Transcriptomic Analysis of Hub Genes Reveals Associated Inflammatory Pathways in Estrogen-Dependent Gynecological Diseases

Simple Summary

Gynecological diseases still make up a large percentage of the overall global disease burden. While oral contraceptives and gonadotropin-releasing hormone drugs for endometriosis and gynecological cancers exist, their known inflammatory side effects can counteract progress in therapy. With this, the present study made use of a systems biology approach to identify correlations between gynecological diseases using gene expression data from DNA microarray samples that contain endometriosis, ovarian cancer, cervical cancer, and endometrial cancer. The highly preserved gene modules and their top interacting hub genes were determined to provide a further understanding of the signaling pathways and biological processes affected. Potential drugs were screened based on the upregulated and downregulated hub genes, which identified drug candidates that have known anti-inflammatory effects, implying the potential of specific inflammatory pathways in estrogen-dependent gynecological diseases as a therapeutic avenue.

Abstract

Gynecological diseases are triggered by aberrant molecular pathways that alter gene expression, hormonal balance, and cellular signaling pathways, which may lead to long-term physiological consequences. This study was able to identify highly preserved modules and key hub genes that are mainly associated with gynecological diseases, represented by endometriosis (EM), ovarian cancer (OC), cervical cancer (CC), and endometrial cancer (EC), through the weighted gene co-expression network analysis (WGCNA) of microarray datasets sourced from the Gene Expression Omnibus (GEO) database. Five highly preserved modules were observed across the EM (GSE51981), OC (GSE63885), CC (GSE63514), and EC (GSE17025) datasets. The functional annotation and pathway enrichment analysis revealed that the highly preserved modules were heavily involved in several inflammatory pathways that are associated with transcription dysregulation, such as NF-kB signaling, JAK-STAT signaling, MAPK-ERK signaling, and mTOR signaling pathways. Furthermore, the results also include pathways that are relevant in gynecological disease prognosis through viral infections. Mutations in the ESR1 gene that encodes for ERα, which were shown to also affect signaling pathways involved in inflammation, further indicate its importance in gynecological disease prognosis. Potential drugs were screened through the Drug Repurposing Encyclopedia (DRE) based on the up-and downregulated hub genes, wherein a bacterial ribosomal subunit inhibitor and a benzodiazepine receptor agonist were the top candidates. Other drug candidates include a dihydrofolate reductase inhibitor, glucocorticoid receptor agonists, cholinergic receptor agonists, selective serotonin reuptake inhibitors, sterol demethylase inhibitors, a bacterial antifolate, and serotonin receptor antagonist drugs which have known anti-inflammatory effects, demonstrating that the gene network highlights specific inflammatory pathways as a therapeutic avenue in designing drug candidates for gynecological diseases.

1. Introduction

Currently, gynecological diseases still make up approximately 4.5% of the overall global disease burden, even surpassing major health concerns, such as malaria, ischemic heart disease, tuberculosis, and maternal conditions [1]. These diseases pertain to conditions related to the dysfunction of the female reproductive system. One of the most common gynecological diseases is endometriosis (EM). It is a condition in which endometrium grows outside the uterus, preventing the blood and tissue it releases during the menstrual cycle from exiting the body. It leads to inflammation, scarring, and abnormal tissue growth that can be a factor in increased risk for gynecological cancer [2]. Particularly, chronic inflammation is a major hallmark of cancer, and it is often experienced by patients suffering from the most common gynecological cancers—endometrial cancer (EC), ovarian cancer (OC), and cervical cancer (CC) [3], making it considered a biological mechanism that underpins carcinogenesis due to its ability to induce excessive production of pro-inflammatory cytokines which promote cellular proliferation and reduce apoptosis that contributes to tumor growth. Obesity and metabolic syndromes are well-established risk factors for endometrial and ovarian cancers due to visceral fats and adipokines promoting a pro-inflammatory state within the microenvironment [4,5,6]. On the other hand, chronic inflammation of the cervical epithelium is a result of human papillomavirus (HPV) infection and often leads to cervical carcinogenesis and progression [7].

Given that gynecological diseases are estrogen-dependent conditions, oral contraceptives and gonadotropin-releasing hormone (GnRH) agonists that maintain steady hormone levels are commonly given as medications. However, reports have shown that oral contraceptive use by women with polycystic ovarian syndrome (PCOS) increases the risk of inflammatory and coagulatory disorders, while aberrant GnRH secretion may even cause inflammation-induced ovarian dysfunction [7,8]. Concerning this, estrogen plays a major role in regulating immune and inflammatory responses, further making it plausible to investigate the molecular underpinnings of inflammation and estrogen production in gynecological diseases. This can be made possible through weighted gene co-expression network analysis (WGCNA) [9,10], which is a systems biology approach that characterizes the gene associations in the gene expression data of several diseases to identify significant gene clusters (modules) and elucidate the key hub genes and main pathways that are dysregulated. The identification of upregulated and downregulated hub genes further allows for the scanning of potential already-existing drugs that can induce countering effects based on the provided gene signature while determining significant pathways opens therapeutic opportunities. Several studies making use of WGCNA to explore gynecological cancers have further expanded our understanding, such as uncovering a new prognostic genetic marker for cervical cancer and identifying important pathways in endometrial cancer tumorigenesis [11,12].

With the growing evidence of significant associations between endometriosis and gynecologic cancers [13], this study aims to further investigate the role of inflammation in the prognosis of gynecological diseases using the gene expression dataset of EM (GSE51981), EC (GSE17025), OC (GSE63885), and CC (GSE63514) that were sourced from the Gene Expression Omnibus (GEO). Through WGCNA, five highly preserved modules were identified across all the datasets, which were then found to be strongly involved in transcription regulation, protein binding, and cell cycle via functional annotation and pathway enrichment analysis. The construction of a protein–protein interaction network revealed the key hub genes within each module that allowed the screening of potential repurposed drugs.

