Exploring Molecular Signature and Prognostic Biomarkers in Ovarian Cancer: Insights From Late‐Stage, Recurrent, and Metastatic Tumors

ABSTRACT

Ovarian Cancer is a leading cause of mortality among women globally, primarily due to lack of specific and sensitive early‐stage diagnostic tools. This study aims to identify hub genes associated with recurrent, late‐stage, and metastatic tumors as potential prognostic biomarkers and drug targets. Gene expression data from eight National Center for Biotechnology Information (NCBI)–Gene Expression Omnibus (GEO) datasets were categorized by recurrence, tumor‐stage, and metastasis. Differential gene expression and enrichment analyses were performed. Hub genes were identified by protein–protein interaction networks and validated by the University of Alabama at Birmingham Cancer Data Analysis Portal (UALCAN), GEPIA2, pROC, and Kaplan–Meier plotter databases. Genetic alterations, immune cell infiltration, miRNA prediction, and drug–gene interactions were assessed using cBioPortal, CIBERSORTx, Encyclopedia of RNA Interactomes (ENCORI), and Drug–Gene Interaction Database (DGIdb), respectively. Eight hub genes (FN1, COL1 A1, COL1A2, COL3A1, POSTN, LUM, IGF1, and CXCL8) were identified, with COL1A2 common across all tumor categories. Note that 19.6% of cases showed mutations in these genes, primarily COL3A1. Overexpression of most hub genes and reduced expression of CXCL8 correlated with worse survival outcomes. COL1A1 and FN1 showed strong diagnostic ability. Late‐stage tumors showed elevated M2 macrophages and neutrophils. hsa‐miR‐29a‐3p, hsa‐miR‐29b‐3p, and hsa‐miR‐29c‐3p were identified as the most interactive miRNAs. Ocriplasmin and pamidronate were identified as potential therapeutics. Our findings highlight the therapeutic relevance of these hub genes and identify them as potential drug targets and prognostic biomarkers in ovarian cancer.

Abbreviations

DEGs

differentially expressed gene

GEO

Gene Expression Omnibus

GO

Gene Ontology

KEGG

Kyoto Encyclopedia of Genes and Genomes

LIMMA

Linear Models for Microarray Data

NCBI

National Center for Biotechnology Information

OC

ovarian cancer

OS

overall survival

PPI

protein–protein interaction

TCGA

The Cancer Genome Atlas

UALCAN

University of Alabama at Birmingham Cancer Data Analysis Portal

1. Introduction

Ovarian cancer (OC) is the fifth most prevalent cause of death among gynecological cancers in women globally [1]. Throughout the last few decades, significant research has been conducted in the field of cancer diagnosis, prognosis, and therapy. Early detection of cancer and cutting‐edge treatment strategies have improved the survival of cancer patients [2]. Despite the improved understanding, the rate of 5‐year survival for women diagnosed with OC is still very low, making it a very fatal disease. OC has a dismal prognosis due to treatment resistance and a high recurrence rate [3]. In the majority of cases, OC is detected at a very late stage (International Federation of Gynaecology and Obstetrics III‐IV) and metastasizes to the abdomen. Nonsignificant early symptoms and a lack of efficient screening methods contribute to this late detection, leading to poor patient outcomes [4]. Despite optimal initial treatment, including surgery and chemotherapy, 70%–90% of patients with progressed OC still experience disease recurrence. Recurrent OC poses significant treatment challenges, particularly due to resistance to platinum‐based chemotherapy, the standard initial treatment. Research into resistance mechanisms and innovative treatment strategies is essential to improve outcomes for these patients [5]. Metastatic disease, either at diagnosis or during recurrence, is related to a significant decrease in survival rates.

Transvaginal ultrasound and quantitation of serum CA125 biomarker levels are the currently available methods for the diagnosis of OC. However, both these methods lack specificity and sensitivity [6]. Elevated levels of serum CA125 are observed only in approximately half of the patient population with Stage I disease. Diagnosis of the disease using this biomarker has been challenging in pre‐ and perimenopausal patients as elevated levels have been observed in other gynecological conditions like ovarian cysts, endometriosis, fibroids, etc. [7]. Identifying more sensitive and specific biomarkers and drug targets of OC is important as the lack of these majorly contributes to the high mortality rate associated with this condition.

There has been a remarkable progress in understanding tumor biology with the development of bioinformatics techniques and next‐generation sequencing (NGS), in recent years. These sequencing technologies combined with advanced computational analysis, such as gene network analysis, detection of variants, and prognostic and prediction model analysis, enable us to understand intricate mechanisms involved in the development of cancer in a relatively short time [8]. The National Center for Biotechnology Information (NCBI) (https://www.ncbi.nlm.nih.gov/) and The Cancer Genome Atlas (TCGA) (https://portal.gdc.cancer.gov/) databases are important public repositories for cancer biomarker identification. These databases provide an extensive collection of standardized gene expression data from diverse cancer studies for accessing and analyzing a wide range of datasets encompassing different cancer types, stages, and treatment responses that allow comprehensive analyses [9].

Advanced technologies like NGS have significantly increased the research in identifying novel biomarkers for OC [10]. Several researchers have demonstrated the relevance of NGS in the identification and quantification of circulating DNA in OC patients [11]. A study by Vogel et al. highlighted the importance of mutations in genes associated with homologous recombination deficiency (HRD) for early screening of treatment‐resistant OCs [12]. Drug resistance in OC has also been attributed to loss of function variation in genes like BRCA1, BRCA2, RAD51D, and RAD51C, which are involved in HDR pathways [13, 14]. Similarly, mutations in TP53 have also been linked with resistance to platinum‐based chemotherapeutic drugs in serous ovarian carcinoma [15].

Researchers have analyzed the genomic data from publicly available sources like TCGA and NCBI to uncover differential gene expression signatures and co‐expression networks. Recently, high throughput sequencing data such as RNA sequencing have been used widely to identify sensitive biomarkers and gene signatures linked with various hallmarks of cancers. In a recent study, Yu et al. identified two potential biomarkers—PIK3R1 and SPP1 for lung adenocarcinoma [16]. Chen et al. used microarray and RNA seq datasets, to construct a predictive gene signature of nine genes to determine survival outcomes in colorectal cancer patients [17]. In yet another study, researchers used RNA seq data from TCGA and identified four genes—CXCL8, IRAK2, REG1A, and EGF as potential biomarkers in OC [18]. Wu et al. identified a seven gene signature (LRRC17, KIAA2022, ZHX3, AKR1B10, CD38, LYPD6B, and CMBL) linked to platinum resistance and predicted survival consequences for patients with high‐grade serous OC [19].

Microarray facilitates the study of the expression of thousands of genes simultaneously and provides a thorough understanding of their behavior. In molecular oncology, it has been used as an efficient tool for studying gene expression patterns, gene regulation, and molecular mechanisms underlying tumorigenesis and development. Several types of biomarkers such as coding genes, miRNAs, long‐noncoding RNAs, and circRNAs have been studied by researchers in OC through in silico analysis of microarray datasets. In one such study, researchers identified CDCA5, FOXM1, KIF15, MCM2, and ZWINT genes to be linked with progressed stage and worse prognosis of epithelial OC [20]. Zheng et al. identified FGF13 as a predictive biomarker for survival outcomes of OC patients and CDC20, KIF20A, CCNB1, BUB1B, BIRC5, CAV1, CFH, and MEIS2 as putative diagnostic biomarkers using integrated in silico analysis technique [21]. In epithelial OC, Gui et al. identified CDCA5 and ESPL1 as hub genes, predicted to have a significant role in tumor progression [22]. Collagen family genes such as COL11A1, COL1A1, and COL1A2 and THBS1 and THBS2 have been linked with critical pathways like extracellular matrix (ECM) receptor interaction, PI3K‐AKT transduction, and cell adhesion, highlighting their role in metastasis of OC [23]. Similarly, in a recent study, Liang et al. identified a six gene signature comprising KIF26B, VSIG4, COL6A6, FOXJ1, MXRA5 and CXCL9 that is crucial for migration and invasion of tumor cells [24].

