR support target engagement and a direct interaction between doxorubicin and MKK6; however, additional specificity experiments (e.g., competition assays or site-directed mutagenesis) will further strengthen the selectivity of this interaction.

MKK6 overexpression attenuates DOX-induced lipid peroxidation and ferroptosis

To establish a causal contribution of MKK6 suppression to doxorubicin (DOX)-induced ferroptosis, we performed rescue experiments by restoring MKK6 expression and assessed ferroptosis-associated lipid peroxidation and biochemical markers in Ishikawa and RL-95-2 cells. Lipid peroxidation was quantified using the ferroptosis-sensitive probe BODIPY 581/589-C11, expressed as the oxidized/reduced fluorescence intensity ratio, and ferroptosis-related biochemical indices (MDA, Fe²⁺/Fe, and GSH/GSSG) were measured in parallel.

We first confirmed that DOX robustly elevates lipid peroxidation relative to control transfection conditions. Following DOX treatment, cells were replated into 96-well plates and subjected to BODIPY 581/589-C11 staining. In Ishikawa cells, lipid peroxidation in the DOX group (28.739% ± 2.011%) was significantly higher than that in the DOX+DMSO-oe control group (18.850% ± 2.209%; U = 106, P < 0.001). Similarly, in RL-95-2 cells, DOX increased lipid peroxidation to 48.445% ± 2.075%, compared with 26.341% ± 1.680% in the DOX+DMSO-oe group (U = 4, P < 0.001) (Supplementary Fig. 3A). These data verified that DOX induces lipid oxidative damage under our experimental settings, while also controlling for potential confounding introduced by transfection procedures.

We next tested whether restoring MKK6 could mitigate DOX-driven lipid peroxidation. Cells were exposed to DOX for 12 h, transfected with an MKK6 overexpression construct, and cultured for an additional 24 h prior to BODIPY 581/589-C11 staining. In Ishikawa cells, MKK6 overexpression significantly reduced the DOX-induced lipid peroxidation signal (DOX: 59.895% ± 4.479% vs. DOX+MKK6-oe: 32.206% ± 1.724%; t = 5.769, P < 0.001). Consistently, in RL-95-2 cells, lipid peroxidation was also significantly decreased in the DOX+MKK6-oe group compared with DOX alone (10.880% ± 1.501% vs. 23.206% ± 2.317%; U = 15, P < 0.001) (Supplementary Fig. 3B). Together, these findings indicate that restoring MKK6 attenuates DOX-induced lipid peroxidation, supporting a functional role for MKK6 suppression in the initiation of ferroptotic lipid damage.

To further support that the observed lipid peroxidation changes were ferroptosis-related, we evaluated the effect of the canonical ferroptosis inhibitor ferrostatin-1 (Fer-1). Cells were pretreated with Fer-1 (or vehicle control) for 12 h prior to DOX exposure and analyzed by BODIPY 581/589-C11 staining. In Ishikawa cells, Fer-1 significantly reduced DOX-induced lipid peroxidation (DOX: 33.347% ± 1.989% vs. DOX + Fer-1: 26.631% ± 1.945%; t = 2.414, P = 0.0226), further confirming that DOX-triggered lipid oxidation in this context is sensitive to ferroptosis inhibition (Supplementary Fig. 3C).

Given that CEBPB functions downstream of the MKK6/p38 axis and transcriptionally regulates SLC7A11, we additionally assessed whether restoring CEBPB could modulate DOX-induced lipid peroxidation. In RL-95-2 cells, CEBPB overexpression significantly decreased lipid peroxidation compared with DOX treatment alone (DOX: 49.650% ± 1.711% vs. DOX+CEBPB-oe: 40.874% ± 1.224%; t = 4.173, P < 0.001), consistent with the involvement of the MKK6/p38/CEBPB pathway in regulating ferroptotic lipid oxidation (Supplementary Fig. 3D).

We next examined ferroptosis-associated biochemical markers after restoring MKK6 expression. In Ishikawa cells, MKK6 overexpression significantly reduced DOX-induced lipid peroxidation and iron dysregulation, as evidenced by a marked decrease in MDA (DOX+MKK6-oe: 1.137 ± 0.010 vs. DOX: 1.759 ± 0.016; t = 32.197, P < 0.001) and a reduced Fe²⁺/Fe ratio (30.906 ± 0.769 vs. 60.635 ± 2.062; t = 13.510, P < 0.001). In parallel, the GSH/GSSG ratio was significantly increased in the DOX+MKK6-oe group (13.591 ± 0.298) compared with DOX alone (12.690 ± 0.052; t = 2.977, P = 0.041), indicating partial restoration of redox homeostasis. Consistent findings were observed in RL-95-2 cells, where MKK6 overexpression significantly decreased MDA (0.542 ± 0.006 vs. 1.165 ± 0.021; t = 28.990, P < 0.001) and lowered the Fe²⁺/Fe ratio (16.532 ± 4.072 vs. 52.443 ± 3.252; t = 6.892, P = 0.002), while markedly increasing the GSH/GSSG ratio (18.543 ± 0.165 vs. 12.696 ± 0.198; t = 22.704, P < 0.001) (Supplementary Fig. 4A).

