A prognostic model based on Scissor⁺ cancer-associated macrophages identified from bulk and single cell RNA sequencing data in ovarian cancer

Introduction

Ovarian cancer (OC), the most common pathological subtype of gynecological malignancies, comprises 70–80% of all OC cases (1,2). Its clinical management has long been hindered by three principal challenges: difficulty in early diagnosis, high recurrence rates, and therapy resistance (3,4). The 5-year survival rate remains consistently below 50%, positioning OC among the gynecological cancers with the poorest prognosis (5). Current clinical prognostic parameters, such as serum markers carbohydrate antigen 125 (CA125) and human epididymis secretory protein 4 (HE4) along with pathological staging, exhibit considerable limitations (6,7). Specifically, CA125 demonstrates a diagnostic sensitivity of only 43.5–65.7% in early-stage OC, which limits its utility in stratifying patients by prognostic risk (8,9). Although HE4 offers higher specificity, it fails to adequately reflect the immune status of the tumor microenvironment (TME), despite the established link between TME immune remodeling and OC progression, recurrence, and immunotherapy response (10). Consequently, identifying potential biomarkers is essential for improving OC diagnosis and treatment.

Tumor-associated macrophages (TAMs), the most abundant immune cells in the TME, critically maintain the functional equilibrium between polarized M1 pro-inflammatory/anti-tumor and M2 immunosuppressive subtypes (11,12). M2-type TAMs are known to suppress CD8⁺ T cell cytotoxicity by secreting inhibitory cytokines like transforming growth factor-β (TGF-β) and interleukin-10 (IL-10), thereby accelerating tumor progression and immune evasion (13,14). Conventional bulk RNA sequencing, however, fails to resolve the functional heterogeneity among TAMs, which complicates the precise identification of TAM subpopulations with true prognostic relevance (15,16). Consequently, progress in developing prognostic markers and therapeutic targets based on TAMs has been slow (17). Furthermore, existing multi-gene prognostic models depend primarily on generalized gene-based screening rather than targeting specific functional cell subpopulations, limiting their clinical applicability and mechanistic interpretability (6,18,19).

Recent progress in single-cell RNA sequencing (scRNA-seq) has enabled the identification of distinct cell subpopulations at single-cell resolution. This advancement provides a more precise view of tumor biology (20-22). This study integrates scRNA-seq with Scissor analysis, machine learning methods—including least absolute shrinkage and selection operator (LASSO)-Cox regression and random survival forests—and multi-cohort validation. Applying Scissor analysis to The Cancer Genome Atlas (TCGA) survival data, we identified Scissor⁺ macrophage subpopulations directly linked to poor prognosis, extracted their key differentially expressed genes (DEGs), and constructed a prognostic risk model. Finally, we investigated the functional mechanisms of core genes by examining immune microenvironment associations and molecular regulatory networks. This integrated strategy aims to pinpoint macrophages and their genetic features associated with unfavorable outcomes in OC, offering new theoretical and experimental support for precision prognostic stratification and immunotherapy target development in OC. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2710/rc).

Methods

Dataset acquisition and preprocessing

The datasets used in this study were obtained from TCGA and the Gene Expression Omnibus (GEO) database. Specifically, bulk RNA-sequencing data of ovarian serous cystadenocarcinoma (TCGA-OV) were downloaded from the Genomic Data Commons (GDC) data portal (https://portal.gdc.cancer.gov/), which included 429 ovarian tumor tissue samples (supplementary tables 1 and 2 available at https://cdn.amegroups.cn/static/public/tcr-2025-1-2710-1.xlsx; https://cdn.amegroups.cn/static/public/tcr-2025-1-2710-2.xlsx). In addition, microarray expression datasets were retrieved from the GEO database (https://www.ncbi.nlm.nih.gov/geo/), including GSE26712, GSE63885, and GSE140082. GSE26712, based on the Affymetrix Human Genome U133A Array (GPL96 platform) (supplementary tables 3-5 available at https://cdn.amegroups.cn/static/public/tcr-2025-1-2710-3.xlsx; https://cdn.amegroups.cn/static/public/tcr-2025-1-2710-4.xlsx; https://cdn.amegroups.cn/static/public/tcr-2025-1-2710-5.xlsx), comprised 185 primary ovarian tumor samples and 10 normal ovarian surface epithelium control samples. GSE63885, generated using the Affymetrix Human Genome U133 Plus 2.0 Array (GPL570 platform), contained 101 OC samples. GSE140082, derived from the Illumina HumanHT-12 WG-DASL V4.0 R2 expression beadchip (GPL14951 platform), included 380 ovarian tumor tissue samples. Furthermore, the scRNA-seq dataset GSE184880, obtained from the GEO database, contained 7 OC tissues and 5 normal ovarian tissues, and was used for cellular-level transcriptional analysis. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Cross-platform data processing and normalization

RNA-seq data (TCGA cohort) were analyzed using TPM-normalized expression values. For microarray datasets (GPL96, GPL570, GPL14951), raw expression matrices were normalized using the limma package (version 3.56.2), including background correction and quantile normalization.

To minimize platform-induced batch effects, prognostic analyses were performed independently within each dataset. Risk scores were calculated using the same gene signature and coefficients, and survival and ROC analyses were conducted separately for each cohort.

