The role of cellular senescence in immune-metabolic features and prognosis of ovarian cancer: an integrated analysis based on single-cell sequencing and multi-omics data

Abstract

Ovarian cancer is characterized by a high recurrence rate, and platinum resistance often contributes to poor patient outcomes. Cellular senescence can influence the tumor microenvironment, metabolic status, and immune regulation. However, the role of senescence-associated genes (SAGs) in the heterogeneity, prognosis, and treatment-related features of ovarian cancer remains to be further clarified. This study integrated data from TCGA, GEO, and single-cell RNA sequencing to systematically analyze the expression patterns, single-cell distribution, molecular subtypes, and associations of SAGs with prognosis and the immune microenvironment in ovarian cancer. Consensus clustering was used to identify senescence-related molecular subtypes, and an SAG-related prognostic risk model was constructed and validated. Associations between the risk score and platinum response, tumor mutational burden (TMB), immune cell infiltration, and immune checkpoint expression were further evaluated. SAGs were differentially expressed between ovarian cancer and normal ovarian tissues and were mainly enriched in cell cycle, DNA damage response, and immune-related pathways. Single-cell analysis showed that senescence signals were mainly distributed in cancer-associated fibroblasts, malignant cells, and selected immune cell populations. According to SAG-based consensus clustering, patients were classified into three molecular subtypes with differences in metabolic activity, DNA repair, immune microenvironment, and genomic stability, although no significant differences in OS were observed among subtypes. An eight-gene senescence-related risk model showed prognostic relevance in the TCGA cohort, and additional analyses in two GEO cohorts supported the prognostic value of the selected SAGs. The high-risk group was associated with greater TMB, platinum resistance-related features, increased immune and stromal scores, reduced tumor purity, a lower proportion of CD8⁺ T cells, and higher expression of regulatory T cells, macrophages, and immune checkpoints, suggesting a link with an immunosuppressive tumor microenvironment. The current findings suggest that senescence-associated molecular features are associated with the biological heterogeneity, prognostic stratification, platinum response, and status of the immune microenvironment in ovarian cancer. The SAG-related risk model may provide clues for prognostic assessment and future studies on treatment-response stratification in ovarian cancer, but further experimental and clinical validation is required.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s11357-026-02343-3.

Background

Ovarian cancer is one of the most lethal gynecological malignancies. Most patients are diagnosed at an advanced stage with extensive metastasis, and the overall 5-year survival rate remains only 25–35%, imposing a substantial disease burden [1–3]. Although surgery followed by platinum-based chemotherapy remains the standard first-line treatment, most patients eventually develop resistance and experience poor outcomes [4]. A deeper understanding of the molecular mechanisms underlying ovarian cancer progression and treatment resistance is therefore needed to inform new therapeutic strategies.

Cellular senescence is a stable state of cell-cycle arrest that has traditionally been regarded as a tumor-suppressive mechanism [5]. During tumor progression, however, senescent cells can also remodel the tumor microenvironment (TME) through persistent secretion of pro-inflammatory factors via the senescence-associated secretory phenotype (SASP), thereby contributing to immune evasion and therapeutic resistance [6]. Dysregulated senescence-associated genes (SAGs) have been linked to chemotherapy resistance in ovarian cancer [7]. The ovarian cancer TME is enriched with cancer-associated fibroblasts (CAFs) and immunosuppressive cell populations, both of which may be closely related to senescence signaling [8, 9]. However, the cellular distribution of senescence signals in ovarian cancer and their associations with the TME, molecular heterogeneity, and clinical outcomes remain to be systematically clarified.

Therefore, this study integrated TCGA, GEO, and scRNA-seq datasets to systematically investigate the expression patterns and biological and clinical relevance of SAGs in ovarian cancer through molecular subtyping and prognostic modeling. We further explored their associations with the TME, platinum resistance, and patient survival to provide clues for risk stratification and future mechanistic studies in ovarian cancer.

Methods

Data acquisition and preprocessing

The present analysis utilized publicly accessible data. scRNA-seq data of ovarian cancer originated from the GEO dataset GSE154600 (n = 5) (https://www.ncbi.nlm.nih.gov/geo). A gene set related to cellular senescence (n = 278) was sourced from the CellAge database (https://genomics.senescence.info/cells/). Bulk RNA sequencing data for patients (n = 420), corresponding mutation profiles, and normal ovarian tissue specimens (n = 88) were derived from TCGA and GTEx projects via the UCSC Xena platform (https://gtexportal.org/home/). To minimize batch effects, the UCSC Toil RNA-seq pipeline was employed for processing all RNA sequencing data uniformly. External validation was carried out via the GEO datasets GSE63885 (n = 101) and GSE14764 (n = 80) (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi).

