Single-cell transcriptome analysis indicated that immune-related programmed cell death modification features could predict the clinical outcomes of patients with ovarian cancer

Highlight

• •Our study develops a novel prognostic model IPCDS for OV.
• •IPCD-related gene PDGFRA serves as a prognosis factor play in predicting survival outcomes of OVs.
• •We provide new targets and ideas for the immunotherapy of ovarian cancer.

Abstract

Background

Ovarian cancer (OV) is characterized by the highest mortality rate among gynecological malignancies. Suboptimal early diagnosis and ineffective prognostic prediction of OV contribute to poor survival outcomes for most patients. This study aimed to identify immune-related programmed cell death (IPCD) signatures and find the valuable biomarker for predicting OV prognosis.

Methods

All OV datasets were downloaded from public databases of TCGA, GEO and ICGC. Prognostic genes from IPCD-related differential genes were screened by univariate cox regression analysis. The construction of the IPCDS model was performed via 101 algorithm combinations. The prognostic performance of IPCDS model were examined by Kaplan-Meier analysis and timeROC curves. TIDE algorithm was used to predict the immune response of TCGA data.

Results

88 IPCD-related prognostic genes were screened for modeling IPCDS. A significantly higher OS in low-IPCDS group was observed among most OV datasets than in high-IPCDS group. 2-, 3-, and 5-year timeROC results of each OV dataset revealed the excellent predictive value of IPCDS for OV, showing a relatively higher AUC after treated 2 years. The low IPCDS group had a higher TIDE value, while the no response group had a higher IPCDS value. In the pan-cancer immunotherapy dataset, patients in the low IPCDS group had a longer overall survival period and a significant immunotherapy effect. Besides, model gene PDGFRA were found to have predictive performance for OV. We validated the upregulation of PDGFRA and C-MYC in ovarian cancer tissues through Western blot and confirmed their co-localization and immunofluorescence analyses.

Conclusion

Our study develops a novel prognostic model IPCDS for OV. IPCD-related gene PDGFRA serves as a prognosis factor play in predicting survival outcomes of OVs.

Graphical abstract

Introduction

Ovarian cancer (OV) remains the most lethal gynecological malignancy [1]. According to GLOBOCAN 2020, there are about 310,000 new cases of ovarian cancer worldwide each year, with 200,000 deaths, and the 5-year survival rate is only 39.1%, and the 5-year survival rate for advanced patients is even lower to <30% [2,3]. Its epidemiological characteristics showed significant geographical, age and genetic differences. OV tends to occur in women over 50 years of age, with high malignancy and early metastasis. About 10% of ovarian cancers are genetically related, with carriers of BRCA1/2 germline mutations having a lifetime risk of 54% and 23%, respectively [4]. The incidence is higher in developed countries than in developing countries, such as North America and Europe [5]. Early symptoms of ovarian cancer are insidious, and most of patients are diagnosed with advanced cancer (stage III-IV). Existing screening methods such as CA125 [6] and transvaginal ultrasound [7] are not sensitive enough to early lesions, resulting in a long-term low-level survival outcomes. Novel markers such as human epididymis protein 4 (HE4) [8] and lncRNA (such as TLR8-AS1 and LBX2-AS1) have shown potential in studies, but have not been widely used. Despite advancements in surgical debulking, platinum-based chemotherapy, and PARP inhibitors, the limited efficacy of current therapies is attributed to tumor heterogeneity, chemoresistance, and immunosuppressive tumor microenvironment (TME). In recent years, the rise of immunotherapy has brought new hope for precision medicine in ovarian cancer. Among them, the regulatory mechanism of immune-related Programmed Cell Death (IPCD) pathways and the exploration of their genetic markers have become research hotspots.