2. Materials and Methods

2.1. Dataset Acquisition and Pre-Processing

Through National Center for Biotechnology Information: Gene Expression Omnibus (NCBI GEO) [https://www.ncbi.nlm.nih.gov/geo/ accessed on 30 October 2023], DNA microarray datasets containing expression data of endometrial tissue and tumor samples from patients who have endometriosis (EM), ovarian cancer (OC), cervical cancer (CC) and endometrial cancer (EC). The sampling of this study was limited to the datasets carried out through GPL570-HG-U133 Plus 2 Affymetrix Human Genome U133 Plus 2.0 Array to prevent inconsistencies caused by differences in probe design or normalization [14,15]. Overall, a total of 77 samples were for EM, 101 samples were used for OC, 104 samples for CC, and 91 samples for EC.

Table 1. GEO Datasets Summary.

Accession No.	GSE51981	GSE63885	GSE63514	GSE17025
Condition	Endometriosis	Ovarian Cancer	Cervical Cancer	Endometrial Cancer
Year Published	2014	2014	2015	2011
Type	Expression Profiling by Array
Platform	GPL570-HG-U133 Plus 2 Affymetrix Human Genome U133 Plus 2.0 Array
Source	Endometrial Tissue	Primary Cancer Tissue
No. of Samples	77	101	104	91

presents the summarized information of the datasets used for the analysis.

Through the getGEO function of Bioconductor in RStudio (http://www.bioconductor.org accessed on 30 October 2023), microarray datasets were extracted and underwent quantile normalization using robust multi-array average (RMA). The phenodata that contain information on the control samples were used to remove normal cell expression data to retain only the samples from endometrial and cancer tissues. Then, expression values were log₂ transformed, and genes without expression values were removed using the goodSamplesGenes function, as well as outliers in the sample clustering dendrograms. After log₂ transformation, gene expression values falling below the minRelativeWeight threshold were denoted as missing entries in the data and removed. This is performed to achieve a stable list of good samples and genes and avoid errors in the calculation [16]. The expression data were then filtered to remove genes with expression values lower than the 20% percentile cut-off in the mean and variance. All genes that were consistently common across datasets were used for the subsequent analysis.

2.2. Weighted Gene Co-Expression Network Analysis (WGCNA)

2.2.1. Scale-Free Network

In approximating the scale-free network, the pickSoftThreshold function of the WGCNA R [17] package (v4.3.1.) was first used to plot the scale-free topology fit versus power index (1–20) to estimate the appropriate soft-thresholding power (β), which is the lowest power where the scale-free topology criterion is met. This is when the distribution of the number of gene connections follows the power-law distribution, and it is usually evaluated by plotting the scale-free topology fit versus soft-thresholding power to measure the relativity of the co-expression network to the linear relationship between log of connectivity and log of connectivity probability. The power at which the scale-free topology fit reaches a plateau often indicates a good fit [18]. Hence, the softConnectivity function was used to plot the linear model fitting the R² index to quantify the goodness-of-fit of the expression data of each dataset at the chosen soft-thresholding power. The dataset exhibiting the best fit is selected as the reference dataset for the WGCNA.

2.2.2. Network Construction and Module Identification

After selecting the reference dataset, it is transformed into adjacency matrices calculated using Pearson’s correlation for network construction. Then, the adjacency matrices were converted into Topological Overlap Matrices (TOM) using the TOMsimilarity function to calculate the similarities between genes for clustering. The hclust function was used to conduct hierarchal clustering and produce gene dendrograms using highly correlated genes. Module identification and generation of TOM dissimilarity dendrogram were performed through the cutreeDynamic function, where the deep split parameter was set to 2 (medium sensitivity) and the minimum cluster size was set to 30. Lastly, eigen correlation was performed to merge the modules with similar expression profiles to produce more significant results.

2.2.3. Module Preservation Analysis

Using the module preservation function, the gene co-expression network preservation of EM, OC, CC, and EC modules was analyzed with the network type “signed” and the number of permutations set to 100. This is to measure the degree to which the connectivity patterns of the reference network align with the other datasets to evaluate the biological relevance of the gene modules identified.

2.3. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) Pathways

Genes in each highly preserved module were extracted from RStudio and sent to the Database for Annotation, Visualization, and Integrated Discovery (DAVID) web-based tool for functional annotation and pathway enrichment analysis to understand the biological foundations of the identified modules. Prior to analysis, the Affymetrix gene IDs resulting from WGCNA were converted into Entrez IDs for database identification. Functional annotation was performed via the Gene Ontology (GO) database to investigate the Biological Processes (BP), Molecular Functions (MF), and Cellular Components (CC) involved. Furthermore, the Kyoto Encyclopedia of Genes and Genomes (KEGG) database was used to provide information on the biological pathways that are affected by the modules. The classification stringency was set to “medium,” and GO Terms and KEGG Pathways with p-values lower than 0.05 were considered statistically significant in the analysis.

2.4. Protein-Protein Interaction (PPI) and Hub Genes

The Entrez IDs of the genes in each highly preserved module were then sent to the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) database to identify genes with potential protein-protein interaction (PPI) networks. Afterward, the networks constructed were exported to Cytoscape, and through the CytoHubba feature, the top 10 genes with the highest interactions within their corresponding modules were identified by ranking them based on degree centrality, classifying them as hub genes.

2.5. Signature-Based Approach for Drug Repurposing

Before scanning for potential drugs, the identified hub genes were first grouped into upregulated and downregulated hub genes through differential expression analysis in GEO2R [https://www.ncbi.nlm.nih.gov/geo/geo2r/ accessed on 20 November 2023]. Then, each group of genes was exported to the Drug Repurposing Encyclopedia (DRE) to search from the Molecular Signatures Database (MSigDB) and Connectivity Map (CMap) for already-existing drugs that can be repurposed based on the provided gene signatures. The resulting drug candidates were ranked based on their Tau scores, and candidates with false discovery rates (FDR) of less than 0.05 were considered.