Collectively, BODIPY 581/589-C11–based lipid peroxidation measurements and ferroptosis-associated biochemical indices consistently demonstrate that restoring MKK6 expression attenuates DOX-induced ferroptotic alterations, thereby strengthening a causal link between MKK6 inhibition and DOX-triggered ferroptosis in endometrial cancer cells.

MKK6 overexpression partially rescues DOX-suppressed proliferation

To determine whether restoration of MKK6 can functionally counteract the antiproliferative effects of doxorubicin (DOX), rescue experiments were performed in Ishikawa and RL-95-2 cells. Cells were exposed to DOX for 12 h and subsequently transfected with an MKK6 overexpression construct; after an additional 24 h of culture, cell growth was monitored by CCK-8 assays over a time course (0–96 h), and DNA synthesis was evaluated by EdU incorporation.

Consistent with a functional rescue effect, MKK6 overexpression significantly increased cell viability compared with DOX treatment alone. In Ishikawa cells treated with DOX (1 µM), OD450 values were comparable at baseline between groups (DOX: 0.480 ± 0.003 vs. DOX+MKK6-oe: 0.482 ± 0.003; t = 0.448, P = 0.661). However, MKK6 overexpression markedly improved viability at subsequent time points, with higher OD450 values observed at 24 h (0.757 ± 0.004 vs. 0.606 ± 0.002; t = 30.861, P < 0.001), 48 h (0.487 ± 0.007 vs. 0.415 ± 0.008; t = 6.748, P < 0.001), 72 h (0.540 ± 0.004 vs. 0.494 ± 0.007; t = 5.709, P < 0.001), and 96 h (0.472 ± 0.003 vs. 0.415 ± 0.003; P < 0.001), indicating that MKK6 reconstitution partially alleviated DOX-induced growth suppression (Supplementary Fig. 4B). A similar trend was observed in RL-95-2 cells treated with DOX (1.6 µM). Baseline OD450 values did not differ between the DOX and DOX+MKK6-oe groups (0.485 ± 0.013 vs. 0.486 ± 0.010; P = 0.798). In contrast, MKK6 overexpression significantly restored cell viability at 24 h (0.536 ± 0.012 vs. 0.442 ± 0.006; t = 7.235, P < 0.001), 48 h (0.376 ± 0.007 vs. 0.265 ± 0.004; t = 14.527, P < 0.001), 72 h (0.681 ± 0.009 vs. 0.610 ± 0.013; t = 4.629, P < 0.001), and 96 h (0.798 ± 0.024 vs. 0.715 ± 0.018; t = 2.722, P = 0.017), further supporting a consistent rescue effect across both endometrial cancer cell lines (Supplementary Fig. 4B).

To corroborate these viability findings at the level of DNA synthesis, EdU incorporation assays were performed. In Ishikawa cells, MKK6 overexpression significantly increased the proportion of EdU-positive cells compared with DOX treatment alone (DOX+MKK6-oe: 16.92% ± 1.052% vs. DOX: 6.161% ± 0.152%; t = 10.124, P < 0.001). Likewise, in RL-95-2 cells, the EdU-positive fraction was significantly higher in the DOX+MKK6-oe group than in the DOX group (11.777% ± 1.022% vs. 10.276% ± 0.705%; t = 5.410, P < 0.001) (Supplementary Fig. 4C).

Together, these results demonstrate that MKK6 overexpression partially rescues DOX-suppressed proliferative capacity in endometrial cancer cells, consistent with the notion that inhibition of MKK6 signaling contributes to the growth-inhibitory effects of DOX.

CEBPB activates SLC7A11 promoter activity and is associated with increased SLC7A11 expression

Given that CEBPB is a downstream effector of the MKK6/P38 signaling pathway, we investigated whether it functions as a transcriptional regulator of SLC7A11. Bioinformatic prediction using the hTFtarget and JASPAR databases identified CEBPB as a putative transcription factor of SLC7A11, with a high-scoring binding motif (GAATTTGCATCATCA) located within its promoter region. To experimentally validate this prediction, we first confirmed the efficiency of CEBPB overexpression in Ishikawa and RL-95-2 cells. qPCR analysis demonstrated a robust induction of CEBPB mRNA in the overexpression group compared with mock controls, increasing from 1.036 ± 0.202 to 2064 ± 53.43 in Ishikawa cells (t = 38.61, P < 0.001) and from 1.025 ± 0.155 to 69.30 ± 3.907 in RL-95-2 cells (t = 17.73, P < 0.001) (Fig. 7A). Consistent with this, Western blotting showed a marked increase in CEBPB protein levels, from 1.157 ± 0.194 to 3.792 ± 0.176 in Ishikawa (t = 10.06, P < 0.001) and from 1.000 ± 0.138 to 1.716 ± 0.092 in RL-95-2 (t = 4.307, P = 0.01) (Fig. 7B).

Fig. 7.