Dysregulation of IPCD pathways not only impacts tumor cell survival but also facilitates immune escape by remodeling the tumor microenvironment (TME) [9,10]. For instance, the PD-1/PD-L1 pathway exemplifies this phenomenon, where tumor cells suppress T cell activation and proliferation through high expression of PD-L1 and binding to PD-1 on T cell surfaces, thereby establishing an immunosuppressive microenvironment [11]. Clinical trials have demonstrated that combining PD-1/PD-L1 inhibitors with chemotherapy significantly extends progression-free survival in patients with recurrent ovarian cancer with the overall response rate (ORR) of 36% [12]. Nevertheless, only a subset of patients exhibit sensitivity to immunotherapy, and the mechanisms underlying drug resistance are multifaceted, necessitating the identification of advantageous populations via in-depth genetic analysis. IPCD plays a significant role in the immune microenvironment shaping and treatment response prediction of ovarian cancer by regulating apoptosis, immune cell infiltration and immune signaling pathways. Models constructed based on IPCD-related genes provide new tools for the precise stratified treatment of ovarian cancer, especially having significant guiding significance in the selection of individualized immunotherapy regimens

During the process of exploring the prognostic signature, a single algorithm may have limitations. The combination of multiple algorithms, through cross-validation and evaluation, can mine data characteristics from different perspectives, screen out the optimal model, and better conduct risk stratification and prognosis prediction for patients with OV. The integration of machine learning technology has markedly enhanced the efficiency of gene marker discovery. Specifically, the gradient boosting decision tree model surpassed traditional single-gene models with a chemotherapy resistance prediction accuracy [13]. Another study leveraged multi-omics data to construct a prognostic model, revealing that exhausted CD8+ T cells are enriched in high-risk OV patients and demonstrating significant correlations between the model's predictions and both immune treatment responses and drug sensitivities [14]. Furthermore, random forest and deep learning algorithms have excelled in identifying ovarian cancer-specific PCD genes (e.g., BIM, BAX), which modulate platinum drug sensitivity by regulating the mitochondrial apoptosis pathway [15,16]. However, the current research still faces challenges. Firstly, the molecular mechanism of immune-related PCD is not yet fully clear. For example, the specific pathway nodes by which Bax affects platinum resistance by regulating the Akt signaling pathway still need further verification. Secondly, the generalization ability of machine learning models needs to be improved, and their predictive efficacy across races and pathological types needs to be verified in a multi-center cohort. Furthermore, the integration methods of multi-omics data such as single-cell and spatial transcriptomes still need to be optimized to more accurately analyze the heterogeneity of the TME.

Single-cell RNA sequencing (scRNA-seq) technology provides a high-resolution perspective for analyzing the cellular heterogeneity of the TME of ovarian cancer [17]. The development of high-throughput sequencing technology has made it possible to systematically analyze IPCD-related genes in OV. These genes are enriched in the regulation of apoptotic signaling pathways and cell chemotaxis, suggesting that IPCD-related genes affect the progression of ovarian cancer by regulating cell death and immune-related pathways. In-depth analysis at the single-cell level reveals that the regulation of immune infiltration by IPCDS is not the effect of a single cell type, but is achieved through the interaction of cross-cellular gene expression networks and signaling pathways [18]. This high-resolution association analysis not only verified the differences in immune infiltration at the bulk level, but also revealed the core roles of effector immune cell activation, myeloid cell polarization and matrix-immune cell communication in the IPCD pathway. Future studies can locate the spatial distribution of these cell interactions through spatial transcriptome technology, and verify the causal relationship of key ligand-receptor pairs in immune infiltration through functional experiments, providing precise intervention targets for targeting IPCD-related cell communication in the TME.

Here, we explored the role of IPCD-related genes in ovarian cancer pathogenesis and developed a novel prognostic model based on 101 machine learning combinations, suggesting the potential of targeting IPCD-related pathways to overcome treatment resistance. This model performs well in differentiating the survival differences between patients in the high and low-risk groups, and also has significant efficacy in some pan-cancer immunotherapy datasets. Meanwhile, the model-related genes, tumor immune infiltration and immunotherapy responses were studied and analyzed. It was found that IPCDS was related to multiple immune indicators, and the PDGFRA gene had an important prognostic role. The PDGFRA gene is associated with CD8A expression and various immune-infiltrating cells, which means it may affect the function of immune cells in the tumor microenvironment. By regulating the PDGFRA gene, it may be possible to regulate the activity and quantity of immune-infiltrating cells, enhance the body's immune response to tumors, and provide new targets and ideas for the immunotherapy of ovarian cancer.