The complexities of late‐stage, recurrent, and metastatic OC significantly impact patient outcomes, treatment strategies, and research directions. To the best of our knowledge, there are few reports on identifying molecular gene signatures for the prognosis of late stage, recurrent, and metastatic tumors through analysis of microarray data. In the current study, we have identified common differentially expressed genes that are relevant in disease progression, recurrence, and metastasis. Here, we have analyzed 809 OC patient samples from eight microarray datasets that are available in the NCBI Gene Expression Omnibus (GEO) database. These datasets were classified into three categories based on disease stage (stages 1 and 2 were classified as early, while stages 3 and 4 as late), the occurrence of tumor (nonrecurrent tumor and recurrent tumor), and the migration of tumor (primary tumor and metastatic tumor). These classification categories represent important events in disease progression and provide insights into molecular mechanisms, disease relapse, and treatment responses. Differential gene expression (DEG) analysis for three categories was carried out to identify the gene expression profiles. Further, hub genes were explored and validated to study their potential prognostic and diagnostic role, correlating their expression patterns with tumor stage, epigenetic modifications, and patient survival outcomes. Tumor immune infiltration analysis was performed to understand how immune modulation influences tumor progression, recurrence, and metastasisby comparing immune cell composition across three categories.Potential microRNAs were predicted for hub genes, and drug–gene interaction was studied.

2. Materials and Methods

2.1. Microarray Gene Expression Profiles

The gene expression data and the clinical information regarding recurrence, tumor stage, and metastasis of patients diagnosed with OC were retrieved from the NCBI GEO. Microarray data of 809 OC patient samples were extracted from eight GEO datasets [25]. The datasets were classified into three categories: nonrecurrent and recurrent tumors, early‐ and late‐stage tumors, and primary and metastatic tumors. These categories were chosen as they significantly affect the therapeutic modality and patient survival. The datasets used in the study, the number of samples in each of these datasets, and the clinical feature for which it was analyzed are mentioned in Table 1. The framework of this study is represented in Figure S1. The comprehensive list of all samples used for the study has been provided in Supporting Information S2.

TABLE 1.

Details of microarray datasets employed in the study.

GEO accession	Number of samples used in the study	Categories of samples used in the study	Description	References
GSE44104	60	Nonrecurrent = 40 Recurrent = 20	Nonrecurrent tumor—Cancerous growth that has not reappeared or returned. Recurrent tumor—Tumor that has reappeared after a period of remission following treatment.	[26]
GSE17260	110	Nonrecurrent = 34 Recurrent = 76		[27]
GSE32063	40	Nonrecurrent = 13 Recurrent = 27		[27]
GSE14764	80	Early stage = 9 Late stage = 71	Early stage—refers to initial phase of disease, encompasses stages I and II. Late stage—refers to advanced progression of disease, encompasses stages III and IV.	[28]
GSE49997	194	Early stage = 9 Late stage = 185		[28]
GSE9899	291	Early stage = 42 Late stage = 249		[29]
GSE73168	16	Primary = 8 Metastatic = 8	Primary tumor—The initial site where cancer cells first form and develop into a cancerous mass. Metastatic tumor—A cancerous growth that has spread from its original site to other parts of the body.	[30]
GSE30587	18	Primary = 9 Metastatic = 9		[30]

2.2. Identification of DEGs

Gene expression in recurrent, late‐stage, and metastatic tumors was analyzed and compared with those in nonrecurrent, early‐stage, and primary tumors, respectively, to identify the DEGs. Microarray datasets were analyzed using R software version 4.2.1 (https://cran.r‐project.org/bin/windows/base/old/4.2.1/). Gene expression profiles were retrieved from NCBI‐GEO using GEOquery as processed series matrix files provided by the original studies [31]. Probes were mapped to gene symbols and collapsed to one probe per gene by retaining the highest variance probe. Values were log₂‐transformed where necessary, and no additional within‐ or between‐array normalization was performed. Differential expression was then analyzed with Linear Models for Microarray Data (LIMMA) package using linear modeling and empirical‐Bayes moderation [32]. A log fold‐change (logFc) threshold of greater than 1 was used to identify upregulated genes, while a threshold of less than −1 was used for downregulated genes and statistical significance was determined with a p‐value < 0.05 [33].

2.3. Topological and Functional Enrichment Analysis

The biological functional roles of most interacting DEGs of each category of samples were determined through functional enrichment study performed using the g:Profiler tool ( https://biit.cs.ut.ee/gprofiler/) [34]. Gene Ontology (GO) analysis encompassed categories of cellular components, molecular functions, and biological processes. Enrichment of molecular pathways was assessed using ShinyGO 0.80 (http://bioinformatics.sdstate.edu/go/), and statistical significance was determined (p <0.05) [35].

Topology‐based pathway analysis was carried out using signaling pathway impact analysis (SPIA) on Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways to identify biologically relevant signaling pathways based on gene orientation and interactions. SPIA integrates enrichment significance(pNDE) and perturbation (pPERT), accounting for pathway topology. DEGs from each category (late vs. early, recurrent vs. nonrecurrent, and metastatic vs. primary) were mapped to Entrez IDs, and pathways with FDR < 0.05 were regarded as significant. The top 20 enriched pathways for each category were represented as −log₁₀(FDR) bar plots [36].

2.4. Analysis of Protein–Protein Interaction (PPI) Network

The STRING database (http://string‐db.org/) was used to develop a functional protein association network, and Cytoscape (version 3.9.1) software (https://cytoscape.org/) was used to visualize and analyze the network. The DEGs in each datasets were analyzed employing the STRING database and Cytoscape [37]. The initial step involved selecting the DEGs with the maximum degree of interaction in the Cytoscape network analysis from each dataset and the ones with the highest number of connections were chosen. These selected DEGs were merged and subjected to further analysis using the STRING database and Cytoscape plugin Network Analyzer, employing the node degree method to explore gene interactions.

2.5. Hub Genes Identification and Validation

The hub genes for all three categories were identified through the dual metric approach of betweenness centrality and node degree method using the plug‐in of CytoHubba (https://apps.cytoscape.org/apps/cytohubba) from cytoscape [38].

A validated University of Alabama at Birmingham Cancer Data Analysis Portal (UALCAN) database (https://ualcan.path.uab.edu/) was used to study the correlation between the expression of intersecting hub genes and the stage of tumor. It is a web‐based, intuitive platform designed for data analysis of gene expression, tumor stage, node metastasis, and survival, based on TCGA data across various cancer types [39]. Additionally, hub genes expression was cross‐validated in GEPIA2 (http://gepia2.cancer‐pku.cn/) using TCGA tumor and GTEx normal ovary tissue samples, and summary boxplots are provided in Figure S2 [40].

2.6. Analysis of Genetic and Epigenetic Modifications in Hub Genes

The genetic mutations associated with identified hub genes in OC were analyzed using the cBioPortal for Cancer Genomics (http://www.cbioportal.org/). Specifically, the analysis focused on the ovarian serous cystadenocarcinoma (TCGA and PanCancer Atlas) cohort comprising 585 patient samples. The genetic alterations evaluated included point mutations, structural variants, copy number variation (CNV), and the effect of mutations on the survival outcomes of the OC patients. DNA methylation levels were analyzed using merged illumina HM27 and HM450k dataset in cBioPortal, restricting to only primary tumors with paired RNA‐seq (RNA‐Seq V2 RSEM) and methylation data. The relationship between gene expression and DNA methylation was assessed using the portal's “mRNA vs. Methylation” tool (Pearson and Spearman statistics). The visualization and interpretation of these genomic profiles provided insights into the mutational and epigenetic landscape of the hub genes in OC [41].

