Single‐Cell Transcriptomic Analysis Reveals Ferroptosis‐Related Signatures and Tumor Microenvironment Heterogeneity in Epithelial Ovarian Cancer Authors: Jie Liu, Jing Li, Bingxin Li, Hui Liang Published: 2026 ABSTRACT ABSTRACTEpithelial ovarian cancer (EOC) is the deadliest gynecological malignancy. Ferroptosis, an iron‐dependent lipid peroxidation‐associated cell death mode, provides a new target for tumor therapy. Nevertheless, its impact on EOC tumorigenesis is not well defined. The quality control, dimensionality reduction, and EOC clustering for the gene expression data were performed through the Scanpy pipeline for single‐cell RNA sequence from GEO (GSE165897). Ferroptosis pathway scoring, computation of transcription factor activity, and analyzes for cell–cell communication were conducted. An independent validation cohort (GSE300897, n = 19,002 cells) was analyzed to confirm ferroptosis signature robustness. Expression of ferroptosis‐related genes in SKOV3 and OVCAR3 cells was confirmed using RT‐qPCR and immunoassays after treatment with erastin, RSL3, and ferrostatin‐1 rescue. NRF2 siRNA knockdown and co‐culture experiments were performed for functional validation. A total of 51,702 cells from 11 patients were analyzed and clustered into 18 clusters that included EOC cells (16.9%), immune cells (67.6%), and stromal cells (15.6%), encompassing 27 fine‐grained cell subtypes. KEGG enrichment revealed that ferroptosis, fatty acid metabolism, and glutathione metabolism were the major pathways. Transcription factor profiling indicated increased NRF2 activation in EOC cells, which implied an augmented ferroptosis‐resistance capacity. Signaling between cells also revealed three key ligand‐receptor interactions, SPP1‐CD44, VEGFA‐KDR, and CCL2‐CCR2, which are involved in the regulation of ferroptosis and were functionally validated in a triple co‐culture model. NRF2 siRNA knockdown synergistically enhanced erastin‐induced ferroptosis sensitization (C11‐BODIPY: 4.52‐fold), rescued by ferrostatin‐1. Platinum‐refractory patients in GSE300897 showed significantly altered ferroptosis scores (Mann–Whitney p = 3.22e‐24), linking ferroptosis to clinical platinum sensitivity. Our work uncovers ferroptosis dysregulation as a key driver of EOC pathogenesis, with NRF2 being a master regulator of resistance to ferroptosis and suggesting new therapeutic approaches for ovarian cancer. Introduction Epithelial ovarian cancer (EOC) is the most fatal gynecologic malignancy in the world, making up approximately 90% of all the cases of ovarian carcinoma and ranking as the fifth leading cause of cancer deaths in women [1]. Even so, the number of deaths has been declining at only about 1% each year [2], and the 5‐year survival rate for advanced‐stage EOC is still less than 30%, primarily because it tends to present as an advanced disease and to acquire resistance to chemotherapy [2, 3]. To date, the majority of patients are diagnosed in an advanced disease stage with peritoneal seeding and, although initial response to platinum‐based chemotherapy is quite good, a relapse usually occurs, and most patients will develop drug‐resistant disease [4]. The tumor microenvironment (TME) of ovarian cancer is a complex network that consists of stromal cells, immune cells, extracellular matrix ingredients, and cancer cells which contribute to the regulation of tumor progression, metastasis, and response to therapies [5]. Immune cells in the TME such as tumor‐associated macrophages, T lymphocytes, and NK cells possess opposing effects on tumorigenesis based on their polarization states that influence interaction with cancerous cells [6, 7]. Stromal cells in the tumor, including cancer‐associated fibroblasts, also provide an immunosuppressive microenvironment and create conditions supporting the survival of tumors, and they do so through metabolic adaptation as well as by releasing pro‐tumorigenic signals [8]. Cellular heterogeneity and intercellular communications in the OvCa TME are critical for the design of more effective therapeutic interventions. Ferroptosis is a newly identified form of RCD with features characterized by iron‐dependent lipid peroxidation; it is independent of apoptosis, necrosis, and autophagy [9]. Implementation of ferroptosis is caused by the persistence of lipid ROS generation beyond cellular antioxidant capability. The central negative regulator of ferroptosis is glutathione peroxidase 4 (GPX4), which reduces lipid hydroperoxides to non‐toxic lipid alcohols [10]. Meanwhile, cystine is supplied by the cystine/glutamate antiporter system Xc‐ (SXC) gene SLC7A11 to sustain glutathione synthesis for GPX4 enzyme function and ferroptosis inhibition [11]. Other regulators are FTH1 for iron storage, NCOA4 for ferritinophagy, TFRC for iron uptake, and ACSL4 to favor lipid peroxidation [12]. Recently, available data started to indicate that ferroptosis exerts both cancer‐suppressing roles by clearing damaged cells and protumorigenic functions in favoring the survival of ferroptosis‐resistant cell populations