Development and validation of a macrophage-related prognostic model for overall survival in ovarian cancer via integrated RNA sequencing analysis

Introduction

Ovarian cancer (OC) is the third most common malignant tumor of the female reproductive system. Due to the lack of effective early detection and diagnostic techniques, 60–70% of patients with OC are diagnosed at an advanced stage (1). While most patients recover following primary cytoreductive surgery and standard first-line chemotherapy, approximately 70% of advanced-stage patients relapse within 2–3 years, eventually developing resistance, resulting in a 5-year survival rate of less than 40% (2).

In recent years, immunotherapy has emerged as a potential treatment strategy for multiple treatment-refractory cancers. However, immunotherapies, including programmed death 1 (PD-1) [programmed death ligand 1 (PD-L1)] immune checkpoint inhibitors, have shown limited efficacy in OC (3). Even in combination with first-line chemotherapy, these therapies have failed to show meaningful improvement in progression-free survival (PFS) (4). This low response rate primarily stems from the highly complex immunosuppressive tumor microenvironment (TME) of OC, which allows immune evasion and unrestrained tumor development (5).

The suppressive cellular microenvironment includes tumor-associated macrophages (TAMs), regulatory T cells (Tregs), myeloid-derived suppressor cells (MDSCs), and tumor-associated dendritic cells (DCs) (6). Among these, TAMs represent the major infiltrating immune subgroup in ovarian tumors and ascites, promoting the formation of an immunosuppressive microenvironment in OC, and facilitating tumor growth, invasion, angiogenesis, and metastasis (7). TAMs exhibit strong pro-tumorigenic functions and have been consistently associated with poor clinical outcomes (8). A review study has shown that the functional plasticity of TAMs in OC allows a continuum from anti-tumoral (M1) to pro-tumoral (M2) states, with the latter predominating to establish an immunosuppressive microenvironment that fuels disease progression (9).

The mechanisms through which TAMs drive OC progression have been increasingly elucidated. TAMs promote tumor cell proliferation through the secretion of factors such as vascular endothelial growth factor (VEGF), interleukin-6 (IL-6), and IL-10, which activate survival signaling pathways in cancer cells (10). Regarding metastatic dissemination, TAMs facilitate the formation of multicellular spheroids in ascitic fluid, a process dependent on epidermal growth factor (EGF) secreted by TAMs, and enhance adhesion at peritoneal metastatic sites (10). This process is particularly important in OC, where the presence and volume of ascites are associated with poor clinical outcomes (8). Furthermore, TAMs interact with tumor protein p53 (TP53), exosomes, and other immune cells such as cancer-associated fibroblasts (CAFs) to support the progression of OC (11).

Growing evidence (12-16) indicates a correlation between high levels of TAM infiltration and poor patient prognosis, with macrophage-related genes influencing the immunotherapy response. Previous studies on OC (17,18) have indicated that the expression of the alternative activation marker CD163 in malignant TAMs in ascites is closely associated with the early recurrence of serous OC following first-line treatment. Emerging evidence suggests that TAMs play a crucial role in driving platinum resistance through exosomal signaling and metabolic reprogramming (19). A review summarized dual therapeutic strategies: targeting TAMs (via recruitment inhibition, subset depletion, or phenotype reprogramming), and engineering TAMs as therapeutic agents such as chimeric antigen receptor macrophages (20). However, the impact of macrophage-associated genes on OC prognosis and treatment response remains poorly understood. Thus, exploring the molecular characteristics of macrophages and their association with prognosis and immunotherapy responses in OC may reveal new prognostic markers and enhance treatment options.

Single-cell RNA sequencing (scRNA-seq) has become an indispensable approach for analyzing the TME (21,22). Compared to bulk RNA sequencing (bulk RNA-seq), which explores the average gene expression in a cell population, scRNA-seq analyzes most transcripts in individual cell profiles using high-throughput sequencing, providing a comprehensive perspective on the molecular diversity and tumor heterogeneity of cell populations (23). Ongoing research is combining scRNA-seq and bulk RNA-seq to identify potential biomarkers for precise patient stratification and clinical benefit group selection. Wang et al. (23) used scRNA-seq and bulk RNA-seq to construct a dual-gene (CXCL13 and IL26) signature prognostic system, which suggested that the heterogeneity of OC poses a challenge for immunotherapy targeting. Hornburg et al. (24) integrated scRNA-seq data from 15 cases of ovarian tumors and bulk RNA-seq to identify immune microenvironment traits associated with T-cell infiltration patterns, suggesting chemokine receptor-ligand interactions as potential mediators of immune infiltration. Guo et al. (25) used scRNA-seq and bulk RNA-seq, along with experimental validation, to identify cell clusters and the key gene RAB13, which is closely associated with OC metastasis. Ding et al. integrated RNA-seq data to a construct cell-characterized gene signature (CCIS), a random forest-based prognostic model for OC (26). Wang et al. developed an inflammation-related gene model for OC through integrated multi-omics analysis, revealing its connections with the tumor immune microenvironment, M2 macrophage infiltration, and therapeutic sensitivity (27).

Using RNA-seq data from The Cancer Genome Atlas (TCGA) and scRNA-seq data from the Gene Expression Omnibus (GEO), we identified independent prognostic genes associated with macrophages and built a predictive model for patients with OC. The predictive performance of this model was subsequently validated using multiple approaches. Additionally, we examined the relationship between the risk model and clinical characteristics, immune infiltration landscape, immunotherapy response, and so on, to appropriately predict patient prognosis risk. Overall, our study identified clinically relevant macrophage-related indicators, described macrophage immunogenomic characteristics in OC, and provided novel insights into targeted therapies in clinical practice. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0237/rc).

Methods

Data acquisition and selection

OC scRNA-seq data were sourced from the GEO database of the National Center for Biotechnology Information (NCBI) (accession number: GSE154600) (28). This dataset included five samples: GSM4675273, GSM4675274, GSM4675275, GSM4675276, and GSM4675277. Subsequently, the cells were filtered based on the following criteria: feature count <500, unique molecular identifier count >20,000, and mitochondrial proportion >5%. After applying these filters, 20,914 cells were obtained for the subsequent single-cell analysis.

RNA-seq data [log₂-transformed fragments per kilobase of transcript per million mapped reads (FPKM)], clinical details (including age, tumor stage, and neoplasm histological grade), and survival information [overall survival (OS) and OS times] of OC patients were obtained from the University of California, Santa Cruz (UCSC) Xena platform (29) (https://toil.xenahubs.net). In this analysis, samples with the identifier “−01A” were designated as cancer tissue samples. Normal ovarian tissue samples were obtained from the Genotype-Tissue Expression (GTEx) database (30) (https://www.gtexportal.org/). Information on 442 samples, including 88 normal and 354 cancerous tissue samples, was downloaded. Among these, 342 cancer tissue samples with available prognostic information were included in the subsequent model construction analysis.