Processing of single-cell RNA-seq data

The scRNA-seq dataset GSE184880 was processed using the Seurat package (version 5.0.1) in R. Low-quality cells and potential doublets were excluded using threshold-based quality control (QC) metrics, including nFeature_RNA, nCount_RNA, and percent.mt. Cells with abnormally low gene counts, excessively high UMI counts suggestive of multiplets, or mitochondrial transcript content greater than 25% were removed. No additional computational doublet detection algorithm was applied. Instead, stringent QC thresholds combined with downstream inspection of canonical lineage marker expression were used to minimize potential doublet contamination. Violin plots of nFeature_RNA, nCount_RNA, and percent.mt before and after filtering are provided in Figures S1,S2. Genes were filtered according to minimum expression levels and feature count thresholds to ensure data robustness. The expression matrix was log-normalized, and the top 2,000 highly variable genes were identified for downstream analysis. After data scaling using the ScaleData function, principal component analysis (PCA) was performed for dimensionality reduction. Cell clustering was carried out using the FindNeighbors and FindClusters functions, and cell types were annotated based on canonical marker genes. To minimize technical variation across samples, batch effects were corrected using the Harmony algorithm, and residual low-quality cells were further excluded before subsequent analyses.

Cell-cell interaction analysis

The cell-cell interactions between different cell types were evaluated via CellChat R package (version 1.6.0). Normalized count data from each condition were used to create a CellChat object, and the recommended preprocessing functions for the analysis of individual datasets were applied with default parameters.

Scissor algorithm for identifying phenotypically related cells

The Scissor R package (version 2.0.0) was employed to identify single-cell populations associated with poor clinical outcomes in OC. Scissor was performed with the following parameters: family = “cox”, alpha =0.05. By integrating scRNA-seq data (GSE184880) with bulk RNA-seq and survival data from the TCGA-OV cohort, individual cells were classified into Scissor⁺ (poor-prognosis-associated) and Scissor⁻ (non-poor-prognosis-associated) groups. Differential expression analysis between Scissor⁺ and Scissor⁻ macrophages [|log₂ fold change (FC)| >1, adjusted P<0.05] revealed key genes linked to unfavorable prognosis. These genes were further refined by intersecting with DEGs identified between TAMs from tumor and normal tissues, ultimately yielding 94 candidate prognostic genes.

It should be noted that survival association was modeled at the single-cell level through Scissor’s penalized Cox regression framework. The 94 intersecting genes were identified through differential expression analysis and were not subjected to individual Cox regression testing at this stage. Prognostic gene selection was subsequently performed using machine learning approaches.

Functional enrichment analysis

The 94 key DEGs were subjected to functional enrichment analysis using the clusterProfiler R package (version 4.8.3). Gene Ontology (GO) enrichment, including biological process (BP), cellular component (CC), and molecular function (MF) categories, as well as Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment, were conducted to elucidate the potential biological functions of these genes. An adjusted P value <0.05 was considered statistically significant. The enrichment results were visualized using the ggplot2 package (version 3.5.1).

Machine learning algorithms and survival analysis

To identify prognostically relevant genes among the 94 candidate DEGs derived from single-cell analysis, we first matched these genes to the TCGA-OV bulk RNA-seq dataset. A total of 92 genes were detectable in the TCGA cohort and were included in subsequent survival analyses.

Univariate Cox proportional hazards regression was performed for the 92 genes using the survival package (version 3.5-7) in R (version 4.3.1). Hazard ratios (HRs), 95% confidence intervals (CIs), and corresponding P values were calculated. Genes with P<0.05 were considered statistically significant. Given that this analysis was conducted on a biologically pre-selected candidate gene set rather than a genome-wide screening context, multiple testing correction was not applied at this stage. The proportional hazards assumption was assessed using Schoenfeld residual tests.

To further identify robust prognostic genes and reduce overfitting, we applied two complementary machine learning approaches: LASSO-Cox regression and Random Survival Forest (RSF). LASSO-Cox regression was implemented using the glmnet package (version 4.1-8) to perform variable selection and regularization in a high-dimensional setting. The optimal penalty parameter (λ) was determined using 10-fold cross-validation. The random seed was set to 408 (set.seed [0408]) to ensure reproducibility. To account for potential nonlinear effects and gene-gene interactions not captured by linear Cox models, an RSF model was constructed using the randomForestSRC package (version 3.3.3). The RSF model was tuned using 10-fold cross-validation with the same random seed (set.seed [0408]). Genes consistently identified by both LASSO-Cox and RSF were considered robust prognostic candidates and retained as core genes for subsequent model construction. Finally, a prognostic nomogram was developed using the rms package (version 6.7-1) to facilitate clinical application.

Survival analysis

Overall survival (OS) was compared between groups using the log-rank test and visualized with Kaplan-Meier survival curves. The predictive accuracy of the model was evaluated through time-dependent receiver operating characteristic (ROC) analysis, assessing the area under the curve (AUC) at 1-, 3-, 5-, and 10-year time points. External validation of the prognostic model was performed using the GSE26712, GSE63885, and GSE140082 datasets, following the same analytical pipeline.

Immune infiltration

Immune infiltration levels between high-risk and low-risk groups (stratified by median risk score) were assessed using the CIBERSORT algorithm, which deconvolutes bulk transcriptomic data to estimate the relative proportions of 22 immune cell types (LM22 signature). CIBERSORT has been widely applied to characterize immune landscapes in solid tumors. CIBERSORT was performed in relative mode to estimate immune cell fractions in the TCGA-OV cohort. The results were visualized with the “ggplot2” package (version 3.5.1) for compositional bar plots, and “pheatmap” (version 1.0.12) for correlation heatmap.