Methods

Acquiring the transcriptomic profiles

The study treated RNA expression profiles and corresponding clinical data of ovarian cancer cohort (n = 378) from The Cancer Genome Atlas (TCGA) as the training cohort to develop model. As the validation set, the OV_AU cohort (n = 93) from the International Cancer Genomics Consortium (ICGC) and GSE17260 (n = 110), GSE26193 (n = 107), GSE26712 (n = 185), GSE30161 (n = 58), GSE49997 (n = 194), GSE63885 (n = 75), GSE9891 (n = 278), and GSE140082 (n = 380) assessed model stability and accuracy. Differential expression analysis was performed using GSE26712 (Normal = 10, Tumor = 185) and GTEx (Normal = 88) datasets. Quality control and normalization of data were conducted using the normalizeBetweenArrays function in the limma package, followed by removal of batch effect via the combat function from the sva package (Suppl Fig. 1A).

Fig. 1.

The results of IPCD-related prognostic genes analysis. A. Heat map of IPCD-related DEGs. B. Forest map of prognostic genes. C-D. Histograms of GO and KEGG enrichment analyses. E. Plots of ssGSEA score among 20 cell death types. F. Distribution map of prognostic genes on human chromosome. G. CNV plots of prognostic genes.

All datasets were standardized to transcripts per million (TPM) expression values and log2-transformed for downstream analyses.

Additionally, six immunotherapy-specific datasets were analyzed, including GBM_PRJNA482620 (n = 34), Melanoma_GSE100797 (n = 25), Melanoma_GSE78220 (n = 28), Melanoma_GSE91061 (n = 109), RCC_Braun_2020 (n = 311), and IMvigor210 (n = 348) in Suppl Fig. 1B

Processing single-cell RNA sequencing (scRNA-seq) data

scRNA-seq data of GSE235931 was, encompassing 15 ovarian tumor samples. Data analysis was conducted using R software (version 4.1.3) with the Seurat package. Cells were filtered based on the following criteria: mitochondrial content <25%, hemocyte contamination <3%, and UMI counts between 200 and 10,000. Genes with expression levels <200 or <4000 across cells were excluded. Normalization was performed using Seurat's NormalizeData function to account for global expression differences. Highly variable genes were identified using FindVariableFeatures (n = 2000 genes). Cell cycle effects were regressed out using the ScaleData function with parameters vars.to.regress = c("S.Score", "G2M.Score"), which quantifies mitotic and G2/M phase activity. Batch effect correction was implemented using the Harmony algorithm to integrate datasets from distinct experimental conditions.

Dimensionality reduction and clustering methods UMAP and Louvain were applied to the normalized data to identify cellular subpopulations. These analyses were performed using functions of RunUMAP and FindClusters, with resolution parameters optimized for distinct cluster separation. Cell annotation was performed. Differentially expressed genes (DEGs) between clusters or cell types were identified using FindAllMarkers with the following thresholds: p-value <0.05, absolute log2 fold change (log2FC) >0.25, and minimum expression proportion >10% across cells.

Mutation analysis

Copy number variation (CNV) analysis was performed using GISTIC 2.0 software, while tumor mutational burden (TMB) was quantified via the maftools package. These analyses provided genomic instability metrics for downstream clinical correlation studies.

Prognostic genes analysis

Limma software was performed to calculate DEGs (|logFC| > 1 and P < 0.05) for tumor and adjacent tumors, and filter out DEGs associated with immune-related programmed cell death (IPCD). Univariate Cox analysis was deployed to calculate the prognostic genes among them (p < 0.01), and the IPCDS model was established for these 88 prognostic genes.

Establishment of risk features

A prognostic model IPCD signature (IPCDS) was developed for 88 prognostic genes using ​101 algorithm combinations to generate individualized risk scores for OV patients. The surv_cutpoint function was employed to determine optimal cutoff thresholds for stratifying patients into high-risk and low-risk groups across the TCGA cohort and all validation datasets. Predictive performance between the two risk groups was systematically compared, and model accuracy was comprehensively evaluated using clinical outcome metrics.