Transcriptional data for OC from the GSE19829 and GSE26193 datasets were obtained from the NCBI GEO database (31). Samples with available prognostic information were extracted, batch effects were removed, and the two datasets were merged. Ultimately, 135 OC samples were included in the study as validation datasets for subsequent modeling. We directly downloaded the preprocessed and standardized probe expression matrices and the corresponding platform annotation files for gene symbol conversion. For probes corresponding to the same gene symbol, the maximum value was used as the gene expression value in subsequent analyses.

Cell annotation

For the Seurat objects of the OC samples in the single-cell data, uniform manifold approximation and projection (UMAP) visualization revealed 20 clusters. Nine distinct cell types were manually annotated based on marker genes. These included macrophages marked by C1QA and C1QB, B cells marked by CD79A and IGHG1, CD4⁺ T cells marked by CD4, CD8⁺ T cells marked by CD8A and CD8B, natural killer (NK) cells marked by NKG7, endothelial cells marked by PECAM1, Tregs marked by FOXP3, epithelial cells marked by PECAM, and fibroblasts marked by COL1A and BGN.

Gene expression differential analysis

The “FindAllMarkers” function was used to compute differential genes among all cell clusters. Specifically, we focused on macrophages, and genes meeting the criteria of |log₂fold change (FC)| >0 and P<0.05 were selected as single-cell differentially expressed genes (scDEGs) for further investigation in the macrophage cell cluster.

Further, we employed the limma package (32) (version 3.10.3, http://www.bioconductor.org/packages/2.9/bioc/html/limma.html), which uses linear regression and empirical Bayesian methods, to perform tumor-versus-normal differential expression analysis on TCGA transcriptome data, resulting in gene-specific P values and log₂FC information. Additionally, we conducted multiple testing corrections using the Benjamini-Hochberg method, yielding adjusted P values (adj.P values). We assessed both the levels of multiplicity and significance of differences, with differential expression thresholds set as follows: adj.P<0.05 and |log₂FC| >0.263.

Protein-protein interaction (PPI) network analysis

Using the Search Tool for Recurring Instances of Neighboring Genes (STRING) database (33) (version 11.0, http://string-db.org/), which contains human PPI relationships, we obtained a PPI network for the intersection of differential genes in macrophages and differential genes in the regular transcriptome. The species used in this study was Homo sapiens. Network construction was performed using Cytoscape (34) (version 3.6.1, https://cytoscape.org/).

Selection of prognostic-related genes

In TCGA samples, based on the obtained intersection of differential genes and considering clinical survival prognosis information, we conducted a univariate Cox regression analysis using the survival package (35) (http://bioconductor.org/packages/survivalr/) in R 4.3.1. Genes with P values <0.05 were considered significant prognostic-related genes.

Construction and validation of prognostic feature models for intersection genes

Based on the intersection of differential genes significantly associated with survival obtained in the previous step, we used survival prognosis information from the training set samples. Combined with the gene expression values in each sample, we applied the least absolute shrinkage and selection operator (LASSO) Cox regression model (36) using the glmnet package (37) (version 2.0-18, https://cran.r-project.org/web/packages/glmnet/index.html) in R. We employed five-fold cross-validation to select gene combinations relevant to prognosis.

Subsequently, a risk score model was constructed based on the regression prognostic coefficients of the genes in the gene combinations and the expression levels of the genes in TCGA samples, which was expressed as follows:

RiskScore=∑βgene×Expgene[1]

where β_gene represents the LASSO regression coefficient of the gene, and Exp_gene represents the expression level of the gene in TCGA dataset.

To validate the accuracy of the model, we calculated the risk score values for each sample in the GEO dataset using the same regression coefficients. Based on the optimal risk score cut-off value, all the GEO samples were divided into high- and low-risk groups. The association between high- and low-risk groupings and actual survival prognosis information was evaluated using the Kaplan-Meier method in the survival package (version 2.41-1).

Association between clinical features and risk score

In TCGA samples, the Wilcoxon test in R (version 4.3.1) was used to statistically compare and assess the association between variables such as tumor stage, neoplasm histologic grade, age, OS, and risk score.

Independence analysis of the prognostic model and nomogram construction

First, to determine whether the risk score model could serve as an independent prognostic factor, uni- and multivariate Cox regression analyses were conducted for tumor stage, neoplasm histologic grade, age, and risk score, respectively. Variables with P<0.05 were selected. A nomogram was developed to increase the interpretability of the results of multiple-factor regression. Calibration curves were plotted to examine the accuracy of the model.

Association of high- and low-risk groups with the immune microenvironment

This analysis employed three algorithms to evaluate the immune microenvironment of the OC samples. Using cell type identification by estimating relative subsets of RNA transcripts (CIBERSORT) (38) (https://cibersort.stanford.edu/index.php), the proportions of 22 immune cell types were calculated based on their expression levels in The Cancer Genome Atlas Ovarian Serous Cystadenocarcinoma (TCGA-OV) tumor samples. CIBERSORT deconvolves the expression matrix of immune cell subtypes based on the principle of linear support vector regression. The analyses were conducted using both relative and absolute modes. (Note: absolute mode refers to the absolute proportions of each immune cell; for example, if the overall immune cell proportion is 3%, then the absolute values for the 22 major immune cells may be less than 0.1%, but the relative mode results in proportions of up to 1.) The ESTIMATE algorithm (39) was used to calculate stromal and immune scores, and ESTIMATE scores based on expression data were used to represent the presence of stromal and immune cells. The differential P values between the high- and low-risk groups were calculated using the Wilcoxon rank-sum test and visualized in box plots. Immune gene sets were employed to calculate single-sample gene set enrichment analysis (ssGSEA) scores for each sample (40). This method quantifies the enrichment of gene sets in each sample based on gene expression data and biological processes (BPs). Unlike traditional gene set enrichment analysis (GSEA) methods, ssGSEA does not require data from multiple samples and can be directly applied to single-sample gene expression data, making it suitable for small-sample analyses.

Drug sensitivity analysis

The sensitivity of each patient to drugs was estimated using the Genomics of Drug Sensitivity in Cancer (GDSC) database (41) (https://www.cancerrxgene.org/). The half-maximal inhibitory concentration (IC₅₀) was quantified using the pRRophetic package (42) in R. Wilcoxon tests were used to compare differences in drug sensitivity between the high- and low-risk groups.

Prediction of immunotherapy response

Tumor Immune Dysfunction and Exclusion (TIDE; http://tide.dfci.harvard.edu/) is a transcriptome-based immunotherapy prediction tool that predicts patterns of interaction between tumor and immune cells (43). TIDE aims to identify the biological mechanisms that cause tumor immune dysfunction and rejection, providing predictions for the responsiveness of tumor immunotherapy. The TIDE scores between the high- and low-risk groups were compared using the Wilcoxon test.

