Comprehensive molecular characterization identifies therapeutic vulnerabilities in ovarian clear cell carcinoma.

The clinical specimens were systematically profiled using a multi-omics approach. All 82 tumor samples collected from Fudan University Shanghai Cancer Center (FUSCC) were underwent bulk RNA sequencing (RNA-seq) and HER2 immunohistochemistry (IHC) analysis. For genomic analysis, 69 of the 82 cases were selected based on the availability of matched peripheral blood samples, allowing for germline–tumor paired WES. Additionally, a subset of 13 tumor specimens was profiled using scRNA-seq. Two independent OCCC gene expression datasets served as validation cohorts to assess the robustness and prognostic value of the identified immune subtypes. The GSE65986 dataset [ 18 ] consisted of 25 OCCC tumor specimens profiled using the Affymetrix Human Genome U133 Plus 2.0 Array ( GPL570 ), and the GSE73614 dataset [ 19 ] consisted of 37 OCCC samples analyzed via the Agilent Whole Human Genome 4 × 44 K microarray ( GPL6480 ). Differential expression analysis (DEA) was performed using an integrated statistical strategy, concurrently employing the DESeq2 [ 20 ] and edgeR algorithms [ 21 ] and the limma-voom method [ 22 ]. To control the false positive rate in multiple hypothesis testing, raw p-values were adjusted using the Benjamini-Hochberg method, with the false discovery rate (FDR) utilized as the primary threshold for significance. Only genes consistently identified as significant across all three algorithms were defined as high-confidence differentially expressed genes (DEGs). For clinical association analyses, continuous variables were compared using the Student’s t-test (for parametric data) or the Wilcoxon rank-sum test (for non-parametric data), depending on the data distribution. Categorical variables were evaluated using the chi-square test, with Fisher’s exact test applied when any expected cell frequency in the contingency table was less than 5. Survival probabilities were estimated via the Kaplan-Meier method and compared using the log-rank test. Independent prognostic determinants were identified through univariate and multivariate Cox proportional hazards regression models. All statistical tests were two-tailed, and the significance threshold was set at alpha = 0.05. Bioinformatics analyses and core statistical computations were primarily implemented in R software (version 4.4.2). For biological data of two groups, multiple t-tests (one per row) to rigorously assess the differences between two groups at each time point, explicitly annotating the P-values at the study endpoint to robustly demonstrate the statistical significance using GraphPad Prism 8.0 software. For three or more groups, One-way ANOVA followed by Tukey’s post hoc test for multiple comparisons was used when the data met the assumptions of normality and homogeneity of variance. Values represent the mean ± SEM, with statistical significance thresholds set at α = 0.05 (* p  < 0.05; ** p  < 0.01; *** p  < 0.001) unless otherwise specified. OCCC cell line ES2 and TOV21G were purchased from Shanghai Cell Bank Type Culture Collection (Shanghai, China). RMG1, OVTOKO, SKOV3 and OVISE were purchased from YaJi Biological (Shanghai, China). All OCCC cells were cultured in DMEM (BasalMedia, L110KJ, Shanghai, China) culture media supplemented with 10% FBS (ExCell Bio, FSP500, Shanghai, China), 1% penicillin-streptomycin (BasalMedia, S110JV). All cell lines were grown in incubator at 37 °C and 5% CO 2 and were authenticated by short tandem repeats sequencing. Routine mycoplasma testing was carried out to ensure cells free of contamination before proceeding with further experiments. The antibodies against FOXA2 (22474-1-AP, WB), HER2 (60311-1-Ig, WB and IHC for OCCC tissue microarray), ACTB (20536-1-AP, WB), BAF170 (12018-1-AP, IP), BRG1 (21634-1-AP, IP), ARID1A (30304-1-AP, IP), GAPDH (10494-1-AP, WB), CD4 (67786-1-Ig, IF), CD56 (14255-1-AP, IF), HRP-conjugated Affinipure Goat Anti-Mouse IgG(H + L) (SA00001-1, WB), HRP-conjugated Affinipure Goat Anti-Rabbit IgG(H + L) (SA00001-2, WB) were purchased from Proteintech (Wuhan, China). HER2 (2165, IHC for OCCC-PDX), Calreticulin (12238, FCM) was purchased from Cell Signaling Technology (Danvers, MA, USA). Ki-67 (GB121141-100, IHC) and cleaved Caspase-3 (GB115600-100, IHC) were purchased from Servicebio (Wuhan, China). γ-H2AX (ab26350) purchased from Abcam (Cambridge, England). Alexa Fluor™ 488-conjugated secondary antibody (A-11008, FCM) were purchased from Thermo Fisher Scientific. For PCDH-FOXA2-WT-Flag plasmids, the wild-type FOXA2 coding sequence was PCR-amplified from a template plasmid (Miaolingbio, Wuhan, China) and cloned into a PCDH-Flag-puro vector. PCDH-FOXA2-MUT-Flag plasmids was generated using a Fast Site-Directed Mutagenesis Kit (KM101, Tiangen, China). ES2 cells (1 × 10 7 ) were formaldehyde-crosslinked (1%, 10 min), quenched with glycine (0.125 M, 5 min), and washed with ice-cold PBS. After centrifugation (1,000 g, 4 °C), pellets underwent sequential lysis in ChIP buffers A/B (Cell Signaling Technology) with protease inhibitors. Chromatin digestion used Micrococcal Nuclease (37 °C, 20 min), terminated by EDTA. Following sonication (300 bp fragments), clarified lysates were incubated overnight at 4 °C with antibody-conjugated protein G magnetic beads. Beads were washed (20 mM Tris-HCl pH 7.4, 500 mM NaCl, 1 mM EDTA, 1% NP40, 0.05% SDS) and treated with RNase A/proteinase K. Purified DNA (Qiagen) was quantified by Qubit. Libraries prepared using KAPA HyperPrep (Roche) underwent paired-end sequencing (Illumina NovaSeq 6000), with alignment to hg38 via Bowtie2. Input controls (1% lysate) accompanied all ChIP reactions. Raw reads were aligned to the human reference genome (GRCh38). MACS2 (v2.2.7) [ 23 ] was employed by comparing samples against matched input controls, with a significance threshold of q-value < 0.1. To facilitate cross-sample comparison and visualization, fragment pileup tracks were normalized by sequencing depth using the --SPMR (Signal Per Million Reads) option. Subsequently, deepTools (v3.5.1) [ 24 ] was used to calculate signal intensities around transcription start sites (TSS) via the computeMatrix command, followed by the generation of profile plots and heatmaps to characterize global occupancy patterns. Genomic features and associated genes of the peak summits were annotated using the annotatePeaks.pl script from the HOMER (v4.11) suite. To evaluate the biological specificity of the immunoprecipitation, findMotifsGenome.pl was utilized for motif enrichment analysis. Potential target genes were defined as those harboring a FOXA2 binding peak within \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\pm\:$$\end{document} 2 kb of the TSS. These candidates were then intersected with DEGs (|log_2FC| > 0.322). The resulting overlapping genes were subjected to downstream functional enrichment and pathway analysis. For the cell proliferation assay, cells were plated in triplicate (1–3 × 10³ cells/well) in 96-well plates. At specified intervals, CCK-8 solution (HY-K0301, MedChemExpress) diluted 1:9 in complete medium was added. After 1.5–2 h incubation (37 °C, 5% CO 2 ), absorbance at 450 nm was quantified using a microplate reader (BioTek Synergy H1). For the colony formation assays, cells (2–5 × 10³/well) were seeded into 6-well plates and cultured for 7–10 days. Colonies were fixed with methanol and stained with 1% crystal violet (C0775, Sigma-Aldrich) for 20 min. Colonies were enumerated using ImageJ v1.53 with optimized threshold settings. SKOV3 and TOV21G cells were seeded into 6-well plates and treated the following day with NC18 at concentrations of 100 ng/mL or 300 ng/mL. After 24 h of treatment, the cells were harvested, resuspended in 100 µL of FACS buffer (PBS containing 2% FBS and 1 mM EDTA), and incubated with an anti-Calreticulin primary antibody (CST, 12238) on ice for 40 min. Following a washing step, the cells were stained with an Alexa Fluor™ 488-conjugated secondary antibody (Thermo Fisher Scientific, A-11008) for 30 min on ice. The cells were subsequently washed again, resuspended in 300 µL of FACS buffer, and labeled with eBioscience™ Fixable Viability Dye eFluor™ 450 (Thermo Fisher Scientific, 65-0863-14) immediately prior to flow cytometry analysis. Extracellular ATP levels were quantified using an ATP Assay Kit (MCE, HY-K0314, China). Briefly, SKOV3 and TOV21G cells were seeded into 6-well plates and treated the following day with NC18 at concentrations of 100 ng/mL or 300 ng/mL. After 48 h of treatment, cell culture supernatants were harvested and centrifuged at 2000 × g for 3 min. Subsequently, 100 µL of the cleared supernatant and 100 µL of ATP detection working solution were added to each well of a 96-well plate, and luciferase luminescence was measured using a microplate reader. For extracellular HMGB1 detection, a Human/Mouse HMGB1 ELISA Kit (Proteintech, KE00343, China) was utilized. Following identical cell seeding and 48-hour NC18 treatment conditions, the supernatants were similarly collected and centrifuged at 2000 × g for 3 min. Next, 100 µL of the supernatant was added to each well of a 96-well plate and incubated at 37 °C for 2 h. After discarding the liquid and washing the wells, 100 µL of HRP-conjugated detection antibody was added, followed by a 40-minute incubation at 37 °C. Following a subsequent washing step, 100 µL of TMB substrate solution was added to each well and incubated at 37 °C in the dark for 15–20 min for color development. The reaction was terminated by adding 100 µL of stop solution per well, and the optical density was immediately measured at a wavelength of 450 nm using a microplate reader. The animal trials were conducted in the specific pathogen-free (SPF) facility of the Laboratory Animal Center at FUSCC. All procedures were performed in accordance with guidelines approved by the Animal Care and Use Committee of FUSCC (approval numbers: FUSCC-IACUC-2024699 and FUSCC-IACUC-2025710). Fresh OCCC specimens were rinsed in ice-cold PBS, mechanically dissected into 1–2 mm³ fragments, and orthotopically engrafted subcutaneously into 6-week-old female NOD/SCID/IL2Rγnull mice using a 16-gauge trocar. Following successful first-generation tumor formation (≥ 500 mm³), tissue fragments were serially passaged into secondary recipients ( n  = 8/group) for pharmacological studies. TOV21G female mouse models (BALB/c-nu/nu, 6-week-old) were randomly divided into two groups, with each group containing 7 mice. Each mouse received a subcutaneous injection of 2 × 10 6 cells in 200 µL of DMEM mixed with Matrigel (Corning, 356234, USA) into the lower back region. Treatment with NC18 (Novacyte Therapeutics, China) commenced when tumors reached 0.5 × 0.5 cm dimensions, administered via intraperitoneal injection (2 mg/kg, twice). Mice were humanely euthanized