Telomerase-Associated Genes as Prognostic Markers in High-Grade Serous Ovarian Cancer

Introduction

High-grade serous ovarian cancer (HGSOC) accounts for approximately 70% of global ovarian cancer deaths, making it the deadliest form of epithelial ovarian cancer [1–3]. Regrettably, less than 30% of patients with advanced HGSOC survive past 5 years despite existing treatments [4]. This ongoing challenge highlights the critical need for identifying novel molecular targets and creating more precise prognostic tools to enhance patient outcomes and alleviate the significant personal and societal burdens associated with managing this severe disease.

Telomerase, a ribonucleoprotein complex, consists of a catalytic core formed by telomerase reverse transcriptase (TERT) and non-coding human telomerase RNA, which serves as a template for adding telomeric repeats to chromosome ends [5]. Telomerase plays a crucial role in tumor biology, as it maintains telomere length and promotes cell immortality [6]. In most somatic cells, telomerase is typically in an inactive state, leading to progressive shortening of telomeres with each cell division, thereby suppressing tumorigenesis [7]. Approximately 85% to 90% of human cancers reactivate telomerase by upregulating TERT, enabling them to circumvent replicative senescence and achieve limitless proliferation [8]. Telomerase reactivation occurs through various mechanisms, such as TERT promoter mutations, epigenetic modifications, gene amplifications, and chromosomal rearrangements, making telomerase a universal oncogenic driver in malignancies [9]. Emerging research suggests that telomerase-associated genes (TAGs) influence critical cancer processes, such as genomic instability, epithelial-mesenchymal transition, and immune microenvironment remodeling [10]. TERT promotes cancer cell proliferation through multiple pathways, including NF-κB, NRF2, MYC, P21, and NOP2. Notably, TERT upregulates NF-κB, NRF2, and MYC, which in turn enhance TERT expression, forming a self-reinforcing positive feedback loop [11–16]. Telomerase inhibitors, including oligonucleotide inhibitors and small molecule inhibitors, exhibit antitumor activity. They achieve this effect by regulating gene transcription, protein synthesis, and the activity of genes undergoing post-translational modifications [17].

Although TAGs are increasingly acknowledged for their role in cancer, substantial gaps persist in understanding their functions within the complex tumor microenvironment of HGSOC. Previous bulk transcriptomic analyses have failed to resolve the cellular heterogeneity that exists in HGSOC, which can potentially obscure critical telomerase-associated molecular signatures in rare cell populations. Furthermore, the interplay between telomerase dysfunction and tumor immune microenvironment remodeling remains largely unexplored in HGSOC. These challenges underscore the necessity of employing single-cell RNA sequencing (scRNA-seq) to clarify the spatial network and functional architecture of TAGs in HGSOC.

In this study, we used an integrated bioinformatics approach combining scRNA-seq with bulk transcriptomic data from public repositories, including the Gene Expression Omnibus (GEO) and the Cancer Genome Atlas (TCGA) databases. ScRNA-seq technology provides detailed insights into cellular diversity, facilitating the detection of rare cell subpopulations and the mapping of developmental pathways in the HGSOC microenvironment. By integrating bulk RNA sequencing analysis, this multi-omics strategy establishes a comprehensive framework to investigate TAGs expression patterns across diverse cellular contexts. This study aims to explore the role of TAGs in HGSOC progression, develop a prognostic model using multi-omics data, and provide new insights into the pathogenesis of HGSOC.

Material and Methods

RESEARCH METHODOLOGY AND DATA ACQUISITION:

This study analyzed publicly available genomic data from 2 primary sources. The scRNA-seq dataset was obtained from the GEO database (https://www.ncbi.nlm.nih.gov/geo/) under accession GSE184880 (platform: GPL24676). Samples were selected based on the following criteria: inclusion of HGSOC tissue and normal ovarian controls, with all patients having undergone primary surgical resection and no prior chemotherapy, immunotherapy, or targeted therapy. The dataset included 7 treatment-naïve HGSOC tumors (across various stages) and 5 age-matched nonmalignant ovarian controls.

Bulk RNA sequencing and clinical data from the TCGA-OV cohort were downloaded via the Genomic Data Commons (https://portal.gdc.cancer.gov/) portal. After applying exclusion criteria, based on insufficient follow-up or missing clinical data, 420 samples from the original set of 617 patients with HGSOC constituted the final analytical cohort. The corresponding clinical characteristics are detailed in Table 1.