Statistical analysis

Before model construction, the “FindAllMarkers” function was employed to calculate DEGs across all cell clusters, with a specific focus on macrophages. Genes meeting the criteria of |log₂FC| >0 and P<0.05 were identified as scDEGs, allowing for an in-depth exploration of DEGs in the macrophage cluster. Additionally, the Limma package was used to perform a differential gene expression analysis of TCGA transcriptomic data, followed by Benjamini-Hochberg correction for multiple testing, resulting in adj.P values. The threshold for differential expression was set at adj.P<0.05 and |log₂FC| >0.263 based on assessments of FC and significance. For the subsequent construction of the prognostic feature model, the glmnet package in R was employed to build the LASSO Cox regression model. The survival data were analyzed using Kaplan-Meier curves, and both uni- and multivariate Cox regression analyses were conducted to identify independent prognostic risk factors. Finally, the Wilcoxon test in R (version 4.3.1) was employed to assess the statistical differences in the categorical variables between the different risk groups. The workflow of the bioinformatics analysis is depicted in Figure 1.

Figure 1 Bioinformatics analysis workflow. DEGs, differentially expressed genes; GDSC, Genomics of Drug Sensitivity in Cancer; GEO, Gene Expression Omnibus; LASSO, least absolute shrinkage and selection operator; PCA, principal component analysis; PPI, protein-protein interaction; scDEGs, single-cell differentially expressed genes; scRNA, single-cell RNA; TCGA, The Cancer Genome Atlas; TME, tumor microenvironment; UMAP, uniform manifold approximation and projection.

Ethics approval and consent to participate

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. All data used in this research were obtained from publicly available databases. The original contributing studies in these databases received ethical approval and obtained informed consent from participants. As this study involves the secondary analysis of de-identified, publicly accessible data, it is exempt from additional institutional ethics review.

Results

Single-cell heterogeneity

We analyzed the expression of 12 marker genes for single-cell subtypes across different cell clusters and presented the results using violin (Figure 2A-2L) and bubble plots (Figure 2M). The annotation revealed the presence of nine cell types (CD4⁺ T cells, CD8⁺ T cells, macrophages, fibroblasts, NK cells, Tregs, B cells, endothelial cells, and epithelial cells) (Figure 2N). The proportion of each cell type was visualized using bar charts (Figure 2O).

Figure 2 Single-cell heterogeneity. (A-L) Violin plots depicting the expression of marker genes across different cell clusters. (M) Bubble plots illustrating the expression of marker genes in different cell clusters. (N) Annotation results for the cell clusters. (O) Proportion of cell clusters. Tregs, regulatory T cell; UMAP, uniform manifold approximation and projection.

Single-cell versus conventional transcriptome differential gene analysis

As described in the Methods section, the differential analysis of single-cell data identified 845 DEGs in macrophages (available online: https://cdn.amegroups.cn/static/public/10.21037tcr-2026-1-0237-1.xlsx). We generated a heatmap displaying the differences in the expression of the top 20 differential genes in macrophages (Figure 3A; P<0.05, |log₂FC| >0).

Figure 3 Single-cell vs. TCGA transcriptome differential genes. (A) Heatmap depicting the expression of differential genes in macrophages from single-cell transcriptome data (top 20 genes, with red indicating upregulation and blue indicating downregulation). (B) Volcano plot displaying differential gene expression in TCGA transcriptome data (red represents upregulation and blue represents downregulation). (C) Venn diagram illustrating the intersection of DEGs between macrophages and TCGA transcriptome data. (D) The top 20 genes in the PPI network (darker colors indicate stronger interactions). DEGs, differentially expressed genes; PPI, protein-protein interaction; scDEGs, single-cell differentially expressed genes; TCGA, The Cancer Genome Atlas; Tregs, regulatory T cells.

Following the described methods, the differential analysis of TCGA transcriptome data for tumor versus normal samples yielded a total of 18,020 differential genes (adj.P<0.05 and |log₂FC| >0.263; available online: https://cdn.amegroups.cn/static/public/10.21037tcr-2026-1-0237-2.xlsx). The volcano plot is illustrated in Figure 3B.

For the differential genes identified in the macrophages and transcriptome data, we obtained 470 overlapping differential genes (P<0.05, Figure 3C). Based on these overlapping genes, we used the online tool STRING to predict PPI relationships. Subsequently, we imported the interactions into the Cytoscape software and employed the CytoHubba plugin with the model confrontation and collaboration (MCC) algorithm to visualize the top 20 genes in the PPI network, as shown in Figure 3D.

Construction of a prognostic model

Based on the aforementioned intersecting genes, we initially conducted a univariate Cox regression, resulting in 30 genes with a P<0.05, as shown in Figure 4A (Table S1). Among these genes, AP1S2, ARPC5, TMSB4X, PFN1, STAT1, TPM3, SH3KBP1, C6orf62, AKIRIN2, GNG5, TAP1, C1orf43, HMGN3, and SDF2L1 had hazards ratio (HRs) <1, indicating a better prognosis, whereas the remaining genes were associated with a worse prognosis. As shown in Figure 4A, all HRs were statistically significant, with none of the 95% confidence intervals (CIs) crossing 1.

Figure 4 Construction and validation of the prognostic model. (A) Forest plot for the univariate Cox analysis. (B,C) Lambda (λ) selection plot in the LASSO model (two dashed lines indicate two specific λ values, with λ.min on the left and λ.1se on the right; any λ value between them is considered suitable). The model constructed with λ.1se uses fewer genes, making it simpler, whereas λ.min uses more genes, slightly increasing its accuracy. λ.min was chosen as λ. (D) Kaplan-Meier survival curves for the prognostic model using TCGA data. (E) Receiver operating characteristic curves for the model’s predictions of 1-, 3-, and 5-year survival (the AUC represents the AUC values). (F) Kaplan-Meier survival curves for the prognostic model in the GEO validation dataset. The results in the validation queue were consistent with those in the training queue. AUC, area under the curve; CI, confidence interval; LASSO, least absolute shrinkage and selection operator; se, standard error; TCGA, The Cancer Genome Atlas.

Next, following the described methodology, we used the 30 genes that were significantly associated with survival prognosis to select an optimal combination of 19 feature genes (AKIRIN2, AP1S2, ARL4C, ARPC5, C1orf43, C5AR1, C6orf62, PIM3, RAB20, RB1, SDF2L1, SH3KBP1, STAT1, TAP1, TGFBI, THEMIS2, TPM3, TREM1, and VSIG4) and determined their corresponding prognostic coefficients using the LASSO Cox regression algorithm, as depicted in Figure 4B,4C (Table S2).

