Introduction 

Epithelial Ovarian Cancer (EOC) is a heterogeneous group of tumors, and High-Grade Serous Ovarian Carcinoma (HGSOC) is its most common histological subtype, accounting for approximately 70% of EOC cases ¹, yet five-year survival rates remain below 50% due principally to late-stage diagnoses ². Strong evidence implicates precursor lesions in the distal fimbriae of the fallopian tube as the origin of HGSOC ³, although the Ovarian Surface Epithelium (OSE) has also been proposed as a possible source ⁴. HGSOC progression involves the development of malignant ascites and omental metastases ⁵, creating an ascites‑rich tumor microenvironment that promotes cellular heterogeneity, chemoresistance, and recurrence ⁶. Chemotherapy‑resistant Ovarian Cancer Stem Cells (OCSCs) have been isolated from malignant ascites and are thought to possibly contribute to relapse ^7,8, yet their cellular origin remains debated. Proposed mechanisms include Epithelial‑to‑Mesenchymal Transition (EMT) ⁹ and putative cell fusion between EpCAM⁺ tumor cells and hematopoietic populations, generating EpCAM/CD45 dual‑positive hybrids ^10,11. However, it is important to note that the EpCAM⁺/ CD45⁺ phenotype alone does not constitute definitive proof of cell fusion; alternative explanations, such as trogocytosis, exosome‑mediated marker transfer, phagocytic uptake, or flow‑cytometric doublets, cannot be excluded based solely on surface‑marker expression.

At present, transcriptomic and genetic features of EpCAM/CD45 Dual positive Ovarian Cancer Stem-like Cells (DpOCSCs) are incompletely defined, partly because available datasets contain only a small number of such samples. Here, the authors performed an integrated transcriptomic, network, and clinical‑ correlation analysis to compare DpOCSCs with reference OCSC populations, identify robust hub genes and pathways, and explored their clinical and immunological correlates in large ovarian cancer cohorts. In particular, SYK, CD74, and MX1 as hub genes associated with the DpOCSCs were identified. 

Materials and Methods 

Data collection and processing

 Gene expression datasets, including GSE90125, GSE216627, and GSE75036, were obtained from the NCBI GEO database. Details regarding platform, sample size, sample data, age, and related publications are presented in table 1. DpOCSC samples (n=2; GSM1941116, GSM1941118) were obtained from GSE75036, which was generated on the Illumina HumanHT-12 V4.0 Expression BeadChip platform (GPL10588). Comparator OCSC samples were obtained from GSE90125 (GSM2398960, GSM2398961) and GSE216627 (GSM6684690, GSM6684691), both generated on the Affymetrix Human Gene 2.1 ST Array platform (GPL17692). All three datasets are microarray-based human gene-expression profiling. The processed expression matrices provided by the original submitters were used. Probe identifiers were mapped to gene symbols using the corresponding platform annotation files, probes without valid gene annotations were removed, and when multiple probes mapped to the same gene, their expression values were summarized by taking the mean. To enable cross-platform integration, only genes shared across all datasets were retained prior to batch correction. Batch effects were corrected using the ComBat function in the sva package, with dataset specified as the batch variable and cell type (DpOCSC vs. OCSC) retained as the biological variable of interest. Sample quality was assessed using boxplots and Principal Component Analysis (PCA), and no samples were excluded after QC. 

Differential gene expression analysis

Differentially Expressed Genes (DEGs) between DpOCSCs and OCSCs were identified using the Limma package. The limma pipeline with empirical Bayes moderation was applied to improve variance estimation despite the small sample size. Linear models were fitted to the normalized microarray expression data using the lmFit () function, followed by empirical Bayes moderation with the eBayes () function to improve the estimation of gene-wise variances and increase the statistical power of differential expression analysis, particularly given the relatively small sample size. Given the small number of DpOCSC samples, DEGs were defined using |log₂FC|>1 and FDR<0.1 to balance sensitivity and specificity. Volcano plots were generated using VolcaNoseR2 for visualization. 