Differential expression analysis, TME assessment, and functional enrichment

Differentially expressed SAGs between tumor and normal tissues were detected with a threshold of |log₂-transformed fold change|> 1 utilizing the limma package (v3.52.4). Copy number variations (CNVs) and somatic mutation spectra for these genes were visually presented via the packages ggplot2 (v3.5.1) and maftools. Protein–protein interaction (PPI) networks were established using the STRING (https://string-db.org/) and visualized with Cytoscape (v3.7.1).

Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed using the clusterProfiler package (v4.9.0.002). StromalScore, ImmuneScore, and ESTIMATE scores were computed utilizing the ESTIMATE package (v1.0.13) to profile the composition of TME. Expression differences for immune checkpoint and human leukocyte antigen-related genes were analyzed with limma. Gene Set Enrichment Analysis (GSEA) was performed using clusterProfiler. Enrichment significance was assessed by normalized enrichment score (NES), nominal p value, and false discovery rate (FDR, reported as q value). Pathways with |NES|> 1, nominal p < 0.05, and q value (FDR) < 0.25 were considered significantly enriched.

scRNA-seq analysis

The Seurat package (v3.1.4) was used to process scRNA-seq data. Quality control for the GSE154600 dataset involved filtering low-quality cells and creating a gene expression matrix. Cells with fewer than 200 or more than 5000 detected genes were excluded, as well as those with mitochondrial gene content exceeding 10%. Following data normalization, uniform manifold approximation and projection (UMAP) was conducted for dimensionality reduction and unsupervised clustering at a resolution of 0.5. Cell clusters were annotated based on established marker genes. The AUCell package (v1.8.0) was used to evaluate the activity levels of the SAG set in each cell. Using this set (n = 278), cells were stratified into two categories based on their AUCell scores: those exhibiting active senescence signals and those with silent signals. The distribution of senescence signals across annotated cell types was visually illustrated.

SAG-based molecular subtyping

Leveraging the expression matrix of SAGs from the TCGA cohort, the ConsensusClusterPlus package (v1.60.0) was used for unsupervised consensus clustering. Cluster stability was evaluated via resampling iterations. The optimal cluster number was determined by analyzing cumulative distribution function (CDF) curves and consensus matrices. Gene Set Variation Analysis (GSVA) was performed with the GSVA package (v1.44.5), and pairwise GSEA was conducted to characterize pathway differences among molecular subtypes. ImmuneScore, StromalScore, ESTIMATE score, tumor purity, aneuploidy score, and fraction of genome altered were compared across subtypes to assess the features of the TME and genomic stability comprehensively. ESTIMATE-related metrics were calculated using the ESTIMATE package (v1.10.13). For comparisons among more than two groups, the Kruskal–Wallis test was applied, followed by pairwise Wilcoxon rank-sum tests with Benjamini–Hochberg correction.

Development of a senescence-associated prognostic model

In the TCGA cohort, univariable Cox regression analysis was first performed to screen for candidate genes associated with overall survival (OS) (p < 0.05). Then, the stepwise Akaike Information Criterion (AIC; stepAIC) approach from the MASS package (v7.3–60) was applied for variable selection to establish a prognostic risk score model, minimize overfitting, and obtain relatively concise gene combinations. During model establishment, the correlations among candidate variables were evaluated to mitigate the influence of multicollinearity on stability. Each patient’s risk score was computed using the following equation based on the standardized expression level (Expi) of candidate genes and their corresponding regression coefficients (Coei):

$$\text{Risk}=\sum _{i=1}^n(\text{Expi}\times \text{Coei})$$

Patients were dichotomized into high- and low-risk groups per the median risk score. Kaplan–Meier (K–M) survival analysis was performed using the survival package (v3.6-4) and visualized using the survminer package (v0.4.9) to evaluate the prognostic capacity of this gene signature. Time-dependent receiver operating characteristic (ROC) curves and the area under the curve (AUC) were employed for quantifying predictive accuracy. Univariable and multivariable Cox regression analyses were conducted to assess the independent prognostic value of senescence-associated risk scores after adjusting for clinical parameters. The identified independent prognostic factors were then applied to establish a nomogram using the rms package (v6.7-0). Its predictive performance was evaluated using ROC curves and calibration plots, and potential clinical net benefit was assessed by decision curve analysis (DCA) with the ggDCA package (v1.2).