CEBPB promotes SLC7A11 expression and transactivates the SLC7A11 promoter. (A, B) qPCR and WB validation of CEBPB overexpression in Ishikawa and RL-95-2 cells; (C, D) qPCR and WB showing increased SLC7A11 mRNA and protein upon CEBPB overexpression; (E) JASPAR-predicted CEBPB binding motif in the SLC7A11 promoter; (F) Dual-luciferase assay in 293T cells showing motif-dependent activation of the SLC7A11 promoter by CEBPB (WT vs. mutant). *P < 0.05, **P < 0.01, ***P < 0.001

We then assessed whether CEBPB overexpression altered SLC7A11 expression. qPCR results revealed that SLC7A11 mRNA was significantly upregulated in both Ishikawa (from 1.001 ± 0.030 to 2.720 ± 0.083, t = 19.50, P < 0.001) and RL-95-2 cells (from 1.007 ± 0.083 to 1.971 ± 0.064, t = 9.234, P < 0.001) (Fig. 7C). Western blot analysis confirmed parallel increases in SLC7A11 protein expression, rising from 1.000 ± 0.033 to 1.250 ± 0.037 in Ishikawa cells (t = 5.057, P = 0.007) and from 1.000 ± 0.029 to 1.302 ± 0.110 in RL-95-2 cells (t = 2.664, P = 0.04) (Fig. 7D).

To further clarify whether CEBPB activates SLC7A11 transcription, we analyzed the SLC7A11 promoter region using the JASPAR database. Motif enrichment identified several putative CEBPB binding sites, among which the highest relative score (R.S.) was assigned to the sequence GAATTTGCATCATCA, suggesting it as the most likely functional element (Fig. 7E). To experimentally validate this prediction, a dual-luciferase reporter assay was conducted in 293T cells. Constructs containing either the wild-type or mutant SLC7A11 promoter were co-transfected with CEBPB overexpression plasmids or empty vectors, alongside internal Renilla controls. The results revealed that co-expression of CEBPB with the wild-type SLC7A11 promoter markedly increased luciferase activity, whereas mutation of the predicted binding site abolished this effect (Fig. 7F). Comparison of PGL3-Basic + CEBPB-oe vs. WT-SLC7A11 + CEBPB-oe showed a significant increase (t = 28.58, p < 0.001), while no significant difference was observed between PGL3-Basic + CEBPB-oe and MT-SLC7A11 + CEBPB-oe (t = 0.9379, p = 0.44) (Fig. 7F). While these reporter assays support motif-dependent promoter transactivation by CEBPB, chromatin-level validation of endogenous CEBPB binding at the native SLC7A11 promoter in EC cells will be required to confirm direct transcriptional regulation.

Together, these results support that CEBPB positively regulates SLC7A11 expression and can transactivate the SLC7A11 promoter in reporter assays in a motif-dependent manner. Chromatin-level binding of endogenous CEBPB to the SLC7A11 promoter in EC cells remains to be validated in future studies.

CEBPB overexpression reverses the anti-proliferative effects of doxorubicin in endometrial cancer cells

To determine whether CEBPB mediates the regulatory effects of doxorubicin on endometrial cancer cell proliferation, rescue assays were performed using CEBPB overexpression. In Ishikawa cells, DOX treatment maintained suppressed proliferation across all time points, with OD450 values of 0.461 ± 0.002 at baseline, 0.505 ± 0.012 at 24 h, 0.521 ± 0.007 at 48 h, 0.539 ± 0.008 at 72 h, and 0.421 ± 0.006 at 96 h. When CEBPB was overexpressed, OD450 values increased to 0.448 ± 0.023 at baseline, 0.674 ± 0.008 at 24 h, 0.627 ± 0.003 at 48 h, 0.660 ± 0.014 at 72 h, and 0.549 ± 0.010 at 96 h. Statistical analysis confirmed that the proliferative recovery in the DOX+CEBPB-oe group was significant at 24 h (U = 0, P < 0.001), 48 h (t = 13.45, P < 0.001), 72 h (U = 0, P < 0.001), and 96 h (t = 10.62, P < 0.001), while no significant difference was detected at baseline (t = 16, P = 0.1). Similarly, in RL-95-2 cells, DOX treatment alone resulted in OD450 values of 0.336 ± 0.005 at baseline, 0.492 ± 0.004 at 24 h, 0.407 ± 0.004 at 48 h, 0.512 ± 0.006 at 72 h, and 0.556 ± 0.007 at 96 h. Overexpression of CEBPB significantly enhanced proliferation, yielding OD450 values of 0.345 ± 0.005 at baseline, 0.507 ± 0.006 at 24 h, 0.482 ± 0.012 at 48 h, 0.706 ± 0.039 at 72 h, and 0.693 ± 0.013 at 96 h. Differences between the groups were not significant at baseline (t = 1.265, P = 0.22) and 24 h (t = 1.962, P = 0.07), but became highly significant at 48 h (t = 5.885, P < 0.001), 72 h (U = 0, P < 0.001), and 96 h (t = 9.132, P < 0.001) (Fig. 8A).

Fig. 8.

CEBPB overexpression attenuates doxorubicin-induced growth suppression in EC cells. (A) CCK-8 time-course showing partial restoration of viability by CEBPB overexpression in DOX-treated Ishikawa and RL-95-2 cells; (B) EdU assay showing increased DNA synthesis in DOX-treated cells upon CEBPB overexpression. **P < 0.01, ***P < 0.001

Consistent with these findings, EdU assays further confirmed that CEBPB overexpression reversed the inhibitory effect of doxorubicin on DNA synthesis. In Ishikawa cells, the percentage of EdU-positive cells rose from 14.61 ± 0.454% in the doxorubicin group to 16.17 ± 0.329% with CEBPB overexpression (t = 2.784, P = 0.0095), while in RL-95-2 cells, positivity increased from 9.057 ± 0.224% to 12.020 ± 0.288% (t = 6.882, P < 0.001) (Fig. 8B). Together, these results indicate that CEBPB restoration effectively counteracts the anti-proliferative effects of doxorubicin in endometrial cancer cells.