via CO 2 inhalation when tumors in the control group reached a size of 1.2 × 1.2 cm. The excised tumors were subsequently subjected to gravimetric analysis, macroscopic documentation, and immunohistochemical profiling using validated antibodies including: anti-HER2 (CST, 2165), Ki-67 (Servicebio, GB121141-100), γ-H2AX (Abcam, ab26350), and cleaved Caspase-3 (Servicebio, GB115600-100). Tumor volumes were calculated using the formula V = (length × width²)/2. Freshly harvested OCCC tissues and established PDX tumors were rinsed twice with ice-cold 1× PBS and fixed in 4% paraformaldehyde for 24 h at room temperature. Fixed specimens were paraffin-embedded, sectioned at 4 μm thickness onto adhesive slides, and baked at 60 °C for 1 h. Deparaffinization was performed through three xylene immersions followed by rehydration in a graded ethanol series (100%-70%) and distilled water rinsing. Antigen retrieval utilized 10 mM sodium citrate buffer (pH 6.0) with microwave irradiation (20 min, 95 °C), followed by endogenous peroxidase blockade using 3% hydrogen peroxide (25 min, RT). Non-specific binding was minimized by 30-min blocking with 3% BSA at RT prior to overnight incubation with primary antibodies at 4 °C in a humidified chamber. After PBS washes, species-matched HRP-conjugated secondary antibodies were applied for 50 min at RT. Chromogenic development employed 3,3′-diaminobenzidine with hematoxylin counterstaining, optimized by real-time microscopic monitoring to ensure signal specificity. For TMA, specimens were collected and embedded in paraffin blocks, which were cut into 4-µm sections and subjected to IHC. Protein expression of HER2 (Proteintech, 60311-1-Ig) on stained slides was assessed by two independent pathologists. All quantifications were evaluated blinded to patient clinical outcomes. The internationally standardized ASCO/CAP guidelines for HER2 assessment in breast cancer and gastric cancer was adopted to accommodate the unique histological architecture and cellular polarity of OCCC, particularly its propensity for basolateral rather than circumferential staining. We established a refined IHC scoring system tailored for glandular malignancies: Score 0 (Negative) indicates no staining or faint/incomplete membrane staining in ≤ 10% of tumor cells; Score 1+ (Low) denotes faint or barely perceptible incomplete membrane staining in > 10% of tumor cells; Score 2+ (Intermediate) represents weak-to-moderate basolateral, lateral, or complete membrane staining in > 10% of tumor cells; and Score 3+ (High) is defined by strong intensity basolateral, lateral, or complete membrane staining in > 10% of tumor cells. The WES was performed on OCCC patients using DNA from both tumor samples and matched noncancerous tissues. Genomic DNA (200 ng) from frozen tissues was extracted using AllPrep DNA/ RNA Kit (Qiagen, Germany) and sheared by Biorupter (Diagenode, Belgium) to acquire 200–300 bp fragments. The ends of DNA fragments were repaired and Illumina Adaptor was added (Fast Library Prep Kit, iGeneTech, Beijing, China). After the sequencing library was constructed, whole exomes were captured using the AIExomeV2 (T192V1T) Enrichment Kit (iGeneTech, Beijing, China) and sequenced on DNBSEQ-T7 platform (MGI, Shenzhen, China) with 150 bp paired-end reads. Raw reads were first filtered and trimmed adapters using fastp (v0.12.2). After quality control, reads were mapped to the human reference genome sequence (hg 19) by BWA-MEM (version 0.7.12) followed by pre-processing steps. Next, GATK MuTect2 was performed to call somatic mutations and tumor-only mutations with frequency higher than 5% were retained for downstream. ANNOVAR (version 201,910) [ 25 ] was used to annotate variants and annovarToMaf converted annovar annotations into MAF for visualization. The smgbp database was selected to be enriched to known oncogenic pathways by the OncogenicPathways function of the maftools R package. Cancer genes were downloaded from NCG6.0 database ( http://ncg.kcl.ac.uk/ ). Somatic CNVss were profiled from WES data using CNVkit (v0.9.1) [ 26 ] with matched normal blood DNA as a baseline. After bias correction and GRCh38/hg38 alignment, log2 ratios were segmented via the CBS algorithm. Copy number states were categorized using discrete thresholds: deep deletion ( 0.7). To distinguish significant genomic aberrations from background noise across the cohort, we employed GISTIC 2.0 on the segmentation outputs, applying a 0.90 confidence level and a Benjamini-Hochberg adjusted q-value < 0.1 to define significant focal peaks. K-means clustering (the ‘‘kmeans’’ function in R) and consensus clustering (the ‘‘ConsensusClusterPlus’’ package in R were performed to determine the optimal number of subtypes. Input data for each sample ( n  = 82) was the ‘‘actual copy change given’’ of each ‘‘peak region’’ obtained from the file ‘‘all_lesions.conf_90.txt’’, where values over 2 were set to 2 (consistent with the cap value in GISTIC 2.0). Consensus clustering was used to assess the robustness of k-means clustering (500 iterations, 0.8 resampling). The optimal number of clusters was determined from the cumulative density function, which plots the corresponding empirical cumulative distribution, defined over a range between 0 and 1, and from a calculation of the proportion increase in the area under the CDF curve. The mixed siRNAs targeting FOXA2 (siFOXA2-1: UUGCAGGGAAGUCUUACUUAATT, siFOXA2-2: GAGGGUUGUACUAUUGUUUAATT, siFOXA2-3: CUCCAUGAACAUGUCGUCGUATT) were ordered from Hippobio (Zhejiang, China). ES2 OCCC cells were seeded in 6-well plates to allow 30% confluency in the next day. Negative control siNC and siRNAs against FOXA2 were transfected into cells using RNAimax (Thermo, 13778150). siRNA was utilized at a working concentration of 50 nM. Cells were harvested at 48 h post-transfection and then performed qPCR analysis or immunoblot to evaluate knockdown efficiency of target genes. Total RNAs (500 ng) from OCCC tissues and matched normal tissues were extracted by Trizol ® Reagent (Thermo, USA) and treated with DNase I (NEB) to remove DNA before constructing the RNA-seq libraries. Strand-specific RNA-seq libraries were prepared using the QIAseq FastSelect–rRNA HMR (Qiagen) and KAPA RNA HyperPrep (Roche) kits. Briefly, QIAseq FastSelect reagent was added to the RNA sample to fragment the RNA at 85°C for 6 min and then stepwise cool the reaction from 75°C to 25°C for 14 min. First- and second-strand cDNA was synthesized with random hexamer primers, treated with DNA End Repair Kit (Qiagen) to repair the ends, then modified with Klenow to add an A at the 3’ end of the DNA fragments, and finally ligated to adapters. Purified dsDNA was further subjected to 11 cycles of PCR amplification. The libraries were quality controlled with Qubit (Thermo Fisher Scientific, USA) and Qsep100 (BiOptic, China) and sequenced by the Illumina sequencing platform (Nova) on a 150 bp paired-end run. FastQC (Babraham Bioinformatics Institute) was used to check the sequencing quality, and high-quality reads were mapped to human reference genome (hg38) along with the gene annotation data (genecode v29) from the Genecode database using STAR (v2.5.3a) [ 27 ]. Raw read counts per gene were obtained using featureCount. Transcripts per million values were calculated with normalisation on a total number of counted reads. Cultured cells transfected with PCDH-GFP, PCDH-FOXA2-WT-Flag and PCDH- FOXA2-MUT-Flag were collected and washed with cold PBS, lysed with IP lysis buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.2% NP40, 20% glycerol) containing 1 × protease inhibitor cocktail (MedChemExpress, HY-K0010, China) on ice for 30 min. The lysate was subsequently clarified by centrifugation at 12,000 g for 10 min at 4 °C. A 1% of the volume of the lysate was taken as an input control, while the remaining lysate was incubated with Flag-nanoab magnetic beads (Lablead, FNM-25-1000) and GFP-nanoab magnetic beads (Lablead, GNM-25-1000) with rotated at 4 °C for 3 h. The magnetic beads bound to the proteins were captured using a magnetic stand, followed by three washes using NT2 buffer (20 mM Tris–HCl pH 7.5, 150 mM NaCl, 1 mM MgCl 2 , 0.05% NP40) and boiled at 95 °C for 5 min with SDS loading buffer for Western blotting analysis. Fusion transcripts were prioritized from RNA-Seq data using a synchronized bioinformatics pipeline integrating STAR-Fusion (v1.12.0) and Arriba (v2.3.0). Raw reads were processed with Trimmomatic for quality control and aligned to the GRCh38 genome via STAR in fusion-sensitive mode. High-confidence fusions were defined by a threshold of ≥ 5 junction reads and were cross-referenced with clinical databases (OncoKB and COSMIC). Common artifacts, read-through events, and germline variants were systematically excluded. To enable visualization and sequence confirmation, FusionInspector (v2.10.0) was employed to perform de novo transcriptome assembly via Trinity, providing reconstructed junction sequences for open reading frame identification and generating localized genomic alignments for inspection in the Integrative Genomics Viewer (IGV). 17 immune-related pathways from the ImmPort database ( https://immport.niaid.nih.gov ) were downloaded. Protein-coding gene expression profiles were transferred to immune-related pathways expression levels using the single sample gene set enrichment analysis GSEA (ssGSEA) tool, and ssGSEA scores were z-score normalised. Clustering was performed using R heatmap package for ssGSEA z-score matrix (clustering method = ward.D). For estimation of immune cell infiltration, the CIBERSORT algorithm in absolute mode was employed to quantify the infiltration levels of 22 distinct immune cell subsets based on the LM22 gene signature. Concurrently, the xCell R package was applied to generate enrichment scores for 64 immune and stromal cell types, facilitating a broader assessment of cellular components within the TME. To identify significant variations in cellular abundance between the predefined high and low immune groups, Wilcoxon rank-sum tests were performed on the xCell enrichment scores. The resulting P-values were adjusted using the Benjamini-Hochberg procedure to control the FDR. Differential infiltration was visualized via volcano plots using the EnhancedVolcano package, with the significance thresholds established at an FDR = 1. To elucidate the biological implications of the identified DEGs, Gene ontology (GO) and kyoto encyclopedia of genes and genomes (KEGG) enrichment analyses were performed using the clusterProfiler R package. GO terms were categorized into biological process (BP), molecular function (MF), and cellular component (CC). For a more comprehensive functional characterization, GSEA was conducted on the complete gene list ranked by a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\uppi\:}$$\end{document} -score, defined