Proteomic profiling and molecular reclassification of high-grade serous ovarian cancer identifies prognostic subtypes and immunotherapy biomarkers

Mengyan Tu; Sangsang Tang; Qiao Zhang; Tianchen Guo; Yixuan Cen; Xiaomeng Xu; Shenglong Wu; Xin Chen; Weiguo Lu; Chen Ding; Junfen Xu

doi:10.1016/j.ebiom.2026.106248

. 2026 Apr 18;127:106248. doi: 10.1016/j.ebiom.2026.106248

Proteomic profiling and molecular reclassification of high-grade serous ovarian cancer identifies prognostic subtypes and immunotherapy biomarkers

Mengyan Tu ^a,^b,^e, Sangsang Tang ^a,^b,^e, Qiao Zhang ^c,^e, Tianchen Guo ^a,^b,^e, Yixuan Cen ^a,^b, Xiaomeng Xu ^c, Shenglong Wu ^a,^b, Xin Chen ^a,^b, Weiguo Lu ^a,^b,^d,^∗, Chen Ding ^c,^∗∗, Junfen Xu ^a,^b,^d,^∗∗∗

PMCID: PMC13101710 PMID: 42001613

Summary

Background

High-grade serous ovarian cancer (HGSOC) is the most lethal histological subtype of ovarian cancer, exhibiting significant heterogeneity and limited therapeutic options. A comprehensive characterisation of proteomic landscape across disease stages is needed to identify actionable biomarkers and therapeutic targets.

Methods

We performed proteomic profiling of 116 primary HGSOC tumours, followed by integrative bioinformatics analyses incorporating clinical annotation. Key findings were validated using multiplex immunohistochemistry, in vitro and in vivo functional assays, and external datasets.

Findings

We identified FIGO stage IIA as a crucial turning point distinguishing early-from advanced-stage disease, marked by a transition from oxidative stress to cell cycle-driven programmes. Trajectory analysis of tumour progression revealed GOSR2 as a key regulator of stage transition. Mechanistically, GOSR2 interacted with SEC24D to inhibit the secretion of CXCL9 and CXCL12, resulting in reduced CD8+ T cell infiltration. Unsupervised clustering defined three reproducible proteomic subtypes (S-I to S-III), which were validated in TCGA and single-cell transcriptomic datasets and associated with distinct clinical outcomes. The S-III subtype was characterised by ECM-receptor interaction, immune evasion, and poor prognosis. Transcription factors network analysis identified regulators potentially driving these phenotypes. In parallel, three immune-contexture subtypes (IC1-IC3) were delineated, reflecting differential tumour immune microenvironment states with prognostic relevance. Advanced-stage HGSOC was further stratified using ISG15, ITGB2, and RELA expression, idenfifying a subgroup with potential susceptibility to immunotherapy.

Interpretation

Our findings provide a framework for biomarker-guided stratification and the development of precision therapeutic strategies in HGSOC.

Funding

Key R&D Program of Zhejiang, NSFC, and 4+X CRP of WHZJU.

Keywords: High-grade serous ovarian cancer, Proteomics, Molecular subtyping, Precision therapy

Research in context.

Evidence before this study

High-grade serous ovarian cancer (HGSOC) is the most lethal gynaecological malignancy worldwide. Large-scale multi-omics efforts, including those by The Cancer Genome Atlas (TCGA) and the Clinical Proteomic Tumour Analysis Consortium (CPTAC), have defined transcriptomic and proteogenomic subtypes of ovarian cancer and revealed substantial molecular heterogeneity. However, the relationship between molecular subtypes and distinct prognostic trajectories remains incompletely defined. In addition, the specific molecular inflection point marking the transition from early to advanced disease has not been systematically characterised at the proteomic level. Given the limited and inconsistent benefit of immune checkpoint inhibitors in HGSOC, a deeper understanding of the tumour immune microenvironment is needed to inform biomarker development and patient stratification strategies.

Added value of this study

We generated an integrated proteomic landscape of HGSOC and characterised molecular features associated with disease progression and clinical outcome. Our analysis identified FIGO stage IIA as a potential molecular “tipping point” distinguishing early-stage from advanced-stage disease, supported by distinct proteomic alterations. Mechanistically, we observed that the GOSR2-SEC24D axis is associated with reduced CD8+ T cell infiltration during the early-to-advanced transition, while ACO2 expression is linked to metabolic regulation in early-stage tumours. We further defined three proteomic subtypes (S-I, S-II, and S-III) with distinct clinical outcomes, which was validated in independent external datasets and supported by multiplex immunohistochemistry (mIHC). In addition, we identified ISG15, ITGB2, and RELA as immune-related proteins entiched in advanced-stage disease, suggesting their potential as biomarkers for disease stratification and candidate targets for therapeutic investigation.

Implications of all the available evidence

Our findings highlight the added value of protein-level profiling in refining the molecular classification of HGSOC and in delineating the biological trajectory of disease progression. By integrating proteomic subtyping with stage-specific molecular features, this study provides a framework for improved prognostic stratification. The identification of immune-associated proteins linked to advanced disease offers hypotheses for future translational research and supports the development of biomarker-informed precision medicine in HGSOC.

Introduction

Ovarian cancer (OC) remains a leading cause of gynaecological cancer-related deaths, with approximately 207,252 fatalities annually worldwide.¹^,² The lack of early-stage symptoms and effective screening methods results in around 75% of patients being diagnosed at advanced stages, leading to a low 5-year survival rate of just 46%.³^,⁴ High-grade serous ovarian cancer (HGSOC), the most common and aggressive subtype, accounts for 70% of all OC and 90% of epithelial OC.⁵^,⁶ HGSOC is characterised by substantial tumour heterogeneity, which poses challenges in patient management. The traditional FIGO staging system, although widely used, fails to account for the heterogeneous outcomes observed within each stage, highlighting the need for a deeper understanding of HGSOC's molecular characteristics to improve clinical management and identify new therapeutic strategies.⁷

While FIGO stage I/II is commonly considered early-stage HGSOC, and stage III/IV as advanced-stage,⁴^,⁸ some studies categorise stage I-IIA as early and stage IIB-IV as advanced.⁹^,¹⁰ Despite these classifications, there is limited molecular research examining the transition from early to advanced stages, which is crucial for understanding the mechanisms driving disease progression. A molecular-based classification could provide a more precise cut-off point, offering new insights into the progression of HGSOC.

Several molecular classification efforts have been made to explore the functional and phenotypic variations in HGSOC. TCGA study classified HGSOC into four subtypes based on gene expression: mesenchymal, immunoreactive, differentiated, and proliferative.¹¹ While proteins are central to cellular functions and therapeutic targets, genomic and transcriptomic changes do not always correlate with proteomic alterations. CPTAC addressed this gap by characterising HGSOC at proteomic, phosphoproteomic, and glycoproteomic levels, identifying five proteomic subtypes: differentiated, metabolic, proliferative, mesenchymal, and stromal subtypes.¹² These investigations have offered high-quality insights into molecular subtyping in HGSOC. However, neither the TCGA nor CPTAC studies revealed distinct prognostic differences among these subtypes, indicating a need for further proteomic research to uncover biomarkers for disease monitoring, prognosis prediction, and therapeutic targeting.

Immunotherapy has shown promise in cancers such as renal cell carcinoma, lung cancer, and hepatocellular carcinoma,13, 14, 15, 16 but its effectiveness in OC, particularly with immune checkpoint inhibitors (ICIs), has been limited.¹⁷^,¹⁸ It has been observed that different T cell infiltration patterns within cancer, namely immune-infiltrated, excluded, and desert, can lead to varying responses to immunotherapies.¹⁹^,²⁰ Previous researches have highlighted the intratumor heterogeneity of HGSOC at the single-cell transcriptomic level.²¹^,²² However, the immune landscape in HGSOC remains poorly understood. Exploring this through proteomics can provide insights into the tumour microenvironment and guide the development of more effective immunotherapies.

In this study, we performed a comprehensive proteomic analysis of 116 primary tumour tissues from patients with HGSOC to identify key stages in disease progression. We found that FIGO stage IIA serves as a critical tipping point, distinguishing early from advanced stages. Our analysis also focused on the transcriptional regulation underlying different proteomic subtypes and their pathophysiological mechanisms. We investigated the tumour immune microenvironment, particularly in “cold tumours”, and developed a machine learning-based predictive model for advanced-stage HGSOC prognosis. These findings offer valuable insights into HGSOC's molecular landscape, which could inform future biological studies, drug discovery, and the development of targeted therapies.

Methods

Clinical samples

Study populations

This study utilised 116 formalin-fixed paraffin-embedded (FFPE) primary tissue samples from patients with HGSOC, collected between January 2013 and December 2018 at Women's Hospital, School of Medicine, Zhejiang University. The study was approval by the Institutional Review Board (IRB-20220255-R). All patients had undergone primary cytoreductive surgery (CRS), and histological diagnoses were confirmed by pathologists. Clinical data, including age, ascites volume, serum CA125, tumour grade, laterality, FIGO stage, residual disease status, lymphatic metastasis, disease-free survival time (DFS), and overall survival time (OS) times, were extracted from the electronic medical records. Optimal CRS was defined as residual disease <1 cm, and R0 CRS as no visible residual disease. Follow up continued for up to 5 years.

Ethics approval and consent to participate

This study had received approval from the Institutional Review Board (IRB) of Women's Hospital, School of Medicine, Zhejiang University (IRB-20220255-R). Written informed consent was obtained from all participants, and the study was conducted in accordance with the Declaration of Helsinki.

Sample preparation

FFPE tissue blocks were sectioned into 10-μm-thick slices, deparaffinized, and H&E stained to assess tumour cellularity. Tumour areas with ≥70% tumour cell and <20% necrosis was selected for proteomic analysis. Divergent histological variants from the same patient were also included. Tissue was scraped, stored at −80 °C, and samples were de-identified.

Peptides preparation for MS analysis

Protein extraction and tryptic digestion

Protein extraction and digestion followed the FFomic strategy. FFPE slides were deparaffinized and lysed with a buffer containing Tris–HCl, DTT (Sigma, 43815), and PMSF (Amresco, M145). Samples underwent sonication, SDS lysis, and acetone precipitation. A filter-aided sample preparation (FASP) procedure was used for protein digestion.²³ Proteins were resuspended in urea, filtered, and digested with trypsin (1:50 enzyme-to-substrate ratio) at 37 °C for 18–20 h. Peptides were collected, washed, and concentrated.

First-dimensional reversed-phase separation

Peptides were redissolved in 0.1% formic acid (FA) and loaded onto a custom-made Durashell Reverse Phase column. A gradient elution buffer was applied, and peptides were eluted in nine steps, followed by combination into three fractions for subsequent vacuum drying.

LC-MS/MS analysis

Peptide separation was performed on a 15 cm C18 column using a 75-min gradient with solvents A (0.1% formic acid in water) and B (0.1% formic acid in 80% acetonitrile). The flow rate was 600 nL/min. Eluted peptides were ionised at 2 kV and analysed on a Q Exactive HF-X mass spectrometer in data-dependent acquisition mode. MS1 spectra were acquired at a resolution of 120,000 (m/z 300–1400), with automatic gain control (AGC) set to 3e+06 and maximum ion injection time of 80 ms. The top 60 precursor ions were selected for MS2 analysis with higher-energy collisional dissociation (HCD) fragmentation (normalised collision energy 27%). Fragment ions were analysed at a resolution of 7500. Dynamic exclusion was applied for 12 s to prevent repetitive scans.

Mass spectrometry data analysis

Peptide and protein identification

Protein quantification was performed using the “Firmiana” proteomic cloud platform, where mzXML files containing identification results and raw data were loaded. Peptide abundances were estimated via the area under the extracted-ion chromatogram (XIC). Protein abundance was calculated using the intensity-based absolute quantification (iBAQ), normalising peptide intensities by the number of observable peptides. The fraction of total (FOT) was determined by dividing each protein's iBAQ by the total iBAQ of all proteins, scaled by 1e+06 for presentation.

Missing value imputation

Proteins with a missing rate <50% were imputed by the K-nearest neighbour (KNN) algorithm using the “impute” R package (v1.74.1) for each clinicopathological subtype.²⁴

Quality control of mass spectrometry data

HEK293T cell lysates (National Infrastructure Cell Line Resource) were measured as a quality control every 3 days, with Spearman's correlation coefficient calculated between quality-control runs in R (v4.3.1) to ensure platform stability.

Global proteomics analysis

Survival analysis

Kaplan–Meier survival curves (log-rank test) were used for overall survival (OS) and disease-free survival (DFS) analysis in 113–114 patients. Cox proportional hazards regression was used to calculate the hazard ratio (HR) and identify significant variables (p < 0.05). The survminer R package (v0.4.9) with “maxstat” (maximally selected rank statistics; http://r-addict.com/2016/11/21/Optimal-Cutpoint-maxstat.html) was employed to determine optimal cut-off point before survival analysis.²⁵

Principal component analysis (PCA)

PCA was performed on 17 early-stage and 96 advanced-stage samples using the “factoextra” R package (v1.0.7) for unsupervised clustering,²⁶ with ellipses indicating 95% confidence intervals.

Differential protein analysis

Differentially expressed proteins (DEPs) were identified using the Wilcoxon rank-sum test, with proteins upregulated or downregulated based on a fold change (FC) > 1.5 or <0.67, respectively, and Benjamini-Hochberg adjusted p-value < 0.05.

Pathway enrichment analysis

DEPs was subjected to Gene Ontology (GO) and Kyoto Encylopedia of Genes and Genomes (KEGG) enrichment using the “clusterProfiler” R package (v4.8.2).²⁷ Gene set enrichment analysis (GSEA) was conducted for the entire protein matrix,²⁸ using hallmark, curated, and GO gene sets (MSigDB). Significance was defined by a Normalised Enrichment Score (NES) > 1, p < 0.05, and false discovery rate (FDR) < 0.25.

Pathway/signature activation scores in different samples

Pathway activity was evaluated using the Gene Set Variation Analysis (GSVA) R package (v1.48.3)²⁹ on a per-sample basis, ranking gene expression to infer activity levels of biological processes and cancer signatures.

Time-series analysis

Time-series data from staged progression (FIGO stage I-IV) were analysed using the Mfuzz R package (v2.60.0),³⁰^,³¹ clustering the data into 8 optimal temporal patterns.

Dynamical Network Biomarkers (DNB) analysis

Dynamical Network Biomarkers (DNB) were identified to predict tumour progression transitions, using the DNB R package (v0.1.1).³²^,³³ They were selected based on significant expression variations across stages, increased standard deviation, and Pearson correlation changes within molecular groups.

Consensus clustering and immune clusters

Consensus clustering of the top 70% of proteins with the highest median absolute deviations (3403 proteins) was performed using the ConsensusClusterPlus R package (v.1.64.0).³⁴^,³⁵ The Partitioning Around Medoids (PAM) algorithm identified immune clusters based on xCell-derived signatures,³⁶ with stromal and immune scores inferred via the Estimation of STromal and Immune cells in MAlignant Tumour tissues using Expression data (ESTIMATE) algorithm.³⁷

Machine learning-based signature generation

A consensus signature was generated using 6 machine learning algorithms, including Ridge, Lasso, SVM, RSF, Xgboost, and Bayes, integrating 17 approaches to analyse 26 feature genes (FC > 3, FDR < 0.05. A few algorithms possessed the ability of feature selection, such as Lasso, RSF, and Xgboost. Thus, we combined these algorithms to generate a consensus model. In total, 17 algorithm combinations were utilised to fit prediction models on our cohort within a repeated 4-fold cross-validation framework, executed across 3 repeats. Classification error (ce) and multiclass area under the curve (mauc) served as the evaluation metrics for each model. In the feature selection and modelling process, we utilised the mlr3 package (v0.16.1) for R to conduct the analyses.³⁸ The “multiclass.roc” function from the pROC package (v1.18.4)³⁹ was employed to compute the area under the receiver operating characteristic (ROC) curve for each model. Prognostic differentiation between subtypes, gauged by model performance on TCGA-OV-2016¹² and PXD010372 datasets,⁴⁰ constituted our model cross-validation strategy.