2.7. Prognostic Significance and Diagnostic Performance of Hub Genes

The prognostic relevance of the identified hub genes was evaluated using the Kaplan–Meier plotter (KM plotter) tool (http://kmplot.com/) [42]. Patients were divided into high and low gene expression cohorts to study the association between gene expression levels and patient survival outcomes. We analyzed overall survival (OS), progression free survival (PFS), and recurrence free survival (RFS). Survival differences were assessed with the log‐rank test, and hazard ratios (HRs) with 95% confidence intervals were obtained from Cox proportional‐hazards models, and p‐value < 0.05 was considered meaningful prognostic associations. To assess the diagnostic efficacy of the hub genes in OC, we generated receiver operating characteristic (ROC) curves using the pROC package in R and calculated the sensitivity and specificity of the area under the ROC curve (AUC). Hub genes with AUC > 0.5 and p‐value < 0.05 were identified as potential diagnostic biomarkers [43].

2.8. Analysis of Immune Cell Infiltration in Tumor Categories

Tumor infiltration analysis was conducted using gene expression and immune cell fraction data from three categories: early versus late stage tumor (GSE14764), nonrecurrent versus recurrent tumor (GSE32063), and primary versus metastatic tumor (GSE73168). This categorization allowed for an in‐depth analysis of immune infiltration across different tumor types. The proportions of 22 immune cell types in each category were quantified using CIBERSORTx, a digital cytometry tool that estimates the fractions of immune cell types from gene expression profiles [44]. The Mann–Whitney U test was employed to evaluate statistically significant differences in immune cell infiltration between groups. For visualization, bar plots were generated to depict the average relative abundance of immune cells across the three categories.

Gene expression levels of eight hub genes were obtained from NCBI‐GEO datasets for each category. The correlation between hub gene expression and immune cell infiltration was assessed using Spearman's rank correlation coefficient. To determine statistical significance, p‐values were calculated, and correlations with p‐values less than 0.05 were considered statistically significant. For visualization, bar plots were generated to depict the average relative abundance of immune cells across the three categories and box plots and scatter plots were generated to compare the tumor infiltration levels of immune cells across the different categories [45].

2.9. Construction of the miRNA–Hub Gene Interaction Network

The Encyclopedia of RNA Interactomes (ENCORI) platform was utilized to predict miRNA–target interactions (http://starbase.sysu.edu.cn/; v3.0). Target miRNAs were selected based on predictions from at least three prediction algorithms, including RNA22, PITA, miRmap, PicTar, TargetScan, miRanda, and microT. Cytoscape was used to construct and display the network illustrating regulatory interactions between miRNAs and hub genes. Additionally, the co‐expression relationships between selected miRNAs and their corresponding hub gene's mRNA were analyzed using the Pearson correlation coefficient (r) in ENCORI. A statistically significant negative correlation was defined as a correlation coefficient (r < 0) with a p‐value less than 0.05 [46].

2.10. Drug–Hub Gene Interaction Analysis

Drug interactions with the hub genes were evaluated using Drug–Gene Interaction Database (DGIdb, v3.0.2). DGIdb (http://www.dgidb.org/search_interactions) integrates information from various sources, including DrugBank, ChEMBL, PharmGKB, NCBI Entrez, Ensembl, and NCBI PubMed. The hub genes and candidate drugs interaction network was constructed and visualized using Cytoscape. To ensure clinical relevance, only Food and Drug Administration (FDA)‐approved drugs were considered for downstream analysis [47]. This approach enabled the identification of potential candidate drugs for future therapeutic interventions targeting hub genes in OC.

3. Results

3.1. DEGs Identification in OCs

In all the analyzed datasets, a total of 2387 DEGs were observed. Of these, 697 genes showed differential expression in recurrent tumor samples as compared to the nonrecurrent ones. In this category, the maximum number of DEGs were observed in GSE44104, followed by GSE32063 and GSE17260. In late‐stage tumors, 965 DEGs were observed as compared to the early‐stage samples, with the maximum number observed in GSE14764, followed by GSE49997 and GSE9899. Metastatic tumors showed 725 DEGs as compared to primary tumors, with GSE73168 showing the maximum number. Detailed information about the number of DEGs in all the datasets is presented in Table 2 and Supporting Information S3 has a complete list of all the genes that are significantly differentially expressed in all the datasets. The volcano plots illustrating the DEGs for each category are displayed in Figure 1.

TABLE 2.

DEGs identified in different categories of the analyzed samples.

Categories	GEO datasets	Total number of DEGs	Upregulated genes	Downregulated genes
Nonrecurrent and recurrent tumor	GSE44104	503	271	232
	GSE17260	53	9	44
	GSE32063	141	68	73
Early‐stage and late‐stage tumor	GSE14764	558	349	209
	GSE49997	225	219	6
	GSE9899	182	102	80
Primary and metastatic tumor	GSE73168	656	345	311
	GSE30587	69	65	4

FIGURE 1.

Determination of genes with differential expression (DEGs) among each category. The volcano plot shows DEGs in (A) nonrecurrent and recurrent tumors, (B) early‐stage and late‐stage tumors, and (C) primary and metastatic tumors. The red‐colored dots indicate genes with elevated expression, while the blue‐colored dots indicate genes with low expression. The genes with no change in expression profiles are represented by black‐colored dots. The ggplot2 package from R Bioconductor was used to draw volcano plots.

3.2. Hub Genes Identification Through Analysis of PPI Network

The PPI networks were created using the top 100 DEGs for each category of the sample, leading to the identification of top 10 hub genes (Figure 2). A multiprotein interaction network of 100 nodes and 743 edges with a maximum node degree of 72 was displayed by the genes that were differentially expressed in recurrent tumors (Figure 2A). Similarly, the PPI network from DEGs in late‐stage tumors showed 100 nodes and 1277 edges with a maximum node degree of 81 (Figure 2B) while that in metastatic tumors represented 100 nodes and 1034 edges with highest node degree of 46 (Figure 2C). IL6, IGF1, STAT1, CXCL8, ESR1, COL1A2, CXCL10, CXCL1, CXCL2, and BMP2 were identified as hub genes in recurrent tumor samples (Figure 2A). FN1, COL1A1, COL1A2, POSTN, COL3A1, THBS2, IGF1, CXCL8, LUM, and DCN were identified as hub genes in late‐stage tumor (Figure 2B) while COL1A1, FN1, COL1A2, COL3A1, POSTN, ITGAM, LUM, MMP2, FCGR3A, and TLR2 were identified as hub genes in primary and metastatic tumor (Figure 2C).

FIGURE 2.

PPI network of top 100 DEGs showing top 10 hub genes for (A) recurrent and nonrecurrent tumor, (B) early‐ and late‐stage tumor, (C) primary and metastatic tumor, (D) Venn diagram illustrating intersecting hub genes among nonrecurrent and recurrent tumor, early‐ and late‐stage tumor, and primary and metastatic tumor. Magenta color circular nodes represent 10 hub genes, light pink color elliptical nodes represent DEGs interacting with hub genes, and Gray color edges represent the relationship between genes in the network based on first‐stage nodes, shortest path, and expended subnetwork.

A dual approach of degree method and betweenness centrality was employed to identify the top 10 hub genes for each of the three categories. Moreover, the common hub gene COL1A2 was detected among all three categories. FN1, COL1A1, COL3A1, POSTN and LUM were common hub genes between metastatic tumor and late‐stage tumor and IGF1 and CXCL8 were common hub genes between recurrent tumor and late‐stage tumor (Figure 2D).

3.3. Functional Enrichment Study of DEGs

DEGs of each category were uploaded to the online software g:Profiler and ShinyGO 0.80 for GO and pathway enrichment analysis, respectively. It was observed that all functions that were enriched were involved in various hallmarks of cancer (Figure 3).

FIGURE 3.

Gene ontology and pathway analysis of differentially expressed genes. (A) Nonrecurrent and recurrent tumor, (B) early‐ and late‐stage tumor, and (C) primary and metastatic tumor.