Protein-protein interaction (PPI) network and identification of hub genes

The Protein-Protein Interaction (PPI) network was constructed from the DEG-derived input gene set using the STRING database plugin in Cytoscape (version 3.10.2). The network was constructed from the DEG-derived input gene set using a confidence score cutoff of 0.4 and combined evidence selected for interaction retrieval. Hub genes were prioritized using CytoHubba (version 0.1) by applying eight centrality algorithms, including Stress, Radiality, MNC, MCC, EPC, Degree, Closeness, and Betweenness. For each algorithm, the top 50 ranked genes were extracted, and genes appearing in all eight ranked lists were defined as common hub genes. These candidates were further evaluated using CentiScape (version 2.1) to support topological prioritization. 

Gene ontology and pathway enrichment analysis

Gene ontology analyses, including Biological Process (BP), Cellular Component (CC), and Molecular Function (MF), as well as MSigDB Hallmark and KEGG pathway ¹² enrichment, were performed with the Enrichr database. The top ten terms for each category were selected using a significance threshold of an adjusted p<0.05. Additional enrichment factor analysis and visualization were performed in R (version 4.3.1). 

Hub gene evaluation

Hub gene expression was supported at the mRNA level using TCGA datasets via GEPIA for tumor (n=426) and normal (n=88) samples (TCGA normal and GTEx data), and at the protein level using UALCAN for tumor (n=100) and normal (n=25) samples. Expression across 33 cancer types was analyzed with GEPIA. 

Association of TCGA-supported hub genes with OvCa stages and genetic alterations

Hub gene expression across ovarian cancer stages was analyzed in TCGA using GEPIA. Genetic alterations were explored using cBioPortal in five ovarian cancer datasets: High-Grade Serous Ovarian Carcinoma (MSK, NPJ Genome Medicine 2021, n=45), TCGA Firehose Legacy (n=617), TCGA Nature 2011 (n=489), TCGA PanCancer Atlas (n=585), and TCGA GDC 2025 (n=602). 

Kaplan-Meier survival analysis

The impact of TCGA-supported hub genes on Overall Survival (OS) and Progression-Free Survival (PFS) was assessed with KM Plotter across multiple datasets. OS analysis included all cases of serous ovarian carcinoma (n=1207), GSE9891, GSE30161, GSE26193, GSE18520, and TCGA OvCa (n=557). PFS analysis included the same datasets, with GSE51373 and TCGA OvCa (n=522) included as well. Cutoff values were selected using the default percentile-based setting in KM Plotter. 

Assessment of predictive biomarkers

Predictive biomarkers were evaluated using Receiver Operating Characteristic (ROC) curve analysis with ROC Plotter for ovarian cancer. The analysis focused on Relapse-Free Survival (RFS) at six months (n=1,347) in three treatment conditions: any chemotherapy, platinum-based therapy, and paclitaxel. Diagnostic performance was assessed individually and in multi-gene combinations, considering an Area Under the Curve (AUC) of 0.7 or higher as acceptable. AUC values were not corrected for multiple testing and are therefore exploratory. 

Correlation with immune checkpoint proteins and immune infiltrating cells

Gene expression data for ovarian serous cystadenocarcinoma from the Cancer Genome Atlas (GDC- TCGA-OV) were retrieved from the UCSC Xena database in the form of STAR-TPM normalized RNA-seq reads. Expression values were subsequently transformed using a log2 (TPM+1) conversion. Pearson correlation coefficients, along with their corresponding p-values, were computed for the selected gene pairs. To correct for multiple hypothesis testing, p-values were adjusted using the Benjamini–Hochberg False Discovery Rate (FDR) procedure. All correlation analyses and associated data visualizations were carried out in R (version 4.2.0). Additionally, the relationship between the identified hub genes and immune infiltrating cell populations within the TCGA-OV cohort was explored using the Timer 2.0 platform. 

Analysis of transcriptional regulation of TCGA-supported hub genes

Transcriptional regulation was investigated for six principal signaling pathways relevant to OCSC: STAT3, TGFβ/Smad2/3/4, Hippo-YAP1/TAZ-TEAD, NF-κB, Notch/RBPJ, and Wnt/β-catenin-TCF/LEF ¹³. The Enrichr transcription panel, including ChEA 2022 and ENCODE TF ChIP‑seq 2015 data, was used for analysis. Additional resources considered included TRUST Transcription Factors 2019, TRANSFAC, and JASPAR position weight matrices. TF–target relationships compiled from these resources include predicted and ChIP‑supported interactions and should be viewed as putative rather than experimentally confirmed in DpOCSCs. 