as: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\pi=\mathrm{Log}_{2}\;\left(\mathrm{Fold}\;\mathrm{Change}\right)\times\left(-\mathrm{Log}_{10}\;{P}-\mathrm{value}\right)$$\end{document} This combined metric accounts for both the magnitude and statistical significance of differential expression. The MSigDB Hallmark and Reactome (c2.cp.reactome) gene sets were utilized as reference libraries. For GSEA, the minimum and maximum gene set sizes were restricted to 10 and 500, respectively. In all analyses, P-values were adjusted using the Benjamini-Hochberg method, and a significance threshold was established at an adjusted P-value < 0.05. Approximately 1,000 nuclei/µL were loaded on a Chromium Controller (10x Genomics) to generate gel bead-in-emulsions. Post-GEM-RT cleanup, cDNA amplification and 3’ gene expression library construction were carried out according to the user guide. The scRNA-seq libraries were generated using the 10x Genomics Chromium Controller Instrument and Chromium Single Cell 3’ v3 Reagent Kits (10x Genomics). The final libraries were quantified using the Qubit High Sensitivity DNA assay (Thermo Fisher Scientific), and the size distribution of the libraries was determined using a High Sensitivity DNA chip on a Bioanalyzer 2200 (Agilent). All libraries were sequenced by Novaseq6000 (Illumina) on a 150 bp paired-end run. Illumina NovaSeq6000 RTA (v.3.4.4) software was used to perform basecalling, and the resulting base call files (.bcl) were converted to standard fastq formatted sequence files using Bcl2Fastq (v.2.20, Illumina). Raw sequencing reads (FASTQ format) from OCCC tumor tissue samples were processed using the cellranger count pipeline (Cell Ranger, v6.0.0), where reads were aligned to the GRCh38 (Gencode v29) reference genome to generate single-cell gene expression matrices and coordinate-sorted BAM files. Subsequently, the gene expression matrices were imported into the Seurat (v5.2.1) R package for stringent quality control and downstream analysis. Initially, a Seurat object was constructed by retaining genes expressed in at least 3 cells and filtering out cells with fewer than 200 detectable genes. To exclude low-quality cells, damaged cells, or potential doublets, dynamic percentile-based filtering was implemented. Specifically, cells were retained only if their number of detected genes (nFeature_RNA) fell between the 5th and 95th percentiles of the total distribution. Additionally, cells with a mitochondrial transcript proportion > 20% were excluded. Following QC, data normalization was performed using the NormalizeData function, and the top 2,000 highly variable genes were identified via FindVariableFeatures. After data scaling with ScaleData, batch effects across samples were mitigated using canonical correlation analysis integration. Dimensionality reduction was performed using principal component analysis, and the top 20 principal components were selected for downstream analysis. Cell clustering was executed using the FindNeighbors and FindClusters functions with a resolution of 1.2, capturing the fine-grained heterogeneity of the TME. Final visualization was achieved through uniform manifold approximation and projection. Cell populations were manually annotated based on the expression of canonical marker genes: B cells (CD79A, MS4A1, CD19); T cells (CD3D, CD3E, CD3G, CD8A, CD8B, CD4); NK cells (NKG7, KLRC1); Monocytes/Macrophages (LYZ, CD14, CD163); Dendritic cells (HLA-DPB1, CST3); Endothelial cells (PECAM1, CLDN5); Stromal cells (DCN, OGN); and Epithelial/Tumor cells (KRT18, KRT19). Chromosomal CNVs were inferred using inferCNV (v1.18.1). We defined epithelial cells ( n  > 2,000) as the observation group, with endothelial and immune cells (T and B cells) serving as the reference baseline. The analysis followed the standard pipeline for 10x Genomics data with a cutoff of 0.1, incorporating denoising and a six-state HMM for CNV region prediction. The final CNV scores were calculated based on the deviation from the reference to quantify the genomic instability of the epithelial subclusters. Intercellular communications were inferred via CellChat (v1.1.3) [ 28 ] using the CellChatDB.human database. Communication probabilities were calculated using the truncated mean approach with default parameters. Significant interactions were determined by permutation tests ( P  < 0.05). The analysis focused on the signaling crosstalk between malignant epithelial cells and immune populations. Survival outcomes were assessed using Kaplan-Meier analysis, with the log-rank test employed to compare differences between groups. To determine the prognostic value of clinical and genomic features, we performed univariate and multivariate cox proportional hazards regression analyses. The multivariate model was constructed by simultaneously incorporating all candidate variables to evaluate the independent impact of each factor.Continuous variables were dichotomized for risk stratification as follows: Tumor mutational burden (TMB) (High: >20 muts/Mb), HRD score (High: >= 42), and Age (High: >52 years). Pathological Stage was categorized as early (IA) or advanced (IB–IV). Additional genomic parameters, including high-frequency mutations, focal CNV events, ERBB2 expression (high: TPM > 100), as well as our proposed immune subtypes and CNV consensus clusters, were integrated into the comprehensive Cox model to calculate adjusted hazard ratios and 95% confidence intervals.

We performed WES, RNA sequencing, and scRNA-seq on 82 OCCC samples, which established the comprehensive genomic and transcriptomic atlas compared with other OCCC studies (Table  1 ). A total of 11,786 somatic mutations were identified, comprising 9,820 missense mutations, 1,079 frameshift insertions/deletions (indels), 614 nonsense mutations, 6 stop-codon mutations, 137 splice-site mutations, 118 in-frame indels, and 12 transcription start-site mutations (Additional file 1: Fig. S1A). Single nucleotide polymorphisms (SNPs) constitute the predominant proportion of genomic variations (Additional file 1: Fig. S1B), with C-to-T transitions being the most frequently observed type (Additional file 1: Fig. S1C-E). To investigate oncogenic drivers in OCCC, we screened for cancer-associated gene mutations and found that both ARID1A and PIK3CA exhibited mutation frequencies exceeding 50% (Fig.  1 A; Additional file 1: Fig. S1F). Among 44 patients with ARID1A mutations, 19 carried frameshift variants while 6 harbored nonsense mutations. Notably, 82% (31/38) of PIK3CA -mutated cases displayed missense mutations (Fig.  1 B), with 30 patients demonstrating co-occurring ARID1A and PIK3CA mutations. Survival analysis initially revealed that patients with ARID1A mutations exhibited a significantly more favorable prognosis compared to the wild-type group (Additional file 1: Fig. S1G). Tumors harboring PIK3CA mutations showed a similar trend toward improved survival, this association did not reach statistical significance (Additional file 1: Fig. S1H). Excitingly, the latest research has confirmed that inavolisib—a selective inhibitor targeting PIK3CA mutations—demonstrates promising efficacy in metastatic breast cancer harboring PIK3CA mutations [ 29 , 30 ], suggesting its potential therapeutic benefit for OCCC patients with these mutations. Table 1 Comparison of current study with major OCCC genomic cohorts Study Sample Population Clinical DNA RNA IHC scRNA PDX Main focus David SP Tan et al. [ 31 ], Clin Cancer Res, 2011 n  = 50 England YES aCGH NO NO NO NO Recurrent amplifications including HER2 in OCCC Yuriko Uehara et al. [ 18 ], PLoS One, 2015 n  = 25 Japan YES SNP Array Microarray NO NO NO Unique chromosomal aberration patterns in OCCC Boris Winterhoff et al. [ 19 ], Gynecologic Oncology, 2016 n  = 37 USA YES NO Microarray NO NO NO Unique transcriptional features of early HGCCOCs Ryusuke Murakami et al. [ 12 ], The American Journal of Pathology, 2017 n  = 39 Japan NO WES NO YES NO NO Key chromosomal regions and mutation gene networks in OCCC Yusuke Shibuya et al. [ 32 ], Genes, Chromosomes and Cancer, 2018 n  = 48 Japan YES WES NO NO NO NO Japanese OCCC WES: ARID1A/PIK3CA mutations, APOBEC/MMR subsets Se Ik Kim et al. [ 13 ], Gynecologic Oncology, 2018 n  = 15 Korean YES WES NO NO NO NO Korean OCCC WES: PIK3CA/ARID1A / KRAS mutations, CNV analysis Joseph J Caumanns et al. [ 33 ], Clin Cancer Res, 2018 n  = 124 Europe, USA NO Kinome Targeted Sequencing, SNP Array NO NO NO YES mTORC1/2 inhibition as effective therapeutic strategy for OCCC Masataka Takenaka et al. [ 11 ], Clin Cancer Res, 2019 n  = 68 Australia, Japan YES Targeted Sequencing NO YES NO NO HNF1B expression as prognostic biomarker for platinum response in OCCC Siyu Xia et al., [ 5 ] J Immunother Cancer, 2024 n  = 44 China YES NO NO YES (31) YES (13) NO Single-cell profiling reveals ARID1A-immune axis and VEGF + PD-1 therapeutic potential in OCCC This study n  = 82 China YES WES (69) Bulk RNA-seq (82) YES (82) YES (13) YES Clinically oriented multi-omics molecular stratification and established a targeted therapeutic subtyping framework Comparison of current study with major OCCC genomic cohorts Kinome Targeted Sequencing, SNP Array Fig. 1 Mutational landscape of OCCC. A Oncoprint displays recurrently mutated genes, with individual tumor samples in columns and specific genes in rows. The top bar plot quantifies the TMB for each patient. The bottom annotation tracks provide clinical and molecular context for each sample, including stage, immune subtype (high vs. low), and patient age. Mutation types are color-coded as indicated in the legend, with “multi-hit” signifying the presence of multiple somatic alterations within the same gene-sample pair. B Lollipop plots of ARID1A and PIK3CA – the two most frequently mutated genes in OCCC – depicting protein domains (colored blocks) and recurrent amino acid alterations; asterisks indicate stop codons, “fs” denotes frameshift mutations with “fs*num” specifying termination codon positions. C Pathway enrichment analysis of pan-cancer driver genes using the “smgbp” database, showing proportion of mutated genes per oncogenic pathway. D TMB (left) and MSI (right) scores across the cohort (log10-normalized; median TMB = 2.87, median MSI = 0.26), each dot representing a single sample. E Mutational signature analysis via non-negative matrix factorization (NMF) of trinucleotide substitution patterns using maftools R package; left heatmap shows cosine similarity between extracted signatures ( n  = 3) and COSMIC database references, with right panel demonstrating best matches (two signatures aligning with SBS30, one with SBS32) based on COSMIC v3.2 classification Mutational landscape of OCCC. A Oncoprint displays recurrently mutated genes, with individual tumor samples in columns and specific genes in rows. The top bar plot quantifies the TMB for each patient. The bottom annotation tracks provide