Gene regulatory network analysis

The microRNA (miRNA) and transcription factor (TF) diagnostic biomarker regulatory networks were explored and constructed using NetworkAnalyst (http://www.networkanalyst.ca). NetworkAnalyst is a comprehensive, web-based tool designed to generate new biological hypotheses by integrating advanced statistical methods with cutting-edge data visualization techniques. It enables the identification of key features, patterns, functions, and interactions within complex gene expression datasets. The final visualization of the networks was performed using Cytoscape (version 3.10.2).

Statistical analysis

All statistical analyses were performed using R version 4.3.2. Continuous variables with a normal distribution are presented as the mean ± standard deviation (SD) and were compared using unpaired Student’s t-tests. Non-normally distributed variables are expressed as the median [interquartile range (IQR)] and were analyzed using the Mann-Whitney U test. All tests were two-tailed, and P<0.05 was considered statistically significant.

Results

Identification of cell types using OC scRNA-seq data

The scRNA-seq data from the GSE184880 dataset were processed through standardization and annotation, yielding a comprehensive dataset of 59,324 cells from seven OC and five non-malignant ovarian control samples. Using the uniform manifold approximation and projection (UMAP) clustering algorithm, all cells were grouped into 10 distinct populations, which included B cells, macrophages, epithelial cells, T/NK cells, plasma cells, proliferating T cells, endothelial cells, fibroblast, myfibroblast, and ovarian stromal cells (Figure 1A,1B). The dominant cell types differed between normal and tumor tissues (Figure 1C): fibroblasts were most abundant in normal ovarian tissue, whereas T/NK cells prevailed in tumors. OC showed a substantial increase in macrophages and a sharp decline in fibroblast proportion relative to normal tissue (Figure 1D). We further investigated intercellular communication patterns and found that interactions in normal ovarian tissue were evenly distributed among cell populations (Figure 1E). In OC samples, by contrast, the communication network was dominated by T/NK cells and macrophages—especially SPP1-positive macrophages, which exhibited markedly intensified crosstalk with epithelial cells, such as through SPP1-CD44 ligand-receptor pairs, and with other cell types (Figure 1E). Ligand-receptor analysis indicated that signaling in normal tissue primarily supported tissue homeostasis (Figure 1F), whereas OC tissue was enriched for immune and inflammatory pathways (Figure 1G), with SPP1-positive macrophages acting as central mediators of these pro-tumor signaling cascades. Collectively, these results indicate that OC substantially remodels the TME and positions macrophages as key regulators of intercellular communication.

Figure 1 Cellular and macrophage subtype landscape in OC. (A) UMAP visualization showing distinct single-cell clusters in normal ovarian and OC tissues. (B) Expression patterns of representative marker genes among major ovarian cell types. (C) Proportional distribution of major cell populations within normal and tumor microenvironments. (D) Comparative analysis of cell-type compositions between normal and tumor samples, with asterisks indicating statistically significant differences. *, P<0.05; **, P<0.01. (E) Intercellular interaction networks in OC tissue. Left: number of interactions; right: interaction strength centered on T/NK cells and macrophages. (F) Ligand-receptor interactions of macrophages with other cell types in OC tissue. Point size indicates significance; color reflects expression correlation. (G) Ligand-receptor interactions of other cell types with macrophages in OC tissue. Point size indicates significance; color reflects expression correlation. OC, ovarian cancer; UMAP, uniform manifold approximation and projection.

Identification of Scissor+ macrophages in OC

Subsequently, we analyzed TAMs and categorized them into M1, M2, and proliferating subsets based on established macrophage markers (Figure 2A,2B). In tumor tissue, the proportion of M2 macrophages was markedly increased, whereas M1 macrophages were decreased (Figure 2C). To determine which macrophage types and subpopulations are linked to poor prognosis, we integrated scRNA-seq datasets with survival data from the TCGA cohort and applied Scissor analysis to identify cells correlated with OC outcomes. This analysis revealed Scissor⁺ cells associated with prognosis across all macrophage subtypes (Figure 2D). Differential gene expression analysis of macrophages from normal and tumor tissues identified 437 DEGs. Separate differential analysis of Scissor⁻ and Scissor⁺ cells yielded 1,014 DEGs, and the intersection of these gene sets resulted in 94 common DEGs (Figure 2E, supplementary table 6 available at https://cdn.amegroups.cn/static/public/tcr-2025-1-2710-6.xlsx). We then performed GO and KEGG enrichment analyses on these 94 DEGs (Figure 2F,2G). KEGG analysis indicated that macrophage-related pathways were primarily immune-related, including coronavirus disease 2019 (COVID-19), the IL-17 signaling pathway, and the Toll-like receptor signaling pathway, which are involved in viral recognition, defense, and inflammatory signaling (Figure 2F). In the GO analysis, BPs were focused on viral responses, such as response to virus and defense response to virus. The CC level included immune complexes like the immunoglobulin complex, and MFs involved cytokine activity, chemokine activity, and G protein-coupled receptor binding, underscoring the central role of macrophages in immune defense, cytokine-mediated signaling, and immune complex-related responses (Figure 2G).