Based on the 19 prognosis-related genes, we performed Gene Ontology (GO) enrichment analysis, including BPs, cellular components (CCs), and molecular functions (MFs), as well as Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis using the clusterProfiler package, with a significance threshold of P<0.05. GO analysis revealed that these genes were significantly enriched in processes related to antigen processing and presentation, such as endogenous peptide antigen processing and presentation via major histocompatibility complex (MHC) class I and cytosol-to-endoplasmic reticulum transport (Figure S1A). In terms of CCs, the genes were primarily localized to the transporter associated with antigen processing (TAP) complex, the MHC class I peptide loading complex, and endoplasmic reticulum-associated protein complexes. Enriched MFs included ATP-binding cassette (ABC)-type peptide antigen transporter activity and TAP2 binding. KEGG pathway analysis indicated that these genes were involved in pathways including Epstein-Barr virus infection, pancreatic cancer, and the complement and coagulation cascades (Figure S1B).

Subsequently, for each element in the selected set of 19 feature genes, we employed Kaplan-Meier survival curves using the R survival package to evaluate the association between high (expression levels greater than or equal to the cut-off value) and low (expression levels lower than the cut-off value) gene expression levels and survival prognosis. In TCGA data, the high-risk patients had shorter OS than the low-risk patients (P<0.001, Figure 4D, available online: https://cdn.amegroups.cn/static/public/10.21037tcr-2026-1-0237-3.xlsx).

To evaluate the performance of the prognostic signature in predicting clinical outcomes, we calculated the area under the curve (AUC) values for 1-, 3-, and 5-year survival, which were 0.71, 0.68, and 0.73, respectively (Figure 4E). The model was subsequently validated using GEO data, where the Kaplan-Meier analysis yielded a significant result (P=0.048, Figure 4F).

Sensitivity analysis

To assess the cellular specificity of the model genes, we performed sensitivity analyses using bubble plots to visualize their average expression across immune cell subsets and violin plots to compare their distribution in macrophages versus other major cell types. The results showed that the 19 genes in the model were significantly highly expressed in macrophage subsets. The bubble plots intuitively reflected the cell type-specific expression patterns of these genes (Figure S2A), and the violin plots further confirmed their enrichment in macrophages (Figure S2B). These findings support that the model genes are predominantly derived from macrophages, rather than being confounded by signals from other cell types.

Differences in risk score clinical parameters

To assess the prognostic value of the risk score and other clinical factors, Wilcoxon tests were used to statistically compare the risk scores between different groups for the following variables: age (Figure 5A), OS (Figure 5B), tumor stage (Figure 5C), and neoplasm histologic grade (Figure 5D). The results revealed that among these four indicators, the risk score was significantly higher in individuals aged >55 years (P=0.002) and those with a deceased survival status (P<0.001). No statistically significant differences were observed in the risk scores among the remaining variables.

Figure 5 Risk score clinical parameter differences. (A) Differences in risk scores among patients with OC in different age groups (with age groups divided at 55 years). (B) Differences in risk scores among patients with OC with different survival statuses. (C) Differences in risk scores among tumors at different stages. (D) Differences in risk scores among tumors with different degrees of metastasis. OC, ovarian cancer; OS, overall survival.

Prognostic model independence analysis and nomogram construction

As described in the methods section, univariate and multivariate Cox regression analyses were performed separately for tumor stage, neoplasm histologic grade, age, and risk score to select variables with a P<0.05. As shown in Figure 6, age and risk score were ultimately considered independent prognostic factors [HR =1.019 (95% CI: 1.006–1.033, P=0.004), and HR =3.685 (95% CI: 2.640–5.143, P<0.001)]. In contrast, neither stage nor grade showed statistical significance. In the multivariate analysis, the corresponding values were 1.014 (95% CI: 1.001–1.028, P=0.03) and 3.692 (95% CI: 2.625–5.192, P<0.001), respectively. The final multivariate Cox proportional hazards model was constructed based on the independent prognostic factors identified in the regression analyses. The model is expressed as:

h(t|X)=h0(t)exp(β1⋅Age+β2⋅RiskScore)[2]

Figure 6 Cox regression analysis of clinical factors. (A) Forest plot of the univariate Cox regression analysis of the clinical factors and risk scores. (B) Forest plot of the multivariate Cox regression analysis of the clinical factors and risk scores. CI, confidence interval.

where h(t|X) represents the hazard function at time t given the covariate vector X, and h0(t) denotes the unspecified baseline hazard function. The maximum partial likelihood estimates of the regression coefficients were: age (continuous, per 1-year increment): β^1=ln(1.014)=0.0139, RiskScore [dichotomized as low-risk (reference) vs. high-risk]: β^2=ln(3.692)=1.306.

Based on these findings, a nomogram containing the above four clinical factors was constructed. By summing the scores from each variable on the point-scale axis, a total score was calculated to predict the 1-, 3-, and 5-year survival rates of patients. (Figure 7A). Additionally, calibration curves (Figure 7B-7D) and decision curve analysis curves (Figure 7E) were plotted to examine the accuracy of the model. The calibration curves for the nomogram showed that the predicted probabilities of OS estimated by the nomogram closely aligned with the actual disease-free survival proportions.

Figure 7 Nomogram construction based on risk score for patients with OC. (A) Nomogram depicting the clinical factors and risk scores. Each variable was assigned a score on a point-scale axis. The total score was calculated by summing the individual scores and projecting them onto a total point scale to estimate survival probability. (B-D) Calibration curves for 1-, 3-, and 5-year nomogram predictions. (E) Decision curve analysis curve for the nomogram model. DFS, disease-free survival; OC, ovarian cancer; OS, overall survival.

Association between high-/low-risk groups and the immune microenvironment

To explore the association between the risk score and immune cell infiltration in tumors, we estimated the relative abundance of immune and stromal cell infiltration in each sample using the CIBERSORT, ssGSEA, and ESTIMATE algorithms (available online: https://cdn.amegroups.cn/static/public/10.21037tcr-2026-1-0237-4.xlsx, Figure 8). Using the CIBERSORT method, the following eight immune microenvironment cells were found to differ significantly between the two groups (P<0.05): CD8 T cells, CD4 naïve T cells, T follicular helper cells, T gamma delta cells, M1 macrophages, M2 macrophages, activated mast cells, and neutrophils (Figure 9A). Using the ssGSEA method, the following 17 immune microenvironment cells were found to differ significantly between the two groups (P<0.05): central memory CD8 T cells, effector memory CD8 T cells, central memory CD4 T cells, effector memory CD4 T cells, T follicular helper cells, Type 1 T helper cells, Tregs, memory B cells, NK cells, MDSCs, activated DCs, plasmacytoid DCs, immature DCs, macrophages, eosinophils, mast cells, and neutrophils (Figure 9B). To further dissect the functional status of macrophages, we performed a quantitative analysis of macrophage polarization based on CIBERSORT-derived infiltration scores. Pearson correlation analysis revealed that M1 macrophage infiltration was significantly negatively correlated with the risk score (r=−0.32, P<0.001) (Figure S3A), whereas M2 macrophage infiltration showed a significant positive correlation (r=0.26, P<0.001, Figure S3B). These findings are consistent with the pro-tumor characteristics of the immune microenvironment in the high-risk group. The ESTIMATE results revealed that the high-risk group had higher stromal and ESTIMATE scores compared to the low-risk group (P<0.001); however, there was no statistically significant difference in the immune scores between the high-risk and low-risk groups (P=0.051). The results are depicted in Figure 9C-9E.