clinical and molecular context for each sample, including stage, immune subtype (high vs. low), and patient age. Mutation types are color-coded as indicated in the legend, with “multi-hit” signifying the presence of multiple somatic alterations within the same gene-sample pair. B Lollipop plots of ARID1A and PIK3CA – the two most frequently mutated genes in OCCC – depicting protein domains (colored blocks) and recurrent amino acid alterations; asterisks indicate stop codons, “fs” denotes frameshift mutations with “fs*num” specifying termination codon positions. C Pathway enrichment analysis of pan-cancer driver genes using the “smgbp” database, showing proportion of mutated genes per oncogenic pathway. D TMB (left) and MSI (right) scores across the cohort (log10-normalized; median TMB = 2.87, median MSI = 0.26), each dot representing a single sample. E Mutational signature analysis via non-negative matrix factorization (NMF) of trinucleotide substitution patterns using maftools R package; left heatmap shows cosine similarity between extracted signatures ( n  = 3) and COSMIC database references, with right panel demonstrating best matches (two signatures aligning with SBS30, one with SBS32) based on COSMIC v3.2 classification Comparative analysis demonstrated that OCCC mutation profiles were significantly enriched in epigenetic remodeling genes (including KMT2C , SMARCA4 , EP300 , ARID1B , SETD1B ) compared to HGSOC. Pathway analysis further revealed predominant enrichment in SWI/SNF chromatin remodeling complexes and PI3K signaling pathways (Fig.  1 C), suggesting crucial roles of epigenetic dysregulation in OCCC pathogenesis [ 34 , 35 ]. Meanwhile, the median TMB was calculated as 2.87 mutations/Mb, with a median microsatellite instability (MSI) score of 0.26 (Fig.  1 D). Three exceptional cases ( OT220056 , O180187, O190154) exhibited both high TMB (> 25 mutations/Mb) and MSI characteristics, indicative of genomic instability and potential responsiveness to immune checkpoint inhibitors. Through non-negative matrix factorization of whole-genome mutational signatures, we identified three distinct patterns. Cross-referencing with the COSMIC database revealed two validated signatures: one associated with NTHL1 -inactivation-mediated base excision repair deficiency, and another reflecting thiopurine treatment-induced immunosuppression (Fig.  1 E). Collectively, these findings highlight the pivotal involvement of epigenetic dysregulation in OCCC tumorigenesis and provide novel insights into its molecular pathogenesis. Recent studies have confirmed that amplifications of specific genomic loci harboring oncogenes such as MET hold potential as therapeutic targets [ 36 , 37 ]. Meanwhile, unlike HGSOC, OCCC is not characterized by high genomic instability, consistent with the lower frequency of germline BRCA1/2 mutations observed in this cancer subtype. To delineate the CNV landscape of OCCC, we employed CNVkit software to characterize the CNV profiles of individual OCCC samples. Our data revealed recurrent amplifications at chromosomal arms 8p, 8q, and 13p, alongside frequent deletions at 14p (Fig.  2 A). Significant focal alterations were detected at the 14q32.33 deletion and 20q13.2 amplification loci, occurring in over 65% of samples. These peaks were annotated with cytogenetic bands and known cancer-associated genes within these regions, including PRKDC , ERBB2 , CEBPB , and ZNF217 (Fig.  2 B). The PRKDC (protein kinase, DNA-activated, catalytic polypeptide) gene encodes DNA-dependent protein kinase catalytic subunit (DNA-PKcs), a core component of the non-homologous end joining (NHEJ) pathway for DNA double-strand break (DSB) repair. DNA-PKcs binds to the Ku70/Ku80 heterodimer to form the DNA-PK complex, facilitating the capture and ligation of broken DNA ends to maintain genomic stability. In tumors, aberrant expression of PRKDC significantly enhances chemoresistance and immune evasion [ 38 ]. Importantly, small-molecule inhibitors targeting PRKDC , such as peposertib and AZD7648, can overcome chemoresistance and have demonstrated therapeutic potential in clinical trials [ 39 ]. These findings suggest that targeting PRKDC holds promise for reversing chemoresistance and improving treatment outcomes in OCCC patients. Meanwhile, HER2-targeting ADCs directed against ERBB2 have demonstrated excellent clinical efficacy in breast cancer and other solid tumors, offering potential targeted therapeutic strategies for OCCC patients [ 40 ]. Fig. 2 CNV landscape in OCCC. A Chromosomal arm-level analysis by GISTIC 2.0 algorithm identifies recurrent amplifications (red) and deletions (blue) across samples, with significant gains at 8p, 8q, 13p, and 20q and focal loss at 14p; horizontal axis displays frequency of copy number gains (red), neutral (gray), and losses (blue) across chromosomal arms. B Focal-level GISTIC analysis reveals localized amplifications (red) and deletions (blue) with genomic coordinates, where -log10-transformed FDR-adjusted q-values (horizontal axis) indicate statistical significance; recurrently altered loci are annotated with known cancer-associated genes. C Genome-wide copy number alteration heatmap displays log2-normalized GISTIC scores (blue-to-red gradient representing deletions to amplifications), with samples ordered by hierarchical clustering (columns) and genomic coordinates arranged from 1p to Xq (rows); top annotation bars indicate pathological stage and CNV-based sample clusters. D The heatmap displays the top 20 genes exhibiting the strongest positive correlation between CNV and mRNA expression. The top panel illustrates the copy number status based on log2-normalized GISTIC scores, while the bottom panel presents the corresponding mRNA expression levels log2(TPM + 1) for the identical gene set. Top annotation bars stratify samples by FIGO stage, immune subtype (high vs. low), and CNV-derived molecular clusters. E Association of CNV burden with clinical and immunological stratification. The cross-tabulation illustrates the distribution of stage and immune subtypes (high vs. low) across OCCC samples with varying CNV status (high vs. low). Statistical significance was assessed using the chi-square test for clinical stage and Fisher’s exact test for immune subtype, as the latter contained cells with an expected frequency of less than 5 CNV landscape in OCCC. A Chromosomal arm-level analysis by GISTIC 2.0 algorithm identifies recurrent amplifications (red) and deletions (blue) across samples, with significant gains at 8p, 8q, 13p, and 20q and focal loss at 14p; horizontal axis displays frequency of copy number gains (red), neutral (gray), and losses (blue) across chromosomal arms. B Focal-level GISTIC analysis reveals localized amplifications (red) and deletions (blue) with genomic coordinates, where -log10-transformed FDR-adjusted q-values (horizontal axis) indicate statistical significance; recurrently altered loci are annotated with known cancer-associated genes. C Genome-wide copy number alteration heatmap displays log2-normalized GISTIC scores (blue-to-red gradient representing deletions to amplifications), with samples ordered by hierarchical clustering (columns) and genomic coordinates arranged from 1p to Xq (rows); top annotation bars indicate pathological stage and CNV-based sample clusters. D The heatmap displays the top 20 genes exhibiting the strongest positive correlation between CNV and mRNA expression. The top panel illustrates the copy number status based on log2-normalized GISTIC scores, while the bottom panel presents the corresponding mRNA expression levels log2(TPM + 1) for the identical gene set. Top annotation bars stratify samples by FIGO stage, immune subtype (high vs. low), and CNV-derived molecular clusters. E Association of CNV burden with clinical and immunological stratification. The cross-tabulation illustrates the distribution of stage and immune subtypes (high vs. low) across OCCC samples with varying CNV status (high vs. low). Statistical significance was assessed using the chi-square test for clinical stage and Fisher’s exact test for immune subtype, as the latter contained cells with an expected frequency of less than 5 To evaluate the prognostic impact of copy number alterations, we focused on the most frequent events, specifically 14q32.33 deletion and 20q13.2 amplification. Both univariate and adjusted multivariate Cox regression analyses indicated that these hotspot CNVs had no significant impact on overall survival (Additional file 1: Fig. S2A-B). We further explored the associations between recurrent CNV hotspots and frequently mutated genes. The results showed that patients harboring high-frequency mutations tended to exhibit a higher incidence of CNVs. However, this correlation did not reach statistical significance (Additional file 1: Fig. S3A-D). K-means and consensus clustering of CNV data optimally segregated samples into two distinct groups with robust concordance (Additional file 1: Fig. S3E-G). The high-CNV group exhibited elevated genomic instability, while the low-CNV group showed reduced variation (Fig.  2 C). Initial univariate survival analysis suggested that the high-CNV group exhibited a more favorable prognosis compared to the low-CNV group (log-rank P  = 0.04; Additional file 1: Fig. S2A). To further elucidate this observation, we examined the clinical and molecular distribution across these subgroups. Cross-tabulation revealed that high-CNV samples were significantly enriched in early-stage cases and the immune-low subtype (Fig.  2 D). Upon adjusting for clinical stage, the prognostic significance of high CNV status completely vanished ( P  = 0.85; Additional file 1: Fig. S2B). These findings demonstrate that the observed survival advantage in the high-CNV group is not an independent prognostic effect but is instead driven by the confounding influence of tumor stage, reflecting the predominance of chromosomal instability in early-stage OCCC. We further integrated CNV data with transcriptomic profiles to identify candidate drivers dictated by genomic alterations. Among 489 cancer-related genes, 175 genes exhibited a significant positive correlation between copy number status and mRNA expression (FDR  0.3), demonstrating a robust gene dosage effect. An integrated heatmap visualized the top 20 genes with the strongest correlations (Fig.  2 E). Within this gene set, we observed not only the upregulation of oncogenes such as ERBB2 and UBR5 in alignment with copy number gains but also a synchronized downregulation of the tumor suppressor SMAD4 in cases with copy number losses. This bidirectional concordance indicates that focal copy number alterations—both amplifications and deletions—are significant factors associated with the transcriptional dysregulation of key cancer-related genes in OCCC. Although OCCC exhibits relatively high chromosomal stability, the occurrence