Clinico-pathological evaluation of DpOCSCs-related hub genes

Associations between hub gene expression and clinical features, such as lymphatic and venous invasion, neoplasm histologic grade, new neoplasm event type, and therapy outcome, were evaluated using UCSC Xena Browser (TCGA OV, n=308). Gene expression was dichotomized based on a Gaussian distribution-derived cutoff, using Cut-off Finder ¹⁴, and chi-square tests were applied as appropriate. A comprehensive data table categorized patients by hub gene expression using TCGA IDs in R (version 4.3.1). 

Statistical analysis

All statistical analyses were performed in R (version 4.3.1). Figures were generated using GraphPad Prism 9 and the SRplot online tool. Differential expression analysis was conducted using the limma package, and False Discovery Rate (FDR) correction was applied to control for multiple testing. Unless otherwise specified, results from exploratory downstream analyses are reported using nominal p-values and should be interpreted as hypothesis-generating. For exploratory analyses, nominal p-values are reported; however, given the hypothesis-generating nature of this study, all results should be interpreted with caution and require independent evaluation in external cohorts. 

Results 

Up-regulated genes in DpOCSCs are related to inflammatory pathways

A total of 347 genes were up-regulated and 23 were down-regulated in DpOCSCs (EpCAM/CD45 dual-positive cells) compared with OCSCs, as revealed by transcriptome analysis of the GSE75036, GSE90125, and GSE216627 datasets (Figure 1A). Volcano plot and heatmap analyses showed a distinct molecular profile, with the top 10 up-regulated and top 5 down-regulated genes in DpOCSCs and OCSCs (GSM1941116, GSM1941118; GSM2398960, GSM2398961, GSM6684690, GSM6684691) shown in figures 1B and 1C.

GO biological process enrichment analysis revealed overrepresentation of interferon-response terms, TNF, NF-κB, IL-6, JAK-STAT3, and CXCR4 chemokine signaling, as well as genes associated with reduced susceptibility to natural killer cell-mediated cytotoxicity, these terms might be interpreted as indicating an immune-evasive, cytokine-rich phenotype for DpOCSCs, along with potential innate immune dysregulation (Figure 2A). Functional studies will be required to determine whether these signatures reflect immune evasion or altered cytokine signaling at the protein level. MSigDB Hallmark analysis highlighted interferon-γ and interferon-α responses as the most enriched signatures, together with TNF-α signaling via NF-κB, IL-6–JAK–STAT3 signaling, hypoxia, and broader inflammatory-response terms (Figure 2B). These modules are consistent with an interferon-inflamed, TNF-associated state in DpOCSCs. Down-regulated genes were mapped to fibroblast growth factor receptor signaling and cellular responses to FGF (Figure S1). The top enriched biological process terms for both up- and down-regulated gene sets, together with the top hallmark pathways for up-regulated genes, are listed in table S1. Cellular component and molecular function terms for these gene sets are provided in table S2. 

PPI network and hub genes

The PPI network among Differentially Expressed Genes (DEGs) is shown in figure S2. The top 50 ranked genes from eight CytoHubba centrality algorithms were intersected, yielding 17 common hub genes that were all up-regulated in DpOCSCs. This intersection strategy was used to increase robustness and reduce algorithm-specific bias (Table S3). These analyses revealed 17 hub genes associated with DpOCSCs, and notably, every single one of them was part of the up-regulated gene set (Figure 3A). Details of the relationships among these 17 hub genes, including network connectivity and key communication points, are presented in table S4.

Gene Ontology (GO) analysis focused on ovarian cancer-relevant terms showed that these hub genes are primarily involved in processes such as chemokine and cytokine production, also immune cell activity. Terms related to macrophage activation and interleukin-6 (IL-6) and interleukin-8 (IL-8) signaling were particularly prominent (Figure 3B). Cellular component terms were frequently linked to vesicle membrane localization, and molecular function terms were predominantly related to MHC class II binding, kinase binding, and GTPase activity (Table S5).