3. Results

3.1. Weighted Gene Co-Expression Network Analysis (WGCNA)

3.1.1. Data Preparation and Scale-Free Networks

In WGCNA, the final gene expression matrices are prepared following a series of pre-processing steps, including normalization, outlier detection, and removal [17,19]. Outliers detected in the sample clustering dendrograms (Figure A1, Figure A2, Figure A3 and Figure A4) were filtered out. A total of 25,138 genes that were common across the datasets were utilized to construct the weighted gene co-expression networks. This process ensures the robustness and reliability of the network analysis, providing valuable insights into the intricate relationships among genes as it emphasizes networks with stronger correlations over weaker ones to capture complex interactions among genes [17,19]. Figure 1 presents the relationship between the scale-free topology fit index and soft threshold (β) values ranging from 1 to 20. As shown, the soft threshold at β > 5 starts to plateau and stabilize across all datasets, indicating that the scale-free topology fit is unaffected by increased power and results in a robust scale-free structure. Hence, it was decided to use a soft threshold power of β = 5 to approximate the scale-free topologies for all datasets. This decision holds significance to ensure the robustness and efficiency of the gene network, since selecting a lower soft-thresholding power results in denser gene networks with more connections. This approach increases the probability that the gene clusters will exhibit biologically relevant relationships between genes. As WGCNA suggests, ideally, the soft threshold power should have an R² greater than 0.8, which can be visually observed in the scale-free topology fit as the point where datasets stabilize and plateau onwards. Thus, selecting a soft-threshold power of 5 would provide a more meaningful result [20,21,22].

With the chosen soft-thresholding power of β = 5, the soft connectivity (k) was calculated to assess the degree of fit of each dataset. This was performed by plotting the soft connectivity and calculating the log-log plot. Shown in Figure 2 is the corresponding log–log plot of the OC dataset, which represents the best fit among the dataset by having an R² value of 0.96. This indicates that the OC dataset should be chosen as the reference dataset for WGCNA network construction and module identification, as the gene co-expression networks generated from the dataset would result in biologically relevant expression patterns [10]. That is based on the assumption that real biological networks tend to exhibit scale-free properties, and the dataset with the highest R² value increases its likelihood of representing functionally relevant biological networks amongst the datasets [23].

3.1.2. Network Construction and Module Identification

In performing WGCNA, one technique is to designate one dataset as a reference for plotting the gene cluster dendrogram and generating modules. In this study, OC was chosen as the reference dataset, which transformed an adjacency matrix utilizing the selected soft threshold power of β = 5. To determine modules within the established network, a cluster dendrogram was generated using a weighted correlation matrix. Grouping genes into modules requires network proximity measurements, translating into similarities between genes within the dataset. Hence, TOM similarity measurement was performed to calculate the relativity between genes across the datasets, and the TOM dissimilarity index was used for plotting the gene dendrogram (Figure 3) that summarizes the identified modules. Overall, genes were segmented to form modules within the network, resulting in the identification of 33 modules using dynamic tree cut. These modules were further merged with similar expression profiles to yield more significant results. The modules in dynamic tree cut can be merged with similar expression profiles into more meaningful modules by calculating eigengene dissimilarity measurement and clustering its module dendrogram. Modules that had greater than 75% correlation value [24] were treated as highly related modules and, thus, will be merged, resulting from the original 33 modules from a dynamic tree cut to 23 eigengene modules. The identified modules were classified arbitrarily based on colors (

Table A1. Summary of identified modules based on TOM dissimilarity dynamic tree cut and merge modules.

Module	Dynamic Tree Cut (Genes)	Merge Modules (Genes)
Cyan	653	1594
Purple	947	947
Pink	1053	1053
Midnightblue	524	1232
Tan	935	935
Blue	1952	1952
Brown	1938	2164
Darkgreen	181	181
Darkgrey	116	116
Darkorange	102	102
Darkturquoise	143	143
Magenta	984	984
Orange	107	107
Paleturquoise	41	41
Royalblue	228	228
Violet	35	35
Yellow	1424	1424
Green	1387	1630
Saddlebrown	57	57
Grey	327	327
Skyblue	66	66
Grey60	354	354
Lightcyan	425	425
Black	1162
Lightgreen	265
Lightyellow	243
Salmon	708
Steelblue	56
Turquoise	6116
Greenyellow	941
Red	1341
Darkred	226
White	101

): blue (1952), brown (2164), cyan (1594), darkgreen (181), darkgrey (116), darkorange (102), darkturquoise (143), magenta (984), midnightblue (1232), orange (107), paleturquoise (41), pink (1053), purple (947), royalblue (228), violet (35), yellow (1424), green (1630), saddlebrown (57), grey (327), skyblue (66), grey60 (354), lightcyan (425), and tan (935). These modules are gene clusters exhibiting similar expression patterns that can provide substantial information about the dysregulated processes or pathways involved in endometriosis and the gynecological cancers analyzed.

3.2. Module Preservation Analysis

In this network analysis, module preservation validates the robustness of identified modules across different gynecological disease datasets. It plays a pivotal role in assessing the reproducibility of co-expression networks and gauging their biological significance. Modules with a Z_summary value exceeding 10 are deemed highly preserved [25]. Moreover, the gene count within each module serves as an indicator of its significance, with modules containing a larger number of genes being more meaningful as it implies more connectivity patterns [25]. The preservation of Z_summary values per dataset is shown in Figure 4, while the highly preserved modules identified across the datasets—cyan, midnightblue, pink, purple, and tan—are detailed in

Table A2. Highly Preserved Modules of OC VS CC.