CEBPB overexpression reverses doxorubicin-induced ferroptosis in endometrial cancer cells

To investigate whether DOX regulates ferroptosis in endometrial cancer cells through CEBPB, we performed rescue experiments by overexpressing CEBPB in Ishikawa and RL-95-2 cells. In Ishikawa cells, DOX treatment significantly elevated lipid peroxidation, as reflected by increased MDA levels compared with controls, whereas co-expression of CEBPB markedly reduced this elevation (H = 20.51, P < 0.05). A similar effect was observed in RL-95-2 cells, where MDA levels were significantly reduced in the DOX+CEBPB-oe group compared with DOX alone (F = 34,509, P = 0.001) (Fig. 9A). Consistently, intracellular iron metabolism was altered by DOX, as the Fe²⁺/Fe ratio was significantly increased, but CEBPB overexpression reduced these levels in both Ishikawa (F = 124.9, P < 0.001) and RL-95-2 cells (F = 55.450, P < 0.001) (Fig. 9B). The redox balance was also restored by CEBPB, as GSH/GSSG ratios were markedly higher in the DOX+CEBPB-oe group than in the DOX group, rising from the DOX-suppressed state in Ishikawa (F = 1245, P < 0.001) and RL-95-2 cells (F = 13,185, P < 0.001) (Fig. 9C). Moreover, ROS accumulation induced by DOX was significantly reduced by CEBPB overexpression in Ishikawa cells (F = 22.04, P = 0.001) and RL-95-2 cells (F = 13.33, P = 0.01) (Fig. 9D). Together, these findings demonstrate that CEBPB overexpression effectively counteracts DOX-induced ferroptosis.

Fig. 9.

CEBPB overexpression counteracts doxorubicin-induced ferroptosis in EC cells. (A) Doxorubicin treatment significantly elevated MDA levels in both Ishikawa and RL-95-2 cell lines compared to controls. CEBPB overexpression markedly reduced MDA levels in both cell types; (B) DOX treatment significantly increased the Fe²⁺/Fe ratio in both Ishikawa and RL-95-2 cells. Overexpression of CEBPB effectively reversed this elevation, restoring iron homeostasis in both cell lines; (C) GSH/GSSG ratio, was significantly restored in the DOX+CEBPB-oe group compared to the DOX-treated cells, with higher GSH/GSSG ratios observed in both Ishikawa and RL-95-2 cells; (D) ROS accumulation induced by DOX was also significantly reduced in both cell lines upon CEBPB overexpression, with marked decreases in ROS levels in Ishikawa and RL-95-2 cells; (E) Expression of ferroptosis regulators SLC7A11 and GPX4 was significantly decreased by DOX in both Ishikawa and RL-95-2 cells. However, CEBPB overexpression significantly restored the expression of both proteins, with higher levels of SLC7A11 and GPX4 in the DOX+CEBPB-oe groups compared to the DOX-only treatment in both cell lines. *P < 0.05, **P < 0.01, ***P < 0.001

To further examine the molecular underpinnings, we evaluated the expression of key ferroptosis regulators SLC7A11 and GPX4. In Ishikawa cells, DOX significantly reduced SLC7A11 (0.655 ± 0.101 vs. 1.000 ± 0.057; F = 18.29, P = 0.01) and GPX4 protein levels (0.828 ± 0.033 vs. 1.108 ± 0.036; F = 6.025, P = 0.04). However, co-treatment with DOX and CEBPB overexpression significantly restored their expression, increasing SLC7A11 to 1.334 ± 0.075 (P < 0.001 vs. DOX) and GPX4 to 1.083 ± 0.098 (P = 0.04 vs. DOX). Similar trends were observed in RL-95-2 cells, where DOX decreased SLC7A11 (0.804 ± 0.043 vs. 1.000 ± 0.024; F = 6.776, P = 0.03) and GPX4 (0.671 ± 0.086 vs. 0.997 ± 0.055; F = 5.937, P = 0.02), both of which were restored by CEBPB overexpression to 1.036 ± 0.067 (P = 0.03 vs. DOX) and 0.978 ± 0.080 (P = 0.02 vs. DOX), respectively (Fig. 9E). These data indicate that CEBPB not only reverses DOX-induced ferroptotic phenotypes but also restores the expression of critical regulators SLC7A11 and GPX4, thereby attenuating ferroptosis in endometrial cancer cells.