To develop the IPCDS model with superior predictive accuracy and stability, we integrated ​10 machine learning algorithms into ​101 algorithmic combinations, including Random Survival Forest (RSF), Elastic Net (Enet), LASSO, Ridge Regression, Stepwise Cox, CoxBoost, Cox Partial Least Squares Regression (plsRcox), Supervised Principal Components (SuperPC), Generalized Boosting Machine (GBM) and Survival Support Vector Machine (survival-SVM). Workflow of generating IPCDS model are: (1) Prognostic gene identification: Univariate Cox proportional hazards regression analysis was performed on the merged dataset (including TCGA) to identify candidate prognostic genes (as described in prior sections). (2) Model training: Each candidate gene set was subjected to ​101 algorithm combinations under a leave-one-out cross-validation (LOOCV) framework within the TCGA-OV cohort. (3) Validation: All trained models were independently tested on external validation datasets. (4) ​Performance evaluation: The Harrell concordance index (C-index) was calculated for each model across all validation cohorts. The model exhibiting the highest average C-index was selected as the optimal solution.

Cell-to-cell communication analysis

Intercellular communication was assessed using the CellChat package. A CellChat object was created by importing the normalized gene expression matrix into the CellChat function. Data preprocessing was performed using default parameters with the identifyOSAerExpressedGenes, identifyOSAerExpressedInteraction, and ProjectData functions. Potential ligand-receptor interactions were identified through computeCommunProb to calculate communication probabilities between cell types, filterCommunication to retain significant interactions, and computeCommunProbPathway to map interactions to biological pathways.

Finally, a cell communication network was constructed using the aggregateNet function to visualize intercellular signaling pathways.

Correlation with tumor immunity and prediction of immunotherapy response

Immune infiltration levels in TCGA cohort were determined using the IOBR software, which integrates seven immunoscore algorithms. Heatmaps were generated to quantify the relative proportions of immune cell subsets within the tumor microenvironment (TME), enabling correlation with clinical outcomes and model predictions.

The TIDE (Tumor Immune Dysfunction and Exclusion) algorithm (https://www.harvard.edu/) was applied to TCGA data to evaluate immune response signatures. This framework assessed allele-specific (AS) differences in immunotherapy efficacy, providing insights into tumor-immune dynamics.

Immunofluorescence (IF) and hematoxylin-eosin (HE) staining

Clinical ovarian tumor tissues and adjacent normal tissues were obtained from three patients. For histopathological evaluation, tissue sections were fixed in 4% paraformaldehyde, embedded in paraffin, and sectioned at 4 μm thickness. HE staining was performed according to standard protocols to evaluate tissue morphology. For immunofluorescence staining, sections were deparaffinized, rehydrated, and subjected to antigen retrieval using citrate buffer (pH 6.0). After blocking with 5% bovine serum albumin (BSA) for 1 h at room temperature, sections were incubated overnight at 4 °C with primary antibodies against PDGFRA and C-MYC. Following three washes, sections were incubated with Alexa Fluor 488-conjugated anti-rabbit secondary antibody (green) and Alexa Fluor 594-conjugated anti-mouse secondary antibody (red) for 1 h at room temperature in the dark. Nuclei were counterstained with 4′,6-diamidino-2-phenylindole (DAPI).

Western blotting (WB)

Total protein was extracted from tumor and adjacent normal tissues using RIPA lysis buffer supplemented with protease and phosphatase inhibitors. Equal amounts of protein (30 μg per lane) were separated by 10% sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and transferred onto polyvinylidene difluoride (PVDF) membranes. Membranes were blocked with 5% non-fat milk in Tris-buffered saline containing 0.1% Tween-20 (TBST) for 1 h at room temperature, followed by overnight incubation at 4 °C with primary antibodies against PDGFRA (1:2000, 84,383-2-RR, Proteintech), C-MYC (1:2000, 10,828-1-AP, Proteintech), and ACTIN as a loading control. After washing with TBST, membranes were incubated with horseradish peroxidase (HRP)-conjugated secondary antibodies for 1 h at room temperature. Protein bands were visualized using enhanced chemiluminescence (ECL) reagent and imaged using a chemiluminescence imaging system.