Statistical analysis

Statistical analyses and data visualization were performed using R software (v4.1.0). Subgroup comparisons were conducted using the Wilcoxon rank-sum test or Kruskal–Wallis test, as appropriate. Pearson correlation analysis was performed to assess associations between variables. For K–M survival analysis, the log-rank test was applied, with p < 0.05 indicating statistical significance. Hazard ratios (HRs) and 95% confidence intervals (CIs) were reported for Cox regression analyses. Statistical significance was denoted as *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001.

Results

SAGs are broadly dysregulated in ovarian cancer

Based on the TCGA-GTEx dataset and CellAge-derived SAGs, 135 differentially expressed SAGs (68 upregulated and 67 downregulated) were identified (Fig. 1A). Their expression patterns distinguished ovarian cancer tissues from normal ovarian tissues (Fig. 1B). PPI network analysis showed that GAPDH, SRC, CDKN2A, CCND1, CDKN1A, and FOS occupied central positions in the network (Fig. 1C). GO analysis showed that differentially expressed SAGs were mainly enriched in biological processes related to cell-cycle regulation, DNA damage response, p53 signaling, and cellular senescence (Fig. 1D). KEGG analysis showed enrichment in pathways including cellular senescence, p53 signaling, cell cycle, carbon metabolism, endocrine resistance, and AGE-RAGE signaling (Fig. 1E).

Fig. 1.

Dysregulation of SAGs in ovarian cancer. A–B Volcano plot (A) and heatmap (B) of SADEGs in ovarian cancer and normal tissues. C PPI network of DEGs. D–E Bubble plot for GO (BP) analysis (D) and bar chart for KEGG analysis (E) of DEGs. F Waterfall plot of the top 20 most frequently mutated genes among DEGs. G The top 20 genes with the highest CNV alteration frequency among DEGs

Somatic mutation analysis of SAGs revealed mutations in 142 samples (32.57%). TP53, CSMD3, MUC16, TTN, and PIK3CA exhibited the highest mutation frequencies (Fig. 1F), with missense mutations being the predominant type. CNV analysis indicated notable amplifications or deletions for most DEGs. GAPDH, CCND1, and CDKN2A were frequently amplified. FOS, SRC, ATM, and SIRT1 were predominantly affected by deletions (Fig. 1G). NOTCH3 displayed a high-mutation frequency with diverse mutation types and occupied a relatively central position within the PPI network. These results suggested that genomic alterations may contribute to the further dysregulation of senescence-associated pathways in ovarian cancer.

Single-cell analysis reveals activities of cell-type-specific SAGs in ovarian cancer

Distinct cell types may differentially regulate senescence signaling during cancer initiation, progression, and therapeutic response. Analysis of ovarian cancer scRNA-seq data initially delineated 30 cellular subclusters (Supplementary Fig. S1A). These subclusters were subsequently annotated into 11 major cell types, including fibroblasts, malignant cells, macrophages, T cells, B cells, and plasma cells, based on specific marker genes (Fig. 2A). Using the AUCell algorithm, a threshold score of 0.11 was applied to distinguish cells with active senescence signals from those with silent signals, identifying 3083 active cells (Fig. 2B). These active cells were distributed across multiple cell types, with higher frequencies observed in fibroblasts and certain immune cell populations (Supplementary Fig. S1B). Violin plot analysis further showed differences in senescence AUCell scores across cell types, with fibroblasts exhibiting relatively high senescence signal activity, followed by malignant cells and specific immune subsets such as monocytes, macrophages, and T cells (Fig. 2C). Given the limited sample size of the single-cell dataset, these results primarily illustrate relative distribution patterns of senescence signals across cell types rather than providing precise quantitative comparisons.

Fig. 2.

Distinct single-cell distribution patterns of SAG signals in ovarian cancer. A UMAP dimensionality reduction and clustering of ovarian cancer single cells, with major cell types annotated based on specific marker genes. B UMAP visualization showing cells classified as active or silent according to senescence-related activity. C Violin plot showing the distribution of senescence AUCell scores across different cell types

SAG-based molecular subtyping reveals biological heterogeneity in ovarian cancer

To evaluate whether the expression patterns of SAGs reflected molecular heterogeneity among ovarian cancer patients, consensus clustering analysis was performed in the TCGA cohort. Evaluation of CDF curves and cluster consensus indices indicated optimal clustering stability at k = 3 (Supplementary Fig. S2A–B). Accordingly, patients were stratified into three molecular subtypes: clusters A, B, and C. Principal component analysis showed relatively separated transcriptomic distributions among the three subtypes (Supplementary Fig. S2C). The expression heatmap of SAGs showed distinct expression patterns across clusters A, B, and C, together with different distributions of clinical annotations, including stage, grade, age, and OS status (Fig. 3A). Further K–M survival analysis showed no significant differences in OS among the three molecular subtypes (Supplementary Fig. S3).