3.3.1. Nonrecurrent and Recurrent Tumors

The DEGs in recurrent tumors were markedly enriched with molecular function related to receptor binding. The functions such as signaling receptor binding, signaling receptor activator activity, cytokine, G protein‐coupled, and chemokine receptor binding were found to be enriched. The biological processes enriched included response to various stimuli such as cytokine, organic substances, and other external factors. Activation of immunologic and defense responses was also found to be enhanced. Extracellular region, matrix, and space were found to be prominently enriched among the cellular components. Several pathways associated with cancer and immune receptor interactions were also found to be significantly enriched (Figure 3A).

3.3.2. Early‐ and Late‐Stage Tumors

The advancement of the cancer stage is facilitated by the enriched molecular functions and biological processes like ECM structural constituent and multicellular organismal process, respectively. Other molecular functions enriched include binding of glycosaminoglycan, heparin, and signaling receptor while the biological processes constitute those associated with the development and response to a stimulus. Similar to that observed in recurrent tumors, the ECM‐related genes were among the cellular components that were enriched in late‐stage tumors. Pathway analysis of DEGs showed their enrichment of genes associated with pathways like PI3K Akt transduction signaling and cancer disease progression (Figure 3B).

3.3.3. Primary and Metastatic Tumors

The spread of the disease is promoted by enriched molecular functions like extracellular matrix structural constituent and binding of integrin, glycosaminoglycan, and collagen. Key cellular processes include cell adhesion, regulation of multicellular organismal processes, collagen fibril organization, etc. The DEGs in metastatic tumors were prominently involved in ECM, cell surface, and vesicles. KEGG pathway analysis of DEGs indicates their significant role in phagosome formation (Figure 3C).

3.3.4. Topology‐Based Pathway Analysis (SPIA)

SPIA highlighted distinct pathway perturbations corresponding to the progressive molecular changes observed across tumor stages and conditions. In late‐stage tumors, activation of the ECM–receptor interaction and focal adhesion pathways indicated enhanced ECM remodeling and adhesion process. In recurrent tumors, strong activation of chemokine/cytokine, NF‐κB, and MAPK/mTOR pathways highlighted inflammatory and receptor‐mediated signaling.

In contrast, metastatic tumors showed suppression of innate immune and cell trafficking pathways, including leukocyte transendothelial migration and FcγR‐mediated phagocytosis, alongside activation of developmental signaling such as Notch and Wnt. Overall, SPIA results emphasized a shift from matrix remodeling in late stages to inflammatory signaling in recurrence and reduced immune surveillance with pro‐metastatic signaling in metastatic state. The bar plots of significant pathways for all three categories are presented in Figure S2, and the complete SPIA output tables are provided in Supporting Information S4.

3.4. Validation of Identified Hub Genes Expression

The association between the expression pattern of eight intersecting hub genes (FN1, IGF1, COL1A2, COL1A1, COL3A1, POSTN, LUM, and CXCL8) with tumor stages was analyzed using UALCAN to further validate their significance. UALCAN database, a web‐based tool, allows researchers to access and analyze the TCGA data, providing understandings into gene expression patterns, survival analysis, and clinical characteristics of various cancer types [48]. UALCAN facilitates the exploration of differential gene expression between normal and tumor tissue and among different tumor subtypes and stages.

In recurrent tumors, upregulation of the genes IGF1 and COL1A2 was observed, while the expression of COL1A1, COL3A1, POSTN, and FN1 remained unaltered. CXCL8 exhibited low expression in recurrent tumor while no significant change was observed in late‐stage and metastatic ovarian tumors. Conversely, significant upregulation of FN1, COL1A1, COL1A2, COL3A1, LUM, and POSTN was observed in late‐stage and metastatic tumors, according to our DEG analysis.

Further validating their expression patterns in TCGA dataset using UALCAN database demonstrated that the expression level of CXCL8 (p value = 2.86E‐03 [III–IV stage]) was found to be significantly decreasing as cancer progresses and FN1 (p value = 7.40E‐04 [II–III stage], p value = 5.28E‐02 [II–IV stage]), IGF1 (p value = 7.78E‐05 [II–III stage], 4.66E‐02 [II–IV stage]), POSTN (p value = 3.51E‐07 [II–III stage], p value = 7.30E‐03 [II–IV stage]), COL1A1 (p value = 3.36E‐04 [II–III stage], p value = 2.35E‐02 [II–IV stage]), COL1A2 (p value = 6.35E‐04 [II–III stage], p value = 1.81E‐02 [II–IV stage]), COL3A1 (p value = 1.79E‐05 [II–III stage], p value = 8.98E‐03 [II–IV stage]), and LUM (p value = 1.26E‐07 [II–III stage], p value = 1.80E‐03 [II—IV stage]) were significantly correlated with increased expression in stage III and IV and decreased expression in stage II (Figure 4A). Differential expression of these genes was observed with respect to clinical stages of OC. The heat map in Figure 4B illustrates the expression pattern of these eight genes.

FIGURE 4.

(A) Box‐whisker plots and (B) the heatmap represent an expression pattern of eight intersecting hub genes across various clinical stages of ovarian cancer using data from the TCGA. The significance of expression was defined as p‐value < 0.05. **p < 0.05, ***p < 0.001.

Complementing the UALCAN findings, hub gene expression analysis using GEPIA2 showed higher expression of COL1A1, COL1A2, COL3A1, FN1, POSTN, and CXCL8 in tumors, while LUM showed higher expression in normal ovary, and IGF1 displayed no significant difference between the two groups. The box plots showing eight hub genes expression between tumor and normal groups are presented in Figure S3.

3.5. Genomic Alterations and Prognostic Implications of Hub Genes

Genetic mutations are closely related to tumorigenesis, influencing cancer initiation and progression. To explore this, we assessed genomic alteration in the eight identified hub genes in TCGA OV cohort, within OC samples using the cBioPortal database. Our analysis revealed that approximately 19.6% of OC samples exhibited alterations in at least one of the hub genes (Figure 5A). Among these, mRNA amplification emerged as the most dominant alteration, observed in about 11.06% of cases (44 samples).

FIGURE 5.

Alteration frequency and prognosis of the eight hub genes. (A) The summary of the cancer types in the cBioPortal was used to calculate the percentages of OC cases with the eight altered hub genes. (B) Genetic alteration of the eight hub genes COL1A1, COL1A2, COL3A1, FN1, IGF1, CXCL8, POSTN, and LUM in OC patients. (C) OS and (D) PFS of OV patients with altered (red) and unaltered (blue) mRNA expression of the eight hub genes. OC, ovarian cancer; OS, overall survival; PFS, progression free survival.

The results showed that COL3A1 was the most frequently altered (6% altered frequency) gene out of the analyzed eight hub genes in OC samples. Other hub genes displayed alteration frequencies as follows: COL1A1 (2%), COL1A2 (5%), FN1 (2.5%), IGF1 (1.5%), CXCL8 (4%), POSTN (2.8%), and LUM (0.8%) (Figure 5B). CNV analysis using GISTIC threshold calls showed copy number gains/amplifications as the predominant events across the collagen genes (COL1A1/COL1A2/COL3A1) and FN1, while deletions were less frequent and typically shallow; accordingly, tumors with gains tended to have higher mRNA levels and those with deletions lower expression (Figure S4).

Complementing CNV, interrogation of the merged HM27/HM450 methylation dataset in cBioPortal revealed weak‐to‐modest inverse methylation–expression relationships overall among the six evaluable genes (FN1, COL1A1, COL1A2, COL3A1, IGF1, and CXCL8 / IL8), with the clearest negative trends for IGF1 and CXCL8, and minimal or no coupling for the collagen genes and FN1 (Figure S5). POSTN and LUM could not be assessed because mapped CpG probes were not available in this dataset.

These findings suggest a substantial involvement of these genes in the genomic landscape of OC. To assess the clinical relevance of these genetic alterations, we conducted survival analyses comparing PFS and OS between patients with or without alterations. As revealed in Figure 5C,D, OC cases with altered or mutated hub gene expression did not exhibit a significant change in OS (p = 0.624) and PFS (p = 0.463) compared to those with unaltered or non‐mutated hub gene expression. These results suggests that although alterations in hub genes are relatively frequent, their direct impact on survival outcomes in OC require further investigation.