of therapeutically actionable chromosomal translocations or other structural variations remains unclear. To address this, we conducted a comprehensive analysis using Arriba and StarFusion, revealing that OCCC genomic alterations were primarily characterized by chromosomal duplications (49.34%) and translocations (35.59%; Fig.  3 A). While chromosomal duplications were numerically predominant, we prioritized chromosomal translocations for further investigation. This focus was guided by the translational rationale that such rearrangements frequently yield in-frame gene fusions, giving rise to chimeric proteins with potent oncogenic activity that often serve as high-value, therapeutically actionable targets. Thus, we prioritized chromosomal translocations for further investigation. Subsequent pathway enrichment analysis revealed that these fusion genes were significantly clustered within the Hippo, AMPK, and PI3K signaling pathways (Fig.  3 B). Among the structural variants identified, seven genes were classified as fusion hotspots, each exhibiting ≥ 3 fusion events (Fig.  3 C). Transcriptomic validation revealed that samples harboring these fusion events consistently displayed mRNA expression levels above the cohort median for the respective genes, suggesting that these structural rearrangements may lead to transcriptional activation or stabilization of the chimeric transcripts (Additional file 1: Fig. S4A). We found notable recurrent fusions involving SMAD3 : exon 1 of SMAD3 fused to exon 2 of KAZN , exon 2 of SMAD3 fused to exon 1 of PTRH1 , and exon 2 of SMAD3 fused to exon 2 of GTF2H5 . As a critical regulator of metastasis-associated genes including ZEB , Twist family members, and Snail, SMAD3 participates in tumorigenesis through TGF-β-induced formation of the SMAD2/3/4 complex (Fig.  3 D) [ 41 ]. Additional fusions included EPB41L5 exon 3 fused to PTRF exon 2, and EPB41L5 exon 7 fused to LTBP3 exon 15 (Fig.  3 E). Fig. 3 Fusion gene landscape in OCCC. A Pie chart summarizing genomic rearrangement types underlying fusion events: duplications (49.34%), translocations (35.59%), inversions (9.98%), and deletions (5.09%). B KEGG pathway enrichment analysis of cancer-associated genes involved in translocational fusions demonstrates predominant enrichment in Hippo, AMPK, and PI3K-Akt signaling pathways. C Circos plot highlighting recurrent fusion partners (≥ 3 events), with PTRF participating in four fusion events and FBLN1 , ITCH , STAT3 , SMAD3 , EPB41L5 , and KAZN each involved in three distinct fusions. D Schematic representation of SMAD3 fusion architectures: PTRH1 (exon1)- SMAD3 (exon2), GTF2H5 (exon2)- SMAD3 (exon2), and SMAD3 (exon1)- KAZN (exon2). E EPB41L5 fusion configurations: EPB41L5 (exon3)- PTRF (exon2) and EPB41L5 (exon7)- LTBP3 (exon15). F FGFR2 (exon17)- RPAP3 (exon9) chimeric transcript structure, illustrating the precise breakpoint junction between partner genes Fusion gene landscape in OCCC. A Pie chart summarizing genomic rearrangement types underlying fusion events: duplications (49.34%), translocations (35.59%), inversions (9.98%), and deletions (5.09%). B KEGG pathway enrichment analysis of cancer-associated genes involved in translocational fusions demonstrates predominant enrichment in Hippo, AMPK, and PI3K-Akt signaling pathways. C Circos plot highlighting recurrent fusion partners (≥ 3 events), with PTRF participating in four fusion events and FBLN1 , ITCH , STAT3 , SMAD3 , EPB41L5 , and KAZN each involved in three distinct fusions. D Schematic representation of SMAD3 fusion architectures: PTRH1 (exon1)- SMAD3 (exon2), GTF2H5 (exon2)- SMAD3 (exon2), and SMAD3 (exon1)- KAZN (exon2). E EPB41L5 fusion configurations: EPB41L5 (exon3)- PTRF (exon2) and EPB41L5 (exon7)- LTBP3 (exon15). F FGFR2 (exon17)- RPAP3 (exon9) chimeric transcript structure, illustrating the precise breakpoint junction between partner genes Of particular clinical significance, we identified a novel FGFR2-RPAP3 fusion in an OCCC patient (Fig.  3 F), an event previously unreported in large-scale cohorts of HGSOC. GSEA revealed significant enrichment of the mTORC1 signaling pathway and G2M checkpoint in the FGFR2-RPAP3 tumor, suggesting that this fusion event may drive aberrant cell cycle progression and metabolic reprogramming (Additional file 1: Fig. S4B). To definitively validate this chimera, we first employed the IGV to visualize the junction site, confirming a precise RNA splicing event linking exon 17 of FGFR2 to exon 9 of RPAP3 (Additional file 1: Fig. S4C). This finding was further corroborated by PCR amplification and Sanger sequencing, providing conclusive evidence for the fusion’s authenticity (Additional file 1: Fig. S4D). This finding holds therapeutic implications given the FDA approval and remarkable clinical success of FGFR2 -targeted inhibitors like pemigatinib in cholangiocarcinoma, suggesting potential therapeutic benefit for this OCCC case [ 42 ]. Collectively, our identification of oncogenic fusion events, including the novel FGFR2 fusion, provides valuable insights into potential therapeutic targets and strategies for OCCC treatment. Recent clinical studies have revealed that a subset of OCCC patients exhibit strong and durable responses to immune checkpoint inhibitors (ICIs) [ 5 ]. Further exploration of the immune microenvironment heterogeneity and its molecular characteristics of OCCC could advance immunocombination-targeted therapies and improve clinical outcomes. To address this, we first performed ssGSEA to map protein-coding gene expression profiles to immune-related pathways, with ssGSEA scores normalized using z-scores. Based on immune activity levels, samples were stratified into high- and low-immunity groups (Fig.  4 A). Differential molecular characterization revealed more frequent ARID1A mutations in the low-immunity group, while epigenetic regulators such as SMARCA4 , SETD1B , EP300 , and SETD2 showed higher mutation rates in the high-immunity group, suggesting that targeting these epigenetic modifiers may enhance immunotherapy efficacy in high-immunity group OCCC patients (Fig.  4 B). Fig. 4 Tumor immune microenvironment characterization in OCCC. A Hierarchically clustered heatmap (ward.D method) of immune pathway activity using z-score normalized ssGSEA scores (blue-to-red gradient indicating low-to-high enrichment), with top dendrogram showing sample clustering and color bars designating immune_high versus immune_low subgroups exhibiting distinct immune pathway activation patterns. B The top differentially mutated genes between immune_high and immune_low subtype. C Volcano plot visualizing significant DEGs between the immune_high and immune_low groups. Red and blue dots represent significantly up-regulated and down-regulated genes, respectively, based on the thresholds of |log_2 FC| > 1 and adj.P.Val < 0.05. D t-SNE visualization of cellular composition across all samples (left), immune_high (middle), and immune_low (right) subgroups, with bar plot (far right) quantifying macrophage (green) and T cell (pink) proportions demonstrating significant enrichment in immune_high cases. E Intercellular interaction strength heatmap (blue-to-red gradient representing weak-to-strong interactions) revealing enhanced epithelial-T cell crosstalk. F Circular plots depicting upregulated stromal-immune crosstalk pathways in immune_high group: left panel shows epithelium-to-immune cell signaling, right panel illustrates reciprocal immune-to-epithelium pathways Tumor immune microenvironment characterization in OCCC. A Hierarchically clustered heatmap (ward.D method) of immune pathway activity using z-score normalized ssGSEA scores (blue-to-red gradient indicating low-to-high enrichment), with top dendrogram showing sample clustering and color bars designating immune_high versus immune_low subgroups exhibiting distinct immune pathway activation patterns. B The top differentially mutated genes between immune_high and immune_low subtype. C Volcano plot visualizing significant DEGs between the immune_high and immune_low groups. Red and blue dots represent significantly up-regulated and down-regulated genes, respectively, based on the thresholds of |log_2 FC| > 1 and adj.P.Val < 0.05. D t-SNE visualization of cellular composition across all samples (left), immune_high (middle), and immune_low (right) subgroups, with bar plot (far right) quantifying macrophage (green) and T cell (pink) proportions demonstrating significant enrichment in immune_high cases. E Intercellular interaction strength heatmap (blue-to-red gradient representing weak-to-strong interactions) revealing enhanced epithelial-T cell crosstalk. F Circular plots depicting upregulated stromal-immune crosstalk pathways in immune_high group: left panel shows epithelium-to-immune cell signaling, right panel illustrates reciprocal immune-to-epithelium pathways To delineate the molecular differences between the identified immune subtypes, we first identified a robust set of 1,235 overlapping DEGs via the intersection of three independent algorithms (Fig.  4 C, Additional file 2: Table S1-2). Functional enrichment analysis revealed that genes upregulated in the immune_high group were predominantly involved in immune response regulation, cytokine-cytokine receptor interactions, the PD-L1/PD-1 checkpoint pathway and TNF-alpha signaling via NF-kappaB pathway (Additional file 1: Fig. S5A-D). While these tumors exhibit a “hot” microenvironment with robust activation, they also harbor significant inhibitory feedback, suggesting that the immune_high subtype may represent a prime candidate for ICI therapy. Specifically, this group exhibited significant overexpression of cytotoxic markers ( GZMB , PRF1 ), immune checkpoints ( CD274 , PDCD1 ), and recruitment factors ( CXCL9 , CCL5 ; Additional file 1: Fig. S5E). To validate these transcriptomic signatures and gain deeper mechanistic insights, we selected a representative subset of 13 samples for integrated multiplex immunofluorescence (mIF) and scRNA-seq analysis. These samples were strategically stratified to capture the full biological spectrum of our cohort, balancing clinical stages (5 early vs. 8 advanced), genomic drivers (including 8 ARID1A -mutant cases), and immune landscapes (7 immune-high vs. 6 immune-low) (Additional file 1: Fig. S6A; Additional file 2: Table S3). Consistent with bulk-level predictions (Additional file 1: Fig. S5F; Additional file 2: Table S3), mIF staining of these 13 samples demonstrated a significantly higher density of infiltrating CD4 + T cells and cytotoxic NK cells in the immune-high subgroup (Additional file 1: Fig. S5G-H). Simultaneously, scRNA-seq analysis at single-cell resolution corroborated these findings, revealing richer infiltration of macrophages, T cells, and B cells (Fig.  4 D). Beyond cellular density, scRNA-seq identified intensified epithelial-immune cell crosstalk in the immune-high group (Fig.  4 E), primarily mediated by CD44–CD74 receptor-ligand pairs and the SPP1 signaling axis (Fig.  4 F). Furthermore, enhanced MHC class I/II pathways (e.g., HLA-A, HLA-DRA) indicated a superior antigen presentation capacity in this subtype (Additional file 1: Fig. S6B-E). To evaluate the generalizability of our immune subtyping, we performed external validation using two independent datasets ( GSE73614 and GSE65986 ; Additional file 2: Table S4) [ 18 , 19 ]. ssGSEA-based profiling confirmed that the immune-high and immune-low patterns were highly reproducible in both cohorts (Additional file 1: Fig. S7A, D). Although the multivariate survival analysis did not reach statistical significance ( P  > 0.05; Additional file 1: Fig. S7B-F)—likely due to the constrained sample size and limited number of clinical events—the consistency of the immune landscapes underscores the robustness of our classification. Collectively, our study provides a rationale for optimizing personalized immunotherapy strategies based on the structural and molecular heterogeneity of OCCC. To comprehensively delineate the molecular trajectories underlying OCCC evolution, we first interrogated the transcriptomic divergence and immune microenvironment landscape between early- and advanced-stage tumors. 