Fig. 3.

SAG-defined molecular subtypes and biological states in ovarian cancer. A Heatmap of SAG expression across molecular subtypes, with clinical annotations including stage, grade, age, and OS status. B GSVA heatmap showing differences in pathway activity among molecular subtypes. C–E Enrichment results of GSEA for clusters A, B, and C, presented as NES. Corresponding FDRs are provided in Supplementary Table S1

To further characterize functional differences among the subtypes, GSVA and GSEA were performed. The GSVA heatmap showed differences among the three subtypes in metabolic, DNA repair, cell-cycle, and immune-related pathways (Fig. 3B). Cluster A showed higher activity in immune- and microenvironment-related pathways, including integrin RhoG/Rac signaling, integrin talin/vinculin signaling, MHC-I antigen presentation, interleukin (IL)−2 family/JAK–STAT signaling, and cytokine JAK–STAT signaling, suggesting a more active immune-stromal interaction feature. Cluster B was mainly characterized by activation of translation initiation, mitochondrial electron transport, and oxidative phosphorylation-related pathways, suggesting stronger metabolic and mitochondrial functional features. Cluster C showed relatively lower activity in immune-related pathways, suggesting a microenvironmental feature with lower immune infiltration.

Pairwise GSEA comparisons further supported these results (Figs. 3C–E). Patients in Cluster A showed enrichment mainly in pathways related to immune response, antigen presentation, and cell–microenvironment interactions. Those in cluster B tended to show enrichment in pathways related to DNA replication and repair, drug metabolism, and oxidative phosphorylation. Cluster C exhibited more prominent enrichment of tumor-related pathways, including Wnt signaling and interaction between extracellular matrix (ECM) and receptor, together with metabolic processes such as oxidative phosphorylation. Detailed results of GSEA, including NES and FDRs, are provided in Supplementary Table S1.

Further analysis of the TME showed significant differences in ImmuneScore, StromalScore, and ESTIMATEScore among the three molecular subtypes (Figs. 4A–C). ImmuneScore reflects the relative degree of immune cell infiltration, StromalScore reflects stromal component abundance, and ESTIMATEScore integrates both immune and stromal signals to estimate the overall non-tumor component within the TME. For all three metrics, higher scores generally indicate greater immune and/or stromal content and are usually associated with lower tumor purity. ImmuneScore was higher in cluster A and cluster B than in cluster C, with no significant difference between cluster A and cluster B (Fig. 4A). StromalScore and ESTIMATEScore were highest in cluster A, followed by cluster B and cluster C (Figs. 4B–C). Consistently, tumor purity showed the opposite pattern, with the highest level in cluster C and the lowest in cluster A (Fig. 4D). Cluster C also showed a higher aneuploidy score than cluster A and cluster B, whereas no significant difference was observed between cluster A and cluster B (Fig. 4E). Fraction genome altered was highest in cluster C, intermediate in cluster B, and lowest in cluster A (Fig. 4F).

Fig. 4.

Differences in the characteristics of TME and genomic instability among SAG-based molecular subtypes. A–C Comparisons of ImmuneScore, StromalScore, and ESTIMATEScore among molecular subtypes. D Comparison of tumor purity among molecular subtypes. E–F Comparisons of aneuploidy score and fraction genome altered among molecular subtypes, reflecting chromosomal and genomic instability. Statistical significance was indicated as follows: ns, p ≥ 0.05; *p < 0.05; **p < 0.01; ***p < 0.001; and ****p < 0.0001

Taken together, SAG-based molecular subtyping revealed biological heterogeneity across multiple levels, including expression patterns, functional pathways, features of the TME, and genomic stability. Cluster A was characterized by immune- and microenvironment-active features, cluster B by metabolic reprogramming and features related to DNA damage response, and cluster C by activation of oncogenic signaling, tumor-cell dominance, and genomic instability. However, the three subtypes did not show significant stratification of OS, suggesting that this subtyping framework may be more suitable for describing senescence-related biological states than for independent prognostic classification.