Statistical analysis

All computational analyses, statistical modeling, and visualization were conducted in R software (version 4.1.3). Correlation analysis: Pearson correlation coefficient for continuous variables. Categorical comparisons: Chi-square test for nominal data. Group comparisons: Wilcoxon rank-sum test or t-test for non-parametric/parametric data. Cutoff determination: Optimal thresholds were identified using the survminer package. Survival analysis: Cox proportional hazards regression and Kaplan-Meier curves were generated via the survival package. Statistical significance was defined as p < 0.05. All analyses accounted for multiple testing corrections where applicable.

Results

Analysis of IPCD-related prognostic genes

Based on GSE26712 and GTEx datasets, differential analysis was performed to identify all DEGs between tumor and normal tissues (Fig. 1). A total of 88 prognostic genes were analyzed via univariate Cox analysis, including 20 hazardous factors (HR<1) and 68 protective factors (HR>1) for OV (Fig. 1B). Their expression profiles of prognostic genes between tumor and normal tissues were showed in Suppl Fig. 2. As indicated in Fig. 1C-D, GO and KEGG enrichment analysis was conducted on selected prognostic genes. GO terms illustrated that most genes were enriched in the biological processes (BP) such as regulation of apoptotic signaling pathway and cell chemotaxis. KEGG signaling pathway included MAPK signaling pathway, focal adhesion, fluid shear stress and atherosclerosis, PI3K-Akt signaling pathway, etc. The ssGSEA scores of 20 cell death types based on TCGA tumor data were shown in Fig. 1E, yet Disulfidptosis, Methuosis, Parthanatos, and Entoticcelldeath showed highest scores, and Netoticcelldeath showed negative scores was with blue sunken inwards.

Fig. 2.

Construction of IPCDS model and immunotherapy differences. A. Heat map of C index of models based on 101 algorithm combinations. B-K. Survival analysis results of the OV dataset between IPCDS groups. L-P. Box plots of the immunotherapy dataset between IPCDS groups. Q-T. Violin charts of IPS value differences between IPCDS groups.

Fig. 1F visualized the location of each gene on human chromosome among all included datasets. The CNV plot of each gene located in diverse chromosome was introduced in Fig. 1G, suggesting frequent changes in mitochondria-related prognostic genes. The CNV information of high and low IPCDS groups were visualized in Suppl Fig. 3A-B based on gistic2 software. The mutation results are calculated by maftools software based on TCGA dataset (Suppl Fig. 3C).

Fig. 3.

Comparison of IPCDS models. A. Bar chart comparing IPCDS with C-index of age, Stage and Grade in each dataset. B-C. PCA plots and timeROC results of each OV dataset. D. Comparison of C-index values between IPCDS model and other models reported in 32 literatures.

Construction of IPCDS

We modeled 88 prognostic genes using 101 algorithm combinations and ultimately selected SuperPC as the best model (Fig. 2A). The survival analysis results of model were calculated based on TCGA dataset and all validation datasets (Fig. 2B-K), showing a significantly higher OS in low-IPCDS group than in high-IPCDS group (p < 0.05, except for GSE17260, GSE30161, and GSE140082). Four pan-cancer immunotherapy datasets were validated in K-M curves, and the OS of low-IPCDS samples were larger than those of high-IPCDS samples in GSE78220 and IMvigor 210 (p < 0.05, Fig. 2L-P). Finally, IPS values were used to evaluate immunotherapy differences between groups from the TCIA database (Fig. 2Q-T). The survival outcomes of IPCDS in pan-cancer immunotherapy-specific datasets were shown in Suppl Fig. 1C, indicating the significant immuno-therapeutic performance in GBM datasets (p = 0.00041). The CNV results were conducted to evaluate the differences of Broad and Focal CNV Gain/Loss between IPCDS groups obtained from the gistic2.0 software (Suppl Fig. 3D-G). The comparison of TMB between IPCDS groups and the correlation of TMB with IPCDS were evaluated with no significance (Suppl Fig. 3H-I). The survival time of IPCDS-TMB groups were assessed, showing the worst survival in high IPCDS-low TMB groups compared to other groups (Suppl Fig. 3J).