92 robust DEGs were identified via the intersection of limma, DESeq2, and edgeR (Additional file 1: Fig. S8A-B), CIBERSORT-based deconvolution revealed largely comparable immune cell infiltration patterns across FIGO stages (Additional file 1: Fig. S8C-D). These high-level transcriptomic similarities suggested that the macroscopic immune architecture remains relatively conserved during OCCC progression, prompting us to further investigate the underlying genomic drivers of malignancy. Consequently, we shifted our focus to the genomic landscape. Comparative analysis of genomic alterations between early- and advanced-stage OCCC not only delineates molecular trajectories underlying tumor evolution but also enables identification of critical mutational drivers, thereby informing precision diagnostics and guiding personalized therapeutic strategies. Here, we compared and analyzed the mutation differences between early-stage and advanced-stage samples, finding that PIK3CA mutations were more frequent in advanced-stage samples, whereas ARHGAP36 mutations were more common in early-stage samples. Notably, FOXA2 mutations were exclusively present in advanced-stage samples (19%, Fig.  5 A-C) and primarily exhibited missense mutations and frameshift deletions (Fig.  5 D). Mutations in FOXA2 were significantly associated with poor overall survival in OCCC patients (Fig.  5 E; Additional file 1: Fig. S2A-B). Fig. 5 Differential mutational landscapes between early- and late-stage OCCC. A Comparative oncoplots of early-stage versus advanced tumors showing differential mutation frequencies in PIK3CA , FOXA2 , and ARHGAP36 . B Stage-specific mutational patterns with ARHGAP36 mutations exclusive to early-stage OCCC and FOXA2 alterations restricted to advanced cases. C Forest plot illustrating differential gene mutations between stages using odds ratios (OR), with significance levels denoted. D Lollipop plot mapping FOXA2 amino acid alterations, revealing mutation clustering within the FH domain including recurrent S216L substitutions in two patients. E Kaplan-Meier overall survival analysis based on FOXA2 mutation status. The plot compares survival outcomes between FOXA2 -mutated and wild-type cases, with the log-rank P-value (0.017) indicated. Results from a multivariate Cox proportional hazards model are also provided ( P  = 0.04), including the hazard ratio (HR = 5.47) and 95% confidence interval (95% CI: 1.08–27.6), after adjusting for clinical covariates. F ChIP-seq in ES2 cells showing impaired chromatin binding capacity of FOXA2 S216L mutant compared to wild-type. G-H . Functional validation via CCK-8 proliferation assay G and colony formation assay H in ES2, TOV21G and SKOV3 cell lines, comparing empty vector, wild-type FOXA2 , and S216L mutant overexpression. Values represent the mean ± SEM. * P  < 0.05, ** P  < 0.01, *** P  < 0.001. One-way ANOVA followed by Tukey’s post hoc test for multiple comparisons was uesd in Fig. 5G-H Differential mutational landscapes between early- and late-stage OCCC. A Comparative oncoplots of early-stage versus advanced tumors showing differential mutation frequencies in PIK3CA , FOXA2 , and ARHGAP36 . B Stage-specific mutational patterns with ARHGAP36 mutations exclusive to early-stage OCCC and FOXA2 alterations restricted to advanced cases. C Forest plot illustrating differential gene mutations between stages using odds ratios (OR), with significance levels denoted. D Lollipop plot mapping FOXA2 amino acid alterations, revealing mutation clustering within the FH domain including recurrent S216L substitutions in two patients. E Kaplan-Meier overall survival analysis based on FOXA2 mutation status. The plot compares survival outcomes between FOXA2 -mutated and wild-type cases, with the log-rank P-value (0.017) indicated. Results from a multivariate Cox proportional hazards model are also provided ( P  = 0.04), including the hazard ratio (HR = 5.47) and 95% confidence interval (95% CI: 1.08–27.6), after adjusting for clinical covariates. F ChIP-seq in ES2 cells showing impaired chromatin binding capacity of FOXA2 S216L mutant compared to wild-type. G-H . Functional validation via CCK-8 proliferation assay G and colony formation assay H in ES2, TOV21G and SKOV3 cell lines, comparing empty vector, wild-type FOXA2 , and S216L mutant overexpression. Values represent the mean ± SEM. * P  < 0.05, ** P  < 0.01, *** P  < 0.001. One-way ANOVA followed by Tukey’s post hoc test for multiple comparisons was uesd in Fig. 5G-H FOXA2, a member of the forkhead (FH) box protein family containing a conserved DNA-binding domain, functions as a pioneer transcription factor by preferentially binding chromatin and recruiting chromatin remodeling complexes or transcriptional co-regulators to orchestrate downstream target gene transcription [ 43 ]. The endometrial cancer (EC) genomic profiling project revealed that endometrioid adenocarcinomas (the predominant histological subtype of EC) exhibit a FOXA2 mutation frequency of approximately 9%, with mutations distributed throughout the coding region, suggesting its potential tumor-suppressive role in endometrial carcinogenesis through loss-of-function mechanisms [ 44 ]. Our data demonstrate an even higher overall FOXA2 mutation frequency in OCCC, with recurrent mutations clustering within the DNA-binding FH domain (Fig.  5 D). Notably, a shared serine-to-leucine substitution at position 216 (S216L) of FH domain was identified in two OCCC patients. To elucidate FOXA2 ’s biological function in OCCC, we generated wild-type and S216L mutant constructs. ChIP-seq assays demonstrated substantial chromatin binding capacity of wild-type FOXA2 in OCCC cells, while the S216L mutant exhibited markedly reduced chromatin association (Fig.  5 F). In vitro functional analyses revealed that ectopic overexpression of wild-type FOXA2 in FOXA2-null TOV21G, SKOV3 and ES2 cells with wild-type FOXA2 high expression significantly suppressed OCCC cell growth, whereas the S216L mutant displayed no inhibitory effects (Fig.  5 G-H; Additional file 1: Fig. S9A-D). We also designed specific siRNAs targeting FOXA2 and found knockdown of FOXA2 in RMG1 and ES2 cells enhanced cell growth (Additional file 1: Fig. S9E-H). Concurrently, rescue experiments in FOXA2 -knockdown ES2 cells demonstrated that while the re-introduction of wild-type FOXA2 reversed the pro-tumorigenic effects of the knockdown, the FOXA2-S216L mutant failed to exhibit this restorative capacity (Additional file 1: Fig. S9G-H), collectively confirming that the S216L substitution represents a loss-of-function mutation compromising FOXA2 tumor-suppressive activity. Investigating downstream signaling, ChIP-seq analysis revealed wild-type FOXA2 enrichment primarily in intronic, intergenic, and promoter-TSS regions (Additional file 1: Fig. S10A-B), while integration with RNA-seq data indicated that potential FOXA2 targets are enriched in pathways such as cell-substrate adhesion and negative regulation of growth (Additional file 1: Fig. S10C-E). Moreover, wild-type FOXA2 forms an epigenetic regulatory complex with SWI/SNF core proteins (e.g., BRG1, BAF170), whereas the mutant’s binding affinity was significantly diminished, suggesting that FOXA2-mediated regulation relies on both DNA binding integrity and cooperative modulation via epigenetic complexes (Additional file 1: Fig. S10F). In summary, these findings further underscore the critical role of epigenetic aberrations in OCCC pathogenesis and malignant progression, wherein FOXA2 mutations emerge as advanced-stage OCCC-specific molecular events with biomarker potential. Specifically, the S216L substitution disrupts both FOXA2 chromatin binding and its interaction with SWI/SNF protein complexes, thereby disabling its tumor-suppressive function in restraining malignant progression in OCCC. Recent advances in therapeutic subtyping have revolutionized precision oncology in HGSOC, fundamentally transforming clinical management paradigms. However, the current lack of systematic molecular classification in OCCC severely limits its clinical management and therapeutic efficacy. To reveal this, we firstly integrated the molecular features including TMB, HRD and immune signature and theraputic genes expression and fusion to establish the “targeted therapeutic subtyping” framework in OCCC. Genomic analysis revealed that approximately 70% of OCCC patients displayed concurrent molecular features including elevated TMB, HRD, enhanced immune infiltration scores, ERBB2 , FGFR1-3 overexpression, and FGFR2 fusion events (Fig.  6 A), indicating potential novel targeted therapeutic approaches for this subgroup [ 5 , 40 , 42 , 45 ]. Given the proven efficacy of ERBB2 -targeted therapies in other cancers and the absence of reported outcomes in OCCC, we prioritized investigating the characteristics and therapeutic potential of ERBB2 in this malignancy. We classified samples with high ERBB2 expression (TPM > 100) or ERBB2 amplification (CNV > 0.3) as the ERBB2  + group, while the remaining samples were assigned to the control group. Our analysis revealed a significant correlation between ERBB2 overexpression and amplification ( P  < 0.01, χ² test; Fig.  6 B, Additional file 1: Fig. S11A-B). IHC profiling also identified HER2 expression in almost 70% of OCCC cases, including 42.7% with moderate-to-high HER2 expression, which correlated with aggressive clinicopathological features and dismal prognosis (Fig.  6 C-D; Additional file 1: Fig. S2B). Fig. 6 Therapeutic targeting of ERBB2 in OCCC. A Integrative molecular subtyping of OCCC reveals that