For the construction of the prognostic model, we harnessed a suite of 9 machine learning algorithms including generalised boosted regression modelling (GBM), supervised principal components (SuperPC), survival-SVM, Ridge, elastic network (ENet), partial least squares regression for Cox (plsRcox), Lasso, stepwise Cox (StepCox), RSF, addressing 24 immune-related genes that demonstrate significant correlations with survival in patients with advanced-stage HGSOC. Feature selection was facilitated by GBM, Enet, Lasso, StepCox, and RSF within the ensemble of 72 combined approaches. We applied the same repeated 4-fold cross-validation (3 repeats) strategy. Model performance was assessed using Harrell's Concordance Index (C-index) across the TCGA-OV-2016 and PXD010372 validation sets. The model with the highest average C-index was selected as optimal. In addition to mlr3 package, the StepCox model was implemented via survival package (v3.5-5).⁴¹ A stepwise algorithm using the Akaike information criterion (AIC) was applied, and the direction mode of stepwise search was set to “both”, “backward”, and “forward”, respectively. The plsRcox model was implemented via plsRcox package (v1.7.7).42, 43, 44 The cv.plsRcox function was used to determine the number of components requested, and the plsRcox function was applied to fit a partial least squares regression generalised linear model. The SuperPC model was implemented via superpc package (v1.12), a generalisation of principal component analysis, which generates a linear combination of the features or variables of interest that capture the directions of largest variation in a dataset.⁴⁵^,⁴⁶ The superpc.cv function used a form of leave-one-out cross-validation to estimate the optimal feature threshold in supervised principal components. To avoid problems with fitting Cox models to small validation datasets, it uses the “pre-validation” approach.

Transcription factors (TFs) activity inference

TF activities were inferred using pySCENIC (v0.12.1) in a Python 3.10.6 environment,47, 48, 49 with GRNBoost2 and cisTarget for gene regulatory network inference and motif enrichment. AUCell quantified TF activity, and Wilcoxon rank-sum test identified characteristic TFs based on regulon specificity scores and differential activity.

Weighted gene correlation network analysis (WGCNA) analysis

WGCNA was performed on 4863 proteins, identifying co-regulated gene modules. The “blockwiseModules” function clustered genes, and pathway enrichment analysis was conducted for each module. Hub genes were identified based on node connectivity.

Immunohistochemical (IHC) staining

FFPE tissue sections were deparaffinized, hydrated, antigen retrieved, and then stained for CD3 (Cell Signalling Technology, 85061), CD45 (Cell Signalling Technology, 13917), ITGB2 (Abcam, ab181548), ISG15 (Abcam, ab227541), and RELA (Cell Signalling Technology, 8242) antibodies using standard IHC procedures.

Cell line authentication and cell culture

The identity of the established cell lines (OVCAR8 and HEK293T) was authenticated by Short Tandem Repeat (STR) profiling. For HUVECs, validation was performed through immunofluorescence (IF) staining of endothelial-specific markers (CD31 and vWF). All cell lines were confirmed to be mycoplasma-negative. OVCAR8 cell line were purchased from MeisenCTCC (Cat. CTCC-001-0663) and cultured in RPMI 1640 (Basal Media) with 10% foetal bovine serum (FBS) and 1% penicillin/streptomycin. HEK293T cell line were purchased from MeisenCTCC (Cat. CTCC-001-0188) and cultured in Dulbecco's modified Eagle's medium (DMEM) (Basal Media) containing 10% FBS and 1% penicillin/streptomycin. HUVECs were purchased from MeisenCTCC (Cat. CTCC-0804-PC) and cultured in Endothelial Cell Medium (ECM, ScienCell, Cat.1001) with 5% FBS, 1% endothelial cell growth supplement (ECGS), and 1% penicillin/streptomycin. All cell lines were cultured at 37 °C in a humidified atmosphere with 5% CO2.

Cell viability assays

Cell viability was evaluated by Cell Counting Kit-8 (CCK-8) kit according to the manufacturer's protocols (Yeasen, 40203ES80).

EdU assays

Cells were seeded into 96-well plates and transfected with specific siRNAs. After 48 h, cells were incubated with 10 μM 5-ethynyl-2′-deoxyuridine (EdU) for 2 h at 37 °C. Cells were then fixed with 4% paraformaldehyde for 30 min and permeabilized with 0.5% Triton X-100 for 10 min. The EdU staining was performed using the BeyoClick™ EdU-555 assay (Beyotime, C0075S), and nuclei were counterstained with DAPI. The images of EdU-positive cells were captured using a fluorescence microscope.

Western blot

Whole-cell proteins were obtained using the RIPA lysis (Solarbio). Proteins were separated using SurePAGE gels and transferred to the PVDF membranes, then the membranes were blocked, incubated in primary antibodies: GOSR2 antibody (Proteintech, 66134-1-Ig, RRID: AB_2881533), FGF2 antibody (Abcam, ab92337, RRID: AB_2049652), SEC24D antibody (Proteintech, 13673-1-AP, RRID: AB_2877968), PPP3R1 antibody (Proteintech, 13210-1-AP, RRID: AB_2252760), VEGF antibody (Absin, abs123884, RRID: AB_3096488), EGLN1 antibody (Proteintech, 19886-1-AP, RRID: AB_10641986), PDGFRB antibody (Proteintech, 13449-1-AP, RRID: AB_2162644), ACO2 antibody (ABclonal, A4524, RRID: AB_2863288), β-ACTIN antibody (Proteintech, 66009-1-Ig, RRID: AB_2687938). Next, membranes were probed using secondary antibodies, and finally visualised using imaging system.

Cell transfection

Transient transfection of siRNAs (GenePharm) was performed by DharmaFECT Transfection Reagents. The sequences of siRNAs were listed as following:

si-GOSR2#1 sense: GGGUUGACCAGUUAAAGUATT, antisense: UACUUUAACUGGUCAACCCTT.

si-GOSR2#2 sense: CUCGAACCUUCACCACUAATT, antisense: UUAGUGGUGAAGGUUCGAGTT.

si-SEC24D#1 sense: GCAAAUCAACAGCUAUGGUTT, antisense: ACCAUAGCUGUUGAUUUGCTT.

si-SEC24D#2 sense: GCCCAUUUAUGCAGUUCAUTT, antisense: AUGAACUGCAUAAAUGGGCTT.

si-PDGFRB#1 sense: GAGGGUGACAACGACUAUATT, antisense: UAUAGUCGUUGUCACCCUCTT.

si-PDGFRB#2 sense: GGAGGACCCAUCUAUAUCATT, antisense: UGAUAUAGAUGGGUCCUCCTT.

si-EGLN1#1 sense: GAGAGCACGAGCUAAAGUATT, antisense: UACUUUAGCUCGUGCUCUCTT.

si-EGLN1#2 sense: CCCUCAUGAAGUACAACCATT, antisense: UGGUUGUACUUCAUGAGGGTT.

si-PPP3R1#1 sense: CAGCGAGUAAUAGAUAUAUTT, antisense: AUAUAUCUAUUACUCGCUGTT.

si-PPP3R1#2 sense: GAAGCUUGAUUUGGACAAUTT, antisense: AUUGUCCAAAUCAAGCUUCTT.

si-FGF2#1 sense: GCUACAACUUCAAGCAGAATT, antisense: UUCUGCUUGAAGUUGUAGCTT.

si-FGF2#2 sense: GUUGGUAUGUGGCACUGAATT, antisense: UUCAGUGCCACAUACCAACTT. The ACO2 overexpression plasmids (GenePharm) were transfected into OVCAR8 cells using X-tremeGENE HP DNA transfection reagent (Roche, USA).

RNA extraction and RT-qPCR

Total RNA was isolated using Trizol. HiScript III RT SuperMix for qPCR (+gDNA wiper) (Vazyme Biotech, R323-01) and ChamQ Universal SYBR qPCR Master Mix (Nanjing Vazyme Biotech, Q711-02) were used for the RNA reverse transcription and PCR analysis. Sequences of primers were exhibited as follows: human GOSR2: 5′-CTTCGGGTTGACCAGTTAAAGT-3′ (forward), 5′-AAGGTTCGAGACAGAAGCTCT-3′ (reverse), human PPP3R1: 5′-CCTTTGGAAATGTGCTCACACT-3′ (forward), 5′-GGATTCTGTTGTAACTCAGGCAG-3′ (reverse), ACO2: 5′-CCCTACAGCCTACTGGTGACT-3′ (forward), 5′-TGTACTCGTTGGGCTCAAAGT-3′ (reverse), SEC24D: 5′-GTCAACAAGGTTACGTGGCTA-3′ (forward), 5′-TAGTGCCCATAATGAGGTGGA-3′ (reverse), human β-actin: 5′-TGGTATCGTGGAAGGACTC-3′ (forward), 5′-AGTAGAGGCAGGGATGATG-3′ (reverse).

Transwell assay

Cell migration and invasion abilities were evaluated using transwell chambers. Cells were plated into the upper chamber with Opti-MEM (Thermo Scientific) and the bottom chamber was full of medium containing 10% FBS. For migration, after 24 h culture, cells migrated to the membrane were fixed and staining with ethyl alcohol containing 0.1% crystal violet for 30 min. For invasion, the upper chambers were pre-coated with Matrigel (Corning, Cat#354234) prior to cell seeding, and the membrane were stained after 36-h culture. The images were captured under a light microscope.

Phalloidin staining

After indicated treatments, cells were fixed with 4% PFA, permeabilized with 0.1% Triton X-100, blocked with 3% bovine serum albumin (BSA), stained with F-actin and DAPI, and finally visualised using confocal microscopy.

Isolation of cancer-associated fibroblasts (CAFs)

CAFs were isolated from primary tissues of 3 patients with HGSOC. Tissue samples were surgically excised, washed with PBS, minced on ice, and enzymatically digested. After filtration through a 40-μm strainer, cells were collected by centrifuging (300×g, 5 min). The pellet was treated with red blood cell lysis buffer and then resuspended in DMEM containing 10% foetal bovine serum. CAFs identity was confirmed as previously described.¹⁰

T cell isolation and co-culture with cancer cells

Human peripheral blood mononuclear cells (PBMCs) were isolated from HGSOC blood using a Ficoll density gradient. CD3+ T cells were purified with the MojoSort™ Human CD3 T Cell Isolation Kit. Following the indicated treatments, T cells were seeded and co-cultured with cancer cells.

Flow cytometry analysis of T cells

CD3+ T cells were blocked with human Fc Block (BD Pharmingen™ 564220) and tstained with fluorescence antibodies for 30 min at 4 °C. For intracellular staining, cells were stimulated (BD Pharmingen™ 550583) for 5 h, then stained for surface molecules. After fixation and permeabilizetion using a Fixation/Permeablization Kit (BD Pharmingen™ 554714), intracellular staining was performed. Data were analysed with Flowjo software.

Animal models

Animal study was approved by Laboratory Animal Welfare & Ethics Committee of the Women's Hospital, School of Medicine, Zhejiang University (Ethical Approval Number: AE20250079). All animal experiments were conducted in strict accordance with the institutional animal use guidelines. Humane endpoints were established for the study. Animals were monitored daily for signs of distress, including weight loss (>20%), hunched posture, lethargy, or ulcerated tumours. If any humane endpoint was reached, the animal was immediately euthanised. No unexpected adverse events or mortality occurred during the experimental period.

Twelve 5-week-old female SCID mice (weight 18–20 g) were purchased from Shanghai SLAC LaboratoryAnimal Co, Ltd. Mice were housed in specific pathogen-free (SPF) conditions under a 12-h light/dark cycle with ad libitum access to food and water. Animals were acclimatized for 1 week prior to the experiment. The study consists of 3 groups: the negative control (si-NC), si-GOSR2#1, and si-GOSR2#2. The single mouse was defined as the experimental unit. The sample size (n = 4 per group, total n = 12) was determined based on previous experience with similar xenograft models to ensure adequate power for detecting significant differences in tumour growth while minimising animal usage (3R principles). All healthy mice meeting the age and weight criteria were included. No animals or data points were excluded from the analysis. Mice were randomly assigned to the three groups using a computer-generated random number sequence. To minimise potential confounders, the order of treatments and measurements was randomised across groups, and cages were randomly positioned within the housing rack. Tumour measurements and data analysis were performed by an investigator who was blinded to the group allocation. To establish the xenograft model, OVCAR8 cells were transfected with si-NC, si-GOSR2#1, or si-GOSR2#2. Twenty-four hours post-transfection, 5 × 10⁶ cells suspended in 100 μL of PBS were subcutaneously injected into the right flank of each mouse. Seven days post-cell injection, when palpable tumours were formed, 1 nmol of the respective siRNAs was injected intratumorally. The investigator administering the treatment was aware of the allocation to ensure correct dosing. The primary outcome measure was tumour volume. Tumour dimensions were measured every 5 days using digital callipers, and volume was calculated using the formula: V = 0.5 × length × width². After 20 days, mice were euthanised via CO2 inhalation, and tumours were excised and weighed. Data are presented as mean ± SD for each group (n = 4). Statistical analysis was performed using GraphPad Prism. Normality of the data was assessed using the Shapiro–Wilk test. Differences between groups were analysed using ANOVA or Student's t-test. p < 0.05 was considered statistically significant.

Co-Immunoprecipitation (co-IP) assay

OVCAR8 cells cultured in a 15 cm dish were lysed on ice using 1 mL pre-chilled IP lysis buffer (Beyotime, China) containing protease and phosphatase inhibitor cocktails (Beyotime, China). The cell lysates were then incubated with 5 μg of the relevant primary antibodies (SEC24D antibody, Santa Cruz Biotechnology, sc-101268, RRID: AB_2186376) at 4 °C overnight with gentle rotation. Subsequently, Protein A/G Magnetic Beads (Bimake, USA) were added to capture the antigen–antibody complexes. After thorough washing, the bound proteins were eluted from the beads by boiling in 1× SDS loading buffer (Beyotime, China) at 95 °C for 5 min.

mIHC assay

Tissue sections were stained using the OPAL multiplex IHC kit (Akoya Biosciences, NEL861001KT) according to the manufacturer's protocol. Briefly, after deparaffinization, antigen retrieval, and blocking, slides were sequentially incubated with the primary antibody: GOSR2 (Proteintech, 66134-1-Ig, RRID: AB_2881533), CD8 (Cell Signalling Technology, 98941, RRID: AB_2756376), VEGFR (Abcam, ab315238, RRID: AB_3695631), SEC24D (Proteintech, 13673-1-AP, RRID: AB_2877968), PPP3R1 (Proteintech, 13210-1-AP, RRID: AB_2252760), AKT2 (Proteintech, 28113-1-AP, RRID: AB_2881064), PLCG1 (Proteintech, 84941-1-RR), PRKCA1(Proteintech, 21991-1-AP, RRID: AB_2878965), SLC16A3 (Proteintech, 22787-1-AP, RRID: AB_11182479), TSTA3 (Proteintech, 15335-1-AP, RRID: AB_2211226), PKP2 (Proteintech, 26479-1-AP, RRID: AB_3669539), AMBP (Proteintech, 26716-1-AP, RRID: AB_3085894), HBB (Proteintech, 16216-1-AP, RRID: AB_10598329), CA1 (Proteintech, 13198-2-AP, RRID: AB_2275041) followed by an HRP-conjugated secondary antibody and an OPAL™ fluorescent dye. Antibody-dye complexes were then stripped via microwave heating before proceeding to the next cycle of staining. This process was repeated for all targets. Nuclei were counterstained with DAPI. Finally, multispectral images were acquired using the AKOYA Vectra Polaris system (Akoya Biosciences).

HUVEC tube formation assay

Angiogenesis plates (ibidi, 81506) were pre-coated with growth factor reduced Matrigel at 37 °C for 30 min. HUVECs were seeded onto the coated plates at a density of 1 × 10⁵ cells per well in 50 μL medium. After incubation at 37 °C for 4h, capillary-like tube structures were visualised under a microscope. Images were captured and the total tube length per field was quantified using ImageJ software.

Statistical analysis

Chi-square or Fishers exact tests were used for examining associations between unordered categorical variables (e.g., proteomic subtype and clinical parameters). Ridit analysis compared ordinal and unordered categorical variables. Binary group comparisons were conducted using the Wilcoxon rank-sum test, and multiple group comparisons with the Kruskal-Walli's test. Correlations between continuous variables were assessed using Spearman's coefficients. Statistical tests were two-sided. Statistical significance was defined as p < 0.05. All analyses were performed in R (v4.3.1), with Benjamini-Hochberg FDR correction applied where appropriate.

Role of funders

The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.

Results

Comprehensive proteomic characterisation in HGSOC

We collected primary tumour tissues from 116 patients with HGSOC for proteomic analysis (Fig. 1A). Detailed clinical characteristics were provided in Supplementary Table S1 and Fig. 1B. The patients’ age at diagnosis ranged from 34 to 78 years, with a median age of 54.5 years. A majority (82.7%) were diagnosed at advanced stage (FIGO stage IIB-IV). Of the 116 patients, 90 (77.6%) underwent optimal CRS, including 74 who achieved R0 CRS. The median follow-up duration was 47 months, ranging from 4 to 60 months. As the last follow-up, 65 patients were alive, 48 had died, and 3 were lost to follow-up, with 1 had a first recurrence. This analysis included 113 patients for OS and 114 patients for DFS.