3.6. Prognostic Significance and Diagnostic Performance of Hub Genes

The prognostic relevance of intersecting hub genes was carried out using KM plotter in OC patient samples. It is a widely used online survival analysis tool, containing extensive data across various cancers such as lung cancer, OC, gastric cancer, etc. This tool enables the analysis of survival data, generating log‐rank statistical p‐values and HRs along with an interval of confidence at 95%, all graphically represented on the plot [49].

The patient samples are split into high and low expression cohorts to evaluate a gene's prognostic value. Consistent with an ECM‐remodeling phenotype, COL1A1, COL1A2, COL3A1, FN1, IGF1, POSTN, and LUM showed an adverse association; their higher expression correlated with shorter OS/RFS and PFS, whereas CXCL8 exhibited the opposite trend, higher expression correlated with better OS/RFS/PFS (Figures 6 and 7; Figure S6).

FIGURE 6.

The prognostic significance by overall survival analysis of eight intersecting hub genes, (A) FN1, (B) COL3A1, (C) COL1A1, (D) COL1A2, (E) POSTN, (F) LUM, (G) IGF1, and (H) CXCL8. The red color line represents patient samples with high expression, while the black color line depicts samples with low expression. The X‐axis represents the probability of overall survival, while the Y‐axis shows time interval in months. p‐value < 0.05 was considered significant.

FIGURE 7.

The prognostic significance by recurrence free survival analysis of eight intersecting hub genes, (A) COL1A1, (B) COL1A2, (C) COL3A1, (D) FN1, (E) POSTN, (F) LUM, (G) CXCL8, and (H) IGF1. The blue color line represents patient samples with high expression, while the yellow color line depicts samples with low expression. The X‐axis represents the probability of recurrence‐free survival (RFS), while the Y‐axis shows time interval in months. p‐value < 0.05 was considered significant.

Complementing the prognosis analysis, diagnostic ROC curves generated with pROC (R package) demonstrated per‐gene discrimination between tumor and normal samples, and an eight‐gene ridge‐logistic signature achieved excellent accuracy (AUC = 0.995, 95% CI 0.987–1.000) (Figure S7). Among single genes, COL1A1 showed very good diagnostic ability (AUC = 0.977). Genes with AUC > 0.6, CXCL8 (0.739), IGF1 (0.704), COL1A2 (0.683), and FN1 (0.607) can be considered moderately diagnostic. COL3A1 (0.547) and POSTN (0.525) showed near‐chance discrimination, and LUM (0.359) was not diagnostic (Figure 8).

FIGURE 8.

Diagnostic performance (receiver operating characteristic [ROC]) of hub genes. ROC curves (A–H) for COL1A1, COL1A2, COL3A1, FN1, POSTN, LUM, CXCL8 (IL8), and IGF1 evaluating discrimination of tumor versus normal ovary based on gene expression. The gray dashed line denotes the no‐discrimination reference.

3.7. Tumor Immune Cell Infiltration Analysis

The tumor immune microenvironment was analyzed across three clinical conditions: early and late stage tumor, nonrecurrent and recurrent tumor, and primary and metastatic tumor. Immune cell composition for each of these categories was assessed using CIBERSORTx. The results revealed significant variations in immune cell infiltration between clinical conditions, highlighting their roles in tumor progression and immune evasion. Statistical analyses were conducted using the Mann–Whitney U test for comparisons between two independent groups (early vs. late stage tumor, nonrecurrent vs. recurrent tumor, and primary vs. metastatic tumor).

The mean immune cell composition for each category is shown in Figure 9A, which compares the relative proportion of 22 immune cell types across the three datasets, each dataset representing one category. M2 macrophages were found to be prevalent in late‐stage, recurrent, and metastatic tumors. CD4 memory T cells were found to be prevalent in late‐stage tumors, and monocytes in metastatic tumors. These immune modulation favors a pro‐tumorigenic environment by providing immunosuppressive conditions.

FIGURE 9.

(A) Bar plot depicting the mean relative abundance of 22 immune cell types across three tumor categories: Early‐ versus late‐stage tumor, nonrecurrent versus recurrent tumor, and primary versus metastatic tumor. (B) Box plot comparing the distribution of immune cell infiltration between early‐ and late‐stage tumors. (C) Immune cell infiltration in nonrecurrent versus recurrent tumors. (D) Immune cell infiltration between primary and metastatic tumors, showing significantly higher macrophages M2 infiltration in late‐stage, recurrent, and metastatic tumors, indicating an immune‐suppressive environment.

Figure 9B–D provides a visual comparison of immune cell infiltration across the three categories using box plots. In early‐ versus late‐stage tumor (Figure 9B), significant increases in macrophages M2 and neutrophils were observed in late‐stage tumors, which points to the contribution of these cells in promoting immune suppression and tumor progression during advanced disease stages. Early‐stage tumors, however, had higher infiltration of activated Dendritic cells, which might indicate a more active immune response in the initial stages of cancer. When comparing nonrecurrent versus recurrent tumors (Figure 9C), although immune cells such as M2 macrophages and Tregs were more abundant in Recurrent tumors, these differences were not statistically significant.

Similarly, the comparison between primary and metastatic tumors (Figure 9D), did not show any statistically significant differences in immune cell infiltration. Despite trends suggesting a more immunosuppressive microenvironment in metastatic tumors (with increased levels of MDSCs, Tregs, and M2 macrophages), these differences were not significant. In contrast, nonrecurrent tumors exhibited slightly higher proportion of memory B‐cells, reflecting a more active adaptive immune response and primary tumors showed more infiltration of T cells CD4 (memory resting), which can indicate an early phase of immune activation, but these differences were not statistically significant.

While significant immune cell infiltration differences were observed in the early versus late stage comparison, no significant differences were found in the recurrent versus nonrecurrent and primary versus metastatic tumor comparisons. These results suggest that additional factors beyond immune infiltration, such as tumor microenvironment dynamics, genetic mutations, and treatment history, might influence recurrence and metastasis.

3.7.1. Scatter Plot Analysis of Hub Genes and Immune Cell Infiltration

The scatter plots (Figures 10, 11, 12) illustrate the specific correlation between hub genes expression and key immune cells, providing a detailed view of their individual associations across each category.

FIGURE 10.

Scatter plots showing the correlation between hub gene expression and immune cell infiltration in early‐ versus late‐stage tumor. Each plot represents the relationship between one hub gene (COL1A1, COL1A2, COL3A1, FN1, CXCL8, IGF1, LUM, POSTN) and various immune cells, including memory B‐cell, T cells CD8, T cells CD4, macrophages M2, neutrophils, and dendritic cells. The regression line and shaded area represent the trend and 95% confidence interval. Significant correlations are indicated by the Spearman correlation coefficients ρ (rho), and q represents the adjusted p‐value displayed on each plot.

FIGURE 11.

Scatter plots showing the correlation between hub gene expression and immune cell infiltration in nonrecurrent and recurrent tumor. Each plot represents the relationship between one hub gene (COL1A1, COL1A2, COL3A1, FN1, CXCL8, IGF1, LUM, POSTN) and various immune cells, including memory B‐cell, T cells CD8, T cells CD4, macrophages M2, neutrophils, and dendritic cells activated. The regression line and shaded area represent the trend and 95% confidence interval. Significant correlations are indicated by the Spearman correlation coefficients ρ (rho), and q represents the adjusted p‐value displayed on each plot.

FIGURE 12.

Scatter plots showing the correlation between hub gene expression and immune cell infiltration in primary and metastatic tumor. Each plot represents the relationship between one hub gene (COL1A1, COL1A2, COL3A1, FN1, CXCL8, IGF1, LUM, POSTN) and various immune cells, including memory B‐cell, T cells CD8, T cells CD4, macrophages M2, neutrophils, and dendritic cells activated. The regression line and shaded area represent the trend and 95% confidence interval. Significant correlations are indicated by the Spearman correlation coefficients ρ (rho), and q represents the adjusted p‐value displayed on each plot.