approximately 70% of OCCC patients exhibit high immune, high HRD and high TMB scores along with overexpression of ERBB2 , FGFR1-3 and FGFR2 fusions, suggesting this subgroup may benefit from potential targeted therapies. B Mutational waterfall plot of ERBB2  + samples with top bar plot showing ERBB2 expression levels (TPM), annotated by immune subtype, clinical stage, and copy number amplification status (* indicates samples with ERBB2 amplification). C Differential HER2 protein expression scores across OCCC specimens. D Prognostic survival analysis stratified by high versus low HER2 protein expression in OCCC tumors. E Chemical structure of HER2-targeting antibody-drug conjugate NC18. F Histopathological evaluation via H&E staining and HER2 IHC in HER2-expressing OCCC PDX. G-H Significant tumor growth inhibition in HER2 + PDX models (G; n  = 8 per group) and TOV21G cells (H; n  = 7 per group) treated with NC18 (2 mg/kg, two intraperitoneal doses), and quantification of NC18 therapeutic effects showing reduced tumor volume compared to controls. I The anti-tumor effect of NC18 (2 mg/kg, two intraperitoneal doses) in HER2-negative OOCC-PDX models ( n  = 6 per group). J Retrospective review of two OCCC patients treated with NC18: Case 1 (HER2 2+, platinum-resistant recurrent disease) exhibited progressive inguinal lymph node regression and concurrent CA125 decline to upper limit of normal (ULN) post-treatment; Case 2 (HER2 1+, multiply relapsed metastatic disease) showed splenic lesion reduction and sustained CA125 decrease, with both patients achieving PR per RECIST v1.1 criteria. Values represent the mean ± SEM. *** P  < 0.001. Multiple t-tests (one per row) was used to rigorously assess the differences between treatment and control groups at each time point (Fig. 6G-I), explicitly annotating the P-values at the study endpoint. Scale bar: 50 μm Therapeutic targeting of ERBB2 in OCCC. A Integrative molecular subtyping of OCCC reveals that approximately 70% of OCCC patients exhibit high immune, high HRD and high TMB scores along with overexpression of ERBB2 , FGFR1-3 and FGFR2 fusions, suggesting this subgroup may benefit from potential targeted therapies. B Mutational waterfall plot of ERBB2  + samples with top bar plot showing ERBB2 expression levels (TPM), annotated by immune subtype, clinical stage, and copy number amplification status (* indicates samples with ERBB2 amplification). C Differential HER2 protein expression scores across OCCC specimens. D Prognostic survival analysis stratified by high versus low HER2 protein expression in OCCC tumors. E Chemical structure of HER2-targeting antibody-drug conjugate NC18. F Histopathological evaluation via H&E staining and HER2 IHC in HER2-expressing OCCC PDX. G-H Significant tumor growth inhibition in HER2 + PDX models (G; n  = 8 per group) and TOV21G cells (H; n  = 7 per group) treated with NC18 (2 mg/kg, two intraperitoneal doses), and quantification of NC18 therapeutic effects showing reduced tumor volume compared to controls. I The anti-tumor effect of NC18 (2 mg/kg, two intraperitoneal doses) in HER2-negative OOCC-PDX models ( n  = 6 per group). J Retrospective review of two OCCC patients treated with NC18: Case 1 (HER2 2+, platinum-resistant recurrent disease) exhibited progressive inguinal lymph node regression and concurrent CA125 decline to upper limit of normal (ULN) post-treatment; Case 2 (HER2 1+, multiply relapsed metastatic disease) showed splenic lesion reduction and sustained CA125 decrease, with both patients achieving PR per RECIST v1.1 criteria. Values represent the mean ± SEM. *** P  < 0.001. Multiple t-tests (one per row) was used to rigorously assess the differences between treatment and control groups at each time point (Fig. 6G-I), explicitly annotating the P-values at the study endpoint. Scale bar: 50 μm Leveraging the clinical success of HER2-ADCs in cancer, we investigated NC18—a novel HER2-ADC comprising trastuzumab biosimilar HLX02 conjugated to the microtubule inhibitor auristatin F-hydroxyethoxyacetamide (AF-HEA) via a polymer-based linker (Fig.  6 E). In vivo evaluation using four OCCC PDX models identified a HER2-hyperexpressing PDX (Fig.  6 F) with accelerated proliferative kinetics. NC18 administration induced significant tumor growth suppression ( P  < 0.001, Fig.  6 G), accompanied by DNA damage and apoptosis (Additional file 1: Fig. S11D). Concurrently, efficacy evaluations in TOV21G cells (characterized by low-to-moderate HER2 expression) and a HER2-negative OCCC-PDX revealed that NC18 significantly inhibited subcutaneous tumor growth in the TOV21G model but exerted no discernible suppressive effect on the HER2-negative PDX (Fig.  6 H-I; Additional file 1: Fig. S11C), thereby indicating potential therapeutic benefits for HER2-low OCCC patients while confirming the specificity of its anti-tumor activity. Meanwhile, we investigated whether NC18 triggers immunogenic cell death (ICD). In both TOV21G and SKOV3 cells, NC18 treatment significantly induced the translocation and enrichment of calreticulin on the cell surface (Additional file 1: Fig. S11E), accompanied by the increased accumulation of extracellular ATP and HMGB1 (Additional file 1: Fig. S11F-G) in the culture supernatants. These findings collectively demonstrate that NC18 elicits ICD in OCCC cells, a process that may activate the host immune system and potentially enhance the therapeutic efficacy of immunotherapy in patients with OCCC. To provide preliminary clinical observations on HER2-ADCs in OCCC, we retrospectively reviewed two OCCC patients treated with NC18. Notably, the NC18 inhibitor demonstrated clinical benefits in both recurrent and metastatic OCCC populations. Case 1 involved a patient with platinum-resistant recurrent OCCC exhibiting HER2 2 + expression. Following two cycles of HER2-ADC therapy, serial imaging revealed progressive volumetric regression of inguinal lymph nodes, accompanied by a synchronous decline in tumor marker CA125 to the upper limit of normal (ULN) (Fig.  6 J). Case 2 presented with multiply relapsed metastatic OCCC showing HER2 1 + expression. Comparable therapeutic outcomes were observed: splenic lesions demonstrated gradual dimensional reduction after two treatment cycles, paralleled by a sustained downward trajectory in CA125 levels (Fig.  6 J). Both patients achieved partial response (PR) per RECIST v1.1 criteria post-intervention. These findings collectively suggest that HER2-ADC agents may represent a novel therapeutic strategy for HER2-expressing OCCC.

OCCC is a distinct histological subtype of epithelial ovarian cancer, which accounts for approximately 10% of ovarian malignancies. Originating from the Müllerian ducts, OCCC is characterized by tumor cells with clear or eosinophilic cytoplasm, hobnail morphology, and architectural patterns such as tubulocystic, papillary, or solid structures [ 1 ]. Notably, OCCC exhibits the association with endometriosis (EM) but is genetically distinct from high-grade serous ovarian carcinoma (HGSOC), as it is not linked to BRCA1/2 mutations [ 2 , 3 ]. According to data from the International Agency for Research on Cancer (IARC), the global annual incidence of OCCC ranges between 15,000 and 30,000 cases, with a higher prevalence in East Asian populations and a trend toward younger age of onset [ 4 ]. Currently, surgical resection and postoperative adjuvant chemotherapy remain the standard clinical treatments for OCCC [ 1 , 5 ]. Approximately 50%-70% of OCCC patients are diagnosed at FIGO stage I, a proportion significantly higher than that of HGSOC (typically diagnosed at advanced stages). Complete surgical resection (R0) in early-stage OCCC correlates with favorable prognosis. However, advanced-stage OCCC (FIGO III/IV, with peritoneal or distant metastases) demonstrates primary resistance to platinum-based chemotherapy. Compared to HGSOC, OCCC exhibits more pervasive platinum resistance and lacks effective alternative therapies [ 6 ]. The objective response rate (ORR) to second-line chemotherapy in recurrent OCCC is extremely low, with a 5-year survival rate below 30%, reflecting a dismal prognosis. Critically, unlike HGSOC—where targeted therapies such as PARP inhibitors (e.g., olaparib) and bevacizumab have achieved significant breakthroughs—OCCC currently lacks approved targeted treatments, posing substantial challenges in clinical management and therapeutic efficacy. In recent years, high-throughput sequencing has revolutionized the molecular characterization of cancers and transformed clinical approaches, particularly in breast and lung cancers, where molecular subtyping enables precision targeted therapies and prognostic prediction [ 7 , 8 ]. For HGSOC, extensive genomic data from WES and transcriptomic analyses have elucidated molecular features driven by high-frequency TP53 and BRCA1/2 mutations, including homologous recombination deficiency (HRD), which has facilitated the clinical translation of target therapies such as PARP inhibitors [ 9 , 10 ]. In contrast, the multi-omics research on OCCC remains insufficient, with a severe paucity of systematic molecular profiling and targeted therapy-oriented subtyping. Over the past decade, only a limited number of OCCC cases have undergone base-level analyses via WES [ 11 – 16 ]. For instance, Itamochi et al. conducted WES on 55 OCCC samples, delineating mutation profiles in Japanese patients [ 16 ]. Although existing research has identified recurrent mutations (e.g., ARID1A/PIK3CA ) and preliminary molecular subtypes, these findings have not yet been translated into clinically actionable tools. Meanwhile, sporadic clinical reports suggest that a subset of OCCC patients exhibit robust and durable responses to immune checkpoint inhibitors [ 5 , 17 ]; however, the heterogeneity of the tumor microenvironment (TME) and associated molecular signatures remain poorly characterized. Therefore, a deeper understanding of OCCC’s molecular landscape is essential to transition from conventional chemotherapy to targeted therapies, immunotherapies, and rational combinatorial strategies, ultimately improving clinical outcomes. In the present study, we performed WES, RNA sequencing, and scRNA-seq on 82 OCCC cases, systematically constructing the comprehensive genomic and transcriptomic atlas of OCCC. We identified key molecular aberrations driving malignant progression (serving as prognostic biomarkers) and TME features. Additionally, we uncovered potentially therapeutic vulnerabilities across genetic mutations, gene fusions, chromosomal copy number variations (CNVs), and epitranscriptomic dysregulation, establishing a targeted therapeutic subtyping framework predictive of therapy efficacy. This classification system was preliminarily supported by a retrospective review of two HER2-ADC-treated OCCC cases, demonstrating its potential to guide precision treatment strategies in OCCC.