Further KEGG and MSigDB Hallmark enrichment analysis focused on ovarian cancer-relevant terms indicated that these hub genes are associated with inflammatory and immune processes. Significant pathways included NF-κB signaling, Toll-like receptor signaling, and PD-1/PD-L1 pathway (Table 2). These results were consistent with the MSigDB findings, which showed strong enrichment of interferon-γ response, interferon-α response, TNF-α signaling via NF-κB, and inflammatory response terms, as detailed in table 2. Collectively, these analyses indicate that the hub genes occupy central positions in inflammatory and immune-related networks and may contribute to the DpOCSC transcriptional program, although direct regulatory roles remain to be experimentally supported. 

Assessment of hub genes at mRNA and protein levels

To assess the relevance of the identified hub genes, we examined their expression at both the mRNA and protein levels in OvCa tumors and normal ovarian tissue using TCGA and UALCAN. In TCGA-OV, the median log2(TPM+1) expression of each supported hub gene was at least 1 unit higher in tumors than in normal tissue (FDR <0.01 for all comparisons; Figure 4). Analysis of TCGA datasets showed that 10 of the 17 hub genes, namely SYK, CD74, SPI1, ITGB2, PLEK, TNF, RAC2, MX1, C1QA, and CXCR4, were expressed at substantially higher levels in OvCa tumors than in normal tissues (Figure 4).

To further investigate protein-level expression of these TCGA-supported hub genes, we used UALCAN to analyze CPTAC proteomics data. A significant elevation in the protein expression of SYK, CD74, and MX1 was observed in ovarian tumors compared to normal tissues (Figure 5A), whereas CXCR4 showed a nonsignificant trend toward higher protein levels (p=0.74; Figure 5A). The remaining supported hub genes, ITGB2, PLEK, RAC2, and C1QA, showed lower protein expression in ovarian cancer tumors (Figure 5B). The discordance between mRNA and protein levels for several genes, such as ITGB2, PLEK, RAC2, and C1QA, may reflect post-transcriptional regulation or the contribution of immune and stromal cells in bulk tumor lysates, and therefore highlights the need for single-cell or sorted-cell evaluation. Protein expression data for TNF and SPI1 were not available in the UALCAN ovarian cancer dataset.

Functionally, SYK is a non-receptor tyrosine kinase involved in key processes regulating both immune and non-immune cells ¹⁵. CD74, also known as the   invariant chain, is overexpressed in multiple cancers and promotes tumor cell survival by activating macrophage migration inhibitory factor-dependent signaling pathways such as NF-κB and PI3K/Akt ¹⁶. CD74 also modulates the tumor immune microenvironment, facilitating immune evasion and tumor progression, and is being explored as a therapeutic target in hematologic and solid tumors ^16,17. MX1 is an interferon-induced dynamin-like GTPase with high GTP hydrolysis activity; elevated MX1 expression has been reported in various malignancies and is functionally linked to oncogenic processes ¹⁸. Beyond ovarian cancer, SYK and CD74 are up-regulated in 12 and 15 other cancers, respectively, and MX1 is increased in 10 distinct cancer types (Figures S3 A-C). Collectively, these mRNA, protein, and pan-cancer data support SYK, CD74, and MX1 as candidate components of immune-inflammatory programs in OvCa that merit further functional investigation. 

Genetic alterations and clinical relevance of TCGA-supported hub genes

The authors next investigated genetic alterations affecting 10 of the 17 hub genes, including SYK, CD74, SPI1, ITGB2, PLEK, TNF, RAC2, MX1, C1QA, and CXCR4, all of which showed higher expression in ovarian cancer tumors than in normal tissue, using cBioPortal. Across five serous ovarian cystadenocarcinoma cohorts, these genes exhibited copy-number alterations in approximately 15-25% of tumors (Figures 6A and 6B). Deep deletions were less frequent (Figure 6A) but more common among non-supported hub genes (Figure S4).