Figure 8 Immune infiltration analysis results. Heatmap depicting the relative abundance distribution of all microenvironmental cells estimated by the three algorithms (different colored horizontal bars on the left represent different algorithms, as indicated in the legend, with different colored horizontal bars at the top representing different groups, with blue indicating low risk and red indicating high risk).

Figure 9 Differences in the TME between the high- and low-risk groups. (A) Differences in TME cells estimated using the CIBERSORT method; (B) differences in TME cells estimated using the ssGSEA method; (C-E) differences in stromal, immune, and ESTIMATE scores calcualted using the ESTIMATE method. *, P<0.05; **, P<0.01; ***, P<0.001. ssGSEA, single-sample gene set enrichment analysis; TME, tumor microenvironment.

Drug sensitivity analysis between high- and low-risk groups

As described in the methods section, we estimated the sensitivity of each patient to OC drugs (veliparib, talazoparib, pazopanib, sunitinib, sorafenib, and erlotinib). IC₅₀ was quantified using the pRRophetic package in R. Differences in drug sensitivity between the high- and low-risk groups were compared using the Wilcoxon test (Figure 10). The results revealed significant differences in four of the six chemotherapy drugs, with pazopanib having a higher IC₅₀ in the low-risk group than the high-risk group (P=0.007), sunitinib having a higher IC₅₀ in the low-risk group than the high-risk group (P<0.001), sorafenib having a higher IC₅₀ in the low-risk group than the high-risk group (P<0.001), and erlotinib having a higher IC₅₀ in the low-risk group than the high-risk group (P=0.02). These findings indicate that high-risk patients may be more sensitive to multiple anti-angiogenic agents, specifically pazopanib, sunitinib, and sorafenib, as well as the EGFR inhibitor erlotinib.

Figure 10 Differences in the IC50 of drug sensitivity between the high- and low-risk groups. (A) Veliparib; (B) talazoparib; (C) pazopanib; (D) sunitinib; (E) sorafenib; (F) erlotinib. IC50, half-maximal inhibitory concentration.

Association between TIDE score and risk groups

As described in the Methods section, we calculated TIDE scores for patients online and then compared the differences between the high- and low-risk groups using the Wilcoxon test. The results revealed that the high-risk group had significantly higher TIDE scores than the low-risk group (P=0.04), indicating a lower sensitivity to immune therapy in the high-risk group (Figure 11).

Figure 11 TIDE analysis. Differences in TIDE scores between the high- and low-risk groups (red, high-risk patients; blue, low-risk patients). TIDE, Tumor Immune Dysfunction and Exclusion.

Subcellular localization of prognostic model genes in single cells

To investigate differences in gene expression patterns between tumor and normal cell lines, we visualized the expression of the prognostic model genes in single-cell clusters using UMAP plots. The results demonstrated that genes such as AP1S2, C5AR1, RB1, THEMIS2, TREM1, and VSIG4, exhibited significantly higher expression levels in macrophages than in other cell clusters (Figure 12A-12S).

Figure 12 Expression of prognostic model genes in different cell clusters. UMAP, uniform manifold approximation and projection.

Discussion

Through the scRNA-seq data analysis, we established a prognostic model for OC comprising 19 genes. Validation using TCGA and GEO datasets confirmed the predictive performance of the model, demonstrating that the risk score was an independent prognostic factor for OS in patients with OC. Algorithmic comparisons revealed that the high-risk group exhibited lower sensitivity to immunotherapy, but higher sensitivity to anti-angiogenic drugs, potentially linked to differences in the immune microenvironment.

Functional enrichment analysis revealed the critical roles of the 19 genes in immune response. Specifically, the enrichment of antigen processing and presentation pathways suggests that these genes may influence the recognition and activation of CD8⁺ T cells by modulating the efficiency of tumor antigen presentation, thereby remodeling the tumor immune microenvironment (44). Among these, the TAP complex, as a core component of MHC class I-mediated endogenous antigen presentation, plays a direct role in determining the intensity of immune surveillance (45). Additionally, the enrichment of the complement and coagulation cascades implies that these genes may be involved in tumor-associated inflammation and angiogenesis, potentially affecting drug sensitivity and patient prognosis (46). Collectively, these findings provide functional insights into how the 19-gene signature influences prognosis in OC through the regulation of immune infiltration and drug response, and also highlight promising directions for subsequent experimental validation.

Due to the molecular and cellular heterogeneity of OC, a single traditional classification system is insufficient for accurate prognostic assessment (47). With the continuous refinement of molecular subtyping, traditional histological classifications and binary models for OC are gradually being replaced (48). In 2011, TCGA research group classified patients with high-grade serous OC into four subtypes based on mRNA expression features (49). Building on this, Jonsson et al. further subdivided high-grade serous OC into differentiated, immune-like, proliferative, and mesenchymal-like subtypes by integrating molecular pathological features (50). Patients with immune-like subtype had the longest survival, while those with proliferative and mesenchymal-like subtypes had shorter survival. Another study classified epithelial OC into five subtypes based on gene expression patterns (EPI-A, EPI-B, MES, STEM-A, and STEM-B), each exhibiting distinct pathological features, signaling pathway alterations, and prognoses (51). Molecular subtypes predict prognosis and aid in the identification of specific molecular targets for targeted therapy. Kommoss et al. revealed that patients with the proliferative and mesenchymal subtypes benefited from bevacizumab monotherapy, with median PFS of 10.1 and 8.2 months, respectively (52). Antony et al. reported that AXL was enriched in mesenchymal (Mes) subtype tumor tissues, and the AXL inhibitor R428 reduced the activation of receptor tyrosine kinase (RTK) and extracellular signal-regulated kinase (ERK), inhibiting the movement and proliferation of Mes cells (53), suggesting that patients with Mes subtype OC may benefit from AXL-targeted therapy. Previous molecular subtyping studies have largely focused on gene expression features in the overall TME, providing limited insights into the interactions between prognosis and single-cell populations in the TME. Unlike other models, our unique scRNA-seq-based prognostic model, which comprises macrophage-related genes, comprehensively reflects macrophage function in the TME. Our model, which was constructed using diverse datasets and algorithmic outputs, achieved an AUC of 0.68–0.73 on heterogeneous training sets, demonstrating its enhanced reliability and relevance. While International Federation of Gynecology and Obstetrics (FIGO) stage (5-year survival: 89% stage I vs. 20% stage IV), residual tumor after primary debulking surgery (54) and CA-125 kinetics (HR: 0.24–0.35) are powerful clinical predictors (55), they lack biological insight into the TME. Our macrophage-specific signature provides complementary information by reflecting TAM-mediated immunosuppression and chemoresistance mechanisms.