as well [13, 14]. In ovarian cancer, multiple studies confirmed the chemoresistance and disease progression significance by which cancer cells developed resistance to ferroptosis induction via upregulation of antioxidant systems and perturbed iron metabolism [15]. The transcription factor NRF2 (NFE2L2) is a known master regulator of ferroptosis resistance via induction of antioxidant and iron metabolism modulatory factors at the transcriptional level [16]. Some reports also indicate that the tumor microenvironment can regulate the sensitivity of ferroptosis via paracrine signaling, metabolic competitive effects, and immune cells [17]. Single cell RNA sequencing (scRNA‐seq) has been established as an invaluable tool for characterizing cellular subpopulations within complex tissues at high resolution [18]. By analyzing cellular transcriptomes at the single‐cell level, scRNA‐seq allows discovery of rare cell populations, transition pathways of cell states, and cell–cell communication networks that might be related to disease pathogenesis. The application of scRNA‐seq to OC tissues provides the possibility of systematically outlining the cellular landscape and revealing disease‐causing mechanisms in pathways related to ferroptosis [19]. In the present investigation, single cell transcriptomic data from the GEO database (GSE165897, n = 51,702 cells) was analyzed by integrated bioinformatics to thoroughly describe ferroptosis‐associated signature genes and tumor microenvironment constitution in EOC patients. An independent validation cohort (GSE300897, n = 19,002 cells) was analyzed to confirm robustness and assess clinical platinum sensitivity association. Specifically, we sought to: (1) determine cell type‐specific expression of ferroptosis‐related genes at single‐cell resolution; (2) characterize the composition and heterogeneity of immune and stromal cells across 27 subtypes; (3) identify the transcription factor network orchestrating ferroptosis resistance; (4) functionally validate paracrine signaling axes (SPP1‐CD44, CCL2‐CCR2, VEGFA‐KDR) and NRF2‐mediated ferroptosis regulation using co‐culture and siRNA knockdown experiments; and (5) correlate ferroptosis pathway activity with clinical platinum sensitivity in an independent cohort. Our results highlight coordinated disruption of ferroptosis and tumor microenvironment dysregulation as signature hallmarks in EOC. Methods Data Acquisition and Processing Single‐cell RNA sequencing data were obtained from the Gene Expression Omnibus (GEO accession: GSE165897) [18, 19]. The dataset contains longitudinal scRNA‐seq data from 11 HGSOC patients with paired primary (treatment‐naive) and interval (post‐neoadjuvant chemotherapy) samples, totaling 22 samples across three sequencing platforms (HiSeq 2500, HiSeq 4000, NovaSeq 6000). Quality control parameters were set as follows: cells with fewer than 200 detected genes, more than 8000 genes, or mitochondrial gene content exceeding 20% were removed. Genes expressed in fewer than 3 cells were excluded. After QC, 51,702 cells and 32,846 genes were retained for downstream analysis. The final dataset comprised 51,702 cells with a median of 1303 genes and 4087 UMIs per cell. An independent validation dataset, GSE300897 (19,002 cells from 9 treatment‐naive HGSOC patients), was processed using the identical pipeline for ferroptosis signature validation and platinum sensitivity association analysis. Normalization and Dimensionality Reduction Raw UMI counts were normalized to 10,000 counts per cell followed by log1p transformation. The top 2000 highly variable genes were identified using the Seurat v3 method. Mitochondrial gene percentage was regressed out. Principal component analysis was performed with 50 PCs. Elbow analysis identified PC4 as the optimal inflection point; 15 PCs were used for downstream analysis. Uniform Manifold Approximation and Projection (UMAP, min_dist = 0.3, spread = 1.0) and t‐distributed Stochastic Neighbor Embedding (t‐SNE, perplexity = 30) were used for two‐dimensional visualization. Clustering and Cell Type Annotation Leiden community detection (resolution = 0.8, n_neighbors = 15) identified 18 distinct transcriptional clusters. Cell type annotation was performed using canonical marker gene expression patterns. For enhanced resolution of tumor microenvironment heterogeneity, fine‐grained annotation identified 27 cell subtypes including T‐cells (20664), Macrophages (6917), B‐cells, NK cells, Mast cells, Dendritic cells (DC‐1, DC‐2), Plasma cells, ILC, pDC, multiple CAF subtypes (CAF‐1, CAF‐2, CAF‐3), Mesothelial cells, and Endothelial cells. Ferroptosis Pathway Scoring Gene sets for ferroptosis were curated from the FerrDb database (http://www.zhounan.org/ferrdb) and published literature [9, 10, 12, 14]. Four functional modules were defined: Ferroptosis Drivers (ACSL4, LPCAT3, ALOX5, ALOX12, ALOX15, TFRC, NCOA4, SAT1, ATG5, ATG7), Ferroptosis Suppressors (GPX4, SLC7A11, FTH1, FTL, HMOX1, NFE2L2, SLC40A1, HSPB1, CBS, GCH1), Iron Metabolism, and Lipid Peroxidation [12, 15]. Module