An SAG-based prognostic risk model is associated with survival stratification in ovarian cancer

Although SAG-based molecular subtyping revealed marked biological heterogeneity, its ability to stratify OS was limited. Therefore, we further evaluated the prognostic value of individual SAGs and constructed a quantitative SAG-based prognostic risk model. Univariable Cox regression analysis was conducted in the TCGA cohort, detecting 20 candidate SAGs significantly associated with OS (p < 0.05). The stepAIC approach was applied to determine eight key genes: AAK1, DDB2, ING2, MATK, PMVK, SNAI1, SOX2, and WWP1. Their regression coefficients were used to define the final risk score: (0.2185) * AAK1 + (− 0.3594) * DDB2 + (− 0.3844) * ING2 + (0.2347) * MATK + (− 0.2463) * PMVK + (0.1675) * SNAI1 + (− 0.1070) * SOX2 + (0.3901) * WWP1. Using the median score, TCGA patients were stratified into high (n = 177) and low (n = 176) risk categories. Multiple validations of the model revealed that the low-risk group experienced notably superior OS (median 56.6 vs. 36.7 months, p < 0.001, Fig. 5A). Time-dependent ROC analysis confirmed the predictive accuracy of the model for 1-, 3-, and 5-year survival (0.672, 0.674, 0.711, respectively, Fig. 5B). The distribution of risk scores and survival time also indicated longer OS for low-risk patients (Fig. 5C).

Fig. 5.

An SAG-based prognostic risk model for survival stratification in ovarian cancer. A–C TCGA cohort: OS (A), time-dependent ROC (B), risk score distribution and survival status (C) for high- and low-risk groups. D–F Corresponding analyses in the GSE14764 cohort. G–I Corresponding analyses in the GSE63885 cohort

Additional analyses in two independent GEO cohorts (GSE14764 and GSE63885) supported the prognostic relevance of the selected SAGs. In both cohorts, low-risk patients showed better OS than high-risk patients (GSE14764: median OS, not reached vs. 46 months, p = 0.003, Fig. 5D; GSE63885: median OS, 42.9 vs. 27.4 months, p = 0.018, Fig. 5G). Time-dependent ROC curves demonstrated stable predictive performance in these external sets (Fig. 5E, H). Risk curves and survival status distributions consistently showed more death events in the high-risk groups (Fig. 5F, I). Collectively, these analyses suggest that the selected SAGs retain prognostic relevance across independent datasets from different sequencing platforms.

Integration of the SAG-based risk model with clinical variables improves prognostic prediction

Cox regression was implemented to evaluate the independent prognostic value of the senescence-associated risk score alongside clinicopathological variables, including race, age, tumor stage, grade, and presence of new cancerous lesions. Univariable and multivariable Cox regression analyses identified age, race, presence of new cancerous lesions, and the risk score as significant prognostic factors (Fig. 6A–B). Notably, the senescence-associated risk score remained an independent predictor of OS in the multivariable model. A nomogram incorporating these factors was created to estimate 1-, 3-, and 5-year OS (Fig. 6C). Patients were stratified into high- and low-risk groups according to the overall nomogram scores. The results showed notably different OS (median 67.6 vs. 36.3 months, p < 0.001, Fig. 6D). The predictive performance of the model was evaluated, yielding AUCs of 0.699, 0.720, and 0.799 for 1-, 3-, and 5-year OS, respectively (Fig. 6E). Calibration curves indicated good agreement between predicted and observed outcomes (Fig. 6F). DCA demonstrated positive net benefit across a wide range of threshold probabilities (Fig. 6G). These results suggest that the combined model may have potential value for prognostic stratification.

Fig. 6.

Prognostic prediction improved by integrating the SAG-based risk model with clinical variables. A Univariable Cox regression analysis of clinical variables and senescence-associated risk scores. B Multivariable Cox regression analysis to identify independent prognostic factors. HRs with 95% CIs are shown in panels A and B. C Nomogram based on senescence-associated risk scores and clinical variables to predict 1-, 3-, and 5-year OS. D K–M curves for nomogram-stratified high- and low-risk individuals. E Time-dependent ROC curves of the nomogram for predicting 1-, 3-, and 5-year OS. F Calibration curve evaluating prediction-observation agreement. G DCA for evaluating clinical net benefit

The SAG-based risk model is associated with treatment response, mutational features, and the immune microenvironment

The senescence-associated risk score was significantly higher in platinum-resistant individuals than in platinum-sensitive individuals (p = 0.021, Fig. 7A). This suggested a potential link between this signature and chemotherapy response. High-risk individuals exhibited a significantly elevated tumor mutational burden and higher mutation frequencies in genes such as TP53 and BRCA1 (Figs. 7B–C). Mutation frequencies for certain immune checkpoint-related genes (e.g., PDCD1, CTLA4) were also increased in this group. GSEA revealed marked enrichment in the high-risk group for pathways related to cellular senescence, inflammatory response, and immune modulation, including VEGF–RAGE signaling, cytokine–cytokine receptor interaction, and complement/coagulation cascades (Figs. 7D–E). Detailed results of GSEA, including NES and FDRs, are provided in Supplementary Table S2.