The CCIS model, constructed using random survival forest on scRNA-seq-derived genes, demonstrated superior and consistent prognostic accuracy across multiple cohorts (an AUC of up to 0.875 for 5-year OS), and its ability to predict immunotherapy response was validated (26). Conversely, our LASSO-based model (an AUC of 0.73 for 5-year OS) showed moderate performance and was validated using a single GEO dataset. Both studies successfully integrated single-cell and bulk transcriptomic data to develop prognostic signatures for OC, revealing associations with the tumor immune microenvironment. Wang et al. uniquely constructed a dual-score system [Metabolic Risk Score (MRS) and Genetic Risk Score (GRS)] focused explicitly on macrophage polarization and its crosstalk with CD8⁺ T cells, supported by a detailed cell-cell communication analysis. Their study provided specific mechanistic insights into macrophage-T cell interactions, but the GRS model’s predictive performance (AUCs not explicitly stated) appears to have been less comprehensively validated (27).

The TME plays a crucial role in clinical outcomes and treatment responses. In OC, various immune-infiltrating cells, such as mature DCs, M1 macrophages, NK cells, αβT cells, and γδT cells, exhibit anti-tumor effects. Conversely, immunosuppressive cells such as immature/tolerant DCs, M2 macrophages, Tregs, and MDSCs hinder anti-tumor immunity (56,57). Studies suggest that immune cell infiltration is correlated with OC prognosis. The infiltration of naïve CD4 T cells, resting CD4 memory T cells, M2 macrophages, and eosinophils is associated with a poor prognosis, while the activation of CD4 memory T cells, M0 macrophages, and M1 macrophages is correlated with an improved prognosis (58). Moreover, the gene expression profiles of immune-infiltrating cells in the TME have been linked to the prognosis of patients with OC. CRMP2, secreted by CAFs, promotes OC growth and metastasis, leading to adverse outcomes (59). The intricate functions of macrophages are related to OC prognosis and treatment outcomes (58). The TME in OC often skews macrophage polarization toward the M2-like phenotype, which fosters an immunosuppressive milieu that aids in disease progression (9). Our prognostic model, incorporating 19 macrophage-specific genes, revealed that the risk score was associated with the infiltration of immune cells, based on ESTIMATE scores and the immune cell infiltration analysis. The high-risk group exhibited higher overall immune cell infiltration and lower CD8⁺ T-cell infiltration. The proportion of TAMs was higher, and the ratio of M1 to M2 macrophages was lower in the high-risk group. Therefore, despite the increased infiltration of immune-activated cells in high-risk individuals, these cells may be functionally inhibited, possibly contributing to an unfavorable prognosis.

Chemotherapy remains the primary approach for the treatment of OC; however, its effectiveness is limited due to tumor heterogeneity and complexity. Novel targeted therapies, such as poly (ADP-ribose) polymerase (PARP) inhibitors, have been shown to significantly prolong PFS (60). In immunotherapy, a few Phase III trials have been prematurely terminated due to ineffectiveness (3). Reports suggest that combining PARP inhibitors with anti-angiogenic agents enhances the efficacy of immunotherapy by stimulating adaptive immune responses, turning “cold tumors” into “hot tumors” (61,62). The DUO-O study demonstrated the feasibility of first-line maintenance therapy with a triple combination of anti-angiogenic agents, PARP inhibitors, and immunotherapy (63). The TOPACIO study evaluated the efficacy of niraparib and pembrolizumab in late-stage recurrent OC, with the biomarker analysis revealing a positive immune score that correlated with clinical benefits (64). This suggests a correlation between immune-related genes and features of the immune microenvironment, which may influence the clinical benefits of combined immunotherapy in patients with OC. In our study, the drug sensitivity analysis revealed that the high-risk patients showed greater sensitivity to certain anti-angiogenic drugs (pazopanib, sunitinib, and sorafenib). Hypoxia-skewed macrophages drive tumor vascular hyperpermeability and hinder drug delivery, while targeting adrenomedullin signaling restores vascular integrity and improves chemotherapy (20). Studies have indicated that VEGF and miR-501-3p derived from TAMs directly mediate vascular formation in tumor tissues, which may explain why increased sensitivity is associated with a higher presence of M2 macrophages, inducing vascular formation (65,66). The TIDE score results indicated weaker sensitivity to immunotherapy in the high-risk group, which was possibly linked to the suppressive TME.

Our research revealed the need for more personalized treatment plans for high-risk patients and underscores the potential of drugs such as PARP inhibitors to enhance OC survival rates. However, this study had several limitations. First, the use of retrospective data to construct and validate prognostic models, relatively singular data sources may introduce potential bias. Second, the GDSC database used for drug sensitivity analysis is based on tumor cell lines and may not fully reflect the intrinsic drug sensitivity of patients in vivo. Third, while we identified macrophage-associated genes from scRNA-seq data, the expression signals in bulk RNA-seq represent a mixture of multiple cell types, which may confound the interpretation of macrophage-specific contributions. To address this concern, we performed a sensitivity check, which confirmed that several key model genes are preferentially expressed in macrophages. Nevertheless, further experimental validation and functional studies are needed to elucidate the regulatory mechanisms of these genes in OC progression and treatment response. These investigations will be the focus of our future research.

Conclusions

This study discovered and established a novel prognostic risk model for OC, using macrophage gene characteristics to effectively predict the prognosis and treatment response of patients with OC. The immunological analysis confirmed the correlation between risk score and TME, elucidating diverse prognoses among patients, and providing a basis for further research on biomarkers and anti-tumor treatment strategies. This study is clinically valuable for screening populations that are likely to benefit from treatment and improving the prognosis of patients with OC.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0237/rc