scores were calculated using the Scanpy score_genes function. A ferroptosis sensitivity landscape was constructed by plotting driver scores against suppressor scores, classifying cells into Sensitive (high driver, low suppressor), Resistant (low driver, high suppressor), and Neutral categories based on median thresholds. Transcription Factor Activity Analysis Transcription factor activity was inferred using curated TF‐target gene relationships [16]. Key transcription factors including NFE2L2 (encoding the NRF2 protein, a master regulator of antioxidant response), ATF4, TP53, HIF1A, STAT3, MYC, JUN, FOS, and SP1 were analyzed. NRF2/NFE2L2 are used interchangeably in this manuscript: NFE2L2 refers to the gene, while NRF2 refers to the encoded transcription factor protein [16]. Cell–Cell Communication Analysis Intercellular communication was inferred from ligand‐receptor pair co‐expression [17]. Putative ligand‐receptor interactions were curated from CellChatDB (v1.6.0) and CellPhoneDB (v5.0) databases. Fifteen ligand‐receptor pairs were analyzed, including SPP1‐CD44, VEGFA‐KDR, CCL2‐CCR2, CXCL12‐CXCR4, IL6‐IL6R, TGFB1‐TGFBR1, VEGFA‐FLT1, CCL5‐CCR5, IL1B‐IL1R1, TNF‐TNFRSF1A, EGF‐EGFR, HGF‐MET, FGF2‐FGFR1, WNT5A‐FZD7, and COL1A1‐ITGB1. Communication scores were calculated as the product of mean ligand expression in sender cells and mean receptor expression in receiver cells. Differential Expression and Pathway Enrichment Differential gene expression analysis was performed using the Wilcoxon rank‐sum test. Comparisons included EOC versus Immune cells, EOC versus Stromal cells, and treatment‐naive versus post‐NACT samples within each cell type. Multiple testing correction was performed using the Benjamini‐Hochberg method (adjusted p‐value < 0.05). Significantly differentially expressed genes were subjected to KEGG pathway and Gene Ontology Biological Process enrichment analysis using the Enrichr API via gseapy. Diffusion pseudotime analysis was performed on EOC cells using the Scanpy diffusion map and DPT implementations, with the root cell selected as the cell with the lowest ferroptosis score. In Vitro Experimental Validation Human ovarian adenocarcinoma cell lines SKOV3 (ATCC HTB‐77) and OVCAR3 (ATCC HTB‐161) were used in this study. Cell lines were authenticated by an expert before being used for experimentation. All cell lines were tested and confirmed to be free from mycoplasma contamination. Cells were cultured under cell‐line‐specific conditions. SKOV3 cells were maintained in McCoy's 5A medium supplemented with 10% FBS, whereas OVCAR3 cells were cultured in RPMI‐1640 supplemented with 20% FBS and insulin. To assess ferroptosis induction, cells were treated with: (1) erastin (10 uM, 24 h), a system xc‐ inhibitor; (2) RSL3 (1 uM, 24 h), a direct GPX4 inhibitor; and (3) erastin or RSL3 co‐administered with ferrostatin‐1 (1 uM) as rescue controls. Vehicle controls with matched final DMSO concentrations were included. RT‐qPCR was performed using SYBR Green Master Mix, with validated reference genes used for normalization. In addition to core ferroptosis‐related genes (GPX4, SLC7A11, ACSL4, TFRC, and FTH1), NRF2 pathway‐related genes (NFE2L2, NQO1, and GCLM) were assessed to evaluate NRF2‐associated ferroptosis resistance. Protein expression was quantified using commercial immunoassays validated for human cell lysates, with catalog numbers confirmed from the manufacturers' product manuals. Cell lysates were normalized by total protein concentration using a BCA assay (Pierce 23225) and analyzed in duplicate. Seven‐point standard curves generated by two‐fold serial dilution were included on each plate and fitted using a four‐parameter logistic model. Quality‐control acceptance criteria were intra‐assay CV < 10%, inter‐assay CV < 15%, and spike‐in recovery of 85%–115%. Absorbance was measured at 450 nm using a BioTek Synergy H1 microplate reader. Because SLC7A11 is a multi‐pass transmembrane protein whose recovery may depend on membrane‐protein extraction conditions, its expression was evaluated using an antibody‐based method validated for SLC7A11. Co‐Culture Model and NRF2 siRNA Knockdown A triple co‐culture system was established comprising EOC cells (SKOV3), THP‐1‐derived macrophages (differentiated with 100 ng/mL PMA for 48 h), and primary CAFs isolated from surgically resected tumor tissue. Co‐cultures were maintained for 48 h in RPMI‐1640 with 10% FBS. For NRF2 loss‐of‐function experiments, SKOV3 and OVCAR3 cells were transfected with NRF2‐specific siRNA (siNRF2, 50 nM) or non‐targeting control siRNA (siCtrl) using Lipofectamine RNAiMAX. Knockdown efficiency was confirmed by RT‐qPCR at 48 h post‐transfection. Cells were subsequently treated with erastin (10 uM, 24 h) or DMSO control, and lipid peroxidation was assessed using C11‐BODIPY 581/591 oxidation assay. Results Quality Control and Single‐Cell Transcriptome Overview After stringent quality control, 51,702 cells from 11 patients (22 samples) were retained for downstream analysis (Figure 1). The distribution of detected genes