For proteomic analysis, we utilised label-free quantification on a Q Exactive HF-X Hybrid Quadrupole-Orbitrap Mass Spectrometer. A total of 12,417 proteins were identified, with a peptide FDR of ≤0.05. The protein numbers identified at each FIGO stage were 9127 (stage I), 10,164 (stage II), 11,737 (stage III), and 8447 (stage IV) (Fig. 1C–E). The utilisation of FOT for normalising proteome quantification highlighted the remarkable sensitivity and comprehensive coverage achieved in our proteomics analysis (Fig. 1F). After applying a series of filters to reduce contamination and normalise data, we identified 2513 to 5139 proteins in each sample, with 4862 proteins detected in more than 50% of samples. The robustness of the mass spectrometer was confirmed by a high Spearman's correlation coefficient (r > 0.9) between quality control samples (Supplementary Figure S1). Our study provides a comprehensive proteomic profile of HGSOC across various FIGO stages.

Critical transition from early to advanced stage in HGSOC

Patients diagnosed at earlier stages of HGSOC generally have a better prognosis than those at advanced stages. Identifying the precise transition point is essential for improving early detection and treatment. Our analysis revealed significant proteomic remodelling during HGSOC progression (Fig. 2A), with 28% of proteins exhibiting at least a 2-fold change in expression (Wilcoxon rank-sum test, FDR <0.05). Time-series analysis categorised protein expression patterns into eight gene signatures (GSs) (Fig. 2B), supporting the hypothesis that cancer progression involves non-linear, dramatic transitions.

The most significant proteomic changes occurred between FIGO stage IIA (early stage) and IIB (advanced stage) (Fig. 2B). Metabolic pathways, such as glycolysis, the tricarboxylic acid (TCA) cycle, and oxidative phosphorylation, were enriched in GS 1, 2, and 6 showing a downward trend, while pathways promoting cancer progression, like the cell cycle and sex hormone signalling, were enriched in GSs with an upward trend (Fig. 2C and D). The expression of prognosis-associated genes in cell cycle and oxidative phosphorylation pathways was depicted in Fig. 2E and Supplementary Figure S2A, highlighting their impact on tumour progression. Notably, oxidative stress increased during tumour progression, with a positive correlation between oxidative stress and genes in the cell cycle pathway, such as ANAPC2, ANAPC4, CDKN2C, HDAC2, and STAG1 (Spearman's cor = 0.184, 0.196, 0.186, 0.205, and 0.207, p < 0.05) (Fig. 2F, Supplementary Figure S2B). Notably, previous research had reported that inhibiting HDAC2 could suppress the proliferation of cervical cancer cells by regulating cellular oxidative stress levels.⁵⁰ As the tumour advanced, genes associated with different phases of the cell cycle also exhibited phase-dependent changes, regulating tumour cell proliferation (Fig. 2G). These findings suggest a cooperative role of oxidative stress and the cell cycle in HGSOC progression.

To identify the critical juncture in HGSOC progression, we employed DNB analysis based on nonlinear dynamic theory (Supplementary Figure S2C and D). DNB, which differs from traditional differential expression molecules, comprises a cluster of molecules exhibiting strong correlations and fluctuations. A DNB group represents a dominant cluster within the dataset, and its identification is guided by three specific criteria outlined in our Materials and Methods (Supplementary Figure S2E). As the system nears a tipping point, DNB theory suggests that collective fluctuations among these molecules signal an impending disease transition, positioning them as dynamic biomarkers predictive of progression. Analysis of proteomic profiles from HGSOC tumour samples across five FIGO stages revealed a strong signal of the critical transition to malignancy at FIGO stage IIA (Fig. 2H). Survival analysis confirmed no significant prognostic difference between FIGO stage I and IIA, while advanced stages (FIGO IIB-IV) showed a marked decline in prognosis compared to early stages (Supplementary Figure S2C). The correlation coefficient (cor) and coefficient of variation (cv) between 18 DNBs and DEPs in various FIGO stages further validated that stage IIA marks the tipping point between early and advanced disease (Supplementary Figure S2F).

Our study identified GOSR2 as a key protein involved in HGSOC progression. Based on its network degrees with DEPs, its correlation with prognosis, and its association with specific GSs (Supplementary Figure S2G), GOSR2 was found to increase in expression as HGSOC advanced, indicating its role as an oncogene (Fig. 2I). Among nine DEPs predicted to interact with GOSR2, six exhibited expression changes around stage IIA. Correlation analysis identified SEC24D as a downstream target of GOSR2, with further analysis suggesting a potential relationship between GOSR2, SEC24D (Fig. 2I).

To validate GOSR2's oncogenic role, we transfected OVCAR8 cells with siRNAs targeting GOSR2. Knockdown efficiency was confirmed through protein expression levels (Supplementary Figure S2H). In vitro experiments revealed that GOSR2 knockdown significantly reduced the proliferation, migration and invasion of OVCAR8 cells (Fig. 2J–L). Furthermore, xenograft tumour growth was also inhibited in vivo (Fig. 2M−N, Supplementary Figure S2I). On day 20, the average tumour volume in the si-NC group reached 285.5 ± 76.6 mm³, whereas the si-GOSR2#1 and si-GOSR2#2 groups showed significantly reduced tumour volumes of 21.8 ± 6.9 mm³ and 16.6 ± 6.8 mm³, respectively (Fig. 2N). co-IP assay confirmed the physical interaction between GOSR2 and SEC24D (Fig. 2O), which is consistent with our correlation analysis. This interaction was further verified by mIHC assay, which demonstrated their co-localisation in OC tissue sections and OVCAR8 cell line, respectively (Fig. 2P–Q). Furthermore, GOSR2 and SEC24D were estimated correlated with immune escape (Fig. 2R and Supplementary Figure S2J). We performed mIHC using OC tissue FFPE slides to explore the immunoregulatory function of GOSR2. Our results showed that GOSR2 mediated the immune evasion of CD8+ T cells (Fig. 2S). Co-culture of T cells from patients with OC and OVCAR8 cells demonstrated an increase in the proportion of cytotoxic Granzyme B + CD8+ T cells following GOSR2 knockdown (Fig. 2T), indicating that GOSR2 may play a role in immune escape regulation. We found that GOSR2 and SEC24D contributed to CD8+ T cells immune escape by inhibiting the production of chemokines CXCL9 and CXCL12 (Fig. 2U, Supplementary Figure S2K). Collectively, our results illustrated that the GOSR2-SEC24D complex drove tumour progression while simultaneously mediating immune evasion via suppression of CXCL9 and CXCL12. In summary, our data highlight a distinct proteomic shift at the critical transition point between early and advanced stages of HGSOC, with implications for prognosis and potential therapeutic targeting of GOSR2 and its associated pathways.

Proteomic alterations in early vs. advanced stage HGSOC

The FIGO staging system is widely used to guide treatment and surveillance strategies for patients with OC. In this study, we categorised FIGO stage I-IIA as early stage and FIGO stage IIB-IV as advanced stage. To elucidate the proteomic alterations in HGSOC across different stages, we performed a systematic analysis of proteomic data. Survival analysis demonstrated a significant difference in outcomes between early and advanced stages (log-rank test, p < 0.05) (Fig. 3A). PCA and DEPs analyses revealed substantial proteomic differences between the stages (Fig. 3B–C). A total of 328 DEPs were identified in tumour tissues from early and advanced stages (Wilcoxon rank-sum test, p < 0.05, |FC| ≥ 2) (Fig. 3C, E). Furthermore, KEGG pathway enrichment analysis revealed distinct molecular features in early- and advanced-stage samples. Early-stage samples showed enrichment in metabolic changes, including glycolysis, the tricarboxylic acid (TCA) cycle, pyruvate metabolism, the peroxisome proliferator-activated receptor (PPAR) signalling pathway, oxidative phosphorylation, and carbon metabolism. In contrast, advanced-stage samples were predominantly associated with cancer-related pathways, the hypoxia inducible factor-1 (HIF-1) signalling pathway, proteoglycans in cancer, epidermal growth factor receptor (EGFR) tyrosine kinase inhibitor resistance, choline metabolism in cancer, endocrine resistance, and vascular endothelial growth factor (VEGF) signalling pathway (Fig. 3D). These findings indicated the distinct molecular characteristics of HGSOC at different stages.

Fig. 3 — **Proteomic alterations of HGSOC in early stage and advanced stage**. A. Kaplan–Meier curves for OS between early and advanced samples (n = 113, early = 17, advanced = 96) (log-rank test). B. PCA of proteins in 17 early samples and 96 advanced samples. Blue, early samples, orange, advanced samples. C. Volcano plots displaying the DEPs in early and advanced samples after applying a 1.5-fold change in expression with p < 0.05 (Wilcoxon rank-sum test). Proteins significantly enriched in the early and advanced samples are represented as blue/orange-filled dots. Orange, upregulation in advanced sample; blue, upregulation in early sample; grey, not significant. D. Comparison of enriched pathways in early and advanced samples. E. Heatmap of DEPs between early and advanced samples. F. Diagram representing the significant proteins in carbon metabolism pathway (enriched in early samples) (top panel) and pathways in cancer (enriched in advanced samples) (bottom panel). G. Kaplan–Meier survival curves for OS based on protein expressions of ACO2, ENO1, PDGFRB, and EGLN1 in HGSOC (log-rank test). H. Quantification of ACO2 expression in human OC tissues compared to normal ovary (NO) tissues, as detected by qPCR (Student's t-test). I. Cell proliferation of OVCAR8-ACO2-OE cells was assessed by CCK-8 assay (Student's t-test). **J-K.** Glucose uptake (K) and lactate production (L) in the culture medium were measured in OVCAR8-NC and OVCAR8-ACO2-OE cells (Student's t-test). L. Heatmap of PRKCA, PPP3R1, AKT2, and PLCG1 between early and advanced samples. M. Volcano plot indicating the correlation between the expression of PPP3R1 and the prognosis of HGSOC. N. Kaplan–Meier survival curves for OS based on protein expressions of PPP3R1 in patients with HGSOC (log-rank test). O. Western blot experiment showing the expression level of VEGF in OVCAR8 cells after knockdown of PPP3R1. P. Representative images and quantification of the tube formation assay of HUVEC cells. HUVEC were cultured with OVCAR8-siNC, OVCAR8-siPPP3R1#1 or OVCAR8-siPPP3R1#2 cells (Student's t-test). Scale bar = 50 μm. Q. IF staining of VEGFR in HUVEC cells after co-culture with OVCAR8-siNC, OVCAR8-siPPP3R1#1 or OVCAR8-siPPP3R1#2 cells. Scale bar = 10 μm. R. Representative images of EdU (red) and DAPI (blue) staining in OVCAR8 cells transfected with specific siRNAs of PPP3R1. The right graph shows the quantification of the percentage of EdU-positive cells (Student's t-test). Scale bar = 100 μm. S. CCK-8 assays were performed to evaluate the cell proliferation of OVCAR8 cells transfected with specific siRNAs of PPP3R1 (Student's t-test). T. Representative images of migration and invasion of OVCAR8 cells following GOSR2 knockdown. The right panel represent the quantitative analysis of cell numbers per field (Student's t-test). Scale bar = 100 μm. U. mIHC staining of a representative HGSOC tissue section for PRKCA, PPP3R1, AKT2, PLCG1, and GOSR2. Scale bar = 50 μm. V. Quantification of PRKCA, PPP3R1, AKT2, PLCG1, and GOSR2 staining intensity across human normal, early-stage, and advanced-stage OC tissues (ANOVA). ∗∗∗*p <* 0.001, ∗∗p < 0.01, ∗p < 0.05.

To prioritise stage-specific drivers from these enriched biological processes, we employed a stepwise screening strategy. We first identified core DEPs within the top-enriched pathways of each stage (Fig. 3F). These candidates were then filtered based on their prognostic significance. Kaplan–Meier analysis showed that high expression of aconitase 2 (ACO2) (log-rank test, p < 0.05) and enolase 1 (ENO1) (log-rank test, p < 0.05), were associated with favourable prognosis in HGSOC (Fig. 3G). This suggests that downregulation of carbon metabolism may suppress HGSOC progression. ACO2 is an essential enzyme linking the TCA cycle and lipid metabolism.⁵¹ From other pan-cancer analyses, ACO2 can serve as a diagnostic and prognostic marker in various cancers, with elevated ACO2 expression associates with favourable patient prognosis.⁵² We examined the expression of ACO2 between OC tissues and normal ovary tissues, demonstrating its lower expression in OC (Fig. 3H). To further investigate the role of ACO2 in OC, we constructed an overexpression plasmid and transfected it into OVCAR8 cells (Supplementary Figure S3A). Subsequent experiments demonstrated that ACO2 overexpression not only inhibited cellular proliferation (Fig. 3I) but also reduced glucose uptake and lactate production (Fig. 3J–K), indicating a metabolic shift from glycolysis toward TCA cycle and OXPHOS. Thus, ACO2 functions not only as a prognosis marker but also, more importantly, as a potential therapeutic target. High expression of ENO1 has been previously shown to promote OC growth by enhancing the Warburg effect.⁵³^,⁵⁴ Conversely, higher expression of platelet-derived growth factor receptor-β (PDGFRB) (log-rank test, p < 0.05) and egl-9 Family Hypoxia Inducible Factor 1 (EGLN1) (log-rank test, p < 0.05) was linked to worse OS (Fig. 3F–G). PDGFRB has been suggested as a potential target in epithelial OC,⁵⁵ while EGLN1 is involved in angiogenesis and tumorigenesis.⁵⁶ To further validate the functional impact of these advanced-stage markers, we performed siRNA-mediated knockdown of PDGFRB or EGLN1 in OVCAR8 cells (Supplementary Figure S3B and C). Our results demonstrated that knockdown of PDGFRB or EGLN1 significantly inhibited cell proliferation, as evidenced by the EdU assay (Supplementary Figure S3D and F). Furthermore, Transwell assay showed that the migratory and invasive capabilities of OVCAR8 cells were markedly suppressed following the knockdown of PDGFRB or EGLN1 (Supplementary Figure S3E and G). These findings confirm that these prognostic DEPs serve as functional drivers of disease progression in advanced-stage HGSOC.

WGCNA is a bioinformatic method used to identify co-expressed gene modules and explore their relationship with clinical phenotypes.⁵⁷ In our study, WGCNA identified seven consensus modules of co-expressed genes (Supplementary Figure S3H and I). The modules contained 31, 144, 81, 54, 52, 2420, and 78 proteins in #CCAED0, #F39A9D, #F6C97D, #F6C0CA, #87D8C3, #D0E2B5, and #74AED3 modules, respectively. Pathway enrichment analysis revealed the biological characteristics associated with each module (Supplementary Figure S3I). Notably, the #74AED3 module was enriched in the NOD-like receptor signalling pathway, glycerophospholipid metabolism, and tyrosine metabolism (Supplementary Figure S3I). We also calculated correlations between modules and clinical traits, proteomic subtypes, and immune clusters, including age, FIGO stage, laterality, tumour grade, CA125 levels, ascites, optimal CRS, R0 CRS, lymphatic metastasis, OS, and DFS (Supplementary Figure S3J). The #74AED3 module exhibited the strongest positive correlation with clinical features, particularly FIGO stage (early vs. advanced, cor = 0.44, p < 0.05, Supplementary Figure S3J). The correlation coefficient between GS for FIGO stage and module membership (MM) was 0.56, confirming the robustness of the module construction. Using threshold of GS > 0.1 and MM > 0.4, we identified 78 hub genes associated with FIGO stage within this module (Supplementary Figure S3K). The protein–protein interaction (PPI) network of these hub genes, constructed using the STRING database with a medium confidence cutoff of 0.4, revealed their involvement in key biological processes, including the NOD-like receptor signalling pathway, signalling by receptor tyrosine kinases, myofibril assembly, response to toxic substance, mitochondrial translation, glycerophospholipid metabolism, membrane fusion, mRNA metabolic process, and ribosome biogenesis (Supplementary Figure S3L).

We also investigated the activation of the VEGF signalling pathway in patients with advanced HGSOC. Four genes—PRKCA, PPP3R1, AKT2, PLCG1—were identified as differentially expressed between early and advanced HGSOC (Fig. 3L). Furthermore, 16 potential hub genes associated with HGSOC progression, including PPP3R1, were identified by overlapping the 78 genes in the #74AED3 module with 83 genes upregulated in advanced HGSOC (Supplementary Figure S3M). PPP3R1, which plays a critical role in activating the VEGF pathway, showed a significant correlation with the progression and prognosis of HGSOC (Fig. 3M−N). We found downregulating PPP3R1 in OVCAR8 cells contributes to decreased expression of VEGF (Fig. 3O). Then, we employed a co-culture system of OVCAR8 and HUVEC cells to explore the angiogenic potential of PPP3R1. HUVEC tube formation Assay further revealed that PPP3R1 knockdown in OVCAR8 cells impaired the angiogenic ability of HUVEC cells (Fig. 3P). In line with this, we observed a corresponding downregulation of VEGFR expression levels in the HUVECs co-cultured with PPP3R1-knockdown OVCAR8 cells (Fig. 3Q). Moreover, in vitro experiments confirmed that PPP3R1 knockdown significantly reduced the proliferation, migration, and invasion of OVCAR8 cells (Fig. 3R–T). These findings suggest that increased expression of PPP3R1 is associated with worse prognosis in advanced HGSOC.