Doxorubicin suppresses the growth of subcutaneous endometrial cancer xenografts in nude mice

To investigate the in vivo antitumor efficacy of DOX, a subcutaneous xenograft model was established in nude mice using endometrial cancer cells. Once tumor volumes reached approximately 100 mm³, mice were randomized into two groups: a control group receiving vehicle (DMSO diluted in PBS) and a treatment group administered DOX at 5 mg/kg every four days for a total of four injections. Tumor progression was monitored by caliper measurements throughout the treatment period. In the control group, tumor volumes increased steadily from 33.51 ± 1.924 mm³ on day 10 to 279.6 ± 11.07 mm³ on day 34. By contrast, DOX-treated tumors displayed markedly attenuated growth, rising from 38.34 ± 3.018 mm³ at day 10 to only 119.6 ± 21.47 mm³ by day 34. Statistical analysis confirmed significant differences at later time points, with tumor volumes on day 30 and day 34 being significantly lower in the DOX group compared with controls (t = 3.434, P = 0.009; t = 6.621, P < 0.001, respectively), although earlier time points showed no statistical significance (P > 0.05) (Fig. 10A and B). Consistent with the volumetric data, tumor weights at the endpoint were also significantly reduced in the DOX group relative to controls (t = 4.661, P = 0.002) (Fig. 10C). These findings suggest that DOX may exert inhibitory effects on the growth of subcutaneous endometrial cancer xenografts in vivo, though further validation is necessary.

Fig. 10.

Doxorubicin suppresses EC xenograft growth and is associated with ferroptosis-related changes in vivo. (A-C) Tumor growth curves and endpoint tumor weights in vehicle vs. DOX-treated mice; (D-F) Tumor tissue biochemical assays showing increased lipid peroxidation/iron dysregulation and altered redox status (MDA, Fe²⁺/Fe, GSH/GSSG); (G) WB showing decreased SLC7A11 and GPX4 in DOX-treated tumors; (H–I) WB showing reduced phosphorylation of MKK6 and p38 in DOX-treated tumors. *P < 0.05, **P < 0.01, ***P < 0.001

Doxorubicin is associated with ferroptosis-related changes in subcutaneous endometrial cancer xenografts

To evaluate whether DOX treatment is associated with ferroptosis-related alterations in vivo, tumor-bearing nude mice were treated with DOX (5 mg/kg) or vehicle (DMSO), and tumor tissues were harvested after treatment for biochemical and molecular analyses. Biochemical assessment revealed that, compared with the control group, tumors from DOX-treated mice exhibited a significant increase in malondialdehyde (MDA) content (U = 0, P = 0.008), consistent with enhanced lipid peroxidation (Fig. 10D). Similarly, the Fe²⁺/Fe ratio was markedly elevated in the DOX group (U = 0, P = 0.008), while the GSH/GSSG ratio was also significantly altered (t = 4.551, P = 0.002) (Fig. 10E and F), indicating disturbed redox homeostasis in DOX-treated tumors.

Western blot analyses further demonstrated that ferroptosis-related proteins were substantially altered following DOX exposure. Expression of SLC7A11 was significantly reduced in the DOX group (0.616 ± 0.101 vs. 1.000 ± 0.100 in controls, t = 2.694, P = 0.03), while GPX4 was similarly downregulated (0.487 ± 0.054 vs. 1.000 ± 0.033, t = 8.059, P < 0.001), consistent with ferroptosis-associated molecular changes in vivo (Fig. 10G).

Beyond these ferroptosis markers, we examined upstream signaling events within the MKK6/P38 axis. DOX-treated xenografts displayed pronounced suppression of phosphorylated MKK6 (P-MKK6/MKK6 ratio: 0.337 ± 0.035 vs. 1.000 ± 0.126 in controls; t = 5.067, P < 0.001) and phosphorylated P38 (P-P38/P38 ratio: 0.319 ± 0.046 vs. 1.000 ± 0.032; U = 0, P = 0.008), supporting inhibition of this signaling pathway in vivo (Fig. 10H and I).

The results suggest that DOX treatment is associated with ferroptosis-related biochemical and molecular alterations in tumor tissues, including increased lipid peroxidation markers, reduced SLC7A11/GPX4 expression, and inhibition of the MKK6/p38 signaling cascade. However, further experiments are needed to definitively establish the causality. A schematic model summarizing the proposed DOX–MKK6/p38/CEBPB–SLC7A11/GPX4 cascade is shown in Supplementary Fig. 5.