ETHICS STATEMENT:

This study used publicly available data in accordance with the access policies of each source database. Ethical approval and informed consent were not required, as all data were obtained from public sources.

DATA PROCESSING AND ANALYSIS:

The scRNA-seq data was processed and normalized using Seurat package (v 4.3.0). We applied quality thresholds to exclude cells following these criteria: gene counts under 200 or over 5000, mitochondrial content over 35%, and genes detected in fewer than 3 cells. The Harmony algorithm was used to mitigate batch effects. The NormalizeData function was used to standardize the filtered data, and the FindVariableFeatures function from the Seurat package identified the top 2000 highly variable characteristics. The RunPCA tool was used to scale and conduct principal component analysis on highly variable genes, selecting the top 17 principal components for subsequent analysis. We used the Shared Nearest Neighbor modular optimization and t-distributed Stochastic Neighborhood Embedding (t-SNE) algorithms to improve dimensionality reduction and visualize cluster classification with the chosen principal components. The FindAllMarkers algorithm was used to identify differentially expressed genes (DEGs) in each cell cluster by performing Wilcoxon rank-sum tests using the wilcox.test function from the R stats package. DEGs were screened based on the following criteria: |log₂FC| >0.25 and false discovery rate <0.01. Transcriptome RNA sequencing, copy number variation, and single nucleotide variant data predicted using VarScan2 Variant Aggregation and Masking, along with clinical data for TCGA-OV patients, were acquired through the R package TCGAbiolinks (v 2.25.0). The Combat function from the sva package was used to eliminate batch effects between samples. Principal component analysis was used to evaluate the degree of correction.

IDENTIFICATION OF TAGS AND CELL TRAJECTORY ANALYSIS:

The AUCell algorithm was used to assess the activity score of TAGs through gene set enrichment analysis (GSEA). The ggplot2 R package (v 3.3.5) was used to map scores and visualize activated cell clusters. We used the Monocle (v 2.26.0) R package to identify potential differentiation trajectories within transcriptomically inferred telomerase-associated activated cell subpopulations. Genes with high variability were identified based on a mean expression ≥0.1 and dispersion_empirical ≥1* dispersion_fit. The DDRTree algorithm was used for trajectory reconstruction. We used the branch expression analysis modeling test to identify the significant branch-dependent expression genes.

CELLULAR INTERACTION ANALYSIS:

CellChat (v 1.1.3) was used to examine intercellular communication in each cell type of HGSOC samples. CellChat analysis employed default parameters, and significance was determined using the Benjamini-Hochberg method to adjust P values, with a threshold of ≤0.05. Circle and bubble diagrams depicted the intensity of cell-cell communication networks among diverse cell types.

FUNCTIONAL ENRICHMENT ANALYSIS:

We used the clusterProfiler package (v 4.2.2) to conduct Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses, examining the TAGs functions. Significance thresholds were established at |log₂FC| >1 and an adjusted P value <0.05.

DEVELOPMENT AND VALIDATION OF A PROGNOSTIC MODEL:

Tumor samples from the TCGA-OV cohort, containing expression data and survival information, were divided into a training set (n=279) and a validation set (n=141) at a 7: 3 ratio. Univariate Cox regression analysis was performed to screen for prognostic genes using the coxph function from the survival package (v 3.5.5) to fit the models. We used least absolute shrinkage and selection operator (LASSO) regression via the glmnet R package to select candidate genes for the final risk model. A prognostic model was developed using multivariate Cox proportional hazards regression. Based on this model, a risk score was calculated for each patient using the following formula:

where i: current signature location; n: total number of signatures; Coef(genei): coefficient of gene i; Expression(genei): expression level of gene i.

The training and validation sets were categorized into high-risk and low-risk groups according to the median risk score. Kaplan-Meier survival curves were used to predict the prognosis of patients with HGSOC, with statistical significance assessed via the log-rank test. The discriminant ability of the prognostic model was evaluated by the receiver operating characteristic (ROC) curve. In addition, the validity of the prognostic model was verified by the validation sets. This model was developed and validated internally using only the TCGA-OV cohort, with a rigorous train-validation split, and has not yet been externally validated with an independent cohort. Since internal cross-validation or resampling methods were not used for further refinement, the performance estimates are potentially subject to bias.