To confirm our proteomic results, we performed mIHC assay to examine the presence and localisation of proteins including PRKCA, PLCG1, AKT2, GOSR2, and PPP3R1 (Fig. 3U). Quantitative analysis across normal, early-stage, and advanced-stage OC tissues validated the differential expression of these markers, consistent with the proteomic data (Fig. 3V).

Identification of proteomic subtypes to delineate intratumour heterogeneity of HGSOC

Tumour heterogeneity can be categorised into two types: intertumoral and intratumoral. Intertumoral heterogeneity refers to variations between patients with the same histological subtype, while intratumoral heterogeneity pertains to the differences among cancer cells within a single patient.⁵⁸^,⁵⁹ Intratumoral heterogeneity is a key factor limiting the effectiveness of current therapies for HGSOC. To address this, we aimed to classify HGSOC into molecular subtypes based on proteomic profiles to better manage patients with similar molecular characteristics. Using consensus clustering, HGSOC was classified into three proteomic subtypes: S–I (20 samples), S-II (63 samples), and S-III (33 samples) (Fig. 4A). These subtypes exhibited significant differences in OS (log-rank test, p < 0.05) and DFS (log-rank test, p = 0.091). The S-II subtype demonstrated the best OS, while the S-III subtype exhibited the poorest OS. The S–I subtype was intermediate (log-rank test, p < 0.05, Fig. 4B). DFS analysis showed a similar trend, although the difference between S-I and S-II subtypes was not statistically significant (Fig. 4C). The S-III subtype had a high risk of recurrence, with 78.1% (25/32) of patients relapsing and 50% (16/32) succumbing to the disease, except for one patient lost to follow up. Clinical features associated with the proteomic subtypes of 116 patients with HGSOC are summarised in Fig. 4D and Supplementary Table S2. Notably, the S-III subtype correlated with advanced disease stage and bilateral tumours (Fig. 4D, Supplementary Table S2).

Fig. 4 — **Identification of proteomic subtypes to delineate intratumour heterogeneity of HGSOC**. A. Heatmap illustrating the characterisation of 3 proteomic subtypes. Each column represents a patient sample and rows indicate proteins. B. Kaplan–Meier curves for OS based on the 3 proteomic subtypes (log-rank test). C. Kaplan–Meier curves for DFS based on the 3 proteomic subtypes (log-rank test). D. Heatmap showing the 3 proteomic subtypes and their connections with clinical features. Fisher's exact test and chi-square test were used to evaluate the association of proteomic subtypes with the clinical features. E. Heatmap displaying KEGG terms enriched in the 3 proteomic subtypes. F. DEPs in 3 proteomic subtypes and the relative fold change compared with S–I subtype. G. ROC curves of different algorithms for identifying the 3 proteomic subtypes. H. External validation of the 10-gene-based model in TCGA-OV-2016 cohort (log-rank test). I. mIHC assay validating the protein expression of key biomarkers identified from survival analysis across the three molecular subtypes (S–I, S-II, S-III) of OC. Representative images show distinct expression patterns for: SLC16A3 in S–I subtype, TSTA3 and PKP2 in S-II subtype, and AMBP, HBB, and CA1 in S-III subtype. Scale bar = 50 μm. J. Potential druggable targets and their corresponding agents for S-III subtype. ∗∗p < 0.01, ∗p < 0.05.

Enrichment analysis revealed distinct biological features among the three subtypes. The S-I subtype was enriched in inflammation- and immune-related pathways, such as the NOD-like receptor signalling, IL-17 signalling, C-type lectin receptor signalling, Rap1 signalling, and Fc epsilon RI signalling pathway (Fig. 4E). The S-II subtype was characterised by proteins involved in metabolic and oxidative stress pathways, including chemical carcinogenesis - ROS, TCA cycle, pyruvate metabolism, and biosynthesis of cofactors (Fig. 4E). The S-III subtype lacked immune-related enrichments but featured oncogenic pathways, including extracellular matrix (ECM)-related process, proteasome, DNA replication, and mismatch repair (Fig. 4E). Consequently, the subtypes were defined as immune activation subtype, energy metabolism subtype, and extracellular matrix dynamics subtype (Fig. 4F).

Machine learning models were employed to identify key proteins for each subtype. Lasso regression was identified as the most effective algorithm (ce = 0.164, mauc = 0.933) (Fig. 4G, Supplementary Figure S4A, Supplementary Table S3). This model selected 10 key proteins: SLC16A3, TNC, PKP2, TSTA3, AMBP, CA1, HBB, PRELP, SH4GLB1, and SYNPO2 (Supplementary Figure S4B and C, Supplementary Table S4). The model's reliability was validated in two external cohorts, including the TCGA-OV-2016 dataset¹² (Fig. 4H). These findings demonstrate the potential of this 10-protein model to enhance HGSOC subtype differentiation and improve therapeutic precision. To identify subtype-specific prognostic biomarkers, we performed an overall survival (OS) analysis across the entire cohort. This analysis revealed key biomarkers significantly associated with survival in each subtype: SLC16A3 in S-I; TSTA3 and PKP2 in S-II; and AMBP, HBB, and CA1 in S-III (Supplementary Figure S4D and E). These representative biomarkers were subsequently validated in OC tissue slides using mIHC (Fig. 4I). Given the poor prognosis of the S-III subtype, we identified 13 overexpressed proteins as potential druggable targets, which were further investigated for FDA-approved drugs (Fig. 4J). This bioinformatic observation suggests they represent candidate therapeutic targets in HGSOC, though this requires direct experimental validation.

To contextualise our proteomic subtypes within existing HGSOC classification, we performed a comparative analysis with proteomic data from the TCGA cohort. Using consensus clustering, we independently identified three molecular subtypes in both cohorts, which exhibited coincident prognostic and biological characteristics. The C2 (TCGA) and S2 (our cohort) subtypes were associated with favourable outcomes, while the C3 and S3 subtypes consistently demonstrated the poorest prognosis (Supplementary Figure S5A and B). Biologically, C1 and S1 displayed activated interferon response, C2 and S2 were enriched in unfolded protein response, C3 and S3 showed activation of complement and coagulation cascade signalling pathways and epithelial–mesenchymal transition (EMT) (Supplementary Figure S5E–H). In the poorest-prognosis subtypes, C3 and S3, coagulation factors, fibrinogen family molecules, complement molecules, and their receptors were significantly upregulated (Supplementary Figure S5E and F), suggesting that enhanced complement and coagulation cascade pathways and increased heterogeneity in the immune microenvironment may be associated with poor patient prognosis. Further immune profile analysis revealed that C1 and C2 (TCGA) exhibited stronger immune infiltration, including higher CD4⁺ and CD8⁺ T cell scores, while C3 enriched in stromal components such as platelets, fibroblasts, and macrophages (Supplementary Figure S5I and J). Parallel results were observed in our cohort: S1 and S2 had elevated CD4⁺ and CD8⁺ T cell scores relative to S3. Notably, platelet activation was significantly elevated in both C3 and S3, aligning with protein-level enrichment of complement and coagulation pathways. To further validate our proteomic classification, we mapped TCGA transcriptomic subtypes onto our proteomic subtypes. The Sankey diagram illustrates a substantial correspondence between the two classification systems. Specifically, the Differentiated transcriptomic subtype was predominantly observed in proteomic clusters S-II (n = 17) and S–I (n = 5). The Immunoreactive subtype was mainly enriched in S-II (n = 20). The Proliferative subtype showed a distribution across S-II (n = 18), S-III (n = 8), and S-I (n = 1). Notably, the Mesenchymal (MES) subtype was primarily concentrated in S-III (n = 14) and S-I (n = 9), with only 6 cases falling into S-II (Supplementary Figure S6A). Quantification confirmed significant enrichment of the TCGA MES subtype within our S-III proteomic subtype. Among S-III patients, 48% (14/29) were classified as MES, compared to only 10% (6/61) in S-II and 45% (9/20) in S-I. Fisher's exact test demonstrated strong association (OR = 4.11, p = 0.003), validating that S-III presents the aggressive, high-stromal features associated with poor prognosis. Conversely, S-II tumours showed a mixed composition predominantly mapping to Immunoreactive and Differentiated subtypes, aligning with their better prognosis (Supplementary Figure S6A and B). Additionally, we projected the annotation results of single-cell RNA sequencing from other study onto our cohort.⁶⁰ The results further validated our findings that dysfunctional T cells significantly enriched in S3, while activated T cells were significantly enriched in S1 and S2 (Supplementary Figure S7).

TFs activity identified the three proteomic subtypes

In HGSOC, the complex interaction of TFs plays a crucial role in regulating gene expression and driving tumour progression and heterogeneity. Our pySCENIC analysis identified 75 TFs with significantly elevated activity, which may contribute to HGSOC pathogenesis. Using Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction, we found that TF activity could differentiate three proteomic subtypes (Fig. 5A). Notably, TF activity did not correlate with other clinical features, suggesting that the proteomic subtypes are driven by TF activity (Supplementary Figure S8A). We also examined the relationship between TFs and their target genes (TGs), demonstrating how they correlated with the three proteomic subtypes and the regulatory network between TFs and TGs (Fig. 5B). Further analysis revealed that specific TFs were dominant in each proteomic subtype: UBTF, STAT1, STAT2, STAT3, and CBFB in the S-I subtype; MAX, PML, TFCP2, POLE4, NFYB, WT1, and RB1 in the S-II subtype; and NFIB, NFIC, NFIX, RBBP5, TEAD1, NR2F6, and CHD2 in the S-III subtype. Cox regression analysis showed that these TFs were associated with survival outcomes, with S-II having the best prognosis and S-III the worst. Potential therapeutic agents targeting these specific TFs were also presented (Fig. 5C, Supplementary Figure S8B). STAT1 was identified as a significant protective biomarker. Our results suggest Cisplatin as a potential agent linked to STAT1. Cisplatin has served as a cornerstone in the systemic treatment of advanced epithelial OC for decades. Preclinical studies have shown that Cisplatin treatment can activate STAT1, inducing tumour cell death.⁶¹ This mechanistic connection provides additional molecular support for the established role of Cisplatin as a first-line treatment for HGSOC. Potential natural compounds targeting CBFB, such as Genistein and Resveratrol, were identified. Preclinical studies indicates that Genistein inhibits the stress response in tumour cells and contribute to programmed cell death via modulation of transcriptional complexes NF–Y/CBFB.⁶² Furthermore, a clinical trial illustrating consumption of resveratrol before surgery reduced colorectal cancer cell proliferation by 5% (p = 0.05).⁶³ This evidence highlights the translational promise of natural compounds in HGSOC therapy.

We also identified TGs regulated by these TFs, which were significantly associated with prognosis. Notably, COL21A1, CA1, HBB, and HBD were negatively associated with survival and linked to ECM regulation (Supplementary Figure S8C). The ECM receptor interaction pathway was identified as a key pathway in the S-III subtype (Fig. 4E and F), with a positive correlation between the four TGs and ECM receptor interaction scores (Spearman's cor = 0.305, 0.388, 0.427, and 0.398, p < 0.05) (Supplementary Figure S8D), as well as between corresponding TFs and ECM receptor interaction scores (Spearman's cor = 0.754, 0.729, and 0.668, p < 0.05) (Supplementary Figure S8E). These findings suggest that these TFs regulate ECM receptor interactions via their target genes, promoting tumour invasion and worsening prognosis.

Further analysis of ECM receptor interaction scores revealed a significant negative correlation with prognosis (Fig. 5D). The S-III subtype was more likely to be diagnosed at an advanced stage, with upregulation of the four TGs in advanced cases (Fig. 5E). GSEA indicated significant enrichment of cancer invasion pathways in the S-III subtype (Supplementary Figure S8F). The ECM receptor interaction score and invasion degree were positively correlated (Spearman's cor = 0.814, p < 0.05), with the four TGs correlating with invasion (Spearman's cor = 0.309, 0.188, 0.204, and 0.211, p < 0.05) (Fig. 5F, Supplementary Figure S8G). Moreover, we identified six downstream TGs (APCS, OLFML1, OGN, COL14A1, MTX2, and TINAGL1) in the S-III subtype (Fig. 5G). Among them, OGN is involved in tumour immune responses.⁶⁴ Immune escape was found to be significantly higher in S-III and correlated with poor survival (Fig. 5H and I). Most patients in the S-III subgroup were classified as IC2 (cold tumour subtype), with higher stromal and ESTIMATE score (Supplementary Figure S8H). These results suggest that the TGs regulate immune cells, contributing to immune escape. TME analysis revealed significant correlations between TF activity, TG expression, and immune cells in the S-III subtype (Fig. 5J and K). Endothelial and fibroblasts activity correlated with poor prognosis, while macrophages and NKT cells were favourable (Fig. 5L). Furthermore, significant correlation was found between the activity of those cells and the immune escape index (Spearman's cor = 0.241, 0.386, −0.191, and −0.217, p < 0.05) as well as the expression of those TGs and the immune escape index (Spearman's cor = 0.417, 0.344, 0.393, 0.346, 0.414, 0.266 and 0.403, p < 0.05) (Fig. 5M, Supplementary Figure S8I). These findings support the hypothesis that TFs regulate immune escape via TGs, leading to worse survival outcomes in the S-III subtype (Fig. 5N, right). In conclusion, our study uncovers the role of specific TFs and their regulatory networks in driving progression and immune evasion in HGSOC, highlighting potential therapeutic targets.

The immune clustering of HGSOC revealed three distinct subgroups with varying characteristics in the immune tumour microenvironment

We performed immune clustering of HGSOC and identified three subgroups with distinct immune and stromal features. TME plays a critical role in cancer progression, and while immunotherapy has shown promise in various cancers, its efficacy in OC remains limited. To better understand immune infiltration in HGSOC at the protein level, we used Xcell for cell-type deconvolution analysis. Consensus clustering based on inferred cell proportions identified three subgroups: IC1 (n = 45), IC2 (n = 43), and IC3 (n = 28) (Fig. 6A). These subgroups were not correlated with FIGO stages but showed a significant association with OS. IC1 demonstrated the best survival, while IC2 had the worst (Fig. 6B–C). IC1 was characterised by a high proportion of NKT and Th1 cells with pathway enrichment showing upregulation of immune-related pathways, including antigen processing and presentation, and NK cell-mediated cytotoxicity (Fig. 6A). In contrast, IC2 exhibited an enrichment of fibroblasts and haematopoietic stem cells (HSC), with upregulation of the TGF-beta signalling pathway, ECM receptor interaction, and regulation of fibroblast migration (Fig. 6A). IC3 had the highest proportion of CD4+ T cells, CD4+ Tcm, and Th2 cells, with pathways related to DNA replication and cell receptor signalling upregulated (Fig. 6A). These characteristics were summarised in Fig. 6D, with subtypes named: NKT/Th1 subtype for IC1, cold tumour for IC2, and CD4+ T/CD4+ Tcm for IC3. IC2, with the worst prognosis, had a lower immune score and a higher stromal score compared to IC1 and IC3. IC3, as the intermediate subtype, showed relatively higher stromal scores, with a similar immune score to IC1 (Fig. 6F). Notably, higher stromal scores corrected with poorer prognosis (Fig. 6E).

Fig. 6 — **The immune clustering of HGSOC revealed three distinct subgroups with varying characteristics in the immune tumour microenvironment**. A. Immune cluster based on xCell scores and clinicopathologic features of samples (top panel), xCell signatures, significant proteins and enriched KEGG pathways in 3 ImmuneCluster subtypes (bottom panel). B. Kaplan–Meier curves for OS based on the 3 ImmuneCluster subtypes (log-rank test). C. Distribution of early and advanced samples in 3 ImmuneCluster subtypes (left panel). Distribution of FIGO stages in 3 ImmuneCluster subtypes (right panel). D. Contour plot of two-dimensional density based on stromal scores (y-axis) and immune scores (x-axis) for different ImmuneCluster subtypes. E. Kaplan–Meier curves for OS based on the stromal score (log-rank test). F. Boxplots showing stromal score in 3 ImmuneCluster subtypes (Spearman's correlation). G. Correlation analysis between the regulation of fibroblast migration and stromal score (Spearman's correlation). H. Venn plot of the 34 genes in regulation of fibroblast migration and the 556 genes upregulated in IC2 subtype. I. Heatmap showing the 4 DEPs (FGF2, ITGB1, GNA12, and AKAP12) among the 3 ICs. J. Volcano plot indicating the correlation between the expression of the DEPs and the prognosis of HGSOC. K. Protein expression levels of FGF2 determined by Western Blot assay in primary CAF following knockdown with specific siRNAs of FGF2. L. Cell Proliferation determined by CCK-8 assay (Student's t-test). M. Cell migration were assessed via transwell assay following knockdown with specific siRNAs of FGF2. Scale bar = 200 μm. N. Representative images of filopodia staining performed in primary CAFs. Scale bar = 10 μm ∗∗∗*p <* 0.001, ∗∗p < 0.01, ∗p < 0.05.