Discussion

In this study, we systematically explored the role of ferroptosis in endometrial cancer and uncovered a novel regulatory mechanism through which doxorubicin exerts its antitumor activity. By integrating transcriptomic analysis and multiple machine learning algorithms, we identified SLC1A5 as a ferroptosis-related hub gene with strong tumor-associated expression and functional relevance. Functional assays demonstrated that doxorubicin significantly inhibited the proliferation of endometrial cancer cells, and this antiproliferative effect was shown to be at least partially dependent on ferroptotic cell death, as pharmacological inhibition with ferrostatin-1 restored proliferative capacity. Notably, the rescue by ferrostatin-1 was incomplete, suggesting that ferroptosis contributes to doxorubicin-induced growth suppression but is unlikely to be the sole mode of cytotoxicity. Thus, ferroptosis likely acts in parallel with established anthracycline mechanisms, including DNA damage and topoisomerase II–dependent stress responses. Mechanistically, our data support a direct interaction between doxorubicin and MKK6 and suggest that doxorubicin is associated with suppression of the MKK6/p38/CEBPB signaling axis. Downstream, our data support that CEBPB positively regulates SLC7A11 expression and transactivates the SLC7A11 promoter in reporter assays; furthermore, CEBPB overexpression partially reversed the antiproliferative and ferroptosis-associated effects of doxorubicin. Importantly, these findings were supported in vivo, where doxorubicin treatment suppressed xenograft growth and was associated with ferroptosis-related biochemical and molecular alterations, including downregulation of SLC7A11 and GPX4 and inhibition of the MKK6/p38 pathway. Collectively, our data support a mechanistic link between doxorubicin, ferroptosis, and the MKK6/p38/CEBPB/SLC7A11 axis, providing new insights into the therapeutic potential of targeting ferroptosis in endometrial cancer. Importantly, the machine-learning screening was restricted to ferroptosis-related DEGs derived from tumor–normal comparisons, whereas MKK6 was identified in an orthogonal, drug-perturbation transcriptomic context as a signaling mediator of doxorubicin response; therefore, it was not expected to emerge from the disease-centric ferroptosis DEG feature space.Although our data support ferroptosis as an important component of doxorubicin-induced cytotoxicity in EC, we acknowledge that some readouts of oxidative injury (e.g., ROS accumulation and mitochondrial alterations) can also be observed in other regulated cell-death programs. Accordingly, ferroptosis inference here relies on convergent and functionally reversible hallmarks, including Fer-1–sensitive lipid peroxidation (BODIPY 581/589-C11 oxidation and MDA accumulation), iron dysregulation (increased Fe²⁺/Fe), depletion of antioxidant capacity (reduced GSH/GSSG), and coordinated suppression of the SLC7A11–GPX4 defense axis. In this framework, ROS elevation and mitochondrial changes are interpreted as supportive features rather than ferroptosis-specific determinants. Future studies using systematic pathway-specific inhibition and/or genetic approaches will further refine the relative contributions of ferroptosis versus parallel death or growth-arrest programs under doxorubicin stress.

Given the pleiotropic, stress-inducing properties of anthracyclines, we do not exclude additional upstream signaling effects of doxorubicin beyond MKK6. We focused on MKK6 because it emerged as a DOX-responsive node in drug-perturbation transcriptomics and was supported by consistent pathway readouts (reduced p-MKK6/p-p38 and CEBPB) together with direct target-engagement evidence (CETSA and SPR). Moreover, functional rescue by MKK6 reconstitution attenuated DOX-associated lipid peroxidation/ferroptosis phenotypes and partially restored proliferation, providing a rationale for prioritizing this axis while acknowledging potential parallel regulators.

Our findings do not contradict these DNA-centric mechanisms; rather, they suggest that ferroptotic lipid peroxidation represents an additional, complementary vulnerability engaged by doxorubicin in EC. Previous research has extensively characterized the mechanisms through which DOX exerts its antitumor effects, primarily through DNA damage and topoisomerase inhibition. DOX intercalates into DNA, disrupting transcription and replication processes, and inducing double-strand breaks (DSBs). These DNA lesions activate key DNA damage response pathways such as ATM/CHK2 and ATR/CHK1, which control cell cycle arrest and apoptosis [39]. This accumulation of DNA damage is further exacerbated by inhibition of DNA repair proteins like RAD51 and DNA-PK, highlighting the critical role of DNA damage in DOX’s cytotoxic effects [40]. Furthermore, DOX functions as a potent topoisomerase II (Top2) poison by stabilizing the Top2–DNA cleavage complex, preventing the religation of DNA strands, and thereby inducing DSBs [41]. This mechanism sets DOX apart from catalytic inhibitors, as it directly traps Top2 on the DNA, heightening the drug’s cytotoxic stress [42].

While these DNA-centric mechanisms have been well-established, our study introduces a novel perspective by identifying ferroptosis as an additional mode of action through which DOX exerts its antitumor effects in endometrial cancer. Unlike traditional DNA damage-driven mechanisms, ferroptosis involves lipid peroxidation and redox imbalance, processes that were robustly induced in vitro and were consistent with ferroptosis-related oxidative and lipid peroxidation stress in vivo following DOX treatment.

Consistent with the concept that ferroptosis is fundamentally driven by membrane lipid peroxidation, we further strengthened the evidence chain by incorporating the lipid peroxidation–specific probe BODIPY 581/589-C11. In both Ishikawa and RL-95-2 cells, DOX increased the oxidized/reduced BODIPY signal, and this increase was attenuated by ferrostatin-1, providing a more direct readout of ferroptotic lipid oxidation in addition to the biochemical (MDA, GSH/GSSG, Fe²⁺/Fe) and ultrastructural (TEM) changes observed. Importantly, restoration of MKK6 markedly reduced the DOX-induced BODIPY oxidation signal and concomitantly ameliorated multiple ferroptosis-associated biochemical alterations, supporting a functional role for the MKK6/p38 axis in shaping the lipid peroxidation state under DOX stress. The observation that CEBPB overexpression similarly decreased BODIPY oxidation is also consistent with our model in which MKK6/p38 signaling regulates the CEBPB–SLC7A11 antioxidant program, thereby modulating cellular susceptibility to lipid peroxidation and ferroptotic cell death. As a limitation, we did not systematically evaluate additional ferroptosis regulators such as ACSL4, ALOX15, and PTGS2, which will be addressed in future studies to further refine the ferroptosis regulatory landscape in endometrial cancer. This finding not only extends the understanding of DOX’s biological impact but also suggests that ferroptosis could be a key player in its therapeutic efficacy, particularly in cancers where other traditional mechanisms may not fully account for its effects.