per cell showed a median of 1303 genes (mean: 1862). GPX4‐ACSL4 co‐expression analysis (Figure 1A) revealed coordinated regulation of ferroptosis‐related pathways. Functional category analysis of differentially expressed genes (Figure 1B) between EOC and Immune compartments showed distinct patterns across ferroptosis, immune, ECM/stromal, epithelial/tumor, metabolism, and signaling gene categories. Ferroptosis score distribution across cell types (Figure 1C) showed distinct patterns among EOC, Immune, and Stromal populations. Patient‐specific cell type composition analysis (Figure 1D) demonstrated inter‐patient heterogeneity. Cell cycle phase distribution (Figure 1E) and cell type proportion analysis (Figure 1F) revealed the overall composition: EOC cells (8723, 16.9%), Immune cells (34,935, 67.6%), and Stromal cells (8044, 15.6%). Quality control and single‐cell transcriptome overview. (A) GPX4‐ACSL4 co‐expression hexbin density plot with Pearson correlation. (B) DEGs grouped by functional category (Ferroptosis, Immune, ECM/Stromal, Epithelial/Tumor, Metabolism, Signaling). Red: Up in EOC; Blue: Up in Immune. (C) Ferroptosis score ridge distribution across EOC, Immune, and Stromal cell types. (D) Patient cell type composition stacked bar chart. (E) Cell cycle phase polar bar chart (G1/S/G2M‐like classification by gene count). (F) Cell type and treatment phase proportion horizontal stacked bar charts. Cell Clustering and NRF2 Regulatory Network Analysis UMAP visualization identified 18 distinct transcriptional clusters (Figure 2A). The NRF2 regulatory network analysis (Figure 2B) revealed NRF2 (encoded by NFE2L2) as a central hub connecting FTH1, GPX4, SLC7A11, HMOX1, NQO1, and GCLM, indicating NRF2‐mediated transcriptional control of ferroptosis resistance. Quality control metrics (Figure 2C) confirmed that the majority of cells passed QC thresholds. Ferroptosis gene expression radar analysis (Figure 2D) demonstrated heterogeneous patterns across cell types. Comparison between high and low ferroptosis activity groups (Figure 2E) revealed differential expression of GPX4, SLC7A11, ACSL4, and TFRC. Treatment response analysis (Figure 2F) demonstrated altered expression of ferroptosis genes between treatment‐naive and post‐NACT EOC samples, with GPX4 and SLC7A11 showing decreased expression following chemotherapy, while ACSL4 and TFRC showed increased expression, suggesting chemotherapy‐induced ferroptosis sensitization. Cell clustering and NRF2 regulatory network analysis. (A) UMAP clustering visualization showing 18 clusters. (B) NRF2 regulatory network with FTH1, GPX4, SLC7A11, HMOX1, NQO1, and GCLM targets. (C) UMI count versus mitochondrial percentage scatter density plot with QC thresholds. (D) Polar radar chart of ferroptosis gene expression across cell types. (E) Violin plots comparing ferroptosis gene expression between high and low ferroptosis activity groups. (F) Treatment response comparison of ferroptosis gene expression between treatment‐naive and post‐NACT EOC samples. Pathway Interaction Network and Transcription Factor Activity Pathway interaction network analysis (Figure 3A) positioned ferroptosis as a central hub connecting apoptosis, autophagy, oxidative stress, lipid metabolism, and inflammation pathways. Ferroptosis score analysis by cell type (Figure 3B) revealed distinct distributions among EOC, Immune, and Stromal populations. Transcription factor activity analysis (Figure 3C) demonstrated elevated NRF2/NFE2L2 activity in EOC cells, consistent with enhanced ferroptosis resistance. Stromal cells showed distinct activation of HIF1A and ATF4. Sample distribution (Figure 3D) and top DEG analysis (Figure 3E) identified key EOC‐associated genes. GPX4 expression dynamics (Figure 3F) revealed cluster‐specific patterns. Pathway interaction network and transcription factor activity. (A) Pathway interaction network with ferroptosis as central hub. (B) Ferroptosis score beeswarm plot by cell type. (C) Transcription factor activity heatmap (NFE2L2, ATF4, TP53, HIF1A, STAT3, MYC, JUN, FOS, SP1). (D) Sample cell distribution lollipop chart. (E) Top differentially expressed genes lollipop chart (EOC vs. Immune). (F) GPX4 expression streamgraph across clusters. Cell–Cell Communication and TME Interactions RNA velocity field analysis (Figure 4A) indicated directional cell state transitions toward the EOC population. Sample‐cluster association analysis (Figure 4B) demonstrated relationships between patient samples and transcriptional clusters. Cell–cell communication network analysis (Figure 4C) identified strong interactions among EOC, Immune, and Stromal populations. Ligand‐receptor interaction analysis (Figure 4D) identified significant communication pathways, with SPP1‐CD44, VEGFA‐KDR, CCL2‐CCR2, CXCL12‐CXCR4, IL6‐IL6R, and TGFB1‐TGFBR1 being the most prominent pairs. Lipid peroxidation gene expression (Figure 4E) showed cluster‐specific patterns. Ferroptosis gene parallel coordinate analysis (Figure 4F) revealed coordinated expression across cell populations. Cell–cell