Across late‐stage tumors, ECM/remodeling hubs coupled with innate and antigen‐presenting compartments. COL1A1 positively correlated with CD8 T cells (ρ≈0.30, q≈6.8×10^{− 4}) and M2 macrophages (ρ≈0.28, q≈1.1×10^{− 3}). COL1A2 and COL3A1 each associated with dendritic cells (resting) (ρ≈0.27 and 0.26; q≈1.4×10^{− 3} and 2.9×10^{− 3}) and with M2 macrophages (ρ≈0.20 and 0.19; q≈0.021 and 0.038), FN1 positively correlated with M2 macrophages (ρ≈0.29, q≈7.7×10^{− 4}) and CD8 T cells (ρ≈0.21, q≈0.018). Consistent with a pro‐inflammatory innate axis, CXCL8 (IL‐8) correlated strongly with mast cells (activated) (ρ≈0.35, q≈6.5×10^{− 5}) and neutrophils (ρ≈0.30, q≈5.8×10). LUM aligned with dendritic cells (resting) (ρ≈0.32, q≈3.9×10^{− 4}), and POSTN with M2 macrophages (ρ≈0.23, q≈9.4×10^{− 3}). Collectively, these interactions suggests that late‐stage ovarian tumors are characterized by an ECM‐driven myeloid response accompanied by neutrophil/mast‐cell activation and selective T‐cell involvement.

In recurrent tumors, the ECM‐related hub genes showed strong associations with CD4‐memory (activated) T cells and myeloid populations, while regulatory T cells (Tregs) declined. COL1A1, COL1A2, and COL3A1 correlated positively with CD4‐memory (activated) (ρ≈0.29–0.33; q≈0.011–0.0045) and with M2 macrophages (ρ≈0.26–0.32; q≤0.011) and were negatively associated with Tregs (ρ≈−0.30 to −0.37; q ≤ 0.0093). FN1 showed the broadest range of interaction, showing positive correlation with CD4‐memory (activated) (ρ≈0.42, q≈3.2×10^{− 4}), M2/M1 macrophages (ρ≈0.34–0.35; q≤0.0036), and neutrophils (ρ≈0.31, q≈7.7×10^{− 3}) and negative with Tregs (ρ≈−0.36, q≈1.7×10^{− 3}). CXCL8 showed positive correlations with mast cells (activated) (ρ≈0.50, q≈4.0 × 10^{− 6}) and neutrophils (ρ≈0.33, q≈4.5 × 10^{− 3}), with depletion of mast cells (resting) (ρ≈−0.32, q≈4.9 × 10^{− 3}). LUM and POSTN showed similar pattern, positively correlating with activated CD4‐memory (LUM ρ≈0.43, q≈2.8×10^{− 4}; POSTN ρ≈0.39, q≈9.3×10^{− 4}) and M1/M2 macrophages (q ≤ 0.0015), while IGF1 showed a selective negative link to NK (resting) (ρ≈−0.25, q≈0.040). Together, these findings indicate that recurrent tumor exhibit a matrix‐remodeled, myeloid‐rich immune environment. The ECM‐related hub genes, particularly COL1A1–COL3A1, FN1, POSTN, and CXCL8, are closely linked with macrophage and neutrophil infiltration, highlighting their potential role in shaping an immunosuppressive tumor microenvironment.

Although fewer pairs reached FDR significance in the metastatic cohort, two clean signals emerged. CXCL8 correlated positively with mast cells (activated) (ρ≈0.81, q≈0.0118) and negatively with monocytes (ρ≈−0.76, q≈0.0325), consistent with its chemotactic, pro‐inflammatory role in metastases. FN1 showed a negative association with T‐follicular helper (Tfh) cells (ρ≈−0.85, q≈0.0056), whereas IGF1 exhibited a positive association with Tfh cells (ρ≈0.75, q≈0.0325), suggesting divergent links of stromal versus growth‐factor signaling to adaptive immunity. The remaining hub genes showed no FDR‐significant associations.

The scatter plots show that the eight hub genes are not merely markers of tumor status; they map functionally distinct immune ecologies across disease progression. Late stage concentrates an ECM–M2–neutrophil/mast‐cell activation response, recurrence amplifies this with CD4‐memory activation and Treg loss; and metastasis preserves a potent CXCL8–mast‐cell axis. These findings suggest progressive immune remodeling from active innate responses to a subdued immune landscape during tumor advancement.

3.8. miRNA‐Hub Gene Network Analysis

ENCORI, an integrated resource that consolidates miRNA—target predictions from multiple established tools, including microT, RNA22, miRmap, PicTar, miRanda, PITA, and TargetScan, was employed to investigate the regulatory relationship between hub genes and the miRNAs. In this study, only miRNAs predicted by three or more of these prediction tools were considered as the targeted miRNAs of the hub genes. Using this criterion, a total 68 miRNAs were predicted for eight hub genes. A comprehensive list of predicted miRNAs targeting the eight hub genes, along with supporting prediction algorithms and phyloP conservation scores, is provided in Supporting Information S4.

Cytoscape software was used to visualize the miRNA‐hub gene network. As illustrated in Figure 13, the network comprises 61 nodes and 68 edges, representing the complex regulatory interplay between the miRNA and their target genes. Using CytoHubba, the most interconnected nodes in the network were identified and ranked by degree. FN1 (degree  =  15), COL1A1 (degree =  11), and CXCL8 (degree   =  11) were the three most interconnected hub genes, indicating that they are the primary targets for multiple miRNAs (Figure 13). Other hub genes such as COL3A1, POSTN, LUM, IGF1, and COL1A2 also demonstrated notable interactions with lower connectivity degrees.

FIGURE 13.

Interaction network between hub genes and targeted miRNAs. Hub genes are presented in red circles, whereas targeted miRNAs are shown in blue diamonds. The interaction between the hub genes and related miRNAs is shown in the form of arrows.

Among the miRNAs, has‐miR‐29a‐3p (degree =  4), has‐miR‐29b‐3p (degree =  4), and has‐miR‐29c‐3p (degree =  4) showed the highest interaction degrees, suggesting their central role in regulating several hub genes simultaneously (Figure 13). Furthermore, based on their presence across multiple prediction databases and higher phyloP conservation scores, several miRNAs were prioritized as key regulators, including has‐miR‐338‐3p (targeting COL1A1), has‐miR‐25‐3p (targeting COL1A2), has‐miR‐29b‐3p (targeting COL3A1), has‐miR‐200c‐3p (targeting FN1), has‐miR‐106b‐5p (targeting CXCL8), has‐miR‐520d‐5p (targeting LUM), and has‐miR‐26b‐5p (targeting POSTN).

We further validated the co‐expression relationship between these miRNAs and their corresponding target mRNAs using ENCORI. Significant negative expression relations were observed for miRNA–mRNA pairs such as has‐miR‐338‐3p/COL1A1, (r  = −0.114, p  = 2.74e‐02), has‐miR‐25‐3p/COL1A2 (r  = −0.159, p  = 1.92e‐03), has‐miR‐200c‐3p/FN1 (r  = −0.165, p  = 1.31e‐03), has‐miR‐29b‐3p/IGF1(r  = −0.174, p  = 7.07e‐04), and has‐miR‐26b‐5p/POSTN(r  = −0.126, p  = 1.45e‐02), supporting their predicted regulatory interactions. In contrast, no significant reverse relationship were observed for has‐miR‐29b‐3p/COL3A1 (r  = −0.065, p  = 2.11e‐01), has‐miR‐520d‐5p /LUM (r  = −0.065, p  = 2.08e‐01), and has‐miR‐106b‐5p/CXCL8 (r  = −0.045, p  = 3.80e‐01) (Figure S8). These results underscore the potential of the identified miRNA to act as critical posttranscriptional regulators of key genes involved in OC progression.