Figure 2 Identification of Scissor⁺ macrophages and functional characterization of DEGs in OC. (A) Characteristic gene expression profiles across M1, M2, and proliferating macrophage subsets, with dot size indicating expression proportion and color denoting relative expression level. (B) UMAP projection illustrating macrophage subtypes, including M1, M2, and proliferating macrophages. (C) Distribution patterns of macrophage subtypes across normal and OC tissues. (D) UMAP plot depicting Scissor⁺ (red), Scissor⁻ (blue), and unclassified (grey). (E) Venn diagram showing the intersection between macrophage DEGs and Scissor DEGs, identifying 94 overlapping key genes. (F) KEGG pathway enrichment analysis of the 94 genes. (G) GO enrichment analysis of the 94 genes across biological process, cellular component, and molecular function categories. BP, biological process; CC, cellular component; COVID-19, coronavirus disease 2019; DEGs, differentially expressed genes; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; MF, molecular function; OC, ovarian cancer; UMAP, uniform manifold approximation and projection.

To quantitatively assess their distribution, we calculated the proportion of Scissor⁺ cells within each macrophage subset. As shown in Figure S3, Scissor⁺ cells were present in M1, M2, and proliferating macrophages, indicating that prognosis-associated macrophages are not confined to a single TAM subtype. To further characterize subtype-specific transcriptional programs, we performed differential expression analysis comparing Scissor⁺ and Scissor⁻ cells separately within M1 and M2 macrophages, followed by GO and KEGG enrichment analyses (Figures S4,S5). Within M1 macrophages, Scissor⁺ cells showed enrichment in inflammatory and bacterial-response pathways, including IL-17 and TNF signaling, whereas antiviral-related pathways were significantly downregulated.

Within M2 macrophages, Scissor⁺ cells were enriched in stress- and hormone-response pathways, such as response to steroid hormone and integrated stress response signaling, accompanied by suppression of antiviral and mitochondrial-related processes.

Establishment of Scissor⁺ macrophage gene prognostic model for OC

To further refine prognostically relevant genes from the 94 DEGs strongly associated with OC Scissor⁺ macrophage function, we first matched these genes to the TCGA-OV cohort. Among them, 92 genes were detectable in the bulk RNA-seq dataset and were subjected to univariate Cox proportional hazards analysis. A total of 13 genes showed significant associations with OS (P<0.05; supplementary table 7 available at https://cdn.amegroups.cn/static/public/tcr-2025-1-2710-7.xlsx). To identify the most robust prognostic candidates and minimize model overfitting, we subsequently applied two complementary machine learning approaches: LASSO-Cox regression and RSF (Figure 3). The LASSO-Cox regression selected candidate genes based on the optimal penalty parameter determined by cross-validation (Figure 3A-3C). In parallel, the RSF model identified key prognostic genes according to the optimized number of trees (Figure 3D) and variable importance ranking (Figure 3E). Genes consistently retained by both algorithms were considered core prognostic genes for subsequent model construction. A Venn diagram integrating the results from both methods revealed five core prognostic genes: IL1B, ISG20, ABCC3, ZFP36, and IL2RG (Figure 3F). Visualization of the gene weights (Figure 3G; the calibration curve and the forest plots of these five genes are shown in Figures S6,S7) indicated distinct scoring ranges for each core gene, and their summed values produced a total risk score. Higher scores corresponded to progressively worse 1-, 3-, 5-, and 10-year survival, demonstrating that this five-gene signature derived from Scissor⁺ macrophages effectively stratifies OC prognosis.

Figure 3 Identification of core prognostic genes and construction of the predictive model. (A) Coefficient profiles of genes obtained from LASSO-Cox regression analysis. (B) LASSO coefficient trajectories plotted against log (λ). (C) Partial likelihood deviance curves across different λ values. (D) Error rate evaluation across the number of trees in the RSF model. (E) Ranking of gene importance based on RSF variable importance measures. (F) Intersection of genes identified by LASSO-Cox and RSF analyses, resulting in five core prognostic genes. (G) Prognostic framework of the five-gene model. Lines two through six: scoring range of each gene; the last four lines: 1-, 3-, 5-, and 10-year survival probabilities across total risk scores. LASSO, least absolute shrinkage and selection operator; RFS, random survival forest.

Validation of the prognostic value of core genes and risk models

Using these five genes and their LASSO-Cox regression coefficients, we constructed a prognostic risk score model for OC: risk scores = (0.062526244 × IL1B expression) + (−0.114213820 × ISG20 expression) + (0.068002281 × ABCC3 expression) + (0.065914917 × ZFP36 expression) + (−0.082467297 × IL2RG expression).

In this model, elevated expression of IL1B, ABCC3, and ZFP36 increases patient risk, whereas high expression of ISG20 and IL2RG decreases it. The absolute value of each coefficient indicates the corresponding gene’s contribution strength. We performed survival and ROC curve analyses using the TCGA-OV cohort (Figure 4). For each core gene, survival analysis using the median expression as the cutoff revealed significant OS differences between high- and low-expression groups: ISG20 (P<0.001), ABCC3 (P=0.02), ZFP36 (P=0.03), IL2RG (P=0.04), and IL1B (P=0.049). Patients with high ISG20 or IL2RG expression, or low ABCC3, ZFP36, or IL1B expression, exhibited relatively longer OS, aligning with the risk trends indicated by the model coefficients (Figure 4A-4E). The prognostic performance of the risk model itself was also evaluated. Survival analysis showed a significant difference in OS between high- and low-risk groups in the TCGA-OV cohort (P<0.001), with the high-risk group experiencing shorter survival (Figure 4F).