	Module	Size	Zsummary Preservation
Common High Preserve Modules	cyan	1594	21.86149527
	midnightblue	1232	26.52630433
	pink	1053	25.86352918
	purple	947	15.59037483
	tan	935	20.96990849
High Preserve Modules in OC VS CC only	black	2604	12.59707559
	green	1630	15.81619288
	lightgreen	265	12.46321267
	royalblue	228	14.71890028
	turquoise	6116	16.18415255

,

Table A3. Highly Preserved Modules of OC VS EC.

	Module	Size	Zsummary Preservation
Common High Preserve Modules	cyan	1594	91.87686929
	midnightblue	1232	21.42695751
	pink	1053	44.18140552
	purple	947	13.8939192
	tan	935	35.24655182
High Preserve Modules in (OC VS EC) only	black	2604	14.84965178
	blue	1952	11.75575271
	brown	2164	13.40832857
	grey60	354	38.92112853
	lightcyan	425	14.51331585
	lightgreen	265	15.06853233
	orange	107	12.18371176
	paleturquoise	41	12.3852009
	royalblue	228	17.13450004
	saddlebrown	57	10.25197434
	skyblue	66	10.88052848
	yellow	1424	22.08498616

and

Table A4. Highly Preserved Modules of OC VS EM.

	Module	Size	Zsummary Preservation
Common High Preserve Modules	cyan	1594	15.67053707
	midnightblue	1232	12.67510038
	pink	1053	12.42696169
	purple	947	12.43896986
	tan	935	15.75233647
High Preserve Modules in (OC VS EM) only	darkturquoise	143	14.73601775
	green	1630	15.24494125
	magenta	984	10.04291952
	saddlebrown	57	10.13071697
	skyblue	66	11.86774714
	turquoise	6116	18.54903814
	yellow	1424	37.50425118

. These findings underscore the robustness and significance of the identified modules in elucidating common molecular mechanisms underlying various gynecological diseases.

3.3. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) Pathways

Following the conversion of Affymetrix gene IDs to Entrez IDs using the DAVID web server, the genes in highly preserved modules underwent functional annotation and pathway enrichment analysis to determine functional GO, focusing on BP, MF, CC, and KEGG pathway domains. The enriched GO and KEGG terms were examined for each module and compiled in

Table A5. Summary of top GO term and KEGG pathway analysis for the Cyan Module.

Category	Term		Count	p-Value
BP	GO:0000122	negative regulation of transcription from RNA polymerase II promoter	25	4.2 × 10−3
	GO:0006355	regulation of transcription, DNA-templated	21	3.4 × 10−2
	GO:0016310	phosphorylation	15	3.8 × 10−2
	GO:0006338	chromatin remodeling	13	1.8 × 10−3
	GO:0006468	protein phosphorylation	12	1.3 × 10−2
CC	GO:0005829	cytosol	116	1.5 × 10−8
	GO:0005634	nucleus	111	3.1 × 10−5
	GO:0005737	cytoplasm	105	4.0 × 10−5
	GO:0005654	nucleoplasm	77	2.5 × 10−4
	GO:0005783	endoplasmic reticulum	28	2.4 × 10−3
MF	GO:0005515	protein binding	193	2.6 × 10−2
	GO:0046872	metal ion binding	60	4.2 × 10−4
	GO:0003723	RNA binding	33	8.3 × 10−3
	GO:0005524	ATP binding	33	1.5 × 10−2
	GO:0005085	guanyl-nucleotide exchange factor activity	11	1.6 × 10−3
KEGG	hsa04810	Regulation of actin cytoskeleton	10	3.7 × 10−3
	hsa05208	Chemical carcinogenesis—reactive oxygen species	8	3.1 × 10−2
	hsa04660	T cell receptor signaling pathway	6	2.4 × 10−2
	hsa05100	Bacterial invasion of epithelial cells	5	2.0 × 10−2
	hsa04012	ErbB signaling pathway	5	2.8 × 10−2

,

Table A6. Summary of top GO term and KEGG pathway analysis for the Midnightblue Module.

Category	Term		Count	p-Value
BP	GO:0006355	regulation of transcription, DNA-templated	26	6.0 × 10−3
	GO:1905515	non-motile cilium assembly	6	2.3 × 10−3
	GO:0070588	calcium ion transmembrane transport	6	3.5 × 10−2
	GO:0030212	hyaluronan metabolic process	3	5.9 × 10−3
	GO:1905606	regulation of presynapse assembly	3	8.7 × 10−2
CC	GO:0005829	cytosol	98	4.7 × 10−2
	GO:0005576	extracellular region	46	1.7 × 10−2
	GO:0005739	mitochondrion	29	9.1 × 10−2
	GO:0098978	glutamatergic synapse	13	2.4 × 10−2
	GO:0005764	lysosome	10	3.5 × 10−2
MF	GO:0005515	protein binding	218	1.4 × 10−2
	GO:0004222	metalloendopeptidase activity	6	3.0 × 10−2
	GO:0019838	growth factor binding	4	2.3 × 10−2
	GO:0004709	MAP kinase kinase kinase activity	3	4.2 × 10−2
KEGG	hsa05168	Herpes simplex virus 1 infection	15	3.3 × 10−2
	hsa05010	Alzheimer disease	11	8.6 × 10−2
	hsa05160	Hepatitis C	7	4.0 × 10−2
	hsa04614	Renin-angiotensin system	4	5.5 × 10−2

,

Table A7. Summary of top GO term and KEGG pathway analysis for the Pink Module.