communication and TME interactions. (A) RNA velocity field visualization on UMAP. (B) Sample‐cluster polar radial dot plot. (C) Cell–cell communication network among EOC, immune, and stromal populations. (D) Ligand‐receptor pairing diagram: SPP1‐CD44, VEGFA‐KDR, CCL2‐CCR2, CXCL12‐CXCR4, IL6‐IL6R, TGFB1‐TGFBR1. (E) Lipid peroxidation gene stacked area plot across clusters. (F) Ferroptosis gene parallel coordinates plot. Dimensionality Reduction and Ferroptosis Pathway Enrichment PCA analysis (Figure 5A) showed that PC1 through PC15 captured the majority of transcriptional variance. Voronoi spatial tessellation (Figure 5B) illustrated the spatial relationships among cellular populations. Gene detection distribution (Figure 5C) showed a median of 1303 genes per cell. Marker gene expression matrix (Figure 5D) displayed cluster‐specific patterns of ferroptosis‐related genes. GSEA analysis (Figure 5E) confirmed significant enrichment of the ferroptosis gene set. SLC7A11 expression (Figure 5F) demonstrated cluster‐dependent variation. PCA analysis and ferroptosis gene expression. (A) PCA variance explained polar area chart. (B) Voronoi spatial tessellation with KDE contour overlay. (C) Gene detection density distribution with median and mean annotations. (D) Ferroptosis marker gene bubble matrix across clusters. (E) GSEA enrichment analysis of ferroptosis gene set (3‐layer: ES curve, hit marks, ranked metric). (F) SLC7A11 expression beeswarm plot across clusters. Gene Expression Patterns and KEGG Pathway Enrichment Gene expression cascade analysis (Figure 6A) revealed cluster‐specific gene expression patterns. KEGG pathway enrichment (Figure 6B) identified significantly enriched pathways including ferroptosis, fatty acid metabolism, glutathione metabolism, HIF‐1 signaling, oxidative phosphorylation, PI3K‐Akt signaling, MAPK signaling, and p53 signaling. Cluster distribution (Figure 6C) and cell type waffle chart (Figure 6D) summarized the composition: 34,935 Immune (67.6%), 8723 EOC (16.9%), and 8044 Stromal (15.6%). Pseudotime analysis (Figure 6E) revealed developmental trajectories. PCA elbow analysis (Figure 6F) confirmed PC4 as the optimal number of components. Gene expression patterns and KEGG pathway enrichment. (A) Cluster‐level gene expression heatmap of key genes. (B) KEGG pathway enrichment Cleveland dot plot. (C) Cluster distribution semi‐circle donut chart. (D) Cell type waffle chart: Immune (34,935), EOC (8723), Stromal (8044). (E) Pseudotime cell type flow showing temporal distribution patterns. (F) PCA elbow scree plot with PC4 as the elbow point. GO Biological Process Enrichment and Ferroptosis Sensitivity GO biological process enrichment (Figure 7A) identified significantly enriched terms including oxidative stress response, iron ion homeostasis, lipid metabolic process, ferroptosis regulation, glutathione metabolism, and ROS metabolic process. Gene‐pathway connection analysis (Figure 7B) illustrated relationships between ferroptosis genes and enriched pathways. Ferroptosis sensitivity landscape (Figure 7C) revealed three distinct populations: Sensitive (8863 cells, 17.1%), Neutral (33,976 cells, 65.7%), and Resistant (8863 cells, 17.1%). MA plot (Figure 7D), cell density contour (Figure 7E), and antioxidant gene expression radar (Figure 7F) provided complementary perspectives. GO enrichment and ferroptosis sensitivity analysis. (A) GO biological process enrichment lollipop chart. (B) Gene‐pathway connection arc diagram. (C) Ferroptosis sensitivity landscape hexbin plot: Sensitive (8863), Neutral (33,976), Resistant (8863). (D) MA plot with hexbin density visualization. (E) Cell density contour plot in UMAP space. (F) Antioxidant gene expression polar radar chart across cell types. Treatment Response and Patient Heterogeneity Treatment phase comparison (Figure 8A) between treatment‐naive (n = 21,699) and post‐NACT (n = 30,003) samples showed altered cell type proportions. t‐SNE visualization (Figure 8B) confirmed clear separation of cell populations with KDE contour overlay. Anatomical location analysis (Figure 8C) revealed samples primarily from Omentum (31,629 cells) and Peritoneum (16,287 cells). Iron metabolism gene expression (Figure 8D) demonstrated elevated FTH1 and FTL in EOC cells, suggesting enhanced iron storage capacity. QC metrics (Figure 8E) and patient heterogeneity analysis (Figure 8F) confirmed inter‐patient variability in ferroptosis gene expression profiles. Treatment response and patient heterogeneity. (A) Treatment phase comparison of cell type proportions (treatment‐naive vs. post‐NACT). (B) t‐SNE visualization of cell types with KDE contour overlay. (C) Polar bar chart of anatomical location distribution: Omentum (31,629), Peritoneum (16,287), Mesentery (2501), Tumor (1285). (D) Iron metabolism gene expression line plot across cell types. (E) Quality control metrics ridge plot across cell types. (F) Patient heterogeneity clustergram with hierarchical clustering. RT‐qPCR Validation of Ferroptosis‐Associated Genes RT‐qPCR analysis validated the single‐cell