To explore the mechanism behind the elevated stromal score in IC2, we found a significant positive correlation between the regulation of fibroblast migration signalling and stromal score (Fig. 6G). Four genes (FGF2, ITGB1, GNA12, AKAP12) involved in this pathway were upregulated in IC2 (Fig. 6H–I). FGF2, in particular, was a prognostic marker for survival (Fig. 6J). Functional assays in CAFs revealed that FGF2 knockdown reduced cell proliferation, migration, and filopodia formation (Fig. 6K–N), suggesting FGF2's role in fibroblast migration and stromal component promotion in IC2, contributing to the worse prognosis.

To investigate the biological and clinical correlations between the proteomic subtypes (S-I, S-II, S-III) and immune subtypes (IC1, IC2, IC3), we compared their prognostic and biological characteristics. We found that the S-III proteomic subtype and the IC2 immune subtype, both linked to the poorest prognosis, demonstrated immunosuppressive phenotype with the enrichment in complement and coagulation cascade pathways. Conversely, the S-I and IC1 subtypes, which shared an immune-activated phenotype, predicted a more favourable prognosis (Fig. 7A–D, Supplementary Figure S5D). This convergence of prognostic and biological features between two independent classification highlights their complementary value in deciphering the heterogeneity of HGSOC. Furthermore, we evaluated the combined prognostic contribution of the proteomic and immune subtypes using a multivariate Cox model. The overall model demonstrated significant prognostic stratification (p = 0.013) (Supplementary Table S5). Within this model, the immune subtype remained an independent prognostic factor (p = 0.045), whereas the proteomic subtype did not reach statistical independence (p = 0.134). While immune subtype offers robust prognostic stratification, proteomic subtyping may provide essential insights on tumour intrinsic biology and direct therapeutic implications. To demonstrate this, we have further focused on the IC-2 immune subgroup and applied our proteomic subtypes. Differential expression analysis and pathway enrichment revealed that within the same immune context (IC2), IC-2/S-II tumours showed significant enrichment in immune-related pathways including antigen processing and presentation, interferon signalling pathway, and mitochondrial oxidative phosphorylation. These pathways indicate an immune-hot phenotype with active antigen presentation and metabolic energy production. In contrast, IC-2/S-III tumours is characterised by extracellular matrix remodelling, glycosaminoglycan metabolism, and absent antigen presentation function. These features characterise a mesenchymal/stromal-rich phenotype consistent with an immune-cold microenvironment with absent interferon signalling and antigen presentation (Supplementary Figure S9A and B). In addition, we examined the expression of actionable drug target proteins within IC-2 subgroups. Among the differentially expressed targets, IC-2/S-II showed significantly higher expression of metabolic enzymes and immune-related targets, while IC-2/S-III exhibited elevated expression of stromal targets. These findings suggest distinct therapeutic strategies: immunotherapy may be more suitable for IC-2/S-II (immune-hot), while anti-stromal/anti-ECM approaches may benefit IC-2/S-III patients (Supplementary Figure S9C). The mechanism diagram summarises the divergent immune-metabolic programs within IC-2 (Supplementary Figure S9D).

Fig. 7 — **TILs determined the prognosis of patients with advanced HGSOC**. A. Correlation analysis among the FIGO stage, immune cluster and proteomic subtypes. B. Kaplan–Meier curves for OS based on the IC1 and IC2 subtypes (log-rank test). C. Volcano plot indicating the prognosis-associated DEPs in advanced-stage HGSOC. D. Performance of the indicated algorithm combinations in the test set and the two external independent cohorts (TCGA-OV-2016 and PXD010372). E. Correlation analysis of the prognosis and protein expression (Pearson correlation analysis). F. Forest plot suggesting the correlation between the prognosis and the protein expression. G. Kaplan–Meier curves for OS based on the model in our dataset and the two external independent cohorts (log-rank test). H. The box plot indicated the difference of the expression in IC1/IC2 subtypes (Wilcoxon rank-sum test). I. Advanced HGSOC can be classified into IC1 and IC2 subtypes based on the expression of ISG15, ITGB2, and RELA. J. Forest plot suggesting the correlation between the prognosis and the cancer signatures (Cox proportional hazards regression). K. GSEA enrichment of the IC1 and IC2 subtypes in advanced stage (up panel), the heatmap of the 10 proteins in IC1 and IC2 subtypes (bottom panel). ∗∗∗*p <* 0.001, ∗∗p < 0.01, ∗p < 0.05.

Tumour-infiltrating lymphocytes (TILs) determined the prognosis of patients with advanced HGSOC

Further analysis of the cohort of patients with advanced-stage HGSOC revealed that immune subtypes IC1 and IC2 were evenly distributed, implying that immune factors may influence prognosis at this stage (Fig. 7A and B). By overlapping 202 immune-related genes associated with prognosis, 24 genes were selected for further analysis (Fig. 7C, Supplementary Figure S10A). These 24 genes were then assessed through 72 combinations of 9 different algorithms (GBM, Lasso, Enet, StepCox, SVM, RSF, ridge, SuperPC, and plsRcox) for feature selection and development of prognostic models (Fig. 7C, Supplementary Figure S10B–D, Supplementary Table S6). The performance of these 72 algorithm combinations were evaluated using the surv-cindex in our dataset and two external cohorts (TCGA-OV-2016¹² and PXD010372⁴⁰) (Fig. 7D, Supplementary Table S6). After averaging the surv-cindex scores from the external cohorts, GBM was selected for feature selection, and SuperPC for modelling, as it demonstrated the best performance (surv-cindex = 0.66, 0.51 and 0.69) (Fig. 7D). In Fig. 7E, five genes (RELA, MAPK14, ITGB2, ISG15, CHP1) were identified for model development. Among these, CHP1 was positively associated with tumour progression and prognosis, while the other genes were associated with favourable tumour progression and improved patient survival (Fig. 7E and F). A prognostic prediction model was conducted using these five genes to classify patients with HGSOC into high- and low-risk subtypes. As shown in Fig. 7G, the high-risk group had significantly worse OS than the low-risk group in both our dataset and the two external validation datasets, highlighting the critical role of these five genes in patients with advanced-stage HGSOC.

To explore the potential mechanism underlying the differing prognosis linked to these five genes, patients with advanced-stage HGSOC were categorised into IC1 and IC2 subtypes. Survival analysis revealed significant differences in both OS and DFS between these subtypes in patients with advanced-stage tumours (log-rank test, p < 0.05) (Fig. 7B, Supplementary Figure S10E and F). Expression levels of three genes (ISG15, ITGB2, RELA) were significantly higher in the IC1 subtype compared to the IC2 subtype (Fig. 7H). As indicated in Fig. 7I and Supplementary Figure S10G, these three genes contributed to the reclassification of advanced-stage HGSOC into two subtypes with distinct molecular features and survival outcomes.

To explore the downstream effects of these three genes, immune signature activity was assessed in 116 HGSOC samples. Fig. 7J shows that the activity of three immune signatures—related to CCR, TILs, and proinflammatory response—favoured survival in patients with advanced-stage HGSOC, with TILs and proinflammatory response significantly enriched in the IC2 subtype. GSEA and heatmap analysis of the 10 core genes in the TILs signature demonstrated significantly higher enrichment in the IC2 subtype compared to the IC1 subtype (Fig. 7K, Supplementary Figure S10H). Additionally, the proinflammatory response signature was also significantly enriched in the IC2 subtype (Fig. 7B).

Previous studies have reported the impact of ISG15, ITGB2, RELA on inflammatory response and their regulation on TILs.65, 66, 67, 68 Consistent with these findings, positive correlations were confirmed between these three genes and the core genes in the TILs signature (Supplementary Figure S10I–K). As shown in Supplementary Figure S11A, IHC staining for CD3 and CD45 indicated immune infiltration in HGSOC tissues. The expression of ISG15, ITGB2, RELA was higher in tissues with greater immune infiltration compared to those with lower immune infiltration (Supplementary Figure S11A). In summary, the high expression of ISG15, ITGB2, RELA in patients with advanced-stage HGSOC significantly activated NKT/Th1 cell infiltration and promoted TILs through an inflammatory response, correlating with better prognosis. Conversely, low expression of these genes led to an immune-exhausted subtype and worse prognosis (Supplementary Figure S11B). This study identified a 3-gene signature that predicted the prognosis of advanced-stage HGSOC by influencing TILs levels and immune responses.

Discussion

Despite significant advancements in precision oncology through genomic profiling, targeting genetic abnormalities in HGSOC has notable limitations. Traditionally, the FIGO staging system has been used to classify HGSOC based on clinical staging, such as FIGO stage II or IIA. However, there is limited understanding of the molecular turning point in HGSOC progression. In this study, we performed a systematic proteomic analysis of 116 patients with HGSOC, uncovering protein profiles and biological insights that could inform clinical management (Fig. 8). Our findings identified FIGO stage IIA as the critical turning point in disease progression and highlighted tumour cell immune escape as a key event in HGSOC progression. Additionally, we established a proteomic subtyping system, explored mechanisms influencing prognosis, and nominated druggable targets for the worst-prognosis subtype, focussing particularly on immune microenvironment factors.

Fig. 8 — **Summary of molecular characteristics based on proteomic clusters in 116 HGSOC samples**.

Previous studies have noted variable prognoses across FIGO stages. Our time-series analysis revealed FIGO stage IIA as a pivotal transition from early to advanced stages, with early-stage cases enriched in oxidative stress pathways, while advanced stages were dominated by cell cycle-related signalling. This aligns with previous research connecting oxidative stress to cell cycle progression,69, 70, 71 where ROS influences mitotic arrest and may synergise with cell cycle events in driving HGSOC progression.⁷² GOSR2 emerged as a critical protein influencing the transition from early to advanced stages. While its role in cancer is understudied,⁷³^,⁷⁴ our findings suggest that GOSR2 inhibition could suppress tumour growth in HGSOC and enhance CD8+ T cell activity, potentially through the regulation of SEC24D. Bioinformatics analysis further supported the association of SEC24D with CD8+ T cell infiltration,⁷⁵ indicating that GOSR2-SEC24D signalling play a pivotal role in HGSOC progression.

Proteome analysis further revealed distinct pathway activation between early and advanced-stage samples. Early-stage samples were primarily driven by metabolic alterations, while advanced-stage samples exhibit enrichment in cancer-related pathways, including VEGF and hypoxia signalling. Notably, PPP3R1 was associated with HGSOC progression and poor prognosis, potentially by activating the VEGF pathway. Our experiments further validated this bio-informative analysis, which underscores the involvement of PPP3R1 in driving the progression of advanced HGSOC.

We classified HGSOC three proteomic subtypes (S-I, S-II, and S-III) based on our proteomic data. The S-III subtype, characterised by high invasiveness, ECM-receptor interaction, and poor prognosis, exhibited a distinct molecular signature, consistent with previous reports linking ECM interactions to tumour invasion.⁷⁶ The S-I subtype was associated with immune activation, while the S-II subtype displayed energy metabolism alterations. Importantly, the S-III subtype may benefit from anti-invasive or ECM-modulating therapies, as the poor prognosis of this subtype is linked to ECM interactions and tumour immune escape. Our analysis of TFs such as NFIX, CHD2, RBBP5, NR2F6, and NFIC, which regulate ECM interactions and immune escape, further supports the potential of targeting these TFs in therapy. NFIX, for example, regulates oxidative stress⁷⁷ and may promote ECM-receptor interaction in HGSOC, while CHD2 and RBBP5 influence migration and invasion through EMT processes.⁷⁸^,⁷⁹

We also identified three immune subtypes (IC1, IC2, and IC3) based on immune cell infiltration. The IC2 subtype, with an immune-desert phenotype, had the worst prognosis, while IC1 subtypes, characterised by NKT and Th1 cell infiltration, were associated with better survival. These immune profiles suggest that enhancing immune cell infiltration, particularly of NKT cells, may improve patient outcomes.80, 81, 82 IC2 subtype, enriched in fibroblasts and HSCs, showed increased TGF-β signalling, which may suppress immune function and promote stromal activation. FGF2, a key regulator of fibroblast migration, was identified as a critical protein influencing poor prognosis in IC2. Its role in OC progression and T-cell exhaustion⁸³^,⁸⁴ suggests that targeting FGF2 could mitigate the adverse effects of stromal activation on immune response. Our study underscores the variability in immune profiles and their influence on prognosis, emphasising the need for tailored immunotherapies. ICIs targeting PD-1, PD-L1, CTLA-4, and LAG3 have shown promise in other cancers, but their monotherapy efficacy in HGSOC is limited. Our data suggest that combining ICIs with agents that modulate the TME may overcome resistance. Patients with advanced-stage tumours characterised by high expression of ISG15, ITGB2, and RELA might exhibit a tumour microenvironment more permissive to immunotherapy, which requires validation in cohorts treated with immune checkpoint inhibitors, while targeting FGF2 may improve outcomes for patients with the immune-desert IC2 subtype.

While our proteomics analysis is powerful for molecular classification and biomarker identification, its clinical implementation is challenging. To bridge this gap, we validated the biomarkers in OC tissue slides for each proteomic subtype (S–I to S-III) using mIHC assay. This assay enables the simultaneous detection of multiple protein markers on a single tissue section, conserving precious patient samples while providing critical spatial information. However, the mIHC assay still has limitations, including high cost and relatively complex protocol, which is a hindrance to its widespread adoption. We believe a simplified and cost-effective mIHC assay will enable patients’ classification in routine diagnostics, thereby directly informing personalised treatment decisions. Specifically, the immune-activated S-I subtype, which frequently aligns with the IC1 immune cluster, exhibits an immunologically active microenvironment. Patients within these subgroups may represent optimal candidates for immunotherapy-based regimens. This subtype could also serve as a biomarker for screening responders in future OC clinical trials of immunotherapy. Conversely, the metabolically driven S-III subtype demonstrates enrichment of pathways related to ECM remodelling and altered cellular metabolism. This molecular profile demonstrating potential vulnerability to metabolic pathway inhibitors or stromal-targeting agents, offering a rational alternative for this patient population that may respond poorly to conventional chemotherapy and immunotherapy.⁸⁵^,⁸⁶

To further guide precision therapy, we combined proteomic subtypes and immune subtypes, conducting analysis between two different proteomic subgroups within the same immune context. The results illustrated that IC2/S-II represents a relatively “immune active” subtype characterised by OXPHOS metabolism, IFN signalling, and active antigen presentation which makes these patients potential candidates for ICIs. IC2/S-III represents an “immune evasive” subtype characterised by ECM remodelling, glycosaminoglycan metabolism, and absent antigen presentation. This phenotype suggests potential benefit from anti-stromal therapies targeting collagen/integrin pathways. Thus, we propose a two-tier stratification system. At the first tier, immune subtypes are used to determine the baseline prognosis of patients. At the second tier, for patients within the IC-2 immune microenvironment, proteomic subtypes including S-II and S-III are further employed to guide the selection of therapeutic strategies.

While this study provides comprehensive insights into the proteomic landscape of HGSOC, we acknowledge certain limitations. Regarding the in vivo validation, we recognise the limitations of the subcutaneous xenograft model, including the small sample size (n = 4 per group) and the use of immunodeficient mice, which restricts the evaluation of immune-mediated antitumour effects. In addition, the cohort included a smaller number of early-stage tumours (FIGO I-IIA) compared to advanced-stage cases, which may cause statistical bias. However, this imbalanced distribution of patient population actually reflects the clinical reality of HGSOC that over 75% of patients are diagnosed at an advanced stage due to the lack of early symptoms and effective screening. To ensure the robustness of our findings, we externally validated key findings by projecting data onto independent single-cell RNA-seq datasets and comparing our proteomic subtypes with those in the TCGA cohort. Beyond this, we will further validate our study in a larger cohort in the future.

Our proteomic analysis provides a comprehensive understanding of HGSOC, highlighting stage-specific molecular features and immune microenvironment contributions to prognosis. The identification of proteomic subtypes and candidate targets reveals molecular heterogeneity that could guide future precision therapy strategies, including potential avenues focussing on ECM interactions, immune modulation, and metabolic pathways. Future prospective studies and clinical trials will be essential to validate these findings and refine therapeutic strategies for HGSOC based on proteomic subtypes. Such efforts may also inform precision medicine approaches in other cancers.

Contributors

J.X. C.D. and W.L. conceived, supervised and administered the study. S.T. and Y.C. collected the samples, the clinical data, extracted the protein, and performed the follow-up of patients. M.T., Q.Z., T.G., and X.X. performed the bioinformatic data analysis and interpreted the results. M.T., S.T., Q.Z., and T.G. wrote the original draft. M.T., S.T., and T.G. assisted with the data analysis, performed the in vitro verification experiments. M.T., T.G., S. W. and X.C. performed the in vivo experiments. J.X, M.T., S.T., and T.G. accessed and verified the underlying data. J.X. C.D. and W.L. revised the manuscript. J.X. acquired funding. All authors read and approved the final manuscript.