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0237/prf

Funding: This work was supported by the Yunnan Revitalization Talent Support Program (No. XDYC-YLWS-2024-0079 awarded to T.Y., No. KH-2025-XDYC-YLWS-07 awarded to J.Z.), and First People’s Hospital of Yunnan Province, Yunnan Provincial Clinical Medicine Research Center for Gynecological and Obstetric Diseases (No. 202505AJ310008 to J.Z.).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0237/coif). T.Y. reports funding support from the Yunnan Revitalization Talent Support Program (No. XDYC-YLWS-2024-0079). J.Z. reports funding support from the Yunnan Revitalization Talent Support Program (No. KH-2025-XDYC-YLWS-07) and First People’s Hospital of Yunnan Province, Yunnan Provincial Clinical Medicine Research Center for Gynecological and Obstetric Diseases (No. 202505AJ310008). The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Schem C, Bauerschlag DO, Meinhold-Heerlein I, et al. Benign and borderline tumors of the ovary. Ther Umsch 2007;64:369-74. [Crossref] [PubMed]
2. Torre LA, Trabert B, DeSantis CE, et al. Ovarian cancer statistics, 2018. CA Cancer J Clin 2018;68:284-96. [Crossref] [PubMed]
3. Maiorano BA, Maiorano MFP, Lorusso D, et al. Ovarian Cancer in the Era of Immune Checkpoint Inhibitors: State of the Art and Future Perspectives. Cancers (Basel) 2021;13:4438. [Crossref] [PubMed]
4. Konstantinopoulos PA, Cannistra SA. Immune Checkpoint Inhibitors in Ovarian Cancer: Can We Bridge the Gap Between IMagynation and Reality? J Clin Oncol 2021;39:1833-8. [Crossref] [PubMed]
5. Cai DL, Jin LP. Immune Cell Population in Ovarian Tumor Microenvironment. J Cancer 2017;8:2915-23. [Crossref] [PubMed]
6. Balkwill FR, Capasso M, Hagemann T. The tumor microenvironment at a glance. J Cell Sci 2012;125:5591-6. [Crossref] [PubMed]
7. Colvin EK. Tumor-associated macrophages contribute to tumor progression in ovarian cancer. Front Oncol 2014;4:137. [Crossref] [PubMed]
8. El-Arabey AA, Alkhalil SS, Al-Shouli ST, et al. Revisiting macrophages in ovarian cancer microenvironment: development, function and interaction. Med Oncol 2023;40:142. [Crossref] [PubMed]
9. Cheng KC, Lin YH, Wu DS, et al. Macrophages and neutrophils in ovarian cancer microenvironment. Front Immunol 2025;16:1677441. [Crossref] [PubMed]
10. Yin M, Li X, Tan S, et al. Tumor-associated macrophages drive spheroid formation during early transcoelomic metastasis of ovarian cancer. J Clin Invest 2016;126:4157-73. [Crossref] [PubMed]
11. El-Arabey AA, Denizli M, Kanlikilicer P, et al. CORRIGENDUM: GATA3 as a master regulator for interactions of tumor-associated macrophages with high-grade serous ovarian carcinoma. Cell Signal 2022;89:110147. [Crossref] [PubMed]
12. He YF, Zhang MY, Wu X, et al. High MUC2 expression in ovarian cancer is inversely associated with the M1/M2 ratio of tumor-associated macrophages and patient survival time. PLoS One 2013;8:e79769. [Crossref] [PubMed]
13. Lan C, Huang X, Lin S, et al. Expression of M2-polarized macrophages is associated with poor prognosis for advanced epithelial ovarian cancer. Technol Cancer Res Treat 2013;12:259-67. [Crossref] [PubMed]
14. Le Page C, Marineau A, Bonza PK, et al. BTN3A2 expression in epithelial ovarian cancer is associated with higher tumor infiltrating T cells and a better prognosis. PLoS One 2012;7:e38541. [Crossref] [PubMed]
15. No JH, Moon JM, Kim K, et al. Prognostic significance of serum soluble CD163 level in patients with epithelial ovarian cancer. Gynecol Obstet Invest 2013;75:263-7. [Crossref] [PubMed]
16. Zhang M, He Y, Sun X, et al. A high M1/M2 ratio of tumor-associated macrophages is associated with extended survival in ovarian cancer patients. J Ovarian Res 2014;7:19. [Crossref] [PubMed]
17. Reinartz S, Schumann T, Finkernagel F, et al. Mixed-polarization phenotype of ascites-associated macrophages in human ovarian carcinoma: correlation of CD163 expression, cytokine levels and early relapse. Int J Cancer 2014;134:32-42. [Crossref] [PubMed]
18. Yuan X, Zhang J, Li D, et al. Prognostic significance of tumor-associated macrophages in ovarian cancer: A meta-analysis. Gynecol Oncol 2017;147:181-7. [Crossref] [PubMed]
19. Li M, Ma Y, Dai T, et al. Advance in therapies targeting tumor-associated macrophages in ovarian cancer. Front Immunol 2025;16:1677839. [Crossref] [PubMed]
20. Górczak M, Kiraga Ł. Macrophages-Target and Tool in Tumor Treatment: Insights from Ovarian Cancer. Cancers (Basel) 2025;17:3182. [Crossref] [PubMed]
21. Potter SS. Single-cell RNA sequencing for the study of development, physiology and disease. Nat Rev Nephrol 2018;14:479-92. [Crossref] [PubMed]
22. Izar B, Tirosh I, Stover EH, et al. A single-cell landscape of high-grade serous ovarian cancer. Nat Med 2020;26:1271-9. [Crossref] [PubMed]
23. Wang Z, Zhang J, Dai F, et al. Integrated analysis of single-cell RNA-seq and bulk RNA-seq unveils heterogeneity and establishes a novel signature for prognosis and tumor immune microenvironment in ovarian cancer. J Ovarian Res 2023;16:12. [Crossref] [PubMed]
24. 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;39:928-944.e6. [Crossref] [PubMed]
25. Guo J, Han X, Li J, et al. Single-cell transcriptomics in ovarian cancer identify a metastasis-associated cell cluster overexpressed RAB13. J Transl Med 2023;21:254. [Crossref] [PubMed]
26. Ding H, Hu B, Guo R. Comprehensive analysis of single cell and bulk data develops a promising prognostic signature for improving immunotherapy responses in ovarian cancer. PLoS One 2024;19:e0298125. [Crossref] [PubMed]
27. Wang S, Wang X, Xia X, et al. Identification of the immune subtype of ovarian cancer patients by integrated analyses of transcriptome and single-cell sequencing data. Sci Rep 2022;12:13296. [Crossref] [PubMed]
28. Geistlinger L, Oh S, Ramos M, et al. Multiomic Analysis of Subtype Evolution and Heterogeneity in High-Grade Serous Ovarian Carcinoma. Cancer Res 2020;80:4335-45. [Crossref] [PubMed]
29. Clark AJ, Lillard JW Jr. A Comprehensive Review of Bioinformatics Tools for Genomic Biomarker Discovery Driving Precision Oncology. Genes (Basel) 2024;15:1036. [Crossref] [PubMed]
30. The Genotype-Tissue Expression (GTEx) project. Nat Genet 2013;45:580-5. [Crossref] [PubMed]
31. Barrett T, Wilhite SE, Ledoux P, et al. NCBI GEO: archive for functional genomics data sets--update. Nucleic Acids Res 2013;41:D991-5. [Crossref] [PubMed]