Category	Term		Count	p-Value
BP	GO:0051301	cell division	92	1.6 × 10−49
	GO:0007049	cell cycle	58	1.7 × 10−22
	GO:0000122	negative regulation of transcription from RNA polymerase II promoter	53	3.7 × 10−3
	GO:0006281	DNA repair	52	5.2 × 10−21
	GO:0006260	DNA replication	39	4.5 × 10−26
CC	GO:0005634	nucleus	352	1.6 × 10−30
	GO:0005829	cytosol	321	4.0 × 10−26
	GO:0005654	nucleoplasm	296	1.1 × 10−43
	GO:0005737	cytoplasm	268	1.7 × 10−9
	GO:0005813	centrosome	77	7.1 × 10−21
MF	GO:0005515	protein binding	578	3.4 × 10−26
	GO:0003677	DNA binding	96	2.0 × 10−10
	GO:0005524	ATP binding	94	5.2 × 10−7
	GO:0003682	chromatin binding	47	2.7 × 10−9
	GO:0016887	ATPase activity	46	1.4 × 10−10
KEGG	hsa04110	Cell cycle	52	8.7 × 10−35
	hsa04218	Cellular senescence	20	5.7 × 10−6
	hsa05166	Human T-cell leukemia virus 1 infection	20	6.9 × 10−4
	hsa05203	Viral carcinogenesis	15	2.2 × 10−2
	hsa04115	p53 signaling pathway	12	9.1 × 10−5

,

Table A8. Summary of top GO term and KEGG pathway analysis for the Purple Module.

Category	Term		Count	p-Value
BP	GO:0006357	regulation of transcription from RNA polymerase II promoter	58	7.9 × 10−2
	GO:0007165	signal transduction	50	2.5 × 10−2
	GO:0045944	positive regulation of transcription from RNA polymerase II promoter	46	7.8 × 10−2
	GO:0000122	negative regulation of transcription from RNA polymerase II promoter	39	7.4 × 10−2
	GO:0016310	phosphorylation	30	1.1 × 10−2
CC	GO:0005634	nucleus	220	7.5 × 10−5
	GO:0005829	cytosol	204	1.0 × 10−4
	GO:0005737	cytoplasm	203	4.2 × 10−4
	GO:0005654	nucleoplasm	154	1.1 × 10−4
	GO:0005739	mitochondrion	61	2.8 × 10−3
MF	GO:0005515	protein binding	437	6.7 × 10−5
	GO:0046872	metal ion binding	106	1.4 × 10−2
	GO:0003723	RNA binding	60	2.5 × 10−2
	GO:0003700	transcription factor activity, sequence-specific DNA binding	32	4.9 × 10−3
	GO:0003676	nucleic acid binding	18	1.7 × 10−2
KEGG	hsa04371	Apelin signaling pathway	10	3.0 × 10−2
	hsa04926	Relaxin signaling pathway	9	4.9 × 10−2
	hsa03083	Polycomb repressive complex	8	1.6 × 10−2
	hsa00520	Amino sugar and nucleotide sugar metabolism	6	1.8 × 10−2
	hsa05219	Bladder cancer	5	3.8 × 10−2

and

Table A9. Summary of top GO term and KEGG pathway analysis for the Tan Module.

Category	Term		Count	p-Value
BP	GO:0045944	positive regulation of transcription from RNA polymerase II promoter	51	2.1 × 10−3
	GO:0000122	negative regulation of transcription from RNA polymerase II promoter	43	3.2 × 10−3
	GO:0045893	positive regulation of transcription, DNA-templated	38	1.1 × 10−4
	GO:0006355	regulation of transcription, DNA-templated	38	2.0 × 10−2
	GO:0016310	phosphorylation	33	3.5 × 10−4
CC	GO:0005634	nucleus	252	2.0 × 10−17
	GO:0005829	cytosol	219	1.3 × 10−11
	GO:0005654	nucleoplasm	214	1.2 × 10−27
	GO:0005737	cytoplasm	199	2.6 × 10−6
	GO:0005739	mitochondrion	51	2.7 × 10−2
MF	GO:0005515	protein binding	434	3.5 × 10−12
	GO:0003723	RNA binding	99	1.5 × 10−15
	GO:0046872	metal ion binding	95	3.6 × 10−2
	GO:0005524	ATP binding	62	5.4 × 10−3
	GO:0003677	DNA binding	60	6.8 × 10−4
KEGG	hsa03040	Spliceosome	18	1.2 × 10−4
	hsa05014	Amyotrophic lateral sclerosis	18	2.6 × 10−2
	hsa05202	Transcriptional misregulation in cancer	12	1.9 × 10−2
	hsa03015	mRNA surveillance pathway	9	5.6 × 10−3
	hsa00310	Lysine degradation	6	3.1 × 10−2

while the top terms are shown in Figure 5. It can be observed that most of the modules primarily affect transcription regulation, as indicated by the midnightblue, cyan, tan, and purple modules, while the pink module was indicated to affect cell division. The highly preserved modules operate within the nucleus or cytosol, while all of them are associated with protein binding. The KEGG pathway analysis showed that the pink module affects the cell cycle, while the midnightblue module showed high similarity in pathways involved in Herpes Simplex Virus (HSV) infection, and cyan, tan, and purple modules were involved in transcription-related pathways algorithm (Table S1).

3.4. Protein–Protein Interaction (PPI) Networks and Hub Genes

The STRING knowledgebase provided the protein-protein interaction (PPI) for the identified hub genes. The PPI network was specified to focus on protein interaction in humans. [26]. Through STRING, PPI networks were constructed to illustrate the gene interactions for each preserved module between the proteins of the corresponding genes in the highly preserved modules [26]. The threshold for the protein–protein interaction network usage was set at high confidence (>0.7). Each PPI network constructed was imported to Cytoscape, and the CytoHubba feature revealed the top 10 hub genes within each module of interest based on the degree centrality algorithm (Table S2). Figure 6 shows the hub genes for each highly preserved module, where the genes in red color exhibit higher interaction scores. Since the identified hub genes exhibit high interaction scores within the PPI networks of their respective modules, further analysis would be beneficial in elucidating their potential roles in the key pathways and processes associated with gynecological diseases.