Figure 4 Prognostic evaluation of core genes and risk stratification model in OC. (A-E) Kaplan-Meier survival curves showing OS differences based on expression levels of individual core genes (ISG20, ABCC3, ZFP36, IL2RG, and IL1B) in the TCGA-OV cohort, the upper panel of each subfigure displays the Kaplan-Meier curve, and the lower panel shows the number of patients at risk over OS time. (F) Survival comparison between high- and low-risk groups defined by the five-gene prognostic signature. (G-I) Kaplan-Meier survival analyses in the independent GSE26712 (G), GSE63885 (H) and GSE140082 (I) validation cohorts, the upper panel of each subfigure displays the Kaplan-Meier curve, and the lower panel shows the number of patients at risk over OS time. OC, ovarian cancer; OS, overall survival; TCGA-OV, The Cancer Genome Atlas-ovarian serous cystadenocarcinoma.

We further validated the model using external OC cohorts (GSE26712, GSE63885, GSE140082), where significant survival differences between risk groups were observed (P=0.008, P=0.04, P=0.046; Figure 4G–4I). Collectively, the risk model effectively stratified OC patients by prognostic risk across multiple external datasets, confirming its stability and reliability.

To further evaluate the robustness of the prognostic signature, we validated the model in three independent external cohorts profiled on different microarray platforms (GPL96, GPL570, GPL14951). Kaplan-Meier analyses demonstrated consistent survival stratification across datasets (Figure 4G-4I). To assess predictive performance within each cohort separately, ROC analyses were performed independently for all validation datasets. As shown in Figure S8, the model achieved statistically significant predictive performance across cohorts, although the AUC values varied, likely reflecting differences in platform characteristics and sample composition. These results support the reproducibility of the macrophage-associated prognostic signature across heterogeneous transcriptomic platforms.

To further explore the biological basis of their protective association, we analyzed the correlation between ISG20 and IL2RG expression and established cytotoxic immune markers in the TCGA-OV cohort. As shown in Figure S9, both ISG20 and IL2RG were positively correlated with CD8A, IFNG, and GZMB expression. These results suggest that higher expression of ISG20 and IL2RG may reflect an immune-activated TME characterized by enhanced cytotoxic T cell infiltration and interferon signaling, thereby contributing to improved patient survival.

Single-cell expression distribution of core genes and immune microenvironment correlation and co-expression network analysis

Correlation analyses were conducted to clarify the expression patterns of core genes (IL1B, ISG20, ABCC3, ZFP36, IL2RG) in macrophages and their relationship with the immune microenvironment (Figure 5). The UMAP plot reveals elevated IL1B and ZFP36 expression in macrophages, while ISG20, ABCC3, and IL2RG show comparatively lower but more broadly distributed expression, illustrating the heterogeneity of these core genes at single-cell resolution (Figure 5A). Analysis of tumor-infiltrating immune cells in the TCGA-OV cohort revealed significantly greater proportions of immunosuppressive cells—M2 macrophages, regulatory T cells (Tregs), and myeloid-derived suppressor cells (MDSCs)—in high-risk patients than in low-risk patients (P<0.05). In contrast, the low-risk group contained significantly more pro-immune cells, including total CD4⁺ memory T cells, effector CD8⁺ T cells, and dendritic cells (P<0.05) (Figure 5B). The five core genes displayed distinct immune infiltration associations in OC, with pro-risk genes correlating with M2 macrophages and Tregs, and protective genes associating with CD4⁺ memory and effector CD8⁺ T cells, underscoring their influence on immune microenvironment composition (Figure 5C, supplementary table 8 available at https://cdn.amegroups.cn/static/public/tcr-2025-1-2710-8.xlsx). To explore the molecular regulation of core genes in OC and their interactions with miRNAs and TFs, we constructed co-expression networks. These analyses identified co-expression relationships between core genes and miRNAs, as well as between core genes and TFs, visualized as distinct molecular modules in the network diagram (Figure 5D,5E). Several potential regulatory interactions emerged, including has-miR-423-5p regulating IL1B to potentially suppress its expression and attenuate immunosuppression, and YY1 binding the ZFP36 promoter to regulate its transcription, offering network-level insights for future functional studies of these core genes.

Figure 5 Expression landscape and regulatory network of core genes. (A) Single-cell expression distribution of core genes visualized via UMAP. (B) Differences in immune cell infiltration levels between high- and low-risk groups. ns, not significant; ns, P>0.05; *, P<0.05; **, P<0.01; ***, P<0.001; ****, P<0.0001. (C) Heatmap of correlations between 5 core genes and immune cell infiltration profiles in ovarian cancer patients. (D,E) Co-expression networks illustrating interactions between core genes and miRNAs (D) and TFs (E). miRNAs, microRNAs; TF, transcription factor; UMAP, uniform manifold approximation and projection.

Discussion

Macrophages are key cellular constituents of the TME and represent important targets for improving therapeutic outcomes and patient survival (23-25). Understanding macrophage heterogeneity and molecular mechanisms is equally critical for developing new prognostic approaches in OC (20,26,27). This study systematically examines the molecular features of macrophages in OC and their relationship with patient prognosis. We developed a clinically applicable prognostic model that provides fresh insights into OC immune regulation and refines prognostic evaluation.