transcriptomic findings in ovarian cancer cell lines (Figure 9). In SKOV3 cells (Figure 9A), erastin treatment (10 uM, 24 h) significantly downregulated GPX4 (0.38‐fold, p < 0.001), SLC7A11 (0.42‐fold, p < 0.001), and FTH1 (0.51‐fold, p < 0.01), while upregulating ACSL4 (2.87‐fold, p < 0.001) and TFRC (2.45‐fold, p < 0.001). RSL3 treatment (1 uM) produced more pronounced effects: GPX4 (0.32‐fold), SLC7A11 (0.38‐fold), ACSL4 (3.12‐fold), and TFRC (2.78‐fold), confirming GPX4‐dependent ferroptosis induction. Ferrostatin‐1 co‐treatment partially rescued these changes (GPX4: 0.88‐fold, ACSL4: 1.25‐fold), confirming ferroptosis specificity. NRF2 target genes (NFE2L2, NQO1, GCLM) were also significantly modulated, supporting NRF2 pathway involvement in ferroptosis resistance. OVCAR3 cells (Figure 9B) displayed consistent patterns. Cross‐cell‐line comparison (Figure 9C) and correlation analysis (Figure 9D) confirmed high concordance between SKOV3 and OVCAR3 responses. RT‐qPCR validation of ferroptosis‐associated genes in ovarian cancer cells. (A) Relative mRNA expression in SKOV3 cells: Control, Erastin (10 uM, 24 h), RSL3 (1 uM, 24 h). Genes: GPX4, SLC7A11, ACSL4, TFRC, FTH1, NFE2L2, NQO1, GCLM. (B) Relative mRNA expression in OVCAR3 cells under the same treatment conditions. (C) Comparative fold change analysis across cell lines and treatments including ferrostatin‐1 rescue. (D) Correlation analysis between SKOV3 and OVCAR3 fold changes. Data presented as mean ± SD (n = 3). ***p < 0.001, **p < 0.01, *p < 0.05. Protein‐Level Validation of Ferroptosis‐Related Expression Immunoassay analysis (Figure 10) validated RT‐qPCR results at the protein level with all QC metrics within acceptance criteria (intra‐assay CV 4.2%–8.7%, inter‐assay CV 6.5%–13.2%, spike‐in recovery 89%–108%). In SKOV3 cells (Figure 10A), erastin treatment decreased GPX4 (0.42‐fold, p < 0.001), SLC7A11 (0.45‐fold, p < 0.001), FTH1 (0.55‐fold, p < 0.01), and NFE2L2 (0.58‐fold, p < 0.01) protein levels, while increasing ACSL4 (2.65‐fold, p < 0.001) and TFRC (2.28‐fold, p < 0.001). RSL3 produced comparable protein‐level changes with greater magnitude. OVCAR3 cells showed similar trends (Figure 10B). Comparative analysis (Figure 10C) and mRNA‐protein correlation (Figure 10D, r = 0.996, p < 0.001) supported transcriptional regulation as the primary control mechanism. For SLC7A11, a multi‐pass transmembrane protein whose detection may depend on membrane‐protein extraction efficiency, expression was verified using a validated antibody‐based method, showing results consistent with the overall ferroptosis induction pattern. Protein‐level validation of ferroptosis‐related expression. (A) Relative protein levels in SKOV3 cells: Control, Erastin (10 uM, 24 h), RSL3 (1 uM, 24 h). (B) Relative protein levels in OVCAR3 cells: Control, Erastin, Erastin+Ferrostatin‐1 rescue. (C) Comparative protein expression across cell lines and treatments. (D) mRNA‐protein correlation analysis. Data presented as mean ± SD (n = 3). Immunoassays with 7‐point standard curves (4‐parameter fit), intra‐assay CV < 10%, inter‐assay CV < 15%. For SLC7A11, validated antibody‐based method. Co‐Culture Model and NRF2 Functional Validation To functionally validate the paracrine signaling axes identified by scRNA‐seq, we established a triple co‐culture system comprising EOC cells (SKOV3), THP‐1‐derived macrophages, and primary CAFs (Figure 11). In co‐culture conditions, CAFs exhibited significantly elevated SPP1 expression (3.45‐fold, p < 0.001 vs. mono‐culture), which was further enhanced in triple co‐culture with macrophages (4.12‐fold, p < 0.001; Figure 11A). Macrophages in triple co‐culture showed upregulation of CD44 (3.52‐fold), CCR2 (2.68‐fold), and KDR (2.34‐fold) (Figure 11B), confirming activation of the SPP1‐CD44, CCL2‐CCR2, and VEGFA‐KDR signaling axes. EOC cells under triple co‐culture conditions displayed enhanced ferroptosis gene dysregulation: GPX4 (0.55‐fold, p < 0.001) and SLC7A11 (0.48‐fold, p < 0.001) were significantly downregulated compared to mono‐culture, while ACSL4 (2.35‐fold) and TFRC (2.12‐fold) increased (Figure 11C). Co‐culture model and NRF2 functional validation. (A) SPP1 mRNA in CAFs: Mono‐culture, EOC co‐culture, triple co‐culture (CAFs+EOC + Macrophage). (B) Receptor expression (CD44, CCR2, KDR) in THP‐1‐derived macrophages under co‐culture conditions. (C) Ferroptosis gene expression (GPX4, SLC7A11, ACSL4, TFRC) in EOC cells across culture conditions. (D) NRF2 siRNA knockdown efficiency: NFE2L2 mRNA in SKOV3 and OVCAR3 (siCtrl vs. siNRF2). (E) Ferroptosis gene expression after NRF2 KD combined with erastin (siCtrl+Era vs. siNRF2 + Era). (F) C11‐BODIPY lipid peroxidation: SiCtrl+DMSO, siCtrl+Era, siNRF2 + Era, siNRF2 + Era+Fer‐1 rescue. Data: Mean ± SD (n = 3). To establish the causal role of NRF2 in ferroptosis resistance, we performed NRF2 siRNA knockdown in SKOV3 and OVCAR3 cells (Figure 11D). NFE2L2 mRNA was reduced to 0.22‐fold (SKOV3) and 0.28‐fold (OVCAR3) (p < 0.001 for both). NRF2 knockdown combined with erastin treatment (Figure 