3.5. Signature-Based Approach for Drug Repurposing

The drug repurposing was based on the deregulated genes, differentially expressed genes (DEGs), revealed after the GEO2R in contrast with gynecological disease versus normal or control samples [27]. Those with fold-change (FC) of positive values were considered upregulated, while genes with negative values were downregulated. The transcriptional and molecular signatures of the upregulated and downregulated hub genes for each highly preserved module were analyzed through the DRE web server [https://www.drugrep.org/accessed on 16 April 2024].

Table 2. Top drug candidates based on the upregulated and downregulated hub genes.

Expression	Genes	Drug	Mechanism	Tau	FDR
Upregulated	KRAS, HNRNPA1, SOS1, YWHAZ, AKAP6, LARP7, PTK2, UBE2D3, SMAD4, CASP3, SQSTM1, ESR1, NRXN1, ENPEP, GPX7, CDK1, CCNB1, TOP2A, CCNA2, BUB1B, EXO1, AURKB, CDC20, BUB1, CDC45, EGFR, MRPS15, GMPS, MARS1, ELOC, POLR1B, CTNNB1, RBM39, HNRNPA2B1, RBM25, DDX17, SRSF11, RACK1, and STAT3	Thiamphenicol	bacterial 50S ribosomal subunit inhibitor	−99.8	5.24 × 10−3
		Trimethoprim	dihydrofolate reductase inhibitor	−99.7	9.69 × 10−3
		Medrysone	glucocorticoid receptor agonist	−99.1	5.79 × 10−3
		Pentolinium	cholinergic receptor agonist	−98.9	4.72 × 10−4
		Paroxetine	selective serotonin reuptake inhibitor	−98.8	1.50× 10−3
Downregulated	CALML6, ITIH4, CDK6, DISC1, CD9, TP53, PES1, VAMP2, FUS, and SNRNP70	Propofol	benzodiazepine receptor agonist	−99.5	2.11 × 10−3
		Fluconazole	sterol demethylase inhibitor	−99.4	1.74 × 10−3
		Dapsone	bacterial antifolate	−99.1	8.96 × 10−3
		Hydrocortisone	glucocorticoid receptor agonist	−98.8	8.22 × 10−3
		MDL73005EF	serotonin receptor antagonist	−98.7	4.68 × 10−3

presents the top drug candidates along with their relevant mechanisms of action. Emphasis was given on drugs that exhibited more negative Tau values and lower FDR, as compounds with more negative Tau values induce a more countering effect on the genes provided, while those with low FDR are associated with fewer false positive drug discovery results. For the upregulated hub genes, thiamphenicol, trimethoprim, medrysone, pentolinium, and paroxetine were the top drug candidates, whereas propofol, fluconazole, dapsone, hydrocortisone, and MDL73005EF were the top drug candidates for the downregulated hub genes.

4. Discussion

4.1. Gene Co-Expression Modules across the Datasets

Early detection has been one of the leading causes of poor prognosis in patients suffering from gynecological cancers. This led to the development of several techniques for early detection and screening of gynecological cancers, such as the identification of biomarkers or liquid biopsy analyses [28,29,30]. Furthermore, a recent study has also discovered an increased risk for uterine sarcoma and endometrial cancer following endometriosis or pelvic inflammatory disease (PID) [31]. WGCNA analysis [32] was applied to scan for existing disease networks in gynecological cancers, represented by the OC, CC, and EC datasets with regard to the EM dataset. The functional annotation and pathway enrichment analysis through DAVID gene ontology has shown that most of the highly preserved modules, cyan, purple, midnightblue, and tan modules, were primarily involved with transcription dysfunction. The KEGG pathway analysis of these modules showed their involvement in spliceosome activity, actin cytoskeleton regulation, and Herpes simplex virus 1 infection. Given that these cellular events are associated with inflammation and immune response [11,12,33,34], it can be speculated that inflammation is a significant contributor to gynecological diseases. This can be further elaborated in

Table 3. Summary of KEGG pathways related to inflammation and immune response in identified modules.

Module/s	KEGG Pathway	p-Value	Reference
Cyan	Ras signaling pathway	1.0 × 10−1	[36]
Midnightblue, Green	Herpes simplex virus infection	3.6 × 10−7	[40]
Pink	p53 signaling pathway	9.1 × 10−5	[41]
Purple	Apelin signaling pathway	3.0 × 10−2	[42]
Tan	Spliceosome	1.2 × 10−4	[43]
Brown	Cytokine-cytokine receptor interaction	5.3 × 10−8	[44]
Darkgrey	Relaxin signaling pathway	1.01 × 10−1	[37]
Lightcyan	Antigen processing and presentation	9.5 × 10−23	[45]
Magenta, Skyblue	MAPK signaling pathway	2.1 × 10−3	[46]
Orange	IL-17 signaling pathway	4.0 × 10−15	[35]
Steelblue	Viral carcinogenesis	1.1 × 10−18	[47]

, where several identified modules exhibited KEGG pathways that were found to be relevant in inflammation and immune response, including IL-17 signaling, Ras signaling, relaxin signaling, cytokine-cytokine receptor interaction, p53 signaling, and even viral carcinogenesis and HSV infection. The IL-17 signaling is a key pathway in inflammation as it promotes the expression of pro-inflammatory cytokines and chemokines and recruits immune cells to sites of inflammation [35]. On the other hand, Ras signaling is involved in the activation of inflammatory response [36], while relaxin signaling is involved in pathways that suppress inflammation [37]. Cytokine–cytokine receptor interaction refers to molecular pathways that describe the interaction between cytokines and cell surface receptors that regulate inflammation and immune response [38]. Lastly, viral infections among gynecological diseases have been known to cause chronic inflammation that could potentially cause carcinogenesis [39].