A nomogram was developed using the rms R package (v 6.2.0) to predict 1-, 2-, and 3-year overall survival for patients with HGSOC, incorporating prognostic and risk scores. The predictive accuracy of the model was assessed using a time-dependent ROC curve.

GSEA AND GENE SET VARIATION ANALYSIS:

GSEA and gene set variation analysis (GSVA) utilized MsigDB pathway data. Functional enrichment analysis between HGSOC and the control group was conducted using GSEA with the R package clusterProfiler (v 4.2.2). Significant enrichment was identified with a normalized enrichment score (|NES| >1), along with an adjusted P value below 0.05. Biological functional differences between high-risk and low-risk groups were explored using the GSVA package (v 1.42.0) in R. The R package pheatmap (v 1.0.12) was used for visualizing the results.

IMMUNE INFILTRATION ANALYSIS:

Gene expression levels from 28 immune cell were obtained from the Tumor-Immune System Interaction Database (http://cis.hku.hk/TISIDB/). We used single-sample GSEA (ssGSEA) using the GSVA R package to evaluate immune cell infiltration abundance in both the high- and low-risk groups. The ggplot2 package (v 3.3.6) in R was used to visualize differences in immune cell infiltration between risk score groups.

CHEMOTHERAPY DRUG SENSITIVITY ANALYSIS:

Utilizing the half maximal inhibitory concentration (IC50) values of cancer drugs from the Genomics of Drug Sensitivity in Cancer (https://www.cancerrxgene.org/) database, we applied the R package oncoPredict (v 0.2) to predict chemotherapy drug sensitivity differences between the high- and low-risk groups.

SOMATIC MUTATION ANALYSIS:

Genes with the highest 20 mutation frequencies are often considered key drivers of malignant tumors [18]. Somatic mutation data from the TCGA database was obtained via the TCGAbiolinks R package. The maftools R package facilitated the analysis and visualization of single nucleotide variant, insertions and deletions, tumor mutation burden, and mutation frequency across high- and low-risk groups.

STATISTICAL ANALYSIS:

Data processing and analysis were conducted using R version 4.1.2 (The R Foundation, Vienna, Austria). The Wilcoxon rank-sum test was used to evaluate differences in continuous variables between high- and low-risk groups. Proportional differences were assessed using the chi-square test or Fisher exact test. Kaplan-Meier survival curves were created using the ggsurvplot function from the survminer package in R, with statistical significance assessed via the log-rank test. LASSO-Cox regression analysis was used to screen the signature genes and construct the prognostic model. The predictive power of the model was evaluated using ROC and time-dependent ROC curves. Statistical tests with significance set at a P value <0.05.

COMPUTATIONAL-ONLY SCOPE:

In this study, all findings and associations reported herein are derived from the analysis of publicly available genomic and transcriptomic datasets. No experimental validation-including functional assays in cell lines or animal models, biochemical analyses, or prospective clinical assessment was performed. The clinical correlations are based on retrospective data and are therefore hypothesis-generating. The prognostic model and identified mechanisms require rigorous experimental and clinical validation in future studies.

NON-APPLICABILITY TO CLINICAL DECISION-MAKING:

The prognostic model we propose is a research tool derived from retrospective bioinformatics analysis and should not be used for clinical decision-making or direct patient care. The model lacks prospective validation in independent, controlled clinical cohorts and has not been integrated into clinical-grade testing or workflows. Any potential future clinical applications would require extensive standardized technical validation, regulatory approval, and demonstration of utility in improving patient prognosis within clinical trial settings.

DATA AND CODE AVAILABILITY:

All original datasets are publicly accessible: the scRNA-seq data under accessions GSE184880 (GEO) and PRJNA766716 (SRA), and the TCGA-OV cohort data via the Genomic Data Commons portal. All analysis scripts (including R code, parameters, and random seeds), and the patient identifier list have been deposited in permanent public repositories. The code necessary to reproduce the results is available on GitHub (https://github.com/wxh7901/hgsoc-telomerase-prognostic-model/releases/tag/v1.0), and the accompanying data are archived on Zenodo (https://doi.org/10.5281/zenodo.18321919).