Recent studies have highlighted immune cell heterogeneity within the TME as a major research focus (28,29). TAMs, as central TME regulators, have drawn considerable interest due to their functional diversity and involvement in OC progression (30,31). However, accurately identifying prognostic TAM subsets and translating these findings into clinical prognostic tools remains challenging. Using scRNA-seq data, we applied Scissor analysis to identify Scissor⁺ macrophages linked to poor prognosis, addressing TAM heterogeneity. A five-gene prognostic model was subsequently established, offering new directions for OC prognosis evaluation and immune mechanism exploration.

The M1/M2 polarization balance of TAMs serves as a critical determinant of the tumor immune microenvironment (32-34). In OC, M2-type TAMs are generally recognized for promoting immunosuppression (35-37). Conventional bulk sequencing, however, cannot resolve functional heterogeneity among TAMs (38). Our analysis revealed a substantially greater proportion of macrophages in OC tissues than in normal ovarian tissues, with an elevated ratio of M2-type TAMs and a decreased proportion of M1-type TAMs (Figure 2C). This observation is consistent with reports that M2 TAMs foster immune evasion via TGF-β and IL-10 secretion (39,40), supporting the role of TAM polarization imbalance in OC progression.

Scissor offers a unique approach by automatically detecting cell subpopulations associated with clinical phenotypes from scRNA-seq data (41). A major advantage is its independence from unsupervised clustering, avoiding subjectivity in cluster number or resolution selection (42). Our Scissor analysis showed that Scissor⁺ cells related to poor prognosis are not restricted to a single TAM subtype, but appear in M1, M2, and proliferating TAMs (Figure 2D). This challenges the conventional view that only M2 TAMs affect prognosis, indicating that TAMs’ prognostic influence arises from function-specific Scissor⁺ subpopulations distributed across subtypes. These findings address prior limitations in TAM functional subdivision and offer precise targets for future TAM subpopulation-directed therapies.

We identified 94 key DEGs by intersecting TAM differential genes from normal versus tumor samples with those from Scissor⁺ versus Scissor⁻ cells (Figure 2B), and performed GO/KEGG enrichment analysis (Figure 2F,2G). KEGG analysis highlighted immune-related pathways such as COVID-19, IL-17 signaling, and Toll-like receptor signaling, which participate in viral recognition, defense, and inflammatory signaling in macrophages. GO analysis revealed enrichment in viral response and defense (BP), immunoglobulin complex (CC), and cytokine activity, chemokine activity, and G protein-coupled receptor binding (MF). These results clarify the roles of DEGs in macrophage immune defense and signal regulation, establishing a foundation for linking Scissor⁺ cellular traits to prognostic mechanisms and screening core genes.

Current OC prognosis still depends heavily on serum biomarkers like CA125 and HE4, along with pathological staging (43). These methods offer limited predictive accuracy and do not capture immune microenvironment status (44). Although recent multi-gene models show improved performance and robustness across cohorts such as TCGA and International Cancer Genome Consortium (ICGC) (45), most rely on generalized gene screening rather than prognostic functional cell subpopulations. From the 94 DEGs, we selected five core genes—IL1B, ISG20, ABCC3, ZFP36, and IL2RG—using LASSO-Cox regression combined with an RSF model. The model effectively stratified high- and low-risk patients in external cohorts GSE26712, GSE63885, and GSE140082 (Figure 4G-4I), outperforming certain existing models and confirming cross-cohort applicability. Besides prognostic prediction, the model reflects immune microenvironment differences: high-risk patients show enrichment of immunosuppressive cells including M2 TAMs and Tregs, whereas low-risk patients exhibit increased CD4⁺ memory T cells and effector CD8⁺ T cells (Figure 5B). This dual functionality aids both prognostic stratification and immunotherapy response prediction.

We validated core gene expression patterns at the single-cell level: IL1B and ZFP36 were highly expressed in macrophages, while ISG20, ABCC3, and IL2RG showed lower expression and broader distribution (Figure 5A). These patterns suggested IL1B and ZFP36 as potential effector molecules in Scissor⁺ macrophages, while ISG20 and IL2RG may regulate immune balance. Co-expression network analysis also identified potential regulatory links between core genes, miRNAs, and TFs (Figure 5D,5E). For example, it is hypothesized that has-miR-423-5p could regulate IL1B to reduce immunosuppression, and YY1 might transcriptionally regulate ZFP36 via promoter binding. These findings propose potential mechanisms that warrant experimental validation in future studies.

This study has several limitations. It is a retrospective computational analysis based solely on publicly available bioinformatics data, without in vitro or in vivo validation of the biological functions of Scissor⁺ macrophages and the five core genes. Experimental validation with clinical samples from our institution was not performed, and the regulatory mechanisms of these core genes in Scissor⁺ macrophages remain unclear. Additionally, confounding clinical factors such as age, pathological stage, tumor grade, and treatment regimens are important for OC prognosis. A multivariate analysis integrating these factors with the risk score could validate the signature’s independent prognostic value. However, due to inconsistent clinical annotations in the public TCGA and GEO datasets, such an analysis could not be conducted. Future studies will focus on experimental validation and the collection of uniformly annotated clinical data for comprehensive analysis, providing stronger evidence for the clinical applicability of the prognostic model.