As shown in Suppl Fig. 4A-B, single-cell annotation results of OV samples are Epithelial cell marker ("EPCAM", "KRT18", "KRT19", "CDH1"); Fibroblast marker ("DCN", "THY1", "COL1A1", "COL1A2"); Endothelial cell marker ("PECAM1", "CLDN5", "FLT1", "RAMP2"); T cell marker ("CD3D", "CD3E", "CD3G", "TRAC"); NK cell maker ("NKG7", "GNLY", "NCAM1", "KLRD1"); B cell marker ("CD79A", "IGHM", "IGHG3", "IGHA2"); Plasma cell Marker ("JCHAIN"); Myeloid cell marker ("LYZ", "MARCO", "CD68", "FCGR3A"), and mast cell marker ("KIT", "IPCDS4A2", "GATA2"). Among prognostic genes, the distribution of 8 genes were selected and exhibited at single-cell levels in Suppl Fig. 4C. The proportions of various cell types in high and low IPCDS groups were visualized in Suppl Fig. 4D The cell interaction differences and communication network of IPCDS groups were indicated in Suppl Fig. 4E-J.

Fig. 4.

Results of tumor immunoassay. A. Heat maps of immune infiltration differences in 7 types of software between high and low groups. B. Radar maps of correlations between high/low IPCDS and immune-related pathways. C. Results of correlation between IPCDS and immune-related gene expression. D. The probability density plots between the predicted immunotherapy outcome of TIDE algorithm and IPCDS. E. and the correlation scatter plots of IPCDS with StromalScore, ImmuneScore, ESTIMATEScore and TumorPurity from ESTIMATE algorithms.

Comparison of IPCDS models

The C-index comparisons between IPCDS and clinical features like age, Stage and Grade in 10 OV datasets are shown in Fig. 3A. PCA plots indicated the gene expression levels between IPCDS groups could effectively distinguish OV patients (Fig. 3B). 2-, 3-, and 5-year timeROC results of each OV dataset revealed the excellent predictive value of IPCDS for OV, showing a relatively higher AUC after treated 2 years. The IPCDS model was also compared with the C-index values published in 32 other literatures among 10 datasets (Fig. 3D).

Tumor immunoassay

The results evaluating immune cell infiltration differ from seven softwares between high and low IPCDS groups (Fig. 4A). The difference of the correlation between IPCDS and immune-related pathways between the high and low IPCDS groups was shown in Fig. 4B We exhibited the expression of immune regulatory genes in both groups, indicating that the high IPCDS group appeared to activate MHC class II molecules and most co-inhibitory molecules and co-stimulatory molecules (Fig. 4C). After that, TIDE algorithm was used to predict the immune response of TCGA data. There was significant difference between the two groups in terms of TIDE value and IPCDS value. The low IPCDS group had a higher TIDE value, while the no response group had a higher IPCDS value (p < 0.001, Fig. 4D). IPCDS is positively correlated with StromalScore, ImmuneScore, and ESTIMATEScore, and negatively correlated with TumorPurity based on ESTIMATE algorithm (Fig. 4E).

Analysis of model gene PDGFRA

IPCDS is positively correlated with PDGFRA (Fig. 5A). PDGFRA had a significant prognostic effect across 10 cohorts (Fig. 5B-L). There was a significant positive correlation between PDGFRA and CD8A expression levels (r = 0.21, p = 3.6e-05, Fig. 5M). Correlations of PDGFRA with immune-infiltrating cells are predicted by four algorithms (Fig. 5N). Heat map of correlation between PDGFRA gene and immune checkpoint genes were visualized in Suppl Fig. 5A. The correlations between PDGFRA gene and 50 HALLMARK signaling pathways and TIP (tumor immunity) were shown in Suppl Fig. 5B

Fig. 5.

Analysis results of model gene PDGFRA. A. Correlation of IPCDS score and PDGFRA. B. Kaplan-Meier survival analysis results of PDGFRA expression level as a protective factor (HR=1.31). C-L. Survival results of PDGFRA among TCGA and other cohorts. M. Scatterplot of correlation between PDGFRA and CD8A expression levels. N. Heat map of correlations of PDGFRA with immune-infiltrating cells.