Results

IDENTIFICATION OF HGSOC CELL SUBTYPES:

After quality control, the scRNA-seq data yielded 53 671 cells. The cell distribution between the HGSOC and control groups was consistent, suggesting minimal batch effects and suitability for further analysis (Figure 1A). The t-SNE algorithm was applied for nonlinear dimension reduction, clustering the single-cell samples into 30 clusters (Figure 1B). Cell types were manually annotated according to the gene expression profiles of each cluster using t-SNE. Figure 1C identifies 12 distinct cell types, and the percentage distribution of various cell types across each group are displayed in Figure 1D. Differential analysis was performed on the clusters, and the 2 significantly differentially expressed marker genes in each cluster are shown in a heatmap (Figure 1E). The analysis delineates the single-cell transcriptomic atlas of HGSOC, defines its cellular composition and characteristic genes, and lays the data foundation for understanding its tumor heterogeneity and microenvironmental regulation.

TRANSCRIPTOMICALLY INFERRED TELOMERASE-ACTIVE CELL POPULATION AND DIFFERENTIATION TRAJECTORIES IN HGSOC:

Using the AUCell_exploreThresholds algorithm, we identified 13 071 transcriptomically inferred active cells in the telomerase subgroup (Figure 2A). The Seurat package was applied to screen 2089 TAGs. Based on the inflection point in the area under the ROC curve (AUC) value distribution, we determined an AUC of 0.26 as the optimal threshold to distinguish between transcriptomically inferred telomerase-active and -inactive cell populations in this subgroup. The analysis suggested that HGSOC cells progress along 3 distinct trajectories. Pseudotime analysis, cell clustering, and single-cell state estimation are illustrated in Figure 2B–2D, while the distribution of different cell types across these states are shown in Figure 2E. Throughout the differentiation process, enrichment of multiple pathways, including epithelial cell migration, was observed (Figure 2F). We identified a transcriptomically inferred telomerase-active cellular subgroup, reconstructed 3 differentiation trajectories in HGSOC, and linked pseudotemporal progression to the enrichment of migration-related pathways, thereby revealing a dynamic axis of cellular state transition in this malignancy.

CELLULAR COMMUNICATION WAS DIFFERENT IN EACH CELL TYPE OF HGSOC:

CellChat analysis suggested intercellular communication networks among various cell types in HGSOC (Figure 3A, 3B). Owing to the relatively rare interactions of fibroblast cells with other cells, they were filtered out. Subsequently, we determined both the quantity and intensity of interactions (Figure 3A, 3B). Potential outgoing and incoming signals as well as specific receptors within these 11 cell types were further investigated. As shown in Figure 3C and 3D, endothelial cells and macrophages are the main signal providers, while epithelial cells and macrophages are the primary signal receivers. The potential signaling pathways within these cell types include MK, MIF, SPP1, VISFATIN, and GALECTIN.

We further found that the highest frequency of communication among these 11 cell types is from macrophage to endothelial cell through the NAMPT-INSR (VISFATIN) signaling pathway, and from macrophage to smooth muscle cell through the SPP1-(ITGAV+ITGB1) (SPP1) signaling pathway (Figure 4A, 4B). As depicted in Figure 4C and D, the ligand NAMPT is mainly highly expressed in macrophages, epithelial cells, and cancer cells. The receptor INSR is highly expressed in endothelial cells. The ligand SPP1 is highly expressed in macrophages. The receptor ITGAV is predominantly expressed in smooth muscle and mesenchymal cells, whereas ITGB1 is widely expressed across nearly all cell types. These results delineate a specialized intercellular communication network within the HGSOC microenvironment, highlighting macrophages as central signaling hubs via pathways such as VISFATIN (NAMPT–INSR) and SPP1, which may influence tumor-stromal crosstalk and disease progression. These results delineate a specialized intercellular communication network within the ovarian cancer microenvironment, highlighting macrophages as central signaling hubs through pathways such as VISFATIN (NAMPT–INSR) and SPP1, which may influence tumor-stroma crosstalk and disease progression.

BIOLOGICAL FUNCTIONS OF TAGS:

We enhanced the DEGs from both the transcriptomically inferred telomerase-active and -inactive cell populations, as well as from HGSOC and normal control samples in the TCGA-OV database (Figure 5A, 5B). We identified 768 key TAGs by intersecting the 2 groups (Figure 5C). GO analysis indicated that the key TAGs were primarily enriched in biological processes associated with T cell activation, cytosolic ribosome, and ribosome structural components (Figure 5D). KEGG analysis revealed significant enrichment of gene sets associated with COVID-19, ribosome, and Epstein-Barr virus infection (Figure 5E, 5F). Analysis revealed a shared transcriptional program between HGSOC and viral infections, which was captured by pathway database enrichment analysis.