Conclusions

In summary, our study establishes an analytical framework combining single-cell analysis of TAM heterogeneity, Scissor⁻based prognostic subpopulation identification, dual-model gene selection, and multi-cohort validation. Focusing on specific cancer-associated macrophage subpopulations associated with poor prognosis in OC, we developed a prognostic model derived from scRNA-seq data. This study enhances our understanding of macrophage-driven immune regulation in OC and could support the development of macrophage-targeted prognostic and therapeutic strategies.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2710/rc

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2710/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2710/coif). The 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. Bray F, Laversanne M, Sung H, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2024;74:229-63. [Crossref] [PubMed]
2. Webb PM, Jordan SJ. Global epidemiology of epithelial ovarian cancer. Nat Rev Clin Oncol 2024;21:389-400. [Crossref] [PubMed]
3. Kuroki L, Guntupalli SR. Treatment of epithelial ovarian cancer. BMJ 2020;371:m3773. [Crossref] [PubMed]
4. Dai W, Zhou J, Chen T. Unraveling the extracellular vesicle network: insights into ovarian cancer metastasis and chemoresistance. Mol Cancer 2024;23:201. [Crossref] [PubMed]
5. Konstantinopoulos PA, Matulonis UA. Clinical and translational advances in ovarian cancer therapy. Nat Cancer 2023;4:1239-57. [Crossref] [PubMed]
6. Cai G, Huang F, Gao Y, et al. Artificial intelligence-based models enabling accurate diagnosis of ovarian cancer using laboratory tests in China: a multicentre, retrospective cohort study. Lancet Digit Health 2024;6:e176-86. [Crossref] [PubMed]
7. Zhang R, Siu MKY, Ngan HYS, et al. Molecular Biomarkers for the Early Detection of Ovarian Cancer. Int J Mol Sci 2022;23:12041. [Crossref] [PubMed]
8. You B, Purdy C, Copeland LJ, et al. Identification of Patients With Ovarian Cancer Experiencing the Highest Benefit From Bevacizumab in the First-Line Setting on the Basis of Their Tumor-Intrinsic Chemosensitivity (KELIM): The GOG-0218 Validation Study. J Clin Oncol 2022;40:3965-74. [Crossref] [PubMed]
9. Kaaks R, Cooley V, Mukama T, et al. A Prospective Study Consortium for the Discovery and Validation of Early Detection Markers for Ovarian Cancer - Baseline Findings for CA125. Clin Cancer Res 2025;31:2441-53. [Crossref] [PubMed]
10. Ghisoni E, Morotti M, Sarivalasis A, et al. Immunotherapy for ovarian cancer: towards a tailored immunophenotype-based approach. Nat Rev Clin Oncol 2024;21:801-17. [Crossref] [PubMed]
11. Cassetta L, Pollard JW. A timeline of tumour-associated macrophage biology. Nat Rev Cancer 2023;23:238-57. [Crossref] [PubMed]
12. Xiang X, Wang J, Lu D, et al. Targeting tumor-associated macrophages to synergize tumor immunotherapy. Signal Transduct Target Ther 2021;6:75. [Crossref] [PubMed]
13. Xu J, Ding L, Mei J, et al. Dual roles and therapeutic targeting of tumor-associated macrophages in tumor microenvironments. Signal Transduct Target Ther 2025;10:268. [Crossref] [PubMed]
14. Zhang Y, Feng Z, Liu J, et al. Polarization of tumor-associated macrophages by TLR7/8 conjugated radiosensitive peptide hydrogel for overcoming tumor radioresistance. Bioact Mater 2022;16:359-71. [Crossref] [PubMed]
15. Ghisoni E, Benedetti F, Minasyan A, et al. Myeloid cell networks govern re-establishment of original immune landscapes in recurrent ovarian cancer. Cancer Cell 2025;43:1568-1586.e10. [Crossref] [PubMed]
16. Li YQ, Luo CL, Jiang JX, et al. Single-Cell Analysis Reveals Malignant Cells Reshape the Cellular Landscape and Foster an Immunosuppressive Microenvironment of Extranodal NK/T-Cell Lymphoma. Adv Sci (Weinh) 2023;10:e2303913. [Crossref] [PubMed]
17. Karakas D, Ashour A, Mokhlis HA, et al. eEF2K is a poor prognostic factor and novel molecular target in pancreatic cancer: regulating tumor growth and progression via the tumor microenvironment. Cell Death Dis 2025;16:501. [Crossref] [PubMed]
18. Liang L, Zhang Y, Li C, et al. Plasma cfDNA methylation markers for the detection and prognosis of ovarian cancer. EBioMedicine 2022;83:104222. [Crossref] [PubMed]
19. Lei H, Guo K, Shu G, et al. GJB2 as a novel prognostic biomarker associated with immune infiltration and cuproptosis in ovarian cancer. Apoptosis 2025;30:1589-613. [Crossref] [PubMed]
20. 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]
21. Yeh CY, Aguirre K, Laveroni O, et al. Mapping spatial organization and genetic cell-state regulators to target immune evasion in ovarian cancer. Nat Immunol 2024;25:1943-58. [Crossref] [PubMed]
22. Wang C, Yang M, Zhong Y, et al. Immunosuppressive JAG2(+) tumor-associated neutrophils hamper PD-1 blockade response in ovarian cancer by mediating the differentiation of effector regulatory T cells. Cancer Commun (Lond) 2025;45:747-73. [Crossref] [PubMed]