To further clarify the importance of these hub genes in OvCa progression, their expression across disease stages using TCGA OvCa tumor data was evaluated. This analysis showed that TNF was significantly associated with OvCa stage (Figure 7A). In addition, the non-supported hub genes ITGAM and NFKBIA were significantly associated with distinct stages of OvCa (Figure S5A). These stage associations were statistically significant at p<0.05, but were not adjusted for multiple testing and should therefore be interpreted as exploratory. 

MicroRNA regulation of DpOCSC-related hub genes

miRNA-target interactions were compiled from the Enrichr-linked miRTarBase resource and include both predicted and experimentally supported pairs. Also miRNA interactions with the TCGA-supported hub genes were explored. Figure 7B shows an interaction network of the top five miRNAs regulating these genes. Among the identified regulators, hsa-miR-203a-3p and hsa-miR-187-3p were linked to TNF, which was associated with OvCa stage. Additionally, hsa-miR-1468-5p and hsa-miR-203a-3p were linked to SYK. Notably, hsa-miR-146a-5p appeared to be an important epigenetic regulator of both supported and non-supported hub genes, including ITGB2, CXCR4, NFKB1, TLR4, and RHOA (Figure 7B and Figure S5B). Importantly, recent studies have reported overexpression of hsa-miR-146a-5p in primary OvCa tumors and omental metastases ¹⁹. 

Correlation of DpOCSC-related hub genes with immune checkpoint proteins and immune-suppressive infiltrating cells

Since DpOCSCs express the pan-hematopoietic marker CD45 ^10,11, we assessed the relationships between DpOCSC-related hub genes and immune checkpoint proteins to clarify their roles in ovarian cancer immunology. Pearson correlation analysis was performed between the 17 identified hub genes and six immune checkpoint proteins, including PD-L1 (CD274), PD-1 (PDCD1), CTLA-4 (CTLA4), TIGIT, LAG-3 (LAG3), and TIM-3 (HAVCR2), in GDC-TCGA OvCa tumors ²⁰. The GDC-TCGA-OV cohort showed predominantly positive associations among immune-related genes. The strongest correlations were observed between HAVCR2 and several myeloid/ immune markers, including IL10RA, ITGAM, ITGB2, PLEK, and SPI1 (Pearson’s r up to 0.95). Most gene pairs remained statistically significant after Benjamini-Hochberg FDR correction, whereas CXCR4, RHOA, and particularly SYK exhibited relatively weak correlations with several immune checkpoint genes (Figure 7C).

 Various immunosuppressive cell types, including regulatory T cells (Tregs), myeloid-derived suppressor cells (MDSCs), pro-tumoral macrophages (M2 macrophages), and Cancer-Associated Fibroblasts (CAFs), are known to contribute to tumor immune escape and progression ²¹. Using TIMER2.0, the correlation between infiltration levels of immune-suppressive cells and the expression of SYK, CD74, and MX1, which were prioritized in the present analysis was evaluated. Correlation coefficients for Tregs and M2 macrophages typically ranged from r ≈ 0.2-0.4 across algorithms, whereas negative correlations with CAFs and MDSCs were of similar magnitude but opposite direction (Figure 8 and Figure S6). These results indicate widespread positive correlations between DpOCSC-related hub genes and multiple immune checkpoint molecules, consistent with an immunosuppressive expression context but not establishing causality. 

Table 1. Details of OCSC-associated datasets

Survival analysis and prognostic value of hub genes in OvCa

To evaluate the prognostic significance of DpOCSC-related hub genes, univariate survival analysis for OS and Progression-Free Survival (PFS) using KM Plotter for ovarian cancer. The present analysis included data from all available datasets limited to the serous type, TCGA, and four additional datasets focused on advanced ovarian cancer, as described in table 3. Only those hub genes that showed consistent results across these datasets are report here.

The analysis of OS revealed that SYK and CD74 may contribute to improved OS, whereas MX1 was linked poorer OS (Table 3). In contrast, all three genes, SYK, CD74, and MX1, were associated with reduced progression-free survival in both serous ovarian carcinoma and advanced ovarian cancer. For chemoresistant high-grade serous ovarian cancer patients, SYK and MX1 were associated with better PFS outcomes (Table 3). These context-dependent associations between SYK, CD74, and MX1 and OS or PFS may reflect differences in cohort composition, treatment history, or the limitations of univariate survival analyses, and should therefore be considered exploratory. All survival analyses were performed using univariate log-rank tests in KM Plotter; therefore, these associations should be regarded as hypothesis-generating and require evaluation in multivariable models.