DEVELOPMENT AND EVALUATION OF THE PROGNOSTIC MODEL BASED ON TAGS:

Based on univariate Cox analysis of 768 TAGs, 68 prognosis-associated genes were initially identified and refined to 20 genes via LASSO regression. Using the resulting risk model, patients were stratified into high- and low-risk groups based on the median risk score cutoff of 0.02 (Figure 6A, 6B). Kaplan-Meier analysis confirmed significantly poorer overall survival in high-risk patients across both the training and validation cohorts (Figure 6C, 6D). Additionally, the predictive performance of the risk score for overall survival was evaluated by ROC analysis. The AUC values for the training cohort were 0.777 (95% CI: 0.678–0.855), 0.750 (95% CI: 0.674–0.817), and 0.704 (95% CI: 0.625–0.769), while for the validation cohort, they were 0.639 (95% CI: 0.432–0.825), 0.646 (95% CI: 0.513–0.763), and 0.628 (95% CI: 0.509–0.734) for the 1-year, 2-year, and 3-year periods, respectively (Figure 6E, 6F). The model showed significant prognostic discriminatory ability in the TCGA-OV internal validation set, although it must be emphasized that this performance is confined to the results of this internal cohort validation. A prognostic nomogram incorporating the risk score showed good calibration, with 1-, 2-, and 3-year prediction accuracies of 0.718 (95% CI: 0.656–0.795), 0.710 (95% CI: 0.667–0.782), and 0.679 (95% CI: 0.629–0.738) (Figure 6G, 6H). This prognostic model utilized TAGs features to provide valuable insights for risk stratification and predicting the prognosis of patients with HGSOC.

GSEA AND GSVA BETWEEN HIGH-RISK AND LOW-RISK GROUPS:

The most significant GSEA pathways are illustrated in Figure 7A–7F. The 5 most significantly different GSVA pathways between high-risk and low-risk groups were selected to create heatmaps (Figure 7G).

IMMUNE INFILTRATION ANALYSIS BETWEEN HIGH-RISK AND LOW-RISK GROUPS:

To clarify immune landscape differences between risk groups, ssGSEA revealed distinct infiltration patterns of 28 immune cell types (Figure 8A). Notably, myeloid-derived suppressor cells were inversely correlated with most other immune subsets (Figure 8B). This correlation pattern is consistent with the recognized profile of an immunosuppressive microenvironment and was observed more frequently in high-risk patients. Most relevant to our hypothesis, several hub genes showed specific immunomodulatory associations: AAK1, FXYD5, and TRABD correlated positively with effector memory CD8+ T cells, whereas FXYD5, GZMB, and TRABD were negatively associated with type 2 T helper cells (Figure 9A–9I). Collectively, these correlative findings link the prognostic model genes to key antitumor immune feature – namely, a concurrent association with higher cytotoxic T cell memory and lower pro-tumor Th2 activity. This pattern generates the hypothesis that these genes may participate in immune modulation relevant to HGSOC progression, a possibility that warrants future mechanistic investigation.

TUMOR MUTATION BURDEN AND DRUG SENSITIVITY ANALYSIS BETWEEN DIFFERENT GROUPS:

We analyzed gene mutations in HGSOC and visualized the 20 most frequently mutated genes. TP53 exhibited the highest mutation frequency in both groups, with TTN following closely (Figure 10A, 10B). We assessed the discriminative ability of the risk score in predicting chemotherapy sensitivity among patients with HGSOC. The study examined the clinical efficacy of Afatinib_1032, Buparlisib_1873, PD0325901_1060, Pictilisib_1058, Sabutoclax_1849, Trametinib_1372, Entinostat_1593, and AGI-6780_1643 in treating HGSOC. Low-risk patients appeared more responsive to Sabutoclax_1849, AZD5582_1617, Entinostat_1593, and AGI-6780_1643 (Figure 10C–10K), suggesting chemotherapy as a viable treatment option for this group.