Validation of PDGFRA expression and clinical association with the C-MYC pathway

In the analysis of immune cell infiltration and enrichment pathways associated with PDGFRA, we observed a significant correlation between PDGFRA and C-MYC target. Therefore, we performed fluorescence staining for PDGFRA in clinical ovarian tumor tissue samples obtained from three patients. Firstly, we evaluated the expression of PDGFRA in tumor and adjacent normal tissues using western blotting. Our results showed that PDGFRA protein expression was significantly elevated in tumor tissues, which was accompanied by elevated C-MYC protein levels (Fig. 6A). Quantitative analysis of protein expression confirmed that both C-MYC and PDGFRA were significantly upregulated (Fig. 6B). In addition, immunofluorescence co-staining of clinical samples revealed enhanced fluorescence intensity of both PDGFRA and C-MYC in ovarian cancer tissues, with prominent co-localization observed between 2 targets (Fig. 6C&D).

Fig. 6.

Validation of PDGFRA expression and clinical association with the C-MYC pathway. A. Western blot analysis of PDGFRA and C-MYC protein expression in tumor and adjacent para-tumor tissues. B. Quantitative analysis of PDGFRA and C-MYC protein expression levels. C-D. Quantification results (C) and representative IF staining of PDGFRA (green) and C-MYC (red) in ovarian cancer and adjacent normal tissues (D).

Discussion

Ovarian cancer, as the disease with the highest mortality rate among gynecological malignant tumors, poses challenges to precise treatment due to its complex immune microenvironment and heterogeneity [19]. This study focused on the immune-related programmed cell death (IPCD) pathway and constructed a novel prognostic model (IPCDS) by integrating multi-omics data, providing a new perspective for the risk stratification of ovarian cancer and the prediction of immunotherapy response.

The IPCDS model constructed based on 88 genes was optimized through the combination of 101 algorithms, and finally the SuperPC algorithm was selected as the best model. It is worth noting that 88 IPCD-related prognostic genes show specific expression patterns in different cell subsets. For example, pro-apoptotic genes are highly expressed in tumor epithelial cells, while immunomodulatory genes are enriched in T cells and myeloid cells, suggesting that the functional heterogeneity of the IPCD pathway may be driven by cell type-specific expression.

This model can effectively distinguish the overall survival (OS) of patients in the high and low-risk groups. The OS of the low IPCDS group was significantly higher than that of the high-risk group (TCGA and most validation datasets, p < 0.05). Furthermore, the C-index of the IPCDS model is higher than that of traditional clinical indicators such as age, stage, and grade, and it performs well in the ROC curve over a period of 2–5 years (with a high AUC), serving as an independent prognostic marker. In the pan-cancer immunotherapy datasets (such as GBM, melanoma), the low IPCDS group responded better to immunotherapy and had a longer OS, indicating that IPCDS can predict the efficacy of immunotherapy.

In the tumor microenvironment of the low IPCDS group, the infiltration levels of immune effector cells such as CD8 T cells and NK cells were higher, while in the high-risk group, pro-cancer myeloid cells (such as macrophages) were enriched. IPCDS was positively correlated with the immune score (StromalScore, ImmuneScore) and negatively correlated with tumor purity, suggesting that the expression of IPCD-related genes may reshape the tumor microenvironment by regulating immune infiltration. In the high IPCDS group, the expressions of MHC class II molecules, co-suppressor molecules (such as PD-L1), and co-stimulatory molecules were upregulated, which might promote tumor immune escape. The TIDE value was higher in the low-risk group, suggesting a stronger immune rejection reaction and greater sensitivity to immunotherapy. In addition, intercellular communication analysis revealed that IPCD-related genes are involved in regulating the signal interaction (such as ligand-receptor interaction) between immune cells like T cells and B cells and tumor cells, affecting the immune surveillance function.