Table 2. Enrichment analysis of hub genes in KEGG pathways and MSigDB Hallmark 

 Results are limited to OvCa-related terms. 

SYK as a biomarker reveals higher diagnostic value in OvCa

SYK, CD74, and MX1 were evaluated as biomarkers for 6‑month recurrence‑free survival. SYK alone approached significant diagnostic performance in any‑ and platinum‑based chemotherapy, with AUC > 0.7 under paclitaxel (PTX) (Figure 9A). Multi‑gene combinations (SYK+CD74, SYK+MX1, SYK+CD74+ MX1) did not outperform SYK alone in any‑ or platinum‑based chemotherapy. Under PTX, the SYK–CD74 combination improved diagnostic performance (Figure 9B), suggesting utility for predicting 6‑month relapse in this context. 

Clinico-pathological evaluation of hub genes

To assess the clinical relevance of DpOCSC-related hub genes, we examined their associations with selected clinicopathological variables in TCGA ovarian cancer data using chi-square tests of independence. SPI1, ITGB2, PLEK, C1QA, and CXCR4 were significantly associated with lymphatic invasion, whereas ITGB2, PLEK, C1QA, and CXCR4 were also associated with venous invasion (Table 4). SYK was associated with new neoplasm event type, and SYK, TNF, and MX1 were associated with primary therapy outcome success. Because these analyses were based on dichotomized expression values and univariate categorical comparisons, the findings should be considered exploratory. Association for non-supported DpOCSC-related hub genes with these features are presented in Table S6.

Given the clinical challenges posed by ovarian cancer, especially recurrence and chemoresistance, these associations may be relevant to disease progression and treatment response. However, because these analyses were based on dichotomized expression values and univariate comparisons, they should be interpreted as exploratory and hypothesis-generating rather than as evidence of causality. 

Predicted upstream regulatory context of DpOCSC-related hub genes

To explore potential upstream regulatory contexts, we analyzed predicted transcription factor binding for the TCGA-supported hub genes using six signaling pathways relevant to ovarian cancer stem cells: STAT3, TGFβ/Smad2/3/4, Hippo-YAP1/TAZ-TEAD, NF-κB, Notch/RBPJ, and Wnt/β-catenin-TCF/LEF (13). TF–target relationships were compiled from ChEA 2022, ENCODE TF ChIP-seq 2015, TRUST 2019, TRANSFAC, and JASPAR PWM databases (Table 5; Supplementary Table S7). Among the inferred regulators, Wnt/β-catenin signaling was the most frequently predicted, with LEF1, TCF3, TCF4, and TAL1: TCF3 motifs detected in the promoters of SYK, CD74, ITGB2, PLEK, TNF, RAC2, MX1, and C1QA across TRANSFAC, JASPAR, and JASPAR PWM human 2025. NF-κB was also strongly implicated, with NFKB1/NFKB2 sites predicted for CD74, TNF, C1QA, and CXCR4 in TRUST 2019 and TRANSFAC/JASPAR. Hippo/YAP1-TEAD components (TEAD1-4) showed predicted interactions with SYK, ITGB2, TNF, RAC2, C1QA, and CXCR4, whereas Notch (RBPJ) and STAT3 appeared less frequently, targeting SYK, CD74, CXCR4, and ITGB2. TGFβ/Smad2/3/4 predictions were sparse, limited to RAC2 (SMAD4 via ChEA 2022 in A2780 cells), MX1 (SMAD4 via TRANSFAC/JASPAR), and ITGB2/CXCR4 (SMAD2 via JASPAR PWM). These predictions should be interpreted cautiously, because several source ChIP-seq datasets derive from non-ovarian and non-CSC cell lines, and the motifs themselves are database-based inferences rather than direct experimental evidence in DpOCSCs. Thus, while these analyses suggest possible involvement of Wnt/β-catenin, NF-κB, Hippo, and, to a lesser extent, TGFβ and Notch pathways, they remain hypothesis-generating. Experimental follow-up, such as ChIP-qPCR, reporter assays, or CRISPR-based perturbation in DpOCSC or relevant ovarian cancer models, will be required to confirm direct transcriptional regulation.