Discussion

The biological complexity of HGSOC arises from its pronounced inter- and intra-tumoral heterogeneity, as well as the dynamic interactions between malignant cells and diverse stromal components within the tumor ecosystem [19,20]. ScRNA-seq serves as a pivotal tool for resolving such heterogeneity in HGSOC [21,22]. In this study, we used scRNA-seq to deeply interrogate the immune landscape and pivotal gene signatures of HGSOC. Our findings further elucidate its biological intricacy and offer new perspectives for understanding the mechanisms of disease progression.

Telomerase is often activated to maintain telomere length and support unlimited replication in tumor cells, which is a hallmark characteristic of cancer [23,24]. TERT directly regulates EGFR and VEGF transcription and is pivotal in the VEGF signaling pathway through its interaction with the VEGF promoter. Telomerase inhibition has been demonstrated to decrease VEGF expression and angiogenesis [25]. TERT engages with signaling pathways such as Wnt/β-catenin and NF-κB to enhance cell proliferation, adhesion, and migration [26]. TERT also interacts with TGF-β responsiveness, enhancing cancer cell dissemination and disease progression [26,27]. In papillary thyroid cancer, bladder cancer, and gliomas, TERT promoter mutations are associated with higher recurrence rates and reduced survival outcomes [28–30]. Preclinical studies indicate that telomerase inhibition via compounds and gene modifications can suppress tumor growth and metastasis in mouse models, highlighting its therapeutic potential [31]. Telomerase inhibitors BIBR1532 and imetelstat have demonstrated inhibitory effects on tumor growth in xenograft models of oral squamous cell carcinoma and non-small cell lung cancer [32,33]. Although encouraging results have been obtained from studies on telomerase in various types of tumors, data on telomerase in HGSOC remain limited.

Our single-cell RNA sequencing analysis uncovered significant cellular diversity in HGSOC, identifying 12 distinct cell types, such as T cells, macrophages, and cancer cells. The identification of 2089 TAGs and subsequent classification of cells into transcriptomically inferred telomerase-active and -inactive subgroups provides novel insights into tumor biology. Particularly noteworthy is the finding that transcriptomically inferred telomerase-active cells (13 071 cells) exhibited distinct functional states during pseudotime trajectory analysis, with state-specific enrichment in pathways related to focal adhesion, T-cell activation, and extracellular matrix organization. These findings suggest that TAGs may play stage-specific roles in tumor progression, potentially influencing both cancer cell behavior and immune responses. Cell trajectory analysis further supports the dynamic nature of telomerase-related processes in HGSOC, highlighting their distinct roles in different progression stages of HGSOC.

Previous studies have shown that macrophages are important immune cells in the tumor microenvironment, involved in regulating various biological behaviors and influencing the prognosis of patients with ovarian cancer. Furthermore, macrophages play a regulatory role in the ovarian cancer microenvironment by interacting with other immune cells, including T cells and mesothelial cells [34]. We investigated intercellular receptor-ligand interactions and signaling pathways based on CellChat to further elucidate the interactions between HGSOC cells. Our study showed strong cellular communication relationships between macrophages, endothelial cells, tumor cells, and T cells. Endothelial cells and macrophages were the main signal providers, while epithelial cells and macrophages were the main signal receivers. This suggests that macrophages and stromal cells are closely associated with both HGSOC cell proliferation and T cell – mediated immunoregulation, indicating their potential involvement in modulating tumor – immune microenvironment interactions in HGSOC. Macrophages were the key cell type for ligand-receptor interactions with HGSOC cells.

The pathway analysis of 768 key TAGs revealed unexpected connections between telomere biology and immune-viral interactions in HGSOC. The enrichment of T-cell activation and ribosome-related pathways was consistent with established features of HGSOC [35,36]. However, the observed significant association between HGSOC and the COVID-19 and Epstein Barr virus infection pathways cannot be directly equated with evidence of active infection by the corresponding viruses or viral oncogenesis in HGSOC. A more plausible explanation is that the intense immune-inflammatory response, cellular stress, and activation of specific signaling pathways induced by the HGSOC tumor microenvironment show significant transcriptomic overlap with host responses triggered by viral infections. Recent studies indicate that CD8+ T cells persistently stimulated by antigens in the context of progressive tumors and chronic viral infections develop dysfunctional states [37,38]. Therefore, the virus-related pathway enrichment observed here is best interpreted as evidence of a shared transcriptional program identified via pathway databases, not as an indication of direct exogenous viral involvement.