32. Ritchie ME, Phipson B, Wu D, et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res 2015;43:e47. [Crossref] [PubMed]
33. Szklarczyk D, Franceschini A, Wyder S, et al. STRING v10: protein-protein interaction networks, integrated over the tree of life. Nucleic Acids Res 2015;43:D447-52. [Crossref] [PubMed]
34. Shannon P, Markiel A, Ozier O, et al. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res 2003;13:2498-504. [Crossref] [PubMed]
35. Therneau TM, Lumley T. Package ‘survival’. R Top Doc 2015;128:28-33.
36. Shahraki HR, Salehi A, Zare N. Survival Prognostic Factors of Male Breast Cancer in Southern Iran: a LASSO-Cox Regression Approach. Asian Pac J Cancer Prev 2015;16:6773-7. [Crossref] [PubMed]
37. Friedman J, Hastie T, Tibshirani R. Package ‘glmnet’. CRAN R Repository 2021.
38. Chen B, Khodadoust MS, Liu CL, et al. Profiling Tumor Infiltrating Immune Cells with CIBERSORT. Methods Mol Biol 2018;1711:243-59. [Crossref] [PubMed]
39. 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. [Crossref] [PubMed]
40. Yi M, Nissley DV, McCormick F, et al. ssGSEA score-based Ras dependency indexes derived from gene expression data reveal potential Ras addiction mechanisms with possible clinical implications. Sci Rep 2020;10:10258. [Crossref] [PubMed]
41. Yang W, Soares J, Greninger P, et al. Genomics of Drug Sensitivity in Cancer (GDSC): a resource for therapeutic biomarker discovery in cancer cells. Nucleic Acids Res 2013;41:D955-61. [Crossref] [PubMed]
42. Geeleher P, Cox N, Huang RS. pRRophetic: an R package for prediction of clinical chemotherapeutic response from tumor gene expression levels. PLoS One 2014;9:e107468. [Crossref] [PubMed]
43. Jiang P, Gu S, Pan D, et al. Signatures of T cell dysfunction and exclusion predict cancer immunotherapy response. Nat Med 2018;24:1550-8. [Crossref] [PubMed]
44. Ouyang P, Zhang J, He X, et al. Infiltration characteristics and regulatory mechanisms of CD8(+) T lymphocytes in solid tumors: spatial distribution, biological functions, and interactions with the immune microenvironment. Front Immunol 2025;16:1661545. [Crossref] [PubMed]
45. Mantel I, Sadiq BA, Blander JM. Spotlight on TAP and its vital role in antigen presentation and cross-presentation. Mol Immunol 2022;142:105-19. [Crossref] [PubMed]
46. Xu R, Wang J, Zhu Q, et al. Integrated models of blood protein and metabolite enhance the diagnostic accuracy for Non-Small Cell Lung Cancer. Biomark Res 2023;11:71. [Crossref] [PubMed]
47. Veneziani AC, Gonzalez-Ochoa E, Alqaisi H, et al. Heterogeneity and treatment landscape of ovarian carcinoma. Nat Rev Clin Oncol 2023;20:820-42. [Crossref] [PubMed]
48. Köbel M, Kang EY. The Evolution of Ovarian Carcinoma Subclassification. Cancers (Basel) 2022;14:416. [Crossref] [PubMed]
49. Integrated genomic analyses of ovarian carcinoma. Nature 2011;474:609-15. [Crossref] [PubMed]
50. Jönsson JM, Johansson I, Dominguez-Valentin M, et al. Molecular subtyping of serous ovarian tumors reveals multiple connections to intrinsic breast cancer subtypes. PLoS One 2014;9:e107643. [Crossref] [PubMed]
51. Tan TZ, Miow QH, Huang RY, et al. Functional genomics identifies five distinct molecular subtypes with clinical relevance and pathways for growth control in epithelial ovarian cancer. EMBO Mol Med 2013;5:1051-66. [Crossref] [PubMed]
52. Kommoss S, Winterhoff B, Oberg AL, et al. Bevacizumab May Differentially Improve Ovarian Cancer Outcome in Patients with Proliferative and Mesenchymal Molecular Subtypes. Clin Cancer Res 2017;23:3794-801. [Crossref] [PubMed]
53. Antony J, Tan TZ, Kelly Z, et al. The GAS6-AXL signaling network is a mesenchymal (Mes) molecular subtype-specific therapeutic target for ovarian cancer. Sci Signal 2016;9:ra97. [Crossref] [PubMed]
54. Torre LA, Trabert B, DeSantis CE, et al. Ovarian cancer statistics, 2018. CA Cancer J Clin 2018;68:284-96. [Crossref] [PubMed]
55. Carrot A, Corbaux P, Colomban O, et al. Prognostic Value of a Joint K-PD Model With Tumor Size Dynamics and CA-125 Kinetics in Recurrent Ovarian Cancer Patients: BOLD Phase II GINECO Study. CPT Pharmacometrics Syst Pharmacol 2025;14:1310-21. [Crossref] [PubMed]
56. Deng X, Terunuma H, Terunuma A, et al. Ex vivo-expanded natural killer cells kill cancer cells more effectively than ex vivo-expanded γδ T cells or αβ T cells. Int Immunopharmacol 2014;22:486-91. [Crossref] [PubMed]
57. Santoiemma PP, Powell DJ Jr. Tumor infiltrating lymphocytes in ovarian cancer. Cancer Biol Ther 2015;16:807-20. [Crossref] [PubMed]
58. Yan C, Li K, Meng F, et al. Integrated immunogenomic analysis of single-cell and bulk tissue transcriptome profiling unravels a macrophage activation paradigm associated with immunologically and clinically distinct behaviors in ovarian cancer. J Adv Res 2023;44:149-60. [Crossref] [PubMed]
59. Jin Y, Bian S, Wang H, et al. CRMP2 derived from cancer associated fibroblasts facilitates progression of ovarian cancer via HIF-1α-glycolysis signaling pathway. Cell Death Dis 2022;13:675. [Crossref] [PubMed]
60. Farmer H, McCabe N, Lord CJ, et al. Targeting the DNA repair defect in BRCA mutant cells as a therapeutic strategy. Nature 2005;434:917-21. [Crossref] [PubMed]
61. Peyraud F, Italiano A. Combined PARP Inhibition and Immune Checkpoint Therapy in Solid Tumors. Cancers (Basel) 2020;12:1502. [Crossref] [PubMed]
62. Cao Y, Langer R, Ferrara N. Targeting angiogenesis in oncology, ophthalmology and beyond. Nat Rev Drug Discov 2023;22:476-95. [Crossref] [PubMed]
63. Westin SN, Moore K, Chon HS, et al. Durvalumab Plus Carboplatin/Paclitaxel Followed by Maintenance Durvalumab With or Without Olaparib as First-Line Treatment for Advanced Endometrial Cancer: The Phase III DUO-E Trial. J Clin Oncol 2024;42:283-99. [Crossref] [PubMed]
64. Färkkilä A, Gulhan DC, Casado J, et al. Immunogenomic profiling determines responses to combined PARP and PD-1 inhibition in ovarian cancer. Nat Commun 2020;11:1459. [Crossref] [PubMed]
65. Tacconi C, Ungaro F, Correale C, et al. Activation of the VEGFC/VEGFR3 Pathway Induces Tumor Immune Escape in Colorectal Cancer. Cancer Res 2019;79:4196-210. [Crossref] [PubMed]
66. Yin Z, Ma T, Huang B, et al. Macrophage-derived exosomal microRNA-501-3p promotes progression of pancreatic ductal adenocarcinoma through the TGFBR3-mediated TGF-β signaling pathway. J Exp Clin Cancer Res 2019;38:310. [Crossref] [PubMed]