3.9. Drug–Gene Interaction Analysis of Hub Genes

DGIdb was employed to identify potentially therapeutic compounds targeting the eight hub genes (COL1A1, COL1A2, COL3A1, FN1, IGF1, CXCL8, POSTN, and LUM). Using DGIdb, 54 candidate drugs interacted with COL1A1, COL1A2, CXCL8, and FN1, which are promising druggable targets and might help in the development of new treatment strategies for OC therapy (Figure S9). Only FDA‐approved drugs were considered for the analysis. Ocriplasmin was identified as a potential drug for therapy due to its inhibitory effects on COL1A1, COL1A2, and FN1. Pamidronate was found to be targeting both COL1A1 and CXCL8, Dacarbazine was found to be targeting both CXCL8 and FN1, and similarly, COLLAGENASE CLOSTRIDIUM HISTOLYTICUM‐AAES was found to be targeting both COL1A1 and COL1A2. Details of the FDA‐approved candidate drugs interacting with the hub genes, including gene targets, drug names, and interaction types, are presented in Supporting Information S4.

4. Discussion

Despite advancements in medical and surgical treatments of OC, it remains the deadliest among gynecological cancers [50]. Most OC‐related deaths occur due to detection at a late stage, metastasis, high rate of recurrence, and resistance to chemotherapy. Hence, there is a critical need to find robust and effective biomarkers and understand the molecular mechanism and pathways for early diagnosis, prognosis, and appropriate treatment of OC [51]. Over the past few years, Bioinformatics techniques and computational tools have grown rapidly, allowing us to rapidly analyze huge data from microarray and high throughput sequencing, and helping us understand the molecular mechanism behind tumor occurrence and progression [52, 53].

In the current study, we identified DEGs among three classes comprising nonrecurrent and recurrent tumors, early‐stage and late‐stage tumors, and primary and metastatic tumor, using computational analysis. We identified 697 DEGs between nonrecurrent and recurrent tumors, 965 DEGs between early‐stage and late‐stage tumors, and 725 DEGs between primary and metastatic tumors. The diverse genomic landscape that characterizes the various stages and conditions of OC is highlighted by these DEGs. Densely connected networks with 100 nodes each were observed for each category of tumor samples through PPI network analysis. These networks showed 743 edges for nonrecurrent and recurrent tumors, and 1277 and 1034 edges for early‐ and late‐stage tumors and primary and metastatic tumors, respectively. The intricacy and density of these networks show complex molecular interactions that drive the development and metastasis of OC. GO analysis showed that the DEGs were abundant in ECM organization, endoplasmic reticulum lumen, Golgi lumen, and complex collagen trimers among cellular components. Moreover, topology‐based pathway analysis (SPIA) revealed significant activation of ECM–receptor interaction, focal adhesion, and PI3K‐Akt signaling pathways, highlighting the central role of ECM remodeling, adhesion, and receptor‐mediated signaling in OC progression.

Cytokines are crucial in cancer as they regulate immune response and cellular communication, affecting tumor growth and metastasis. Previous studies have shown that cytokines have both anti‐tumorigenic and pro‐tumorigenic roles in cancer [54, 55]. According to a recent review by Browning et al., IL‐6 induces JAK/STAT and epithelial‐to‐mesenchymal transition (EMT) pathways, which leads to migration, proliferation, survival, and chemoresistance of OC cells [56]. Interlukin‐17, a pro‐inflammatory cytokine encourages metastasis of OC by upregulation of MTA1 mRNA and protein and also induces self‐renewal of cancer stem‐like cells (CSLCs) [57]. The balance and interaction between these cytokines influence the microenvironment of tumor and can determine the course of the disease. Analyzing these interactions is crucial for developing targeted cancer therapies that modulate the immune response to effectively combat cancer. Many previous studies have reported the role of ligand‐induced AGE RAGE signaling in the progression of glioma, gastric, and breast [58, 59, 60]. In cancer cells, the binding of AGE with RAGE activates downstream PI3K signaling, which promotes pro‐survival autophagy and suppresses apoptotic signaling [61]. In OC, overexpression of RAGE could be a useful biomarker suggested by Rahimi et al. [62]. The altered PI3K‐AKT pathway could be a major factor for most of the malignancies, therefore focusing on this pathway's effectors is a potential treatment approach [63]. Patients with OC are undergoing clinical trials for PI3K pathway inhibitors, as these inhibitors have demonstrated tremendous promise in the management of OC [64].

These enriched functions and pathways emphasize the role of immune response and extracellular dynamics in the progression and metastasis of OC. Further analysis revealed distinct sets of hub genes for recurrent, late‐stage, and metastatic tumors. For recurrent tumors, the top hub genes identified were IL6, IGF1, STAT1, CXCL8, ESR1, COL1A2, CXCL10, CXCL1, CXCL2, and BMP2. In late‐stage tumors, the top hub genes were COL1A1, FN1, COL1A2, COL3A1, POSTN, THBS2, IGF1, CXCL8, LUM, and DCN and For metastatic tumors, network analysis identified COL1A1, FN1, COL1A2, COL3A1, POSTN, ITGAM, LUM, MMP2, FCGR3A, and TLR2 as the top 10 hub genes. Eight intersecting hub genes including FN1, COL1A1, COL3A1, COL1A2, POSTN, LUM, IGF1, and CXCL8 from the top hub genes of each category were identified. Among these, COL1A2 was a common hub gene among recurrent, late‐stage, and metastatic ovarian tumors, suggesting its pivotal role in OC progression. FN1, COL1A1, COL3A1, LUM, and POSTN were common hub genes between metastatic and late‐stage tumors and IGF1 and CXCL8 were common hub genes between recurrent and late‐stage tumors. FN1, IGF1, POSTN, COL1A1, COL1A2, COL3A1, and LUM were upregulated significantly in the late stages and downregulated in the early stages of OC. Conversely, CXCL8 was significantly downregulated in the late stages. These expression patterns were validated using UALCAN, GEPIA2 and KM Plotter online tools, both of which confirmed the differential expression observed in our DEG analysis. Survival and diagnostic analyses confirmed the clinical relevance of the identified hub genes. High expression of ECM‐related genes (FN1, COL1A1, COL1A2, COL3A1, POSTN, and LUM) was associated with poor OS, PFS, and RFS and strong diagnostic accuracy, particularly for FN1 and COL1A1.

These genes were closely linked to the stage of tumor, patient survival outcome, and immune modulation, suggesting they could be valuable prognostic markers. While previous research has investigated the involvement of these genes individually in OC, this study suggests a molecular signature using these gene clusters as potential prognostic biomarkers through analysis of microarray datasets. Identification of COL1A2 as a gene differentially expressed in recurrent, late‐stage, and metastatic tumor is also unique to this study.

Fibronectin1 (FN1), plays a crucial role in enabling cellular attachment, movement, growth, and differentiation [65]. Previous studies reported that FN1 promotes tumor progression and migration in several cancer types by regulating EMT [66, 67]. Wang et al. found that elevated FN1 levels were correlated with immune cell infiltration and altered immunological microenvironment in gastric cancer [68]. Sponziello et al. demonstrated that silencing of FN1 resulted in decreased adhesion, proliferation, and invasiveness in thyroid cancer cell lines [69]. Several studies demonstrated the role of the FN1 gene in the activation of PI3K/Akt pathway of cell cycle regulation and platinum resistance [70, 71, 72].

IGF1 is a key growth regulator involved in many physiological phenomena and has been associated with the genesis and progression of several malignancies like colorectal cancer, breast cancer, and cervical cancer [73, 74, 75]. Its involvement in cancer progression makes it a good target for research and therapy [76]. According to a study, cancer‐associated fibroblasts increased IGF‐1/ERβ signaling in human bladder cancer cells (BCa), leading to enhanced expression of the Bcl‐2 gene and simultaneously increasing chemoresistance of BCa cell. Blocking of IGF‐1/ERβ/Bcl‐2 signaling reduces chemoresistance of BCa cells [77]. In a recent study, Wang et al. suggested that the Ang2‐TEMs‐IGF1 pathway could be a promising candidate for novel OC therapies as angiogenesis caused by Tie2 overexpressing monocytic cells may be significantly influenced by IGF1‐mediated signaling in OC cells [78].