Fig. 7.

Association of the SAG-based risk model with treatment response, mutational features, and the immune microenvironment. A Platinum-chemosensitivity differences between high- and low-risk groups. B–C Somatic mutation profiles across risk groups. D–E Results of GSEA based on risk groups. Representative pathways are shown according to their NES, and detailed FDRs are provided in Supplementary Table S2. F Comparison of ImmuneScore, StromalScore, ESTIMATE score, and tumor purity between risk groups. G Infiltration levels of immune cell subsets across risk groups. H Expression differences of key immune checkpoints between risk groups

TME analysis displayed that high-risk individuals had notably elevated ImmuneScore (p = 0.00014), StromalScore (p = 8.8 × 10⁻⁸), and ESTIMATE score, coupled with significantly lower tumor purity (p = 4 × 10⁻⁷) (Fig. 7F). In contrast, low-risk individuals exhibited lower immune/stromal scores and higher tumor purity. Immune infiltration analysis suggested increased proportions of regulatory T cells (Tregs) and macrophages, alongside a relative decrease in CD8⁺ T cells, within the high-risk group (Fig. 7G). Furthermore, expression levels of multiple immune checkpoints, such as PDCD1 and CTLA4, were markedly upregulated in high-risk individuals (Fig. 7H).

Discussion

Recent evidence indicates that cellular senescence has complex roles in tumor evolution, immune regulation, and therapeutic resistance [10–15]. In this study, CellAge-derived SAGs were analyzed in ovarian cancer from a multi-omics perspective, focusing on their expression patterns and associations with molecular subtypes, prognosis, treatment response, and the tumor immune microenvironment. The findings suggest that senescence-associated molecular features are associated with multiple biological alterations in ovarian cancer. Notably, these SAGs were curated from human cell gene perturbation experiments as regulators that promote or inhibit cellular senescence, and previous studies have shown their overlap with tumor-suppressor genes and oncogenes. Thus, dysregulated expression of SAGs in ovarian cancer may primarily reflect tumor-local senescence-like programs driven by proliferative stress, DNA damage, genomic instability, and microenvironmental remodeling, rather than systemic aging-related senescence [16].

We first compared the expression profiles of SAGs between ovarian cancer and normal ovarian tissues and found that senescence-associated molecular changes could distinguish tumor from normal samples. Differentially expressed SAGs were mainly enriched in canonical senescence-related pathways, including cell cycle, DNA damage response, and p53 signaling, consistent with their biological nature and with previous findings on the development and platinum resistance of ovarian cancer [17–20]. PPI and CNV analyses further suggested that some key nodes may be influenced by genomic structural alterations and associated with the proliferation control and stress adaptation of tumor cells. Beyond canonical senescence pathways, the enrichment of carbon metabolism, glycolysis/gluconeogenesis, AGE-RAGE signaling, and endocrine resistance pathways suggested that dysregulation of SAGs may also be related to metabolic adaptation and pro-survival signaling. Given that cellular senescence can involve changes in glycolysis, mitochondrial function, redox balance, and SASP-related biosynthetic demands [21], and that the endocrine resistance pathway includes pro-survival modules such as PI3K-Akt, MAPK/ERK, EGFR, ESR-related signaling, and cell-cycle regulation [22], these findings suggest that differentially expressed SAGs may contribute to adaptive proliferation, survival signaling, and microenvironmental regulation under stress conditions.

Further single-cell analysis showed that, in addition to tumor cells, senescence signals were mainly distributed in CAFs and selected populations of immune cells. CAFs are a major component of the TME in ovarian cancer, are increased in tumor tissues, and can promote tumor evolution through ECM remodeling, metabolic regulation, and therapy resistance [8, 9, 23, 24]. In the TME of ovarian cancer, CAF-derived IL-8 can activate adipocytes to supply fatty acids, thereby supporting tumor growth and metastasis [25]. Senescent CAFs may also modulate local immune status and promote immune evasion by secreting SASP factors such as IL-6, IL-8, and transforming growth factor-β [26]. At the cellular level, these findings suggest that senescence-related programs may reflect tumor-intrinsic changes and participate in multicellular interactions between tumor cells and the microenvironment.