23. Mantovani A, Allavena P, Marchesi F, et al. Macrophages as tools and targets in cancer therapy. Nat Rev Drug Discov 2022;21:799-820. [Crossref] [PubMed]
24. DeNardo DG, Ruffell B. Macrophages as regulators of tumour immunity and immunotherapy. Nat Rev Immunol 2019;19:369-82. [Crossref] [PubMed]
25. Tichet M, Wullschleger S, Chryplewicz A, et al. Bispecific PD1-IL2v and anti-PD-L1 break tumor immunity resistance by enhancing stem-like tumor-reactive CD8(+) T cells and reprogramming macrophages. Immunity 2023;56:162-179.e6. [Crossref] [PubMed]
26. Elorbany S, Berlato C, Carnevalli LS, et al. Immunotherapy that improves response to chemotherapy in high-grade serous ovarian cancer. Nat Commun 2024;15:10144. [Crossref] [PubMed]
27. Zhong Y, Wang Y, Wang C, et al. Targeting mesothelin-CD24 axis repolarizes tumor-associated macrophages to potentiate PD-1 blockade therapy in high-grade serous ovarian cancer. J Immunother Cancer 2025;13:e011230. [Crossref] [PubMed]
28. Heras-Murillo I, Adán-Barrientos I, Galán M, et al. Dendritic cells as orchestrators of anticancer immunity and immunotherapy. Nat Rev Clin Oncol 2024;21:257-77. [Crossref] [PubMed]
29. van Weverwijk A, de Visser KE. Mechanisms driving the immunoregulatory function of cancer cells. Nat Rev Cancer 2023;23:193-215. [Crossref] [PubMed]
30. Heath O, Berlato C, Maniati E, et al. Chemotherapy Induces Tumor-Associated Macrophages that Aid Adaptive Immune Responses in Ovarian Cancer. Cancer Immunol Res 2021;9:665-81. [Crossref] [PubMed]
31. Lu X, Ji C, Jiang L, et al. Tumour microenvironment-based molecular profiling reveals ideal candidates for high-grade serous ovarian cancer immunotherapy. Cell Prolif 2021;54:e12979. [Crossref] [PubMed]
32. Chen S, Saeed AFUH, Liu Q, et al. Macrophages in immunoregulation and therapeutics. Signal Transduct Target Ther 2023;8:207. [Crossref] [PubMed]
33. Cheng Y, Zhu Y, Xu J, et al. PKN2 in colon cancer cells inhibits M2 phenotype polarization of tumor-associated macrophages via regulating DUSP6-Erk1/2 pathway. Mol Cancer 2018;17:13. [Crossref] [PubMed]
34. Gunalp S, Helvaci DG, Oner A, et al. TRAIL promotes the polarization of human macrophages toward a proinflammatory M1 phenotype and is associated with increased survival in cancer patients with high tumor macrophage content. Front Immunol 2023;14:1209249. [Crossref] [PubMed]
35. Neri M, Sanna E, Ferrari PA, et al. Divergent Immune-Metabolic Profiles in Endometriosis and Ovarian Cancer: A Cross-Sectional Analysis. Cancers (Basel) 2025;17:2325. [Crossref] [PubMed]
36. El-Arabey AA, Denizli M, Kanlikilicer P, et al. GATA3 as a master regulator for interactions of tumor-associated macrophages with high-grade serous ovarian carcinoma. Cell Signal 2020;68:109539. [Crossref] [PubMed]
37. Hou L, Jiang M, Li Y, et al. Targeting SPP1(+) macrophages via the SPP1-CD44 axis reveals a key mechanism of immune suppression and tumor progression in ovarian cancer. Int Immunopharmacol 2025;160:114906. [Crossref] [PubMed]
38. Yan Y, Lu J, Luo H, et al. Decoding immune low-response states in ovarian cancer: insights from single-cell and spatial transcriptomics for precision immunotherapy. Front Immunol 2025;16:1667464. [Crossref] [PubMed]
39. Wang Y, Ma C, Li X, et al. Unraveling the role of M2 TAMs in ovarian cancer dynamics: a systematic review. J Transl Med 2025;23:623. [Crossref] [PubMed]
40. Liu Z, Kuang W, Zhou Q, et al. TGF-β1 secreted by M2 phenotype macrophages enhances the stemness and migration of glioma cells via the SMAD2/3 signalling pathway. Int J Mol Med 2018;42:3395-403. [Crossref] [PubMed]
41. Sun D, Guan X, Moran AE, et al. Identifying phenotype-associated subpopulations by integrating bulk and single-cell sequencing data. Nat Biotechnol 2022;40:527-38. [Crossref] [PubMed]
42. Zhang G, Yu J, Zhang F, et al. Constructing Scissor+ risk model to predict prognosis and immunotherapy responses in PAAD by integrating bulk and single-cell RNA sequencing data. Front Pharmacol 2025;16:1646840. [Crossref] [PubMed]
43. Yu X, Zhang Z, Zou Y, et al. CT-based radiomics model for noninvasive prediction of progression-free survival in high-grade serous ovarian carcinoma: a multicenter study incorporating preoperative and postoperative clinical factors. BMC Med Imaging 2025;25:320. [Crossref] [PubMed]
44. Chen Z, Ouyang H, Sun B, et al. Utilizing explainable machine learning for progression-free survival prediction in high-grade serous ovarian cancer: insights from a prospective cohort study. Int J Surg 2025;111:3224-34. [Crossref] [PubMed]
45. Li Z, Wang J, Zhang Y, et al. Predicting the prognosis of epithelial ovarian cancer patients based on deep learning models. Front Oncol 2025;15:1592746. [Crossref] [PubMed]