Thus, while earlier studies have emphasized DNA damage and topoisomerase inhibition as the primary drivers of DOX-induced cell death, our research highlights the contribution of ferroptosis in endometrial cancer. This expands the mechanistic repertoire of DOX, offering new insights into its potential for inducing iron-dependent cell death, a pathway that has not been previously linked to the drug’s action in this context [43].

Ferroptosis has emerged as a therapy-relevant vulnerability across cancers, influencing treatment response and resistance through iron-dependent lipid peroxidation and redox failure [44–46]. In endometrial cancer, however, the extent to which conventional cytotoxic agents engage ferroptosis and the upstream signaling circuitry connecting drug exposure to ferroptotic execution have remained insufficiently defined. By demonstrating that doxorubicin activates a ferroptosis program in EC and mapping this response to suppression of the MKK6/p38/CEBPB–SLC7A11 axis, our study provides a focused mechanistic framework that may inform ferroptosis-oriented therapeutic optimization.

The MKK6/P38/CEBPB signaling cascade is well established as a central mediator of cellular stress and inflammatory responses, integrating signals from oxidative stress, DNA damage, and environmental challenges to regulate survival, apoptosis, and immune activation [47]. Activation of p38 MAPK by MKK6 can drive mitochondrial cytochrome c release and stress-induced apoptosis, while downstream CEBPB contributes to inflammatory regulation by modulating cytokine production and immune cell activation [48]. Dysregulation of this pathway has been implicated in diverse pathologies, including glioblastoma, where MKK6 stabilization supports p38 signaling and modulates tumor-associated inflammation, and colorectal cancer, where MKK6/P38/CEBPB promotes macrophage polarization and angiogenesis within a tumor-permissive microenvironment [49, 50]. Beyond oncology, alterations in this pathway contribute to cardiac hypertrophy and dysfunction, underscoring its broader role in stress-associated disease processes [51].

Building on these insights, the present study expands the functional scope of the MKK6/P38/CEBPB axis by demonstrating its involvement in regulating ferroptosis. By linking a classical stress- and inflammation-related signaling pathway to lipid peroxidation and iron-dependent cell death, our findings uncover a novel layer of mechanistic interplay that not only refines the understanding of doxorubicin’s action but also positions this pathway as a potential therapeutic target in ferroptosis-based interventions for endometrial cancer.

Doxorubicin induces ferroptosis through a multifaceted mechanism, starting with its direct binding to MKK6, a key protein in the stress response pathway. We propose a temporal model in which early inhibition of MKK6/p38 signaling precedes downstream transcriptional reprogramming (reduced CEBPB activity and SLC7A11 expression), leading to progressive impairment of GPX4-centered antioxidant capacity and culminating in lipid peroxide accumulation and ferroptosis. This binding is associated with suppression of the MKK6/P38/CEBPB signaling axis, a critical cascade involved in inflammation and cellular stress regulation. As a result, the downstream ferroptosis-related proteins SLC7A11 and GPX4, which are essential for maintaining redox balance and preventing lipid peroxidation, are downregulated. This suppression of SLC7A11 and GPX4 impairs the cell’s ability to manage oxidative stress, leading to an accumulation of reactive oxygen species (ROS) and increased lipid peroxidation, hallmarks of ferroptosis. Moreover, CEBPB, as a transcriptional regulator, typically upregulates SLC7A11, thereby protecting against ferroptosis. However, DOX’s inhibition of CEBPB expression further exacerbates ferroptosis by limiting SLC7A11 expression. To validate this pathway, rescue experiments using ferrostatin-1 (Fer-1) and CEBPB overexpression effectively reversed the effects of DOX, demonstrating the central role of ferroptosis in mediating DOX-induced cell death. Collectively, these findings reveal a novel mechanism by which DOX induces ferroptosis, positioning the MKK6/P38/CEBPB pathway as a crucial regulator of this form of cell death in endometrial cancer.

The findings of this study carry important clinical and translational implications. Our data suggest that the therapeutic efficacy of doxorubicin in endometrial cancer may, at least in part, be mediated through ferroptosis, providing a mechanistic basis for its antitumor activity beyond the classical DNA damage and topoisomerase inhibition pathways. This highlights ferroptosis as a potential determinant of treatment response, with clinical utility as a predictive marker of doxorubicin sensitivity. In particular, SLC1A5 expression emerged as a promising biomarker, as its levels correlated with doxorubicin responsiveness, suggesting that stratification of patients based on SLC1A5 status could improve therapeutic precision and patient outcomes. Furthermore, the identification of the MKK6/CEBPB signaling cascade as a critical regulator of ferroptosis opens new avenues for drug development. Targeting this pathway may enhance the pro-ferroptotic effects of chemotherapy or overcome resistance, thereby providing novel therapeutic strategies in endometrial cancer and potentially other malignancies. Together, these insights underscore the value of integrating ferroptosis biology into clinical oncology, both for biomarker-guided therapy and for the development of innovative treatment approaches. With respect to translational relevance, the doxorubicin concentrations used in vitro were selected around the IC50 range in each EC cell line (approximately 1–2 µM) to enable mechanistic interrogation under biologically active exposure conditions. Clinically, systemic doxorubicin exposure is dynamic and influenced by dosing schedules, protein binding, and tissue distribution, and intratumoral drug levels may not directly mirror nominal plasma concentrations. Therefore, while our findings support a mechanistic link between doxorubicin and ferroptosis signaling in EC cells, further pharmacokinetic/pharmacodynamic–informed studies in clinically relevant models will be important to refine dose–exposure relationships in vivo. Nonetheless, because our current models do not incorporate an intact immune–stromal microenvironment, further studies in clinically relevant systems will be needed to define how stromal and immune modulation influences doxorubicin-associated ferroptosis and to refine translational extrapolation.