Data sharing statement

The raw files and the processed, de-identified protein expression matrix have been deposited to the ProteomeXchange Consortium (https://proteomecentral.proteomexchange.org) via the iProX partner repository⁸⁷^,⁸⁸ with the dataset identifier PXD074808. The datasets generated and/or analysed during the current study are available from the corresponding author upon reasonable request.

Declaration of interests

The authors declared the following financial interests/personal relationships which may be considered as potential competing interests: J.X., S.T., M.T., T.G., and S.W. has patent 2024100926916 submitted to the Chinese Patent Office, J.X., S.T., and M.T. has patent 2024100882316 submitted to the Chinese Patent Office.

Acknowledgements

This research was supported by the Key R&D Program of Zhejiang, China (Grant No. 2026C02A1110 to J.X), the National Natural Science Foundation of China (Grant No. 82472891 to J.X), the Zhejiang Provincial Natural Science Foundation of China (Grant No. LZ24H160001 to J.X), and 4+X Clinical Research Project of Women's Hospital, School of Medicine, Zhejiang University (Grant No. ZDFY2022-4X202 to J.X).

Footnotes

^{Appendix A}

Supplementary data related to this article can be found at https://doi.org/10.1016/j.ebiom.2026.106248.

Contributor Information

Weiguo Lu, Email: lbwg@zju.edu.cn.

Chen Ding, Email: chend@fudan.edu.cn.

Junfen Xu, Email: xjfzu@zju.edu.cn.

Appendix A. Supplementary data

Supplementary Figures and Tables

mmc1.pdf^{(9.2MB, pdf)}

Supplemental Western blots

mmc2.pdf^{(5.1MB, pdf)}

Cell line validation file

mmc3.pdf^{(4.9MB, pdf)}

References

1.Sung H., Ferlay J., Siegel R.L., et al. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2021;71(3):209–249. doi: 10.3322/caac.21660. [DOI] [PubMed] [Google Scholar]
2.Picard M. Why do we care more about disease than health? Phenomics. 2022;2(3):145–155. doi: 10.1007/s43657-021-00037-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
3.Siegel R.L., Miller K.D., Fuchs H.E., Jemal A. Cancer statistics, 2022. CA Cancer J Clin. 2022;72(1):7–33. doi: 10.3322/caac.21708. [DOI] [PubMed] [Google Scholar]
4.Lheureux S., Braunstein M., Oza A.M. Epithelial ovarian cancer: evolution of management in the era of precision medicine. CA Cancer J Clin. 2019;69(4):280–304. doi: 10.3322/caac.21559. [DOI] [PubMed] [Google Scholar]
5.Kuroki L., Guntupalli S.R. Treatment of epithelial ovarian cancer. BMJ. 2020;371 doi: 10.1136/bmj.m3773. [DOI] [PubMed] [Google Scholar]
6.Lheureux S., Gourley C., Vergote I., Oza A.M. Epithelial ovarian cancer. Lancet. 2019;393(10177):1240–1253. doi: 10.1016/S0140-6736(18)32552-2. [DOI] [PubMed] [Google Scholar]
7.Ying W. Phenomic studies on diseases: potential and challenges. Phenomics. 2023;3(3):285–299. doi: 10.1007/s43657-022-00089-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Peres L.C., Cushing-Haugen K.L., Köbel M., et al. Invasive epithelial ovarian cancer survival by histotype and disease stage. J Natl Cancer Inst. 2019;111(1):60–68. doi: 10.1093/jnci/djy071. [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Harter P., Sehouli J., Lorusso D., et al. A randomized trial of lymphadenectomy in patients with advanced ovarian neoplasms. N Engl J Med. 2019;380(9):822–832. doi: 10.1056/NEJMoa1808424. [DOI] [PubMed] [Google Scholar]
10.Xu J., Fang Y., Chen K., et al. Single-cell RNA sequencing reveals the tissue architecture in human high-grade serous ovarian cancer. Clin Cancer Res. 2022;28(16):3590–3602. doi: 10.1158/1078-0432.CCR-22-0296. [DOI] [PMC free article] [PubMed] [Google Scholar]
11.Integrated genomic analyses of ovarian carcinoma. Nature. 2011;474(7353):609–615. doi: 10.1038/nature10166. [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Zhang H., Liu T., Zhang Z., et al. Integrated proteogenomic characterization of human high-grade serous ovarian cancer. Cell. 2016;166(3):755–765. doi: 10.1016/j.cell.2016.05.069. [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Finn R.S., Qin S., Ikeda M., et al. Atezolizumab plus bevacizumab in unresectable hepatocellular carcinoma. N Engl J Med. 2020;382(20):1894–1905. doi: 10.1056/NEJMoa1915745. [DOI] [PubMed] [Google Scholar]
14.Gandhi L., Rodríguez-Abreu D., Gadgeel S., et al. Pembrolizumab plus chemotherapy in metastatic non-small-cell lung cancer. N Engl J Med. 2018;378(22):2078–2092. doi: 10.1056/NEJMoa1801005. [DOI] [PubMed] [Google Scholar]
15.Rini B.I., Plimack E.R., Stus V., et al. Pembrolizumab plus axitinib versus sunitinib for advanced renal-cell carcinoma. N Engl J Med. 2019;380(12):1116–1127. doi: 10.1056/NEJMoa1816714. [DOI] [PubMed] [Google Scholar]
16.Morazán-Fernández D., Mora J., Molina-Mora J.A. In silico pipeline to identify tumor-specific antigens for cancer immunotherapy using exome sequencing data. Phenomics. 2023;3(2):130–137. doi: 10.1007/s43657-022-00084-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
17.Matulonis U.A., Shapira-Frommer R., Santin A.D., et al. Antitumor activity and safety of pembrolizumab in patients with advanced recurrent ovarian cancer: results from the phase II KEYNOTE-100 study. Ann Oncol. 2019;30(7):1080–1087. doi: 10.1093/annonc/mdz135. [DOI] [PubMed] [Google Scholar]
18.Zamarin D., Burger R.A., Sill M.W., et al. Randomized phase II trial of nivolumab versus nivolumab and ipilimumab for recurrent or persistent ovarian cancer: an NRG oncology study. J Clin Oncol. 2020;38(16):1814–1823. doi: 10.1200/JCO.19.02059. [DOI] [PMC free article] [PubMed] [Google Scholar]
19.Hegde P.S., Chen D.S. Top 10 challenges in cancer immunotherapy. Immunity. 2020;52(1):17–35. doi: 10.1016/j.immuni.2019.12.011. [DOI] [PubMed] [Google Scholar]
20.Ni D., Zhou H., Wang P., Xu F., Li C. Visualizing macrophage phenotypes and polarization in diseases: from biomarkers to molecular probes. Phenomics. 2023;3(6):613–638. doi: 10.1007/s43657-023-00129-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
21.Hornburg M., Desbois M., Lu S., et al. Single-cell dissection of cellular components and interactions shaping the tumor immune phenotypes in ovarian cancer. Cancer Cell. 2021;39(7):928–944.e6. doi: 10.1016/j.ccell.2021.04.004. [DOI] [PubMed] [Google Scholar]
22.Olbrecht S., Busschaert P., Qian J., et al. High-grade serous tubo-ovarian cancer refined with single-cell RNA sequencing: specific cell subtypes influence survival and determine molecular subtype classification. Genome Med. 2021;13(1):111. doi: 10.1186/s13073-021-00922-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Wisniewski J.R., Zougman A., Nagaraj N., Mann M. Universal sample preparation method for proteome analysis. Nat Methods. 2009;6(5):359–362. doi: 10.1038/nmeth.1322. [DOI] [PubMed] [Google Scholar]
24.Hastie T.T.R., Narasimhan B., Chu G. 2023. R Package Version 1760. [Google Scholar]
25.Seckinger A., Meissner T., Moreaux J., et al. Clinical and prognostic role of annexin A2 in multiple myeloma. Blood. 2012;120(5):1087–1094. doi: 10.1182/blood-2012-03-415588. [DOI] [PubMed] [Google Scholar]
26.Alboukadel Kassambara F.M. 2020. R Package Version 107. [Google Scholar]
27.Wu T., Hu E., Xu S., et al. clusterProfiler 4.0: a universal enrichment tool for interpreting omics data. Innovation (Camb) 2021;2(3) doi: 10.1016/j.xinn.2021.100141. [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Subramanian A., Tamayo P., Mootha V.K., et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A. 2005;102(43):15545–15550. doi: 10.1073/pnas.0506580102. [DOI] [PMC free article] [PubMed] [Google Scholar]
29.Hänzelmann S., Castelo R., Guinney J. GSVA: gene set variation analysis for microarray and RNA-Seq data. BMC Bioinformatics. 2013;14:7. doi: 10.1186/1471-2105-14-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
30.Futschik M.E., Carlisle B. Noise-robust soft clustering of gene expression time-course data. J Bioinform Comput Biol. 2005;3(4):965–988. doi: 10.1142/s0219720005001375. [DOI] [PubMed] [Google Scholar]
31.Kumar L., Futschik M. Mfuzz: a software package for soft clustering of microarray data. Bioinformation. 2007;2(1):5–7. doi: 10.6026/97320630002005. [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Chen L., Liu R., Liu Z.P., Li M., Aihara K. Detecting early-warning signals for sudden deterioration of complex diseases by dynamical network biomarkers. Sci Rep. 2012;2:342. doi: 10.1038/srep00342. [DOI] [PMC free article] [PubMed] [Google Scholar]
33.Li G. 2023. R Package Version 011. [Google Scholar]
34.Monti S., Tamayo P., Mesirov J., Golub T. Consensus clustering: a resampling-based method for class discovery and visualization of gene expression microarray data. Machine Learn. 2003;52(1–2):91–118. [Google Scholar]
35.Wilkerson M.D., Hayes D.N. ConsensusClusterPlus: a class discovery tool with confidence assessments and item tracking. Bioinformatics. 2010;26(12):1572–1573. doi: 10.1093/bioinformatics/btq170. [DOI] [PMC free article] [PubMed] [Google Scholar]
36.Aran D., Hu Z., Butte A.J. xCell: digitally portraying the tissue cellular heterogeneity landscape. Genome Biol. 2017;18(1):220. doi: 10.1186/s13059-017-1349-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
37.Yoshihara K., Shahmoradgoli M., Martínez E., et al. Inferring tumour purity and stromal and immune cell admixture from expression data. Nat Commun. 2013;4:2612. doi: 10.1038/ncomms3612. [DOI] [PMC free article] [PubMed] [Google Scholar]
38.Lang M., Binder M., Richter J., et al. mlr3: a modern object-oriented machine learning framework in R. J Open Source Softw. 2019;4:1903. [Google Scholar]
39.Robin X., Turck N., Hainard A., et al. pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics. 2011;12:77. doi: 10.1186/1471-2105-12-77. [DOI] [PMC free article] [PubMed] [Google Scholar]
40.Coscia F., Lengyel E., Duraiswamy J., et al. Multi-level proteomics identifies CT45 as a chemosensitivity mediator and immunotherapy target in ovarian cancer. Cell. 2018;175(1):159–170.e16. doi: 10.1016/j.cell.2018.08.065. [DOI] [PMC free article] [PubMed] [Google Scholar]
41.Therneau T. 2023: a package for survival analysis in R. R package version 3.5. https://CRAN.R-project.org/package=survival
42.Maumy-Bertrand F.B.A.M. 2022. Partial Least Squares Regression for Cox Models and Related Techniques, R Package Version 1.7. [Google Scholar]
43.Frederic Bertrand P.B., Meyer N., Maumy-Bertrand M. Proceedings of User2014; Los Angeles: 2014. plsRcox, Cox-Models in a High Dimensional Setting in R; p. 177. [Google Scholar]
44.Bastien P., Bertrand F., Meyer N., Maumy-Bertrand M. Deviance residuals-based sparse PLS and sparse kernel PLS regression for censored data. Bioinformatics. 2015;31(3):397–404. doi: 10.1093/bioinformatics/btu660. [DOI] [PubMed] [Google Scholar]
45.Bair E., Tibshirani R. Semi-supervised methods to predict patient survival from gene expression data. PLoS Biol. 2004;2(4) doi: 10.1371/journal.pbio.0020108. [DOI] [PMC free article] [PubMed] [Google Scholar]
46.Bair E., Hastie T., Paul D., Tibshirani R. Prediction by supervised principal components. J Am Stat Assoc. 2006;101(473):119–137. [Google Scholar]
47.Aibar S., González-Blas C.B., Moerman T., et al. SCENIC: single-cell regulatory network inference and clustering. Nat Methods. 2017;14(11):1083–1086. doi: 10.1038/nmeth.4463. [DOI] [PMC free article] [PubMed] [Google Scholar]
48.Huynh-Thu V.A., Irrthum A., Wehenkel L., Geurts P. Inferring regulatory networks from expression data using tree-based methods. PLoS One. 2010;5(9) doi: 10.1371/journal.pone.0012776. [DOI] [PMC free article] [PubMed] [Google Scholar]
49.Van de Sande B., Flerin C., Davie K., et al. A scalable SCENIC workflow for single-cell gene regulatory network analysis. Nat Protoc. 2020;15(7):2247–2276. doi: 10.1038/s41596-020-0336-2. [DOI] [PubMed] [Google Scholar]
50.Sun X., Shu Y., Ye G., et al. Histone deacetylase inhibitors inhibit cervical cancer growth through Parkin acetylation-mediated mitophagy. Acta Pharm Sin B. 2022;12(2):838–852. doi: 10.1016/j.apsb.2021.07.003. [DOI] [PMC free article] [PubMed] [Google Scholar]
51.You X., Tian J., Zhang H., et al. Loss of mitochondrial aconitase promotes colorectal cancer progression via SCD1-mediated lipid remodeling. Mol Metab. 2021;48 doi: 10.1016/j.molmet.2021.101203. [DOI] [PMC free article] [PubMed] [Google Scholar]
52.Wang Z., Zheng W., Chen Z., et al. Pan-cancer analysis shows that ACO2 is a potential prognostic and immunotherapeutic biomarker for multiple cancer types including hepatocellular carcinoma. Front Oncol. 2022;12 doi: 10.3389/fonc.2022.1055376. [DOI] [PMC free article] [PubMed] [Google Scholar]
53.Xie F., Zhang H., Zhu K., et al. PRMT5 promotes ovarian cancer growth through enhancing Warburg effect by methylating ENO1. MedComm. 2023;4(2) doi: 10.1002/mco2.245. [DOI] [PMC free article] [PubMed] [Google Scholar]
54.Gan L., Zhao S., Gao Y., et al. N6-methyladenosine methyltransferase KIAA1429 promoted ovarian cancer aerobic glycolysis and progression through enhancing ENO1 expression. Biol Direct. 2023;18(1):64. doi: 10.1186/s13062-023-00420-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
55.Matei D., Kelich S., Cao L., et al. PDGF BB induces VEGF secretion in ovarian cancer. Cancer Biol Ther. 2007;6(12):1951–1959. doi: 10.4161/cbt.6.12.4976. [DOI] [PubMed] [Google Scholar]
56.Luukkaa M., Jokilehto T., Kronqvist P., et al. Expression of the cellular oxygen sensor PHD2 (EGLN-1) predicts radiation sensitivity in squamous cell cancer of the head and neck. Int J Radiat Biol. 2009;85(10):900–908. doi: 10.1080/09553000903074104. [DOI] [PubMed] [Google Scholar]
57.Langfelder P., Horvath S. WGCNA: an R package for weighted correlation network analysis. BMC Bioinformatics. 2008;9:559. doi: 10.1186/1471-2105-9-559. [DOI] [PMC free article] [PubMed] [Google Scholar]
58.Dagogo-Jack I., Shaw A.T. Tumour heterogeneity and resistance to cancer therapies. Nat Rev Clin Oncol. 2018;15(2):81–94. doi: 10.1038/nrclinonc.2017.166. [DOI] [PubMed] [Google Scholar]
59.Jamal-Hanjani M., Quezada S.A., Larkin J., Swanton C. Translational implications of tumor heterogeneity. Clin Cancer Res. 2015;21(6):1258–1266. doi: 10.1158/1078-0432.CCR-14-1429. [DOI] [PMC free article] [PubMed] [Google Scholar]
60.Vázquez-García I., Uhlitz F., Ceglia N., et al. Ovarian cancer mutational processes drive site-specific immune evasion. Nature. 2022;612(7941):778–786. doi: 10.1038/s41586-022-05496-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
61.Youlyouz-Marfak I., Gachard N., Le Clorennec C., et al. Identification of a novel p53-dependent activation pathway of STAT1 by antitumour genotoxic agents. Cell Death Differ. 2008;15(2):376–385. doi: 10.1038/sj.cdd.4402270. [DOI] [PubMed] [Google Scholar]
62.Zhou Y., Lee A.S. Mechanism for the suppression of the mammalian stress response by genistein, an anticancer phytoestrogen from soy. J Natl Cancer Inst. 1998;90(5):381–388. doi: 10.1093/jnci/90.5.381. [DOI] [PubMed] [Google Scholar]
63.Patel K.R., Brown V.A., Jones D.J., et al. Clinical pharmacology of resveratrol and its metabolites in colorectal cancer patients. Cancer Res. 2010;70(19):7392–7399. doi: 10.1158/0008-5472.CAN-10-2027. [DOI] [PMC free article] [PubMed] [Google Scholar]
64.Hu X., Li Y.Q., Li Q.G., Ma Y.L., Peng J.J., Cai S.J. Osteoglycin-induced VEGF inhibition enhances T lymphocytes infiltrating in colorectal cancer. eBioMedicine. 2018;34:35–45. doi: 10.1016/j.ebiom.2018.07.021. [DOI] [PMC free article] [PubMed] [Google Scholar]
65.Yeung T.L., Tsai C.C., Leung C.S., et al. ISG15 promotes ERK1 ISGylation, CD8+ T cell activation and suppresses ovarian cancer progression. Cancers (Basel) 2018;10(12) doi: 10.3390/cancers10120464. [DOI] [PMC free article] [PubMed] [Google Scholar]
66.Gu M., Zhou X., Sohn J.H., et al. NF-κB-inducing kinase maintains T cell metabolic fitness in antitumor immunity. Nat Immunol. 2021;22(2):193–204. doi: 10.1038/s41590-020-00829-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
67.Yu H., Lin L., Zhang Z., Zhang H., Hu H. Targeting NF-κB pathway for the therapy of diseases: mechanism and clinical study. Signal Transduct Target Ther. 2020;5(1):209. doi: 10.1038/s41392-020-00312-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
68.Sai Krishna A.V.S., Ramu A., Hariharan S., Sinha S., Donakonda S. Characterization of tumor microenvironment in glioblastoma multiforme identifies ITGB2 as a key immune and stromal related regulator in glial cell types. Comput Biol Med. 2023;165 doi: 10.1016/j.compbiomed.2023.107433. [DOI] [PubMed] [Google Scholar]
69.Shackelford R.E., Kaufmann W.K., Paules R.S. Oxidative stress and cell cycle checkpoint function. Free Radic Biol Med. 2000;28(9):1387–1404. doi: 10.1016/s0891-5849(00)00224-0. [DOI] [PubMed] [Google Scholar]
70.Matés J.M., Segura J.A., Alonso F.J., Márquez J. Intracellular redox status and oxidative stress: implications for cell proliferation, apoptosis, and carcinogenesis. Arch Toxicol. 2008;82(5):273–299. doi: 10.1007/s00204-008-0304-z. [DOI] [PubMed] [Google Scholar]
71.Patterson J.C., Joughin B.A., van de Kooij B., Lim D.C., Lauffenburger D.A., Yaffe M.B. ROS and oxidative stress are elevated in mitosis during asynchronous cell cycle progression and are exacerbated by mitotic arrest. Cell Syst. 2019;8(2):163–177.e2. doi: 10.1016/j.cels.2019.01.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
72.Alexandre J., Hu Y., Lu W., Pelicano H., Huang P. Novel action of paclitaxel against cancer cells: bystander effect mediated by reactive oxygen species. Cancer Res. 2007;67(8):3512–3517. doi: 10.1158/0008-5472.CAN-06-3914. [DOI] [PubMed] [Google Scholar]
73.Praschberger R., Lowe S.A., Malintan N.T., et al. Mutations in Membrin/GOSR2 reveal stringent secretory pathway demands of dendritic growth and synaptic integrity. Cell Rep. 2017;21(1):97–109. doi: 10.1016/j.celrep.2017.09.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
74.Tomoishi S., Fukushima S., Shinohara K., Katada T., Saito K. CREB3L2-mediated expression of Sec23A/Sec24D is involved in hepatic stellate cell activation through ER-Golgi transport. Sci Rep. 2017;7(1):7992. doi: 10.1038/s41598-017-08703-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
75.Sarwar S., Ashraf S., Shafiq M., et al. SEC24D gene as a biomarker in human cancers and its association with CD8+ T cell immune cell infiltration. Am J Transl Res. 2023;15(5):3115–3130. [PMC free article] [PubMed] [Google Scholar]
76.Carey P., Low E., Harper E., Stack M.S. Metalloproteinases in ovarian cancer. Int J Mol Sci. 2021;22(7) doi: 10.3390/ijms22073403. [DOI] [PMC free article] [PubMed] [Google Scholar]
77.Ribeiro V., Martins S.G., Lopes A.S., Thorsteinsdóttir S., Zilhão R., Carlos A.R. NFIXing cancer: the role of NFIX in oxidative stress response and cell fate. Int J Mol Sci. 2023;24(5) doi: 10.3390/ijms24054293. [DOI] [PMC free article] [PubMed] [Google Scholar]
78.Luijsterburg M.S., de Krijger I., Wiegant W.W., et al. PARP1 links CHD2-Mediated chromatin expansion and H3.3 deposition to DNA repair by non-homologous end-joining. Mol Cell. 2016;61(4):547–562. doi: 10.1016/j.molcel.2016.01.019. [DOI] [PMC free article] [PubMed] [Google Scholar]
79.Mitchell A.V., Wu L., James Block C., et al. FOXQ1 recruits the MLL complex to activate transcription of EMT and promote breast cancer metastasis. Nat Commun. 2022;13(1):6548. doi: 10.1038/s41467-022-34239-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
80.Paul S., Chhatar S., Mishra A., Lal G. Natural killer T cell activation increases iNOS(+)CD206(-) M1 macrophage and controls the growth of solid tumor. J Immunother Cancer. 2019;7(1):208. doi: 10.1186/s40425-019-0697-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
81.Molling J.W., Langius J.A., Langendijk J.A., et al. Low levels of circulating invariant natural killer T cells predict poor clinical outcome in patients with head and neck squamous cell carcinoma. J Clin Oncol. 2007;25(7):862–868. doi: 10.1200/JCO.2006.08.5787. [DOI] [PubMed] [Google Scholar]
82.Tachibana T., Onodera H., Tsuruyama T., et al. Increased intratumor Valpha24-positive natural killer T cells: a prognostic factor for primary colorectal carcinomas. Clin Cancer Res. 2005;11(20):7322–7327. doi: 10.1158/1078-0432.CCR-05-0877. [DOI] [PubMed] [Google Scholar]
83.Lau M.T., So W.K., Leung P.C. Fibroblast growth factor 2 induces E-cadherin down-regulation via PI3K/Akt/mTOR and MAPK/ERK signaling in ovarian cancer cells. PLoS One. 2013;8(3) doi: 10.1371/journal.pone.0059083. [DOI] [PMC free article] [PubMed] [Google Scholar]
84.Chen Q.Y., Li Y.N., Wang X.Y., et al. Tumor fibroblast-derived FGF2 regulates expression of SPRY1 in esophageal tumor-infiltrating T cells and plays a role in T-cell exhaustion. Cancer Res. 2020;80(24):5583–5596. doi: 10.1158/0008-5472.CAN-20-1542. [DOI] [PubMed] [Google Scholar]
85.Cabral-Pacheco G.A., Garza-Veloz I., Castruita-De la Rosa C., et al. The roles of matrix metalloproteinases and their inhibitors in human diseases. Int J Mol Sci. 2020;21(24) doi: 10.3390/ijms21249739. [DOI] [PMC free article] [PubMed] [Google Scholar]
86.Zhang S., Yuan L., Danilova L., et al. Spatial transcriptomics analysis of neoadjuvant cabozantinib and nivolumab in advanced hepatocellular carcinoma identifies independent mechanisms of resistance and recurrence. Genome Med. 2023;15(1):72. doi: 10.1186/s13073-023-01218-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
87.Ma J., Chen T., Wu S., et al. iProX: an integrated proteome resource. Nucleic Acids Res. 2019;47(D1):D1211–D1217. doi: 10.1093/nar/gky869. [DOI] [PMC free article] [PubMed] [Google Scholar]
88.Chen T., Ma J., Liu Y., et al. iProX in 2021: connecting proteomics data sharing with big data. Nucleic Acids Res. 2022;50(D1):D1522–D1527. doi: 10.1093/nar/gkab1081. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Figures and Tables