Table 3 Association of mRNA- and protein-evaluated hub genes with overall survival and progression-free survival 

 HGSOC: High-Grade Ovarian Serous Carcinoma.

Table 4. Association of TCGA-supported hub genes with clinico-pathological features 

 a: Yes/No, b: Metastatic/Locoregional disease/Progression of disease/ Recurrence, c: Partial remission/Stable disease/ Progressive disease/Complete remission (* p<0.05, ** p<0.01, *** p<0.001, **** p<0.0001).

Table 5. Predicted modulation of TCGA-supported hub genes by signaling pathways in OCSCs 

 Note: Predictions were generated using motif- and database-based resources, and one ChIP-seq-supported interaction was identified in A2780 ovarian cancer cells. A2780 is an ovarian cancer cell line derived from an untreated patient. These results are predictive and require experimental evaluation in DpOCSCs.

 Discussion

OCSCs are a biologically important subpopulation within ovarian tumors and have been implicated in tumor initiation, metastasis, chemoresistance, and recurrence ^22,23. In this context, stem-like EpCAM/ CD45 dual-positive ovarian cancer cells isolated from malignant ascites may represent a rare and aggressive subpopulation associated with recurrent disease, although their biological origin and emergence remain incompletely defined ¹¹. Recent pilot data further suggested that circulating EpCAM⁺/CD45⁺ extracellular vesicles were associated with platinum response in high‑grade serous ovarian cancer ²⁴. This recent study may suggest that post‑chemotherapy increases in EpCAM⁺CD45⁺ EVs may not only relate to tumor cell death and debris release but could also be influenced by therapy‑associated enrichment of resistant EpCAM⁺CD45⁺ tumor subsets and by immune-tumor interactions that generate EVs carrying both epithelial and leukocyte markers ²⁴. Ascites may further shape this biology by providing a permissive and immunologically active microenvironment that supports tumor adaptation and stemness-associated programs ^25,26.

The present transcriptomic analysis suggests that this DpOCSC population is enriched for inflammatory and immune-related signaling. The prioritized hub genes were linked to interferon- and TNF/NF-κB-associated pathways, and several genes showed positive associations with immune-cell subsets and immune-related markers, consistent with the highly immunomodulatory nature of the ovarian cancer ascites microenvironment ²⁵. Rather than implying a specific mechanism, these findings support a model in which DpOCSCs occupy an immune-inflamed and therapy-resistant state.

Among the prioritized hub genes, SYK, CD74, and MX1 were particularly notable because they integrated inflammatory signaling, immune regulation, and clinical associations in OvCa. SYK is a non-receptor tyrosine kinase involved in immune signaling and is being explored as a therapeutic target in malignancies ¹⁵. In this analysis, SYK was overexpressed and frequently amplified in ovarian tumors, was associated with immune-related marker expression and immune-cell infiltration and showed exploratory associations with overall and progression-free survival, as well as AUC values around 0.7 for short-term relapse prediction. These results are compatible with emerging evidence that SYK contributes to chemoresistance and may regulate stem-like properties in ovarian cancer, but our data remains correlative and hypothesis-generating ^27-29. Recent work also showed that SYK overexpression promotes resistance to DNA-targeted therapy in high-grade serous ovarian cancer, further supporting its relevance to aggressive disease biology ³⁰. CD74, the invariant chain and receptor component for MIF signaling, is overexpressed in ovarian cancer and has been linked to poor clinical outcome and invasion ^16,17,31. CD74 can engage PI3K/Akt and ERK/MAPK pathways and may promote survival programs relevant to chemoresistance, EMT, and spheroid maintenance ^32,33. MX1, an interferon-induced GTPase, was similarly overexpressed, enriched in interferon-response modules, and has been associated with poor outcome or treatment resistance in prior cancer studies ^18,34. Spatial profiling data from ovarian cancer further supports an interferon-rich and immune-suppressive microenvironment, with MX1 appearing within that broader interferon-stimulated context ³⁵. Together, these findings support a model in which SYK, CD74, and MX1 participate in an inflammatory, immune-modulated transcriptional state associated with DpOCSC biology.