This study has several limitations. First, our machine-learning feature-selection workflow was not externally validated at the model-performance level, as no additional independent cohort beyond TCGA-UCEC (used here to validate expression patterns) was included; thus, the generalizability of selected features requires confirmation in further datasets. Second, although supported by in vitro and in vivo experiments, validation in clinical specimens was limited, and additional patient-derived tissue analyses are needed to strengthen translational relevance. Third, the in vivo evidence was obtained from subcutaneous xenografts in immunodeficient nude mice, which may underestimate interactions between ferroptosis and antitumor immunity, leaving immune-mediated contributions to doxorubicin’s effects incompletely defined. The experimental systems used here (established EC cell lines and subcutaneous xenografts in immunodeficient nude mice) do not capture an intact immune–stromal microenvironment, including fibroblasts, endothelial cells, and immune subsets, all of which can modulate ferroptosis sensitivity. Accordingly, our findings primarily reflect tumor cell–intrinsic responses to doxorubicin, and validation in immunocompetent/humanized models and clinical specimens will be important to define microenvironmental contributions. Fourth, we did not perform ferroptosis-inhibitor rescue experiments (e.g., ferrostatin-1) in xenograft models; therefore, although tumor lysate readouts (lipid peroxidation, iron dysregulation, redox imbalance, and GPX4/SLC7A11 suppression) were consistent with ferroptosis-related changes after DOX treatment, definitive causal attribution of tumor growth inhibition to ferroptosis in vivo will require inhibitor-based rescue and/or genetic approaches. Accordingly, our in vivo data should be considered supportive and hypothesis-generating, and larger sample sizes will be needed for definitive quantification. Finally, while motif prediction and luciferase assays support CEBPB-dependent transactivation of the SLC7A11 promoter, we did not perform chromatin-level validation (e.g., ChIP-qPCR, CUT&RUN/CUT&Tag) to demonstrate endogenous CEBPB occupancy at the native SLC7A11 locus in EC cells; thus, direct chromatin binding remains to be confirmed.

Future work should validate these findings in clinical specimens to establish their translational relevance and assess whether SLC1A5, MKK6, and CEBPB can serve as predictive biomarkers in patient populations. In parallel, studies using immunocompetent or humanized models will be necessary to capture the contribution of immune-mediated mechanisms that. In particular, in vivo ferroptosis-inhibitor rescue experiments (e.g., ferrostatin-1) will be important to establish the causal contribution of ferroptosis to DOX-mediated tumor suppression.

Conclusion

This study demonstrates for the first time that doxorubicin exerts its antitumor effects in endometrial cancer not only through its classical actions of DNA damage and topoisomerase inhibition, but also by inducing ferroptosis through suppression of the MKK6/P38/CEBPB signaling pathway. By disrupting redox homeostasis and enhancing lipid peroxidation, doxorubicin positions ferroptosis as a critical component of its therapeutic activity. These insights suggest that ferroptosis-related markers could serve as predictors of drug sensitivity and identify the MKK6/CEBPB axis as a promising therapeutic target, thereby providing a rationale for refining doxorubicin-based regimens and advancing ferroptosis-centered strategies to improve treatment outcomes in endometrial cancer.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Xiaoqin Lu and Jingyan Zhang responsible for designing the study, writing the manuscript, extracting and analyzing data, and interpreting the results. Jingyan Zhang, Yanfang Li, Zhenhui Wang and Panpan Zhao were responsible for the experiments. Yanfang Li, Zhenhui Wang, Panpan Zhao and Dan Ren contributed to reviewing and editing the article. All authors reviewed and approved the final manuscript.

Funding

This research received no external funding.

Data availability

The datasets used or analyzed during the current study are available from the GEO database (https://www.ncbi.nlm.nih.gov/gds/), the NCBI Gene Database (https://www.ncbi.nlm.nih.gov/gene), and the TCGA GDC website (https://portal.gdc.cancer.gov/).

Declarations

Ethics approval and consent to participate

All animal experiments were approved by the Ethics Committee of the Second Affiliated Hospital of Zhengzhou University (Approval No. KY2024200) and were conducted in accordance with the institutional guidelines and the ARRIVE guidelines. No human participants or human-derived samples were included in this study; therefore, informed consent was not required.

Consent to participate

Not applicable.

Consent for publication

All authors have read and approved the final manuscript and consent to its publication.

Competing interests

The authors declare no conflict of interest.