4.2. Module Hub Genes and Their Protein Functions

4.2.1. Inflammatory Pathways Affecting Transcription Dysfunction

Among the hub genes identified (Figure 6), it can be observed that most hub genes with high interaction scores are involved in signaling pathways that are known to regulate inflammation. Such genes that are primarily involved in the NF-kB signaling pathway, whose primary function is to induce the expression of pro-inflammatory cytokines and chemokines, were included [48]. The upregulation of the CASP3 gene indicates increased pyroptosis, which was found to be negatively correlated with the survival of cervical cancer patients [49], while the SQSTM1 gene, known to increase inflammasome activity and oxidative stress-induced inflammation through p62, was also upregulated [48]. The downregulation of the TP53 gene further shows impaired resolution of inflammation, resulting in the failure of the body to effectively halt the inflammatory response and restore tissue homeostasis [50]. The STAT3 gene is a central player in inflammation through the JAK-STAT pathway. Its upregulation further activates the JAK-STAT signaling pathway, which is known to trigger cellular inflammation response, resulting in increased transcription and, eventually, carcinogenesis [51]. On the other hand, the upregulation of the KRAS gene indicates increased activation of the MAPK-ERK signaling pathway that is known to promote cytokine production, inflammasome activation, and amplification of inflammatory signals [52]. The results also indicate the upregulated activation of the mTOR signaling pathway through the upregulated EGFR gene. Overexpressed mTOR signaling is known to increase the production of pro-inflammatory cytokines [53]. All in all, these pathways are also known to regulate RNA splicing and processing [54,55,56,57], which further supports the involvement of inflammatory signals in transcription deregulation observed in gynecological cancer and condition datasets. Conversely, several genes that are involved in RNA splicing and processing, such as HNRNPA1, HNRNPA2B1, RBM39, and SRSF11, were also shown to have a potential impact on inflammation. A recent study showed that the knocked-down function of HNRNPA1 and HNRNPA2B1 genes resulted in decreased expression of IL-6 transcriptional activity [58]. Similarly, the knock-down of the RBM39 gene showed reduced induction of interferon-stimulated genes [59].

4.2.2. Viral Infections Triggering Inflammatory Pathways

In cervical cancer, HPV infection has long been known as a primary risk factor due to its disruption of the normal cellular regulatory mechanisms in cervical epithelial cells [60]. Aside from promoting cell proliferation, it often triggers the persistent release of pro-inflammatory cytokines and chemokines, leading to chronic inflammation. This creates a microenvironment that is prone to mutations and genomic instability, which facilitates the progression of precancerous lesions to invasive cervical carcinoma [7]. Among the highly preserved modules, the midnightblue module exhibited strong similarity with the pathways involved in HSV infection. Like in HPV, HSV infection also causes the release of pro-inflammatory cytokines and chemokines that lead to chronic inflammation, as well as tissue damage and ulceration at the site of infection that triggers immune response [40]. Several studies have indicated a higher risk of cervical cancer with HPV/HSV-2 coinfection [61]. Involved inflammatory pathways that are deregulated upon HPV and HSV infection, such as the NF-kB signaling, MAPK-ERK signaling, and JAK-STAT signaling pathways [40,62,63,64,65,66], were also included in the KEGG pathway results (Table 3). This further indicates the significance of these inflammatory pathways in gynecological disease prognosis.

4.3. Hormonal Imbalance and Inflammation

Estrogen, together with progesterone, are ovarian-derived steroid hormones that participate in facilitating the endometrium cycle in preparation for the menstruation cycle. However, overexpression of estrogen may occur during this cycle, which could potentially lead to progesterone resistance that may cause hormonal imbalance and inflammation sensitivity [67,68,69,70]. Progesterone resistance was thought to reduce cellular responsiveness to progesterone receptors, which may lead to gynecological diseases [67,71]. In endometriosis, debris from menstrual reflux containing endometrial glands and stroma in the fallopian tubes to the pelvic cavity causing lesions and/or extensive adhesion to lesions is its primary cause [67,71] High estrogen levels or overexpression may affect the ratio of estrogen receptor alpha and beta (ERα/ERβ) triggering an increase in pro-inflammatory cytokine IL-6 and overexpression of GREB-1 and c-Myc [72].

In the analysis conducted, the upregulation of the ESR1 gene has been observed (Table 2). Furthermore, several hub genes (Figure 6) relating to signaling pathways that are associated with ERα overexpression were also involved, such as the MAPK-ERK, Ras, and JAK-STAT signaling pathways. This upregulation is thought to further amplify the inflammatory activity of these pathways [73,74,75]. Following the results in the PPI network construction and hub genes, ESR1 upregulation has also been observed in the events of gynecological cancers, such as EC, OC, and CC, where a novel prognostic implication and probable specific endocrine therapy associated with ESR1 expression [76]. High levels of estrogen-related receptors were formerly observed in ovarian cancer with poor overall survival in patients [77], while a GWAS study indicates a strong association between genetic risk variants on chromosome 6q25 and endometriosis, implicating the role of the ESR1 gene and its correlated expression with related genes [78]. Similarly, ESR1 mutations were observed in cervical cancer patients, which were thought to participate in the carcinogenesis of cervical squamous cell carcinoma [79]. The ESR1 p.Y537S hotspot mutation was also observed in low-grade endometrial stromal sarcomas with histologic high-grade transformation and resistance to endocrine treatment [80]. Hence, this aligns with the results and may show a possible connection between ESR1 mutations/ERα activation and signaling pathways in gynecological diseases.