mmc1.pdf^{(9.2MB, pdf)}

Supplemental Western blots

mmc2.pdf^{(5.1MB, pdf)}

Cell line validation file

mmc3.pdf^{(4.9MB, pdf)}

[bib1] 1.Sung H., Ferlay J., Siegel R.L., et al. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2021;71(3):209–249. doi: 10.3322/caac.21660. [DOI] [PubMed] [Google Scholar]

[bib2] 2.Picard M. Why do we care more about disease than health? Phenomics. 2022;2(3):145–155. doi: 10.1007/s43657-021-00037-8. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib3] 3.Siegel R.L., Miller K.D., Fuchs H.E., Jemal A. Cancer statistics, 2022. CA Cancer J Clin. 2022;72(1):7–33. doi: 10.3322/caac.21708. [DOI] [PubMed] [Google Scholar]

[bib4] 4.Lheureux S., Braunstein M., Oza A.M. Epithelial ovarian cancer: evolution of management in the era of precision medicine. CA Cancer J Clin. 2019;69(4):280–304. doi: 10.3322/caac.21559. [DOI] [PubMed] [Google Scholar]

[bib5] 5.Kuroki L., Guntupalli S.R. Treatment of epithelial ovarian cancer. BMJ. 2020;371 doi: 10.1136/bmj.m3773. [DOI] [PubMed] [Google Scholar]

[bib6] 6.Lheureux S., Gourley C., Vergote I., Oza A.M. Epithelial ovarian cancer. Lancet. 2019;393(10177):1240–1253. doi: 10.1016/S0140-6736(18)32552-2. [DOI] [PubMed] [Google Scholar]

[bib7] 7.Ying W. Phenomic studies on diseases: potential and challenges. Phenomics. 2023;3(3):285–299. doi: 10.1007/s43657-022-00089-4. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib8] 8.Peres L.C., Cushing-Haugen K.L., Köbel M., et al. Invasive epithelial ovarian cancer survival by histotype and disease stage. J Natl Cancer Inst. 2019;111(1):60–68. doi: 10.1093/jnci/djy071. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib9] 9.Harter P., Sehouli J., Lorusso D., et al. A randomized trial of lymphadenectomy in patients with advanced ovarian neoplasms. N Engl J Med. 2019;380(9):822–832. doi: 10.1056/NEJMoa1808424. [DOI] [PubMed] [Google Scholar]

[bib10] 10.Xu J., Fang Y., Chen K., et al. Single-cell RNA sequencing reveals the tissue architecture in human high-grade serous ovarian cancer. Clin Cancer Res. 2022;28(16):3590–3602. doi: 10.1158/1078-0432.CCR-22-0296. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib11] 11.Integrated genomic analyses of ovarian carcinoma. Nature. 2011;474(7353):609–615. doi: 10.1038/nature10166. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib12] 12.Zhang H., Liu T., Zhang Z., et al. Integrated proteogenomic characterization of human high-grade serous ovarian cancer. Cell. 2016;166(3):755–765. doi: 10.1016/j.cell.2016.05.069. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib13] 13.Finn R.S., Qin S., Ikeda M., et al. Atezolizumab plus bevacizumab in unresectable hepatocellular carcinoma. N Engl J Med. 2020;382(20):1894–1905. doi: 10.1056/NEJMoa1915745. [DOI] [PubMed] [Google Scholar]

[bib14] 14.Gandhi L., Rodríguez-Abreu D., Gadgeel S., et al. Pembrolizumab plus chemotherapy in metastatic non-small-cell lung cancer. N Engl J Med. 2018;378(22):2078–2092. doi: 10.1056/NEJMoa1801005. [DOI] [PubMed] [Google Scholar]

[bib15] 15.Rini B.I., Plimack E.R., Stus V., et al. Pembrolizumab plus axitinib versus sunitinib for advanced renal-cell carcinoma. N Engl J Med. 2019;380(12):1116–1127. doi: 10.1056/NEJMoa1816714. [DOI] [PubMed] [Google Scholar]

[bib16] 16.Morazán-Fernández D., Mora J., Molina-Mora J.A. In silico pipeline to identify tumor-specific antigens for cancer immunotherapy using exome sequencing data. Phenomics. 2023;3(2):130–137. doi: 10.1007/s43657-022-00084-9. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib17] 17.Matulonis U.A., Shapira-Frommer R., Santin A.D., et al. Antitumor activity and safety of pembrolizumab in patients with advanced recurrent ovarian cancer: results from the phase II KEYNOTE-100 study. Ann Oncol. 2019;30(7):1080–1087. doi: 10.1093/annonc/mdz135. [DOI] [PubMed] [Google Scholar]

[bib18] 18.Zamarin D., Burger R.A., Sill M.W., et al. Randomized phase II trial of nivolumab versus nivolumab and ipilimumab for recurrent or persistent ovarian cancer: an NRG oncology study. J Clin Oncol. 2020;38(16):1814–1823. doi: 10.1200/JCO.19.02059. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib19] 19.Hegde P.S., Chen D.S. Top 10 challenges in cancer immunotherapy. Immunity. 2020;52(1):17–35. doi: 10.1016/j.immuni.2019.12.011. [DOI] [PubMed] [Google Scholar]

[bib20] 20.Ni D., Zhou H., Wang P., Xu F., Li C. Visualizing macrophage phenotypes and polarization in diseases: from biomarkers to molecular probes. Phenomics. 2023;3(6):613–638. doi: 10.1007/s43657-023-00129-7. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib21] 21.Hornburg M., Desbois M., Lu S., et al. Single-cell dissection of cellular components and interactions shaping the tumor immune phenotypes in ovarian cancer. Cancer Cell. 2021;39(7):928–944.e6. doi: 10.1016/j.ccell.2021.04.004. [DOI] [PubMed] [Google Scholar]

[bib22] 22.Olbrecht S., Busschaert P., Qian J., et al. High-grade serous tubo-ovarian cancer refined with single-cell RNA sequencing: specific cell subtypes influence survival and determine molecular subtype classification. Genome Med. 2021;13(1):111. doi: 10.1186/s13073-021-00922-x. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib23] 23.Wisniewski J.R., Zougman A., Nagaraj N., Mann M. Universal sample preparation method for proteome analysis. Nat Methods. 2009;6(5):359–362. doi: 10.1038/nmeth.1322. [DOI] [PubMed] [Google Scholar]

[bib24] 24.Hastie T.T.R., Narasimhan B., Chu G. 2023. R Package Version 1760. [Google Scholar]

[bib25] 25.Seckinger A., Meissner T., Moreaux J., et al. Clinical and prognostic role of annexin A2 in multiple myeloma. Blood. 2012;120(5):1087–1094. doi: 10.1182/blood-2012-03-415588. [DOI] [PubMed] [Google Scholar]

[bib26] 26.Alboukadel Kassambara F.M. 2020. R Package Version 107. [Google Scholar]

[bib27] 27.Wu T., Hu E., Xu S., et al. clusterProfiler 4.0: a universal enrichment tool for interpreting omics data. Innovation (Camb) 2021;2(3) doi: 10.1016/j.xinn.2021.100141. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib28] 28.Subramanian A., Tamayo P., Mootha V.K., et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A. 2005;102(43):15545–15550. doi: 10.1073/pnas.0506580102. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib29] 29.Hänzelmann S., Castelo R., Guinney J. GSVA: gene set variation analysis for microarray and RNA-Seq data. BMC Bioinformatics. 2013;14:7. doi: 10.1186/1471-2105-14-7. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib30] 30.Futschik M.E., Carlisle B. Noise-robust soft clustering of gene expression time-course data. J Bioinform Comput Biol. 2005;3(4):965–988. doi: 10.1142/s0219720005001375. [DOI] [PubMed] [Google Scholar]

[bib31] 31.Kumar L., Futschik M. Mfuzz: a software package for soft clustering of microarray data. Bioinformation. 2007;2(1):5–7. doi: 10.6026/97320630002005. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib32] 32.Chen L., Liu R., Liu Z.P., Li M., Aihara K. Detecting early-warning signals for sudden deterioration of complex diseases by dynamical network biomarkers. Sci Rep. 2012;2:342. doi: 10.1038/srep00342. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib33] 33.Li G. 2023. R Package Version 011. [Google Scholar]

[bib34] 34.Monti S., Tamayo P., Mesirov J., Golub T. Consensus clustering: a resampling-based method for class discovery and visualization of gene expression microarray data. Machine Learn. 2003;52(1–2):91–118. [Google Scholar]