Current molecular databases including The Cancer Genome Atlas (TCGA) lack comprehensive multi-omics datasets for OCCC. Compared to our study, previous investigations exhibit notable limitations. For example, Kim et al. compared genomic features between endometriosis-associated and non-endometriosis-associated OCCC, but their analysis was constrained by limited sample size ( n  = 32) and lacked transcriptomic sequencing or single-cell RNA profiling [ 13 ]. Ye and Yang et al. characterized genomic alterations in OCCC using formalin-fixed paraffin-embedded (FFPE) tumor tissues and peripheral blood samples, though FFPE-derived specimens may introduce artifactual mutations [ 15 , 46 ]. Itamochi et al. focused on mutational profiles of Japanese OCCC patients but lack molecular subtyping framework for predicting targeted therapy efficacy [ 16 ]. By establishing the most comprehensive OCCC cohort to date, this study systematically elucidates the genomic architecture, transcriptomic programs, and therapeutic vulnerabilities of the disease through integrated multi-omics analysis. This resource constitutes an unprecedented molecular landscape that not only deepens our understanding of OCCC pathobiology but also grounds our proposal of a novel precision diagnostic and treatment paradigm termed “targeted therapeutic subtyping” (Fig.  7 ), the potential translational feasibility of which was substantiated through pharmacological validation in PDX models. Fig. 7 The framework of targeted therapeutic subtyping in OCCC. “targeted therapeutic subtyping” workflow is predicated on a “hierarchical and combinatorial” therapeutic algorithm designed to resolve therapeutic conflicts arising from overlapping targets. Specifically, within this strategic framework, a hierarchical stratification based on multi-omic sequencing was conducted, assigning the highest priority to rare mutational events with strong oncogenic driving force—particularly FGFR2 fusions such as FGFR2-RPAP3 —for which we recommend exclusive targeted monotherapy with agents like pemigatinib or FGFR inhibitor combined with other targeted drugs. For patients lacking such dominant driver genes, we proceed to a secondary combinatorial decision phase contingent upon the TME status: In immune-high phenotype OCCC populations, for cases exhibiting moderate-to-high HER2 expression, we recommend combination therapy with HER2-ADC and ICIs; for those with ARID1A or PIK3CA mutations, we recommend anti-VEGF/TKI combined with ICIs; and for cases demonstrating high FGFR1/2/3 expression, we recommend FGFR inhibitors combined with ICIs. In immune-low OCCC populations, when HER2-high expression co-occurs with HRD-high status or BRCA1/2 mutations, HER2-ADC combined with PARPi may be considered; for HER2-negative cases presenting high FGFR1/2/3 expression alongside ARID1A/PIK3CA mutations, we recommend TKI-based combination therapy. For the remaining OCCC patients exhibiting only a single molecular feature, we favor standard treatment approaches—such as HER2-ADC monotherapy for isolated HER2-high expression, or chemotherapy combined with targeted agents where clinically indicated. Note: clinical decision-making of HER2 expression should be driven primarily by IHC (with RNA measures as supportive), and our “targeted therapeutic subtyping” workflow proposed in this study remains in the preclinical exploratory and theoretical speculation stage and is not yet sufficient to immediately guide clinical treatment decisions for OCCC The framework of targeted therapeutic subtyping in OCCC. “targeted therapeutic subtyping” workflow is predicated on a “hierarchical and combinatorial” therapeutic algorithm designed to resolve therapeutic conflicts arising from overlapping targets. Specifically, within this strategic framework, a hierarchical stratification based on multi-omic sequencing was conducted, assigning the highest priority to rare mutational events with strong oncogenic driving force—particularly FGFR2 fusions such as FGFR2-RPAP3 —for which we recommend exclusive targeted monotherapy with agents like pemigatinib or FGFR inhibitor combined with other targeted drugs. For patients lacking such dominant driver genes, we proceed to a secondary combinatorial decision phase contingent upon the TME status: In immune-high phenotype OCCC populations, for cases exhibiting moderate-to-high HER2 expression, we recommend combination therapy with HER2-ADC and ICIs; for those with ARID1A or PIK3CA mutations, we recommend anti-VEGF/TKI combined with ICIs; and for cases demonstrating high FGFR1/2/3 expression, we recommend FGFR inhibitors combined with ICIs. In immune-low OCCC populations, when HER2-high expression co-occurs with HRD-high status or BRCA1/2 mutations, HER2-ADC combined with PARPi may be considered; for HER2-negative cases presenting high FGFR1/2/3 expression alongside ARID1A/PIK3CA mutations, we recommend TKI-based combination therapy. For the remaining OCCC patients exhibiting only a single molecular feature, we favor standard treatment approaches—such as HER2-ADC monotherapy for isolated HER2-high expression, or chemotherapy combined with targeted agents where clinically indicated. Note: clinical decision-making of HER2 expression should be driven primarily by IHC (with RNA measures as supportive), and our “targeted therapeutic subtyping” workflow proposed in this study remains in the preclinical exploratory and theoretical speculation stage and is not yet sufficient to immediately guide clinical treatment decisions for OCCC Our study reveals distinct mutational landscapes between OCCC and HGSOC. OCCC is characterized by high-frequency ARID1A (64%) and PIK3CA mutations (55%), with these mutation rates significantly exceeding those reported in Korean and Japanese OCCC cohorts. In contrast, HGSOC demonstrates near-ubiquitous TP53 mutations (> 90%) as its central driver event, alongside more frequent germline or somatic mutations in homologous recombination repair genes including BRCA1/2 [ 9 ]. Notably, we observed a 13% KRAS mutation rate in OCCC—a finding corroborated in East Asian cohorts—suggesting KRAS activation may represent an early-stage event, whereas KRAS alterations in HGSOC occur at lower frequencies [ 13 , 16 ]. Intriguingly, chromatin remodeling complex components ( ARID1A , KMT2C , SMARCA4 ) exhibit higher mutation prevalence in OCCC compared to HGSOC, with pathway enrichment analyses revealing prominent epigenetic dysregulation. Collectively, OCCC’s molecular profile favors PI3K pathway aberrations and chromatin remodeling dysregulation, whereas HGSOC is dominated by genomic instability and homologous recombination defects. Supporting our findings, Fukumoto et al. demonstrated enhanced therapeutic efficacy through combined HDAC6 inhibition and immune checkpoint blockade in ARID1A -mutant OCCC, underscoring epigenetic targeting’s clinical potential [ 47 ]. More importantly, emerging studies have consistently reported that PIK3CA mutation-targeting agents such as inavolisib demonstrate promising therapeutic efficacy in advanced/metastatic breast cancer harboring PIK3CA mutations with HER2-negative status [ 29 , 30 ]. Subsequent prospective clinical trials should be expedited in OCCC to validate their potential clnical benefits. Our investigation identifies somatic FOXA2 mutations as exclusive participants of advanced-stage OCCC (19%), correlating with poor prognosis—an association distinct from EC [ 48 ]. These mutations predominantly cluster in the forkhead domain. Crucially, we demonstrate that the recurrent S216L variant abrogates tumor suppression via a dual mechanism: impairing DNA-binding capacity and drastically disrupting interaction with the SWI/SNF chromatin-remodeling complex. We propose that wild-type FOXA2 functions as a lineage-specifying tumor suppressor, maintaining the Müllerian differentiation program inherent to OCCC’s endometriotic origin to upregulate cell adhesion and inhibit epithelial-mesenchymal transition. In advanced stages, mutational inactivation bypasses this restraint, facilitating dedifferentiation and peritoneal metastasis. This aligns with mechanisms in pancreatic cancer [ 49 ], where FOXA2 loss drives invasive potential. Collectively, these findings establish FOXA2 mutations as critical biomarkers for progression risk, revealing how the disruption of lineage-specific transcriptional and epigenetic control accelerates OCCC malignancy. At present, there are no approved targeted therapeutic regimens for OCCC. The exploration of targeted therapeutic strategies primarily encompasses inhibitors directed against driver mutations/signaling pathways, immune checkpoint inhibitors, and ADCs. However, all these modalities remain investigational and confined to clinical trial stages, underscoring a critical need to discover novel targeted therapeutic modalities with transformative potential for clinical translation. HER2-directed ADCs have been approved and demonstrated compelling clinical efficacy across malignancies [ 50 , 51 ]. Though HER2-ADCs remain unapproved in ovarian cancer, early-phase trials have reported promising efficacy [ 52 ]. During molecular characterization of OCCC, we found that HER2 expression is common in this malignancy. Functional studies using HER2-expressing PDX models and OCCC cell lines, together with retrospective clinical observations from two NC18-treated OCCC patients, provided preliminary evidence supporting the therapeutic potential of HER2-ADCs in OCCC and a rationale for further clinical evaluation of HER2-targeted strategies. Importantly, our preclinical data showed that HER2-ADCs exerted marked anti-tumor activity and specificity even in OCCC models with low-to-moderate HER2 expression, thereby potentially broadening the population that may benefit from this therapeutic approach. Future studies should prioritize prospective evaluation of HER2-ADCs, alone or in combination with other targeted therapies, in molecularly defined OCCC subgroups to advance precision oncology. In recent years, molecular classification-guided targeted therapy has achieved groundbreaking progress in tumors such as breast cancer and HGSOC, reshaping clinical therapeutic paradigms. However, a systematic molecular classification of OCCC based on targetable pathways remains largely unexplored. To date, only Ye et al. preliminarily categorized OCCC into high-immunity and low-immunity subgroups and investigated their prognostic relevance [ 15 ]. Although Chandler et al. demonstrated that ARID1A/PIK3CA co-mutations drive OCCC pathogenesis and represent a critical molecular subtype, no approved therapies specifically targeting this subgroup currently exist [ 53 ]. In this study, we first propose a novel framework termed “targeted therapeutic subtyping” for stratifying OCCC patients into subgroups with potential therapeutic vulnerabilities, encompassing high-immunity, elevated TMB, HRD scores, ERBB2 , FGFR1-3 overexpression, and FGFR2 fusion events. Notably, clinically approved targeted therapies—including immune checkpoint inhibitors for high-immunity subgroups, HER2-ADCs, pemigatinib ( FGFR2 fusion inhibitor), olaparib (PARP inhibitor for HRD OCCC patients) and lenvatinib (for FGFR1-3 overexpression OCCC patients)—may provide viable therapeutic options for these populations [ 5 , 42 , 45 , 50 , 54 ]. Under this framework, rare yet potent oncogenic drivers, specifically FGFR2 fusions, are assigned the highest hierarchical priority due to oncogene addiction, mandating tyrosine kinase inhibitor (TKI) monotherapy targeting this driver (e.g., pemigatinib), with or without combination agents against co-occurring oncogenic targets, irrespective of concurrent molecular features. For overlapping molecular alterations lacking such target exclusivity, we advocate mechanism-based combination therapies: for instance, combining HER2-ADCs with ICIs in cases with concurrent HER2 overexpression and high immune signatures to harness ADC-induced ICD, or deploying HER2-ADCs with PARP inhibitors for HER2-positive/HRD-high tumors within immune-desert microenvironments to exploit synthetic lethality. This refined, clinically actionable framework transcends simplistic categorization, optimizing therapeutic selection for complex OCCC cases. Notably, this conceptual approach remains under clinical investigation, necessitating rigorous validation through prospective clinical trials to substantiate its proof of concept. Our study has the following limitations: (1) Evidence supporting the therapeutic potential of HER2-ADCs in OCCC is derived mainly from HER2-expressing OCCC PDX models, OCCC cell lines, and retrospective clinical observations from two NC18-treated patients, and therefore requires validation in prospective clinical studies. (2) Although we identified many of potential therapeutic vulnerabilities and proposed a novel precision classification system for OCCC targeted therapy, we specifically validated HER2-targeted efficacy while insufficiently exploring critical events such as FGFR2 fusion in therapeutic contexts. Subsequent investigations should employ additional OCCC animal models and prospective clinical trials to explore innovative targeted approaches; (3) While recent advances in proteomics and metabolomics have provided transformative perspectives and novel targets for cancer therapy [ 55 , 56 ], this study primarily characterized OCCC molecular features through WES, transcriptomics, and scRNA-seq, lacking complementary multi-omics data. Future work will incorporate proteomic profiling and other omics analyses to construct a comprehensive molecular landscape of OCCC and identify translationally valuable therapeutic targets.

In summary, this study has systematically constructed the comprehensive molecular atlas of OCCC, identifying distinct mutational and transcriptomic features while uncovering unique stage-specific mutational disparities between early- and advanced-stage disease. Furthermore, through precision molecular characterization, we have validated the significant therapeutic benefit of HER2-targeted ADCs in HER2-expressed OCCC patients and establishing a molecular subtyping framework. Collectively, these findings not only enhance the fundamental understanding of OCCC biology but also provide therapeutic vulnerabilities for advancing clinical management of this recalcitrant malignancy, offering transformative strategies for molecularly guided diagnosis and treatment.

Supplementary Material 1. Supplementary Material 1. Supplementary Material 2. Supplementary Material 2.