11E) produced synergistic ferroptosis sensitization: GPX4 decreased to 0.28‐fold (vs. 0.42‐fold with erastin alone), SLC7A11 to 0.32‐fold, and ACSL4 increased to 3.85‐fold. Lipid peroxidation measured by C11‐BODIPY oxidation (Figure 11F) increased 4.52‐fold in siNRF2 + erastin conditions, which was partially rescued by ferrostatin‐1 (1.45‐fold, p < 0.01), confirming ferroptosis specificity and establishing NRF2 as a causal mediator of ferroptosis resistance in EOC. Independent Validation Using GSE300897 Cohort To validate the robustness of ferroptosis signatures identified in the discovery cohort, we performed independent analysis on GSE300897 (19,002 cells from 9 treatment‐naive HGSOC patients; Figure 12). Cell type composition analysis (Figure 12A) confirmed consistent TME heterogeneity: ovarian cancer cells (5451 cells), T cells (5568), myeloid cells (4104), and fibroblasts (3186). Independent validation using GSE300897 cohort. (A) Cell type composition in GSE300897 (n = 19,002 cells from 9 patients). (B) Ferroptosis scores by cell type in GSE300897, sorted by score magnitude. (C) Ferroptosis score comparison: Platinum‐Sensitive vs. Refractory patients. Violin plots with Mann–Whitney p‐value. (D) Cross‐dataset ferroptosis score comparison: GSE165897 (discovery) vs. GSE300897 (validation) for matched cell type categories. Ferroptosis scoring across cell types in the validation cohort (Figure 12B) reproduced the pattern observed in GSE165897: Plasma.cell (FS = −0.523), T.cell (FS = −0.594), B.cell (FS = −0.609), Ovarian.cancer.cell (FS = −0.663), Endothelial.cell (FS = −0.709). Critically, platinum‐sensitive and platinum‐refractory patients showed differential ferroptosis scores (Figure 12C, Mann–Whitney p = 3.22e‐24), with refractory patients exhibiting higher ferroptosis scores, providing evidence linking ferroptosis pathway activity to clinical platinum sensitivity in EOC. Cross‐dataset comparison of ferroptosis scores (Figure 12D) showed consistent directionality between the discovery and validation cohorts. Discussion This study provides a comprehensive single‐cell transcriptomic characterization of ferroptosis‐related signature genes and tumor microenvironment composition in epithelial ovarian cancer, revealing coordinated dysregulation of ferroptosis pathways as a central feature of disease pathogenesis. Our findings extend current understanding of ovarian cancer biology by highlighting the critical contribution of iron‐dependent cell death mechanisms and their relationship with tumor microenvironment cellular heterogeneity. The identification of 18 distinct cell clusters comprising EOC cells (16.9%), immune cells (67.6%), and stromal cells (15.6%) reveals the cellular complexity of the ovarian cancer microenvironment. The predominance of immune cells in our dataset is consistent with previous reports demonstrating extensive immune infiltration in ovarian cancer [20]. The significant correlation between GPX4 and ACSL4 expression suggests coordinated regulation of ferroptosis‐promoting and ferroptosis‐suppressing mechanisms. The NRF2 regulatory network analysis revealed that NRF2 acts as a master regulator of ferroptosis resistance in ovarian cancer [11]. NRF2 activation has been previously linked with chemotherapy resistance and poor prognosis in ovarian cancer [21, 22]. Our analysis of transcription factor activity, showing increased NRF2/NFE2L2 activity in EOC cells, provides mechanistic evidence as to how EOC cells escape ferroptotic cell death. The parallel activation of ATF4 and HIF1A in stroma may represent complementary mechanisms by which the tumor microenvironment maintains ferroptosis resistance [23]. NRF2 siRNA knockdown experiments (Figure 11D–F) directly demonstrated that NRF2 depletion synergistically enhances erastin‐induced ferroptosis sensitization, with C11‐BODIPY lipid peroxidation increasing 4.52‐fold and ferrostatin‐1 rescue confirming ferroptosis specificity, establishing NRF2 as a causal mediator. The KEGG pathway enrichment analysis showing significant enrichment of ferroptosis, fatty acid metabolism, glutathione metabolism, and oxidative phosphorylation illustrates the metabolic reprogramming that confers resistance to ferroptosis in ovarian cancer. The activation of PI3K‐Akt and MAPK signaling pathways implies crosstalk between growth factor signaling and ferroptosis regulation that could be exploited by combination therapy [24]. The discovery of p53 signaling enrichment is particularly salient considering that TP53 mutations are among the most frequent alterations in high‐grade serous ovarian cancer [25]. Cell–cell communication network analysis uncovering the top ligand‐receptor interactions (SPP1‐CD44, VEGFA‐KDR, CCL2‐CCR2, CXCL12‐CXCR4, IL6‐IL6R, TGFB1‐TGFBR1) offers insights into the paracrine signaling mechanisms that contribute to ferroptosis resistance in the TME. Our co‐culture experiments (Figure 11A–C) functionally validated that CAF‐derived SPP1 and macrophage‐expressed CD44, CCR2, and KDR are induced under triple co‐culture conditions, directly