The ribosome pathway enrichment may reflect the high protein synthesis demands of rapidly proliferating cancer cells, while the immune-related pathways underscore the complex interplay between telomere maintenance and tumor immunity. These results provide a molecular basis for understanding how telomerase dysfunction might contribute to both cancer cell survival and immune evasion in HGSOC.

The immune landscape analysis revealed significant differences in 12 immune cell populations between risk groups, with particularly strong associations observed between specific genes (AAK1, FXYD5, TRABD) and effector memory CD8+ T cells. These findings extend previous reports of immune dysfunction in HGSOC by linking TAGs to antitumor immune features. The negative correlation observed between myeloid-derived suppressor cells and multiple immune cell populations suggests a potential immunosuppressive pattern in high-risk patients. Notably, PCBP2 was identified as strongly associated with type 2 T helper cells, an association not previously reported in the cancer context. The differential immune profiles between risk groups provide a potential correlation with the observed prognostic differences.

Despite numerous prognostic signatures having been published for HGSOC, their translation into clinical practice remains limited. For instance, the gene signature developed by Maki et al showed prognostic relevance in the original cohort but could not be validated in several large independent cohorts, including TCGA, underscoring the common challenges of overfitting and inadequate validation in models built from retrospective data [39]. Biomarkers derived from a single technology platform frequently lack cross-platform robustness, which compromises the reproducibility of associated models across studies using different experimental or profiling methods [40]. Furthermore, the molecular landscape of HGSOC is characterized by extensive genomic instability, making it intrinsically difficult to identify robust and independent prognostic signals beyond established risk factors, such as clinical stage and residual disease [41]. In the present study, our proposed prognostic model demonstrated consistent risk stratification in both the training and validation cohorts, outperforming previously reported 2-gene models in predicting the prognosis of patients with HGSOC (AUC: 0.777 vs 0.65) [42]. Its potential clinical utility is strengthened by an integrated nomogram, which could facilitate personalized prognostic evaluation. This model demonstrates the correlation between TAGs and the biological characteristics of HGSOC. To substantiate these findings, future work should focus on functional validation of key TAGs through in vitro and in vivo experiments, aimed at elucidating their specific roles in tumor progression.

Drug sensitivity analysis revealed intriguing patterns, with patients at low-risk showing particular sensitivity to Sabutoclax and AZD5582, both targeting BCL-2 family proteins [43,44]. These in silico predictions suggest the hypothesis that TAGs may be associated with altered apoptotic priming in HGSOC cells, although this remains to be validated experimentally. The universal presence of TP53 mutations across risk groups aligns with its established role in HGSOC pathogenesis, while the observed differences in predicted drug responses indicate that additional molecular features beyond TP53 status may influence computational sensitivity estimates. These predictive findings, derived from cell line-based models, are intended primarily for hypothesis generation and do not constitute evidence of therapeutic efficacy or clinical response. Further experimental and clinical validation is required before such associations can inform trial design or stratified patient management.

This study has several limitations. First, the identified TAGs features lack experimental validation, which restricts the mechanistic interpretation of the findings. Second, the sample size of the single-cell RNA sequencing cohort is relatively small, including only 7 tumor and 5 normal specimens, and thus may not fully capture the tumor heterogeneity of HGSOC. Third, although rigorous computational methods were applied to mitigate batch effects, the integration of data from different databases may still introduce potential confounding factors. Fourth, the cell-cell communication network inferred using CellChat is based on a probabilistic model of ligand-receptor co-expression and cannot directly demonstrate functional signaling activation, downstream pathway activity, or causal relationships in intercellular communication. Further experimental validation is required to confirm the functional relevance of the predicted interactions. Finally, although the prognostic model underwent rigorous partitioning and validation within the TCGA-OV cohort, its construction, optimization, and evaluation relied entirely on a single dataset. It has not yet undergone validation using external independent cohorts or clinical-grade assessment. Therefore, the model’s performance and robustness and the generalizability of the biological associations it identifies remain untested, which represents a crucial next step for future research.

Conclusions

In conclusion, this study systematically delineates the role of TAGs in the progression of HGSOC through muti-omics integration. Based on TAGs, we developed and internally validated a prognostic model associated with patients with HGSOC using the TCGA-OV cohort. While the validity of this model in external populations still requires further verification, it provides a novel computational perspective for prognostic prediction in patients with HGSOC.