The collagens (COL1A2, COL3A1, and COL1A1) and POSTN are critical in the restructuring of the ECM, which is critical for tumor infiltration and metastasis [79, 80]. Numerous collagen types have been found to be overexpressed in various tumors, prompting research into their role in tumor progression using gene expression modulation studies. In a related study, Cheon et al. constructed a gene signature comprising AEBP1, TIMP3, POSTN, LOX, SNAI2, COL11A1, VCAN, COL5A1, COL6A2, and THBS2 genes, involved in collagen restructuring that are modulated by TGF‐β1 signaling and were associated with worse survival outcomes in high‐grade serous OC. They demonstrated that the knockdown of COL11A1, one of the signature genes, reduced progression of tumor and invasion in vitro and in mouse models [81]. In another study, researchers conducted a comprehensive expression analysis of COL1A2 across 33 human cancer types using the TIMER database and data from TCGA. They showed that elevated COL1A2 expression was linked to a worse prognosis in colorectal adenocarcinoma (COAD) [82]. Prior studies have demonstrated that elevated levels of COL1A1 can be correlated with altered immune filtration and the worst survival outcome in lung cancer patients [83], and increased expression of COL3A1 leads to enhanced migration, invasion, and growth of esophageal squamous cell carcinoma (ESCC) cells, whereas knocking down COL3A1 reduced these effects [84]. In Eca109 and TE‐1 cell lines of esophageal cancer (ESCA), siRNA‐mediated knockdown of COL1A2 reduced their growth and migration significantly [85]. POSTN is associated with dismal prognosis in multiple cancers as its elevated expression is correlated with enhanced tumor aggressiveness and reduced patient survival [86, 87].

Lumican (LUM) has emerged as a significant putative biomarker in various cancers due to its unique expression patterns and roles in tumor progression [88]. Lumican regulates ECM, which influences key processes such as cellular adhesion, epithelial‐mesenchymal transition, and growth [89]. In pancreatic and breast cancer, higher levels of lumican have been linked with improved patient outcomes. Conversely, in gastric and bladder cancer, lumican expression has been linked with increased metastasis and tumor growth, further emphasizing its pro‐tumorigenic functions [90]. Additionally, lumican has been implicated in modulating angiogenesis, which is essential for neoplastic growth and metastasis [91]. CXCL8, or IL8, plays critical roles in tumor progression and development [92]. As a pro‐inflammatory cytokine, CXCL8 is crucial for the immune response by recruiting neutrophils to areas of inflammation [93]. Numerous studies have highlighted the essential role of CXCL8 in various aspects of tumor biology, contributing to tumor progression, angiogenesis, migration, and the tumor microenvironment [94, 95, 96].

Genetic alteration analysis of these eight hub genes revealed that approximately 19.6% of OC cases exhibited alterations in these genes, with COL3A1 being the most frequently altered (6%). However, these alterations were not significantly associated with OS or PFS, suggesting limited direct prognostic relevance. Methylation analysis revealed mild inverse correlations between promoter methylation and expression of IGF1 and CXCL8, indicating partial epigenetic regulation, while collagen genes showed minimal methylation‐expression correlation, implying transcriptional control independent of DNA methylation.

Further, immune infiltration analysis revealed distinct immune remodeling across tumor stages. Late‐stage tumors showed higher infiltration of M2 macrophages, neutrophils, and activated CD4⁺ T cells, indicating an immunosuppressive microenvironment that facilitates tumor progression. In recurrent and metastatic tumors, selective associations such as CXCL8 with mast cells and FN1/IGF1 with T‐follicular helper cells suggest further immune dysregulation and adaptive evasion. These gene–immune correlations highlight therapeutic entry points (e.g., IL‐8/CXCR1/2, macrophage polarization, and mast‐cell and neutrophil crosstalk) and argue for immune‐context‐aware exploitation of ECM hubs in advanced OC. These findings highlight that ECM‐related genes influence immune cell composition, emphasizing the role of tumor–immune crosstalk in disease advancement and its potential as a target for immunomodulatory therapies in OC.

Moreover, the identification of miRNA interactions further supports the posttranscriptional regulation of these hub genes, which could be explored for targeted therapy. To further enhance the significance of these identified miRNAs, we focused on miRNA with high phyloP scores, which reflect their evolutionary conservation across species. This analysis was crucial as conserved miRNAs are likely to play fundamental roles in regulating essential biological processes, such as cell cycle and apoptosis, which are critical for tumorigenesis. By identifying these highly conserved miRNAs, we not only highlight their potential as key regulatory elements in cancer progression but also strengthen their candidacy as robust biomarkers for early detection, prognosis, and targeted therapy in OC. Additionally, drug–gene interaction analysis identified promising therapeutic drugs, such as ocriplasmin and pamidronate, which target key genes like COL1A1 and FN1. Collectively, these findings not only enhance our understanding of OC biology but also propose these eight hub genes as potential candidates for prognostic, diagnostic, and therapeutic interventions, providing a foundation for personalized treatment strategies.

5. Conclusion

This study provides comprehensive insights into the underlying biological processes driving OC progression by identifying key hub genes (FN1, COL1A1, COL1A2, COL3A1, POSTN, IGF1, LUM, and CXCL8), their associated miRNAs, which are associated with progression, recurrence, and metastasis of OC. The integrated analyses from transcriptomic, epigenetic, and immune perspectives strengthen the understanding of molecular events driving OC progression and highlight the potential of these genes as robust biomarkers for early diagnosis, prognosis, and therapeutic targeting. Notably, COL1A2 emerged as a common hub gene across all three categories, suggesting its pivotal role in OC advancement. The identification of promising therapeutic agents further paves the way for personalized treatment strategies, offering hope for improving outcomes in OC patients.

6. Future Studies

The scope of the current study is limited to the in‐silico identification of the molecular gene signature in OC. Further investigation and validation are necessary to translate these observations into effective prognostic and therapeutic interventions. The hub genes identified here have to be initially validated in both in vitro and in vivo models, followed by patient samples. The study on immune infiltration has to be further elaborated with more samples in silico, and it has to be validated in other tumor models too. A similar validation for miRNA would also add more value. A comprehensive study on hub genes as therapeutic targets and the drug–gene interaction will bring us one step closer to developing novel drugs for OC. The current study attempts to delve into understanding the complex molecular mechanism of OC and serves as a precursor to identify potential biomarkers and therapeutic targets.

Author Contributions

A.S. and R.V. did conceptualization and data curation, microarray data analysis was performed by V.Y., the original draft was written by V.Y. and A.S., and review and editing of the manuscript was performed by R.V.

Funding

The authors received no external funding. The work was supported only by institutional resources.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supplementary figure 1. Workflow of the study.

Supplementary figure 2. Topology‐based pathway analysis (SPIA) of KEGG pathways across categories.

Supplementary figure 3. Differential expression of hub genes in ovarian cancer samples by GEPIA2.

Supplementary figure 4. Copy‐number variation‐expression relationship for hub genes in TCGA OV.

Supplementary figure 5. Methylation–expression correlation of hub genes in TCGA‐OV

Supplementary figure 6. The prognostic significance by progression free survival analysis of eight intersecting hub genes.

Supplementary figure 7. Diagnostic performance (ROC) of the combined 8 hub‐gene signature.

Supplementary figure 8. Predicted miRNA–mRNA relationships in ovarian cancer (starBase v3.0)

Supplementary figure 9. Drug‐Gene interactions network visualized in Cytoscape

Supplementary file 2 (comprehensive list of all samples, xlsx)

Supplementary file 3 (Complete list of DEGs in all datasets, xlsx)

Supplementary file 4 (Topology enrichment, miRNA and Drug Hub gene interaction analysis, xlsx)

Data Availability Statement

The datasets analyzed during the current study are available in NCBI‐GEO (https://www.ncbi.nlm.nih.gov/geo/). The accession numbers of datasets are listed in Table 1 of our article.