The observed copy-number patterns are consistent with an association between genomic alteration and aggressive biology. Prioritized DpOCSC-related hub genes were more amplified, whereas genes outside this core set more frequently harbored deep deletions, a pattern consistent with prior reports linking copy-number changes to chemoresistance in ovarian cancer ^36,37. Although our cross-sectional cBioPortal analysis cannot establish causality, it is consistent with the idea that inflammatory signaling, genomic instability, and aggressive disease biology may co-occur in DpOCSC-enriched tumors.

The prioritized hub genes also showed associations with invasion-related features. Enhanced migratory and invasive behavior is a recognized property of aggressive OCSC populations, and in this analysis several hub genes correlated with lymphatic and venous. This observation is consistent with prior evidence that EpCAM⁺/CD45⁺-positive ovarian cancer cells can exhibit invasive and stem-like traits ¹¹. 

Conclusion 

Taken together, the literature and the present in silico results point to SYK-associated inflammatory signaling as a candidate pathway for further investigation in DpOCSCs. However, the current study does not establish direct functional roles for SYK, CD74, or MX1 in stemness, and their elevated expression may reflect an aggressive immune-inflamed state rather than a direct contributing mechanism. Functional studies in DpOCSCs will be required to determine whether these genes regulate stem-like behavior, invasion, or therapy resistance.

This study has limitations that should be considered when interpreting the findings. The DpOCSC group was small, and the analyses were therefore exploratory. In addition, the use of public datasets from different platforms and cohorts may introduce residual heterogeneity despite normalization and batch correction. The observed survival, immune, and clinicopathological associations are therefore best regarded as supportive signals that warrant evaluation in larger cohorts. Moreover, the EpCAM⁺/CD45⁺ phenotype identifies a distinct ascites-derived cell population, but its precise biological origin and functional significance remain to be clarified. Future experimental work will be essential to define the mechanistic roles of SYK, CD74, MX1, and related hub genes in DpOCSC biology.

In summary, comparative transcriptomic analysis of stem-like DpOCSCs versus other ascites-derived OCSC populations identified an interferon- and TNF-dominated inflammatory module containing 17 up-regulated hub genes. Among these, SYK, CD74, and MX1 were consistently overexpressed in our analyses and may contribute to frequent copy-number gain, immune-related features, and exploratory clinical outcome metrics in large OvCa cohorts (Figure 10). These findings nominate SYK-associated inflammatory signaling as a candidate axis for mechanistic and translational studies in recurrent, ascites-rich ovarian cancer. 

Data availability statement

The datasets analyzed during the current study are available in the National Center for Biotechnology Information (NCBI)- Gene Expression Omnibus (GEO) database (https:/www.ncbi.nlm.nih.gov/geo/) with accession numbers GSE75036, GSE90125 and GSE216627, Enrichr database (https://amp.pharm. mssm.edu/Enrichr/), Bioconductor package (v3.17) for R (version 4.3.1) (https://www.r-ptoject.org), GraphPad Prism9 software (https://www.graphpad. com), SRplot online tool (https://www.bioinformatics. com.cn/en), Volcano plot online tool (https://huygens. science.uva.nl/VolcaNoseR2/), String (v1.7.1), CentiScape (v2.1) and Cytohubba (v0.1) plugins for Cytoscape software (v3.10.0) (https://cytoscape.org/), GEPIA database (http://gepia.cancer-pku.cn), UALCAN database (https://ualcan.path.uab.edu/), cBioPortal database (http://cbioportal.org), KM plotter database (https://kmplot.com), ROC Plotter database (https://rocplot.org), UCSC Xena Browser database (http://xenabrowser.net), Cut-off Finder online tool (https://molpathoheidelberg.shinyapps.io/CutoffFinder_v1/), Timer 2.0 database (http://timer.cistrome.org/) and the R code generated for clinco-pathological analysis on Github (https://github.com/sadra-daneshvari/OvarianCancerProject). All data generated and analyzed in this study can be found in these databases. 

Conflict of Interest 

The authors declare no competing interests.

Funding: This work was supported by grant #268308/k6/034 by university of Tehran.