After observing abnormal distribution of SAGs across ovarian cancer tissues and distinct cell types, we further examined whether these senescence-associated molecular changes reflected biological heterogeneity among patients. SAG-based molecular subtyping identified three subtypes with differences in metabolic activity, DNA damage response, immune microenvironment, and genomic stability. Cluster A showed activation of immune- and stromal-related pathways, including integrin signaling, antigen presentation, cytokine/JAK-STAT signaling, and immune receptor signaling, together with higher ImmuneScore and StromalScore. This suggests that Cluster A may not simply represent an immune-activated state, but rather an SASP-related immune-stromal reactive state. SASP factors can contribute to ECM remodeling, sustained inflammatory signaling, recruitment of immune cells, and functional immune regulation; the ECM of ovarian cancer can also affect tumor-associated macrophages and T-cell function [13, 27, 28]. Cluster B exhibited enrichment mainly in DNA repair, drug metabolism, oxidative phosphorylation, transport of mitochondrial electrons, and translation-related pathways, suggesting a potential association with DNA damage response and mitochondrial metabolic adaptation. Because platinum-based therapy acts mainly by inducing DNA damage, tumor cells may adapt to therapeutic stress by enhancing DNA repair, maintaining mitochondrial OXPHOS, regulating redox balance, and modulating drug metabolism. OXPHOS has also been implicated in the progression and chemoresistance of ovarian cancer, and platinum induced mitochondrial OXPHOS can promote the enrichment of cancer stem cells [29, 30]. In contrast, cluster C showed lower immune infiltration, higher tumor purity, increased aneuploidy score and fraction genome altered, and activation of tumor-related pathways such as Wnt signaling. Chromosomal instability and aneuploidy can influence tumor evolution and are associated with an immunosuppressive microenvironment, while Wnt/β-catenin signaling is linked to reduced immune cell infiltration and poor response to immune checkpoint blockade [31, 32].

Although the three SAG-based subtypes showed distinct molecular, metabolic, and microenvironmental features, no significant differences in OS were observed; pairwise survival comparisons among the three groups also showed no significant differences (Supplementary Fig. S3). Notably, clusters A and B showed higher immune- and stromal-related features, but these did not translate into better survival, suggesting that increased immune infiltration does not necessarily indicate effective antitumor immunity. In cluster A, which showed a more active immune-stromal reactive phenotype, higher immune infiltration may instead reflect accumulation of immunosuppressive cells and functional exhaustion of effector T cells. This is consistent with reports that ovarian cancer is often “immunologically hot” but responds poorly to immunotherapy [33–36]. Cluster C, characterized by higher genomic instability and Wnt/β-catenin-related immune exclusion, may reflect a tumor-intrinsic exclusion mechanism underlying its lower immune infiltration. Sustained activation of the Wnt/β-catenin pathway has been associated with poor prognosis in ovarian cancer and can promote immunotherapy resistance by suppressing immune cell infiltration [32, 37, 38]. Overall, different SAG-based subtypes may represent distinct senescence-related evolutionary routes and immune microenvironment states in ovarian cancer, but these routes may ultimately converge toward similar aggressive clinical phenotypes. Thus, this subtyping framework is better suited for revealing biological heterogeneity and generating hypotheses for future stratification studies than for serving as an independent prognostic classification system.

Based on this consideration, the relationship between SAGs and patient survival was further assessed using a continuous risk-score approach, and an eight-gene senescence-related prognostic model was constructed, comprising AAK1, DDB2, ING2, MATK, and other genes. The predictive performance of this model was comparable to previously reported prognostic models of ovarian cancer based on immune infiltration-related genes, metabolism-related genes, or broad transcriptomic screening [16, 39, 40]. After integrating the risk score with clinicopathological variables such as age, race, and tumor stage, the combined model showed improved predictive performance compared with the molecular risk score alone. In addition, given that BRCA1/2 represents an important DNA repair-related molecular feature in ovarian cancer, the mutation status of BRCA1/2 was incorporated into the multivariable Cox regression model. The senescence-related risk score remained independently associated with OS (Supplementary Fig. S4). Compared with previous models based mainly on single molecular features, the present study combined multicohort validation, clinical-variable integration, and adjustment of BRCA1/2 status, providing a more comprehensive evaluation and improving the stability and clinical interpretability of prognostic stratification.