[bib35] 35.Wilkerson M.D., Hayes D.N. ConsensusClusterPlus: a class discovery tool with confidence assessments and item tracking. Bioinformatics. 2010;26(12):1572–1573. doi: 10.1093/bioinformatics/btq170. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib36] 36.Aran D., Hu Z., Butte A.J. xCell: digitally portraying the tissue cellular heterogeneity landscape. Genome Biol. 2017;18(1):220. doi: 10.1186/s13059-017-1349-1. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib37] 37.Yoshihara K., Shahmoradgoli M., Martínez E., et al. Inferring tumour purity and stromal and immune cell admixture from expression data. Nat Commun. 2013;4:2612. doi: 10.1038/ncomms3612. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib38] 38.Lang M., Binder M., Richter J., et al. mlr3: a modern object-oriented machine learning framework in R. J Open Source Softw. 2019;4:1903. [Google Scholar]

[bib39] 39.Robin X., Turck N., Hainard A., et al. pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics. 2011;12:77. doi: 10.1186/1471-2105-12-77. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib40] 40.Coscia F., Lengyel E., Duraiswamy J., et al. Multi-level proteomics identifies CT45 as a chemosensitivity mediator and immunotherapy target in ovarian cancer. Cell. 2018;175(1):159–170.e16. doi: 10.1016/j.cell.2018.08.065. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib41] 41.Therneau T. 2023: a package for survival analysis in R. R package version 3.5. https://CRAN.R-project.org/package=survival

[bib42] 42.Maumy-Bertrand F.B.A.M. 2022. Partial Least Squares Regression for Cox Models and Related Techniques, R Package Version 1.7. [Google Scholar]

[bib43] 43.Frederic Bertrand P.B., Meyer N., Maumy-Bertrand M. Proceedings of User2014; Los Angeles: 2014. plsRcox, Cox-Models in a High Dimensional Setting in R; p. 177. [Google Scholar]

[bib44] 44.Bastien P., Bertrand F., Meyer N., Maumy-Bertrand M. Deviance residuals-based sparse PLS and sparse kernel PLS regression for censored data. Bioinformatics. 2015;31(3):397–404. doi: 10.1093/bioinformatics/btu660. [DOI] [PubMed] [Google Scholar]

[bib45] 45.Bair E., Tibshirani R. Semi-supervised methods to predict patient survival from gene expression data. PLoS Biol. 2004;2(4) doi: 10.1371/journal.pbio.0020108. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib46] 46.Bair E., Hastie T., Paul D., Tibshirani R. Prediction by supervised principal components. J Am Stat Assoc. 2006;101(473):119–137. [Google Scholar]

[bib47] 47.Aibar S., González-Blas C.B., Moerman T., et al. SCENIC: single-cell regulatory network inference and clustering. Nat Methods. 2017;14(11):1083–1086. doi: 10.1038/nmeth.4463. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib48] 48.Huynh-Thu V.A., Irrthum A., Wehenkel L., Geurts P. Inferring regulatory networks from expression data using tree-based methods. PLoS One. 2010;5(9) doi: 10.1371/journal.pone.0012776. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib49] 49.Van de Sande B., Flerin C., Davie K., et al. A scalable SCENIC workflow for single-cell gene regulatory network analysis. Nat Protoc. 2020;15(7):2247–2276. doi: 10.1038/s41596-020-0336-2. [DOI] [PubMed] [Google Scholar]

[bib50] 50.Sun X., Shu Y., Ye G., et al. Histone deacetylase inhibitors inhibit cervical cancer growth through Parkin acetylation-mediated mitophagy. Acta Pharm Sin B. 2022;12(2):838–852. doi: 10.1016/j.apsb.2021.07.003. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib51] 51.You X., Tian J., Zhang H., et al. Loss of mitochondrial aconitase promotes colorectal cancer progression via SCD1-mediated lipid remodeling. Mol Metab. 2021;48 doi: 10.1016/j.molmet.2021.101203. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib52] 52.Wang Z., Zheng W., Chen Z., et al. Pan-cancer analysis shows that ACO2 is a potential prognostic and immunotherapeutic biomarker for multiple cancer types including hepatocellular carcinoma. Front Oncol. 2022;12 doi: 10.3389/fonc.2022.1055376. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib53] 53.Xie F., Zhang H., Zhu K., et al. PRMT5 promotes ovarian cancer growth through enhancing Warburg effect by methylating ENO1. MedComm. 2023;4(2) doi: 10.1002/mco2.245. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib54] 54.Gan L., Zhao S., Gao Y., et al. N6-methyladenosine methyltransferase KIAA1429 promoted ovarian cancer aerobic glycolysis and progression through enhancing ENO1 expression. Biol Direct. 2023;18(1):64. doi: 10.1186/s13062-023-00420-7. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib55] 55.Matei D., Kelich S., Cao L., et al. PDGF BB induces VEGF secretion in ovarian cancer. Cancer Biol Ther. 2007;6(12):1951–1959. doi: 10.4161/cbt.6.12.4976. [DOI] [PubMed] [Google Scholar]

[bib56] 56.Luukkaa M., Jokilehto T., Kronqvist P., et al. Expression of the cellular oxygen sensor PHD2 (EGLN-1) predicts radiation sensitivity in squamous cell cancer of the head and neck. Int J Radiat Biol. 2009;85(10):900–908. doi: 10.1080/09553000903074104. [DOI] [PubMed] [Google Scholar]

[bib57] 57.Langfelder P., Horvath S. WGCNA: an R package for weighted correlation network analysis. BMC Bioinformatics. 2008;9:559. doi: 10.1186/1471-2105-9-559. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib58] 58.Dagogo-Jack I., Shaw A.T. Tumour heterogeneity and resistance to cancer therapies. Nat Rev Clin Oncol. 2018;15(2):81–94. doi: 10.1038/nrclinonc.2017.166. [DOI] [PubMed] [Google Scholar]

[bib59] 59.Jamal-Hanjani M., Quezada S.A., Larkin J., Swanton C. Translational implications of tumor heterogeneity. Clin Cancer Res. 2015;21(6):1258–1266. doi: 10.1158/1078-0432.CCR-14-1429. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib60] 60.Vázquez-García I., Uhlitz F., Ceglia N., et al. Ovarian cancer mutational processes drive site-specific immune evasion. Nature. 2022;612(7941):778–786. doi: 10.1038/s41586-022-05496-1. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib61] 61.Youlyouz-Marfak I., Gachard N., Le Clorennec C., et al. Identification of a novel p53-dependent activation pathway of STAT1 by antitumour genotoxic agents. Cell Death Differ. 2008;15(2):376–385. doi: 10.1038/sj.cdd.4402270. [DOI] [PubMed] [Google Scholar]

[bib62] 62.Zhou Y., Lee A.S. Mechanism for the suppression of the mammalian stress response by genistein, an anticancer phytoestrogen from soy. J Natl Cancer Inst. 1998;90(5):381–388. doi: 10.1093/jnci/90.5.381. [DOI] [PubMed] [Google Scholar]

[bib63] 63.Patel K.R., Brown V.A., Jones D.J., et al. Clinical pharmacology of resveratrol and its metabolites in colorectal cancer patients. Cancer Res. 2010;70(19):7392–7399. doi: 10.1158/0008-5472.CAN-10-2027. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib64] 64.Hu X., Li Y.Q., Li Q.G., Ma Y.L., Peng J.J., Cai S.J. Osteoglycin-induced VEGF inhibition enhances T lymphocytes infiltrating in colorectal cancer. eBioMedicine. 2018;34:35–45. doi: 10.1016/j.ebiom.2018.07.021. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib65] 65.Yeung T.L., Tsai C.C., Leung C.S., et al. ISG15 promotes ERK1 ISGylation, CD8+ T cell activation and suppresses ovarian cancer progression. Cancers (Basel) 2018;10(12) doi: 10.3390/cancers10120464. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib66] 66.Gu M., Zhou X., Sohn J.H., et al. NF-κB-inducing kinase maintains T cell metabolic fitness in antitumor immunity. Nat Immunol. 2021;22(2):193–204. doi: 10.1038/s41590-020-00829-6. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib67] 67.Yu H., Lin L., Zhang Z., Zhang H., Hu H. Targeting NF-κB pathway for the therapy of diseases: mechanism and clinical study. Signal Transduct Target Ther. 2020;5(1):209. doi: 10.1038/s41392-020-00312-6. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib68] 68.Sai Krishna A.V.S., Ramu A., Hariharan S., Sinha S., Donakonda S. Characterization of tumor microenvironment in glioblastoma multiforme identifies ITGB2 as a key immune and stromal related regulator in glial cell types. Comput Biol Med. 2023;165 doi: 10.1016/j.compbiomed.2023.107433. [DOI] [PubMed] [Google Scholar]

[bib69] 69.Shackelford R.E., Kaufmann W.K., Paules R.S. Oxidative stress and cell cycle checkpoint function. Free Radic Biol Med. 2000;28(9):1387–1404. doi: 10.1016/s0891-5849(00)00224-0. [DOI] [PubMed] [Google Scholar]

[bib70] 70.Matés J.M., Segura J.A., Alonso F.J., Márquez J. Intracellular redox status and oxidative stress: implications for cell proliferation, apoptosis, and carcinogenesis. Arch Toxicol. 2008;82(5):273–299. doi: 10.1007/s00204-008-0304-z. [DOI] [PubMed] [Google Scholar]

[bib71] 71.Patterson J.C., Joughin B.A., van de Kooij B., Lim D.C., Lauffenburger D.A., Yaffe M.B. ROS and oxidative stress are elevated in mitosis during asynchronous cell cycle progression and are exacerbated by mitotic arrest. Cell Syst. 2019;8(2):163–177.e2. doi: 10.1016/j.cels.2019.01.005. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib72] 72.Alexandre J., Hu Y., Lu W., Pelicano H., Huang P. Novel action of paclitaxel against cancer cells: bystander effect mediated by reactive oxygen species. Cancer Res. 2007;67(8):3512–3517. doi: 10.1158/0008-5472.CAN-06-3914. [DOI] [PubMed] [Google Scholar]

[bib73] 73.Praschberger R., Lowe S.A., Malintan N.T., et al. Mutations in Membrin/GOSR2 reveal stringent secretory pathway demands of dendritic growth and synaptic integrity. Cell Rep. 2017;21(1):97–109. doi: 10.1016/j.celrep.2017.09.004. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib74] 74.Tomoishi S., Fukushima S., Shinohara K., Katada T., Saito K. CREB3L2-mediated expression of Sec23A/Sec24D is involved in hepatic stellate cell activation through ER-Golgi transport. Sci Rep. 2017;7(1):7992. doi: 10.1038/s41598-017-08703-6. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib75] 75.Sarwar S., Ashraf S., Shafiq M., et al. SEC24D gene as a biomarker in human cancers and its association with CD8+ T cell immune cell infiltration. Am J Transl Res. 2023;15(5):3115–3130. [PMC free article] [PubMed] [Google Scholar]

[bib76] 76.Carey P., Low E., Harper E., Stack M.S. Metalloproteinases in ovarian cancer. Int J Mol Sci. 2021;22(7) doi: 10.3390/ijms22073403. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib77] 77.Ribeiro V., Martins S.G., Lopes A.S., Thorsteinsdóttir S., Zilhão R., Carlos A.R. NFIXing cancer: the role of NFIX in oxidative stress response and cell fate. Int J Mol Sci. 2023;24(5) doi: 10.3390/ijms24054293. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib78] 78.Luijsterburg M.S., de Krijger I., Wiegant W.W., et al. PARP1 links CHD2-Mediated chromatin expansion and H3.3 deposition to DNA repair by non-homologous end-joining. Mol Cell. 2016;61(4):547–562. doi: 10.1016/j.molcel.2016.01.019. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib79] 79.Mitchell A.V., Wu L., James Block C., et al. FOXQ1 recruits the MLL complex to activate transcription of EMT and promote breast cancer metastasis. Nat Commun. 2022;13(1):6548. doi: 10.1038/s41467-022-34239-z. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib80] 80.Paul S., Chhatar S., Mishra A., Lal G. Natural killer T cell activation increases iNOS(+)CD206(-) M1 macrophage and controls the growth of solid tumor. J Immunother Cancer. 2019;7(1):208. doi: 10.1186/s40425-019-0697-7. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib81] 81.Molling J.W., Langius J.A., Langendijk J.A., et al. Low levels of circulating invariant natural killer T cells predict poor clinical outcome in patients with head and neck squamous cell carcinoma. J Clin Oncol. 2007;25(7):862–868. doi: 10.1200/JCO.2006.08.5787. [DOI] [PubMed] [Google Scholar]

[bib82] 82.Tachibana T., Onodera H., Tsuruyama T., et al. Increased intratumor Valpha24-positive natural killer T cells: a prognostic factor for primary colorectal carcinomas. Clin Cancer Res. 2005;11(20):7322–7327. doi: 10.1158/1078-0432.CCR-05-0877. [DOI] [PubMed] [Google Scholar]

[bib83] 83.Lau M.T., So W.K., Leung P.C. Fibroblast growth factor 2 induces E-cadherin down-regulation via PI3K/Akt/mTOR and MAPK/ERK signaling in ovarian cancer cells. PLoS One. 2013;8(3) doi: 10.1371/journal.pone.0059083. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib84] 84.Chen Q.Y., Li Y.N., Wang X.Y., et al. Tumor fibroblast-derived FGF2 regulates expression of SPRY1 in esophageal tumor-infiltrating T cells and plays a role in T-cell exhaustion. Cancer Res. 2020;80(24):5583–5596. doi: 10.1158/0008-5472.CAN-20-1542. [DOI] [PubMed] [Google Scholar]

[bib85] 85.Cabral-Pacheco G.A., Garza-Veloz I., Castruita-De la Rosa C., et al. The roles of matrix metalloproteinases and their inhibitors in human diseases. Int J Mol Sci. 2020;21(24) doi: 10.3390/ijms21249739. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib86] 86.Zhang S., Yuan L., Danilova L., et al. Spatial transcriptomics analysis of neoadjuvant cabozantinib and nivolumab in advanced hepatocellular carcinoma identifies independent mechanisms of resistance and recurrence. Genome Med. 2023;15(1):72. doi: 10.1186/s13073-023-01218-y. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib87] 87.Ma J., Chen T., Wu S., et al. iProX: an integrated proteome resource. Nucleic Acids Res. 2019;47(D1):D1211–D1217. doi: 10.1093/nar/gky869. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib88] 88.Chen T., Ma J., Liu Y., et al. iProX in 2021: connecting proteomics data sharing with big data. Nucleic Acids Res. 2022;50(D1):D1522–D1527. doi: 10.1093/nar/gkab1081. [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

Proteomic profiling and molecular reclassification of high-grade serous ovarian cancer identifies prognostic subtypes and immunotherapy biomarkers

Mengyan Tu

Sangsang Tang

Qiao Zhang

Tianchen Guo

Yixuan Cen

Xiaomeng Xu

Shenglong Wu

Xin Chen

Weiguo Lu

Chen Ding

Junfen Xu

Summary

Background

Methods

Findings

Interpretation

Funding

Research in context.

Evidence before this study

Added value of this study

Implications of all the available evidence

Introduction

Methods

Clinical samples

Study populations

Ethics approval and consent to participate

Sample preparation

Peptides preparation for MS analysis

Protein extraction and tryptic digestion

First-dimensional reversed-phase separation

LC-MS/MS analysis

Mass spectrometry data analysis

Peptide and protein identification

Missing value imputation

Quality control of mass spectrometry data

Global proteomics analysis

Survival analysis

Principal component analysis (PCA)

Differential protein analysis

Pathway enrichment analysis

Pathway/signature activation scores in different samples

Time-series analysis

Dynamical Network Biomarkers (DNB) analysis

Consensus clustering and immune clusters

Machine learning-based signature generation

Transcription factors (TFs) activity inference

Weighted gene correlation network analysis (WGCNA) analysis

Immunohistochemical (IHC) staining

Cell line authentication and cell culture

Cell viability assays

EdU assays

Western blot

Cell transfection

RNA extraction and RT-qPCR

Transwell assay

Phalloidin staining

Isolation of cancer-associated fibroblasts (CAFs)

T cell isolation and co-culture with cancer cells

Flow cytometry analysis of T cells

Animal models

Co-Immunoprecipitation (co-IP) assay

mIHC assay

HUVEC tube formation assay

Statistical analysis

Role of funders

Results

Comprehensive proteomic characterisation in HGSOC

Fig. 1.

Critical transition from early to advanced stage in HGSOC

Fig. 2.

Proteomic alterations in early vs. advanced stage HGSOC

Fig. 3.

Identification of proteomic subtypes to delineate intratumour heterogeneity of HGSOC

Fig. 4.

TFs activity identified the three proteomic subtypes

Fig. 5.

The immune clustering of HGSOC revealed three distinct subgroups with varying characteristics in the immune tumour microenvironment

Fig. 6.