Figure 7 shows the identified hub genes, signaling pathways, and biological processes that are potentially related to inflammation and ERα activation. KRAS gene mutations may lead to dysregulation in ERα activation that may lead to cancer proliferation, progression, tumor cell survival response, and metastasis [81,82]. In mice, SOS1 knock-down suppresses the Ras-ERK/P38MAPK/JNK pathway and induces epithelial-mesenchymal transition through NF-kB signaling, which may contribute to the metastasis of epithelial ovarian cancer cells [83]. The SOS1 gene is a co-repressor of ERα and coactivates STAT3 [84], which may be influenced by the leptin-induced signal transduction that may result in transactivation of STAT3 [85]. ERα activation may also affect the EGFR gene expression and may result in the upregulation of the mTOR/PI3K/AKT/MAPK signaling pathway and cause tumor growth [86]. CTNNB1 gene mutations are associated with the recurrence of endometrial cancer and overall survival [87], possibly modulated by estrogen [88]. On the other hand, TP53 hotspot mutations may lead to mutated p53 function and were found to be common among high-grade serous ovarian carcinoma undergoing primary treatment [89]. The TP53 gene was also shown to have a strong hormonal influence through the ER signaling pathway, which may serve as a prognostic marker [90], as ERα activity could repress p53 transcription and potentially lead to decreased apoptosis and increased tumor cell proliferation [91,92]. This further supports the results, as the TP53 gene was found to be downregulated in EM, OC, CC, and EC. In TGF-β/SMAD signaling, TGF-B1 inhibits the growth of normal ovarian surface epithelial cells and triggers the nuclear translocation of SMAD4 [93,94]. ERα activation is known to affect the TGF-β signaling pathway, which may act as both a suppressive and promotive pathway for inflammation [95]. Similarly, the Wnt signaling pathway may also promote carcinogenesis, malignant transformation, and inflammation in cancer cells [96]. The PI3K-Akt signaling pathway also contains downstream protein targets, which may promote inflammation and are upregulated upon ERα activation [97]. Lastly, the transcriptional processes of RNA processing and splicing, which may both contribute to inflammation by promoting the expression of inflammatory genes and immune cell activity, were also found to be influenced by ERα activation [98].

4.4. Signature-Based Approach for Drug Repurposing

Several inflammatory pathways involved in transcription regulation, viral infection, and estrogen receptor activation have been highlighted (Table 3). Through the DRE web server, the top 5 hub genes of the upregulated and downregulated hub genes were identified. For the upregulated genes, the leading candidate drug is thiamphenicol, which is primarily known for inhibiting bacterial protein synthesis and used in treating PID. PID is a condition that has long been known to increase the risk of ovarian cancer as it induces sustained inflammation in the ovarian microenvironment. A recent study has reported that a history of PID increases the risk of having borderline ovarian tumors [99]. Hence, thiamphenicol could be a potential therapeutic option in ovarian cancer and other gynecological diseases due to its anti-inflammatory properties [100,101]. On the other hand, propofol was the leading drug candidate based on the downregulated hub genes. It is a benzodiazepine receptor agonist that is commonly used as an anesthetic agent [102], although previous studies have shown that the drug possesses anti-inflammatory effects that may play a role in modulating cancer progression and immune response [103]. This may be due to benzodiazepine derivatives that have been found to exhibit anti-inflammatory and analgesic tendencies [104].

Other drug candidates include trimethoprim, which inhibits the activity of the dihydrofolate reductase enzyme, which has long been a target for anti-inflammatory and anti-cancer treatment [105,106]. Another is a cholinergic receptor agonist, pentolinium, which is commonly used as an antihypertensive. Cholinergic receptors are known to activate anti-inflammatory pathways [107]. Fluconazole, a sterol demethylase inhibitor, is used to treat fungal or yeast infections such as vaginal candidiasis and various candida infections causing inflammation in other parts of the body [108]. On the other hand, dapsone is a bacterial antifolate that is commonly used to treat leprosy and dermatitis herpetiformis but also has known anti-inflammatory effects due to inhibiting reactive oxygen species production [109]. Furthermore, glucocorticoid receptor agonists, such as medrysone and hydrocortisone, were also included. This may be due to glucocorticoid receptor activation repressing gene expression of several biological processes, including inflammation [110]. Lastly, paroxetine, a selective serotonin reuptake inhibitor, and an experimental serotonin receptor antagonist drug, MDL73005EF, were also included. The role of serotonin as a chemotactic agent in inflammation, by increasing pro-inflammatory cytokine secretion, has been studied in the past years [111]. Overall, the identified drug candidates are known to have anti-inflammatory effects that could potentially inhibit the progression of gynecological diseases, as suggested in the analysis conducted. Studies have also indicated that anti-inflammatory drugs are beneficial in decreasing menstrual bleeding, postoperative treatment, and pain management in endometriosis, further supporting these results [112,113].

5. Conclusions

In this study, five highly preserved modules in different estrogen-dependent gynecological diseases that were represented by EM (GSE51981), OC (GSE63885), CC (GSE63514), and EC (GSE17025) datasets were successfully determined. Through WGCNA, the top 10 hub genes per module were identified and used to screen for potential drug candidates. Moreover, the functional annotation and pathway enrichment analysis allowed for the elucidation of the complex molecular basis for endometriosis and gynecological diseases, which highlighted several inflammatory pathways that are linked with transcription regulation. Based on the hub genes, signaling pathways associated with viral infections, particularly HSV, associated with gynecological cancers were also identified. The involvement of the ESR1 gene, which encodes for ERα, a known modulator of signaling pathways involved in inflammation, further indicates its importance in gynecological disease prognosis. The drugs thiamphenicol, a bacterial ribosomal subunit inhibitor, and propofol, a benzodiazepine receptor agonist, were the top candidates identified in the signature-based drug repurposing. Other drug candidates include a dihydrofolate reductase inhibitor, glucocorticoid receptor agonists, cholinergic receptor agonists, selective serotonin reuptake inhibitors, sterol demethylase inhibitors, bacterial antifolate, and serotonin receptor antagonist drugs which have known anti-inflammatory effects, demonstrating that the gene network can be used as a therapeutic avenue to design drug candidates for gynecological diseases. However, further evaluation through wet laboratories is required as the data presented were conducted in silico.