Functionally, the model genes were mainly involved in DNA damage response and repair, cell-cycle regulation, transduction of signals, and ubiquitination. Reduced expression of DDB2 and ING2 usually indicates impaired recognition of DNA damage and cell-cycle control and is associated with poor prognosis [41–43]. Dysregulated AAK1, MATK, and WWP1 may participate in the stress adaptation and survival regulation of tumor cells by affecting receptor endocytosis, transduction of signals, and ubiquitin-mediated degradation [44–46]. SOX2 is commonly associated with stemness and invasiveness [47]; however, in this study, it acted as a protective factor, suggesting that its role may depend on the biological context and mainly reflects its statistical contribution within the integrated model. Regarding treatment response and the immune microenvironment, high-risk patients showed a tumor immune state characterized by immunosuppression and exhaustion. Upregulation of immune checkpoint genes such as PDCD1 (PD-1) and CTLA4 suggested a possible adaptive immune regulatory response under selective pressure [35], but this did not translate into effective antitumor immunity. Tregs and tumor-associated macrophages can highly express ligands for PD-1 and CTLA-4, potentially weakening the efficacy of immune checkpoint blockade [36]. Thus, the high-risk group may reflect a senescence-related immunosuppressive state rather than a simply immune-activated tumor phenotype. This state was associated with poor prognosis and platinum resistance, consistent with the higher risk scores observed in platinum-resistant patients [48, 49].

The current study has several limitations. First, it mainly relied on public multi-omics datasets, and correlation-based analyses alone cannot establish causal roles for key SAGs in the progression, immune regulation, metabolic reprogramming, or treatment resistance of ovarian cancer. Second, the effects of tumor stress-induced senescence and systemic age-related cellular senescence could not be strictly distinguished. Third, the scRNA-seq dataset had a limited sample size, and the related findings require confirmation in larger single-cell cohorts and spatial omics studies. Fourth, differences across public datasets in population composition, sequencing platforms, treatment background, and completeness of clinical annotations may introduce residual batch effects. In addition, incomplete HRD scores and detailed population information in some cohorts may lead to potential confounding from prior treatment exposure. Therefore, findings related to precision therapy or immunotherapy stratification should be regarded as hypothesis-generating clues rather than direct evidence for clinical decision-making. Future in vitro and in vivo functional studies are needed to further validate the underlying mechanisms, as well as the robustness and clinical applicability of the model.

Conclusion

Focusing on senescence-associated molecular features, this study integrated analyses of ovarian cancer biological heterogeneity and its association with clinical outcomes. We characterized senescence-related signals at the bulk transcriptomic and single-cell levels and identified three SAG-based molecular subtypes. These subtypes differed in metabolic, immune microenvironmental, and genomic features; although they did not show significant OS separation, they may reflect distinct senescence-related biological states in ovarian cancer. We further developed an eight-gene SAG-related risk score and integrated it with clinical variables to establish a combined predictive model with improved prognostic stratification ability. Overall, these findings suggest that senescence-associated features are linked to the remodeling of the TME, treatment-related characteristics, and prognostic stratification in ovarian cancer, providing hypothesis-generating clues for future mechanistic validation and treatment-response stratification studies.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AIC

Akaike Information Criterion

AUC

Area under the curve

CAF

Cancer-associated fibroblasts

CDF

Cumulative distribution function

CNVs

Copy number variations

Coef_i

Corresponding regression coefficients

Exp_i

Expression level

DCA

Decision curve analysis

DEGs

Differentially expressed genes

GEO

Gene Expression Omnibus

GO

Gene Ontology

GSEA

Gene Set Enrichment Analysis

KEGG

Kyoto Encyclopedia of Genes and Genomes

K-M

Kaplan–Meier

OS

Overall survival

PPI

Protein-protein interaction

ROC

Receiver operating characteristic

SADEGs

Senescence-associated differentially expressed genes

SAGs

Senescence-associated genes

SASP

Senescence-associated secretory phenotype

scRNA-seq

Single-cell RNA sequencing

TCGA

The Cancer Genome Atlas

TME

Tumor microenvironment

UMAP

Uniform manifold approximation and projection

Author contribution

The study was designed by YL, YHZ, and LPW. SYL, MYB, and YL were responsible for drafting the initial manuscript and verifying the underlying data. Statistical analyses were conducted by WBW and HFW. LPW and ZQZ contributed to data interpretation. All authors reviewed and approved the final version of the manuscript.

Funding

This work was supported by the National Natural Science Foundation of China (No. 82474232); the Basic Research Scheme of Shenzhen Science and Technology Innovation Commission (JCYJ20240813152305008, JCYJ20230807094815031); the Scientific Research Project of Shenzhen Association of Chinese Medicine (No. 2024077 F); and the 2024 Guangzhou University of Chinese Medicine “Jie-bang Gua-shuai” Graduate Innovation Capacity Enhancement Program (A3-0317-24-429-015).

Data availability

This study used publicly available data. scRNA-seq data of ovarian cancer originated from the GEO dataset GSE154600, Bulk RNA sequencing data were derived from TCGA and GTEx projects, and external validation was carried out via the GEO datasets GSE63885 and GSE14764.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Conflict of interest

The authors declare no competing interests.