This study also found that PDGFRA was positively correlated with IPCDS and had a significant prognostic effect in independent cohorts. The role of PDGFRA is significant in tumorigenesis, and it is reported frequently altered at the genomic level in high-grade glioma (HGG) [20]. There were also PDGFRA mutations present in inflammatory fibroid polyps [21]. PDGFRA may become a hub molecule connecting the IPCD pathway with immune activation by regulating the infiltration of CD8 T cells and the expression of immune checkpoint genes in OV. Significant overexpression of miR-140–5p suppresses the proliferation of ovarian cancer cells and triggers apoptosis, showing an inverse relationship with PDGFRA [22]. Its positive correlation with tumor immune pathways (such as IFN-γ response and T-cell co-stimulation) in this study suggests that targeting PDGFRA may synergistically enhance the effect of immunotherapy, consistent with previous results [23]. It is worth noting that the differential expression of PDGFRA in fibroblasts and tumor cells suggests that it may promote tumor immune escape through paracrine signals, providing a new target for the combination of anti-angiogenic therapy and immunotherapy.

Although this study demonstrated the multi-dimensional advantages of IPCDS, there are still certain limitations. For example, no significant survival differences were observed in some validation cohorts such as GSE17260 and GSE30161, which might be related to sample heterogeneity or mutation profiles. The lack of significant correlation between TMB and IPCDS suggests that more genomic features (such as DNA methylation and non-coding RNA) need to be included for integrated analysis. Future research can focus on the following directions: 1) Verify the molecular mechanism of key IPCD genes including PDGFRA in immune escape of ovarian cancer through functional experiments; 2) Analyze the spatial distribution and cellular interaction patterns of IPCD-related genes in the TME by combining spatial transcriptome; 3) Conduct prospective clinical trials based on IPCDS to verify its clinical value in guiding the selection of immune checkpoint inhibitors.

The IPCD pathway has become the core hub for precise treatment of ovarian cancer by regulating the mode of tumor cell death and the immune status of the TME. The machine learning model based on multi-omics data not only reveals the complex regulatory mechanism of the IPCD gene network, but also builds a bridge for the transformation from basic research to clinical applications. Despite challenges such as heterogeneity analysis and model optimization, with the deep integration of single-cell technology, spatial omics and artificial intelligence, individualized treatment targeting the IPCD pathway is expected to break through the diagnosis and treatment bottleneck of ovarian cancer and bring new hope to patients in the era of precision medicine.

Conclusion

In this study, through multi-omics integrated analysis, the first ovarian cancer prognosis model IPCDS based on the IPCD pathway was constructed, revealing its triple value in risk stratification, assessment of the immune microenvironment, and prediction of immunotherapy response. IPCD -related genes have become the key hubs connecting tumor genomic characteristics and immune status by regulating apoptotic signals, immune cell infiltration and intercellular communication. In particular, the discovery of core genes such as PDGFRA has provided new targets for the development of the combined targeted immunotherapy. We validated the upregulation of PDGFRA and C-MYC in ovarian cancer tissues through Western blot and confirmed their co-localization and immunofluorescence analyses. Although mechanism verification and clinical transformation are still needed, the model framework established and the molecular characteristics discovered in this study provide a scalable paradigm for precision medicine in ovarian cancer and are expected to promote the development of individualized treatment strategies based on immune programmed cell death.

Ethics statement

Ethical approval was not required for the studies on humans in accordance with the local legislation and institutional requirements because only commercially available established cell lines were used.

Generative AI statement

The author(s) declare that no Generative AI was used in the creation of this manuscript.

Funding

This work was supported by the Scientific Research Project of Jiangsu Province Maternal and Child Health Care Association (FYX202023) and Science and Technology Project of Xuzhou Health and Wellness Committee (XWKYHT20240126).

Data availability

The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding authors.

CRediT authorship contribution statement

Jieyun Sun: Writing – original draft, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Li Jing: Writing – original draft, Data curation, Conceptualization. Xiangfei Zhu: Data curation, Conceptualization. Huan Yang: Formal analysis, Data curation, Conceptualization. Yu Sun: Methodology. Xiaoyuan Lu: Writing – review & editing, Supervision. Zhao Liu: Writing – review & editing, Supervision, Investigation, Data curation, Conceptualization.

Declaration of competing interest

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