demonstrating TME‐mediated modulation of ferroptosis sensitivity. Osteopontin (SPP1) has been implicated in ovarian cancer progression and chemoresistance [26]. STAT3 signaling induced by the IL6‐IL6R axis can block ferroptosis through transcriptional regulation of antioxidant genes [27]. Ferroptosis sensitivity landscape analysis demonstrated that cells distribute across Sensitive (8863 cells, 17.1%), Neutral (33,976 cells, 65.7%), and Resistant (8863 cells, 17.1%) regions. This heterogeneous resistance profile provides experimental evidence that ferroptosis resistance is a prevalent feature of ovarian cancer and may partially account for the limited clinical efficacy of ferroptosis‐inducing agents [28]. GO enrichment analysis defining oxidative stress response, iron ion homeostasis, and glutathione metabolism underscores the central role of redox biology in ovarian cancer ferroptosis regulation. Upon comparing treatment‐naive and post‐NACT samples, we observed that neoadjuvant chemotherapy alters ferroptosis gene expression patterns. This may reflect selection of ferroptosis‐resistant clones or an adaptive response to chemotherapy‐induced oxidative stress [29]. Patient heterogeneity, showing divergent molecular profiles within samples, highlights the requirement for personalized therapeutic strategies focusing on ferroptosis pathways. Independent validation using the GSE300897 cohort (19,002 cells, 9 patients) confirmed the reproducibility of ferroptosis signatures across datasets and, critically, established an association between ferroptosis scores and clinical platinum sensitivity (Figure 12C, p = 3.22e‐24). These findings suggest several therapeutic strategies: GPX4 inhibitors or SLC7A11 blockers may sensitize resistant cancer cells [30]; targeting NRF2‐mediated antioxidant response may overcome ferroptosis resistance [31]; disrupting cell–cell communication pathways may enhance ferroptosis sensitivity [32]; and combination strategies targeting ferroptosis alongside conventional chemotherapy may provide synergistic benefits [33]. Limitations Several limitations of this study warrant consideration. First, although the analysis was validated using two independent datasets (GSE165897, n = 51,702; GSE300897, n = 19,002), validation in additional independent patient cohorts with complete clinical annotations (GSE192898, GSE213243) would further strengthen the generalizability of our findings. Second, the co‐culture experiments (Figure 11) were performed with a single primary CAF isolate and THP‐1‐derived macrophages. Primary patient‐derived CAFs and tumor‐associated macrophages from multiple EOC patients would strengthen the generalizability of paracrine ferroptosis regulation findings. NRF2 siRNA knockdown was performed in two EOC cell lines; CRISPR/Cas9 NRF2 knockout with rescue experiments would provide definitive causal evidence for NRF2‐mediated ferroptosis resistance. Third, the GSE300897 validation cohort confirmed the association between ferroptosis scores and platinum sensitivity status (Sensitive vs. Refractory, p = 3.22e‐24). However, comprehensive clinical annotations including FIGO stage, pathological subtype, and overall/progression‐free survival were not available in either GEO dataset. Prospective studies with annotated clinical outcomes are needed to establish ferroptosis scores as predictive biomarkers. Fourth, the cross‐sectional nature of the scRNA‐seq data precludes assessment of temporal dynamics in ferroptosis activation during disease progression and treatment response. Conclusions This study establishes ferroptosis dysregulation and altered tumor microenvironment composition as central features of epithelial ovarian cancer pathogenesis through comprehensive single‐cell transcriptomic analysis. The ovarian cancer microenvironment is characterized by enhanced ferroptosis resistance mechanisms, coordinated expression of ferroptosis‐related genes, and complex cell–cell communication networks. NRF2 was identified and functionally validated as a master regulator of ferroptosis resistance, and paracrine signaling axes (SPP1‐CD44, CCL2‐CCR2, VEGFA‐KDR) were shown to modulate ferroptosis sensitivity in co‐culture models. Independent validation linked ferroptosis pathway activity to clinical platinum sensitivity, highlighting the translational potential of ferroptosis‐based biomarkers and therapeutic strategies in EOC. Author Contributions Hui Liang: conceptualization, supervision, funding acquisition, project administration, writing – review and editing. Jie Liu: conceptualization, methodology, investigation, formal analysis, visualization, writing – original draft, writing – review and editing. Bingxin Li: writing – review and editing, formal analysis, visualization. Jing Li: data curation, investigation, resources, validation, writing – review and editing. Funding The authors have nothing to report. Ethics Statement The authors have nothing to report. Consent The authors have nothing to report. Conflicts of Interest The authors declare no conflicts of interest.