In this study, data sets are from the National Center for Biotechnology Information (NCBI) pool of gene expression (GEO, https://www.ncbi.nlm.nih.gov/geo/). The data in the GEO database we used contained preprocessed and normalized data, which is suitable for direct use in downstream analysis. Three endometriosis datasets, GSE7305, GSE23339, and GSE25628, and one endometrial cancer dataset, GSE17025, were used. The GSE7305 dataset contains 10 ectopic endometrium and 10 samples of normal endometrium. GSE23339 contains 10 cases of ectopic endometrium and 9 cases of normal endometrium. GSE25628 contains 7 cases of ectopic endometrium, 9 cases of eutopic endometrium, and 6 cases of normal endometrium. GSE17025 contains 12 cases of normal endometrium and 91 cases of tumor. The downloaded data were processed using the R package limma followed by calibration, standardization, and log2 transformation, and details about the dataset are summarized in Table 1 .

R Studio software was utilized to process and standardize the files. Four GEO datasets were preprocessed. For the genes corresponding to multiple probes, the average value was taken as the expression level of the gene, and the expression levels of all genes were converted into log2 (expression level). The “limma” R package was applied to analyze the differences between the experimental and control groups. The threshold of difference analysis was set as p-value <0.05 and FC >1.5. Overlapping DEGs from the four database screenings were utilized in subsequent GO enrichment, KEGG pathway analysis, and protein-protein interaction (PPI) analyses. We then used Venn software online (https://jvenn.toulouse.inrae.fr/app/example.html) to visualize the common DEGs.

Enrichment analysis of gene sets provides unique biological properties of genes, including biological processes, cellular components, and molecular functions. KEGG pathway enrichment analysis explores the critical pathways involved in disease occurrence and progression. KEGG and GO enrichment analysis of differential genes was conducted with the R package clusterProfiler, with a p-value threshold of less than 0.05. We considered the input dataset to enrich the pathway or term if it met this criterion.

The activity of protein-protein interactions (PPIs) is regarded as a significant target for cell biology research and is vital for elucidating key genes and necessary gene modules during cancer development. Common DEGs (version 11.0, https://string-db.org/) were inputted into STRING, a search tool for retrieving interacting genes, to generate PPI networks with a medium confidence score of 0.400. The STRING database comprises electronically inferred interaction data to ensure the most comprehensive coverage of potential protein interactions possible. The PPIs obtained through the analysis with Cytoscape software (https://cytoscape.org/) were visualized and analyzed using this powerful bioinformatics software. Additionally, the Cytoscape plugin CytoHubba (https://apps.cytoscape.org/apps/cytohubba) was employed to extract Hub genes, and the MCC function of CytoHubba was used to identify the top 10 Hub genes from the PPI network.

To further select precise biomarkers, we performed Gene Expression Profiling (GEPIA, http://gepia.cancer-pku.cn/). This web-based tool provides fast and customizable functionality to display gene expression profiles from the TCGA database while performing survival analysis of genes. To explore the survival differences between different groups. Immunohistochemical (IHC) data for common genes were retrieved in the Human Protein Atlas (HPA) project, which contains detailed information about samples.

NetworkAnalyst (https://www.networkanalyst.ca/) was employed to identify genes, and TF has identified common genetic interactions, the research of transcription factors, and regulation networks between the target genes. NetworkAnalyst is a comprehensive network platform used to perform gene expression on many species, enabling them to be subjected to meta-analysis.TF-gene interaction network of network platform from NetworkAnalyst ENCODE (https://www.encodeproject.org/) contained in the database.The ENCODE database provides abundant information on transcription factor binding sites, which can provide strong support for the study of TF gene interactions.

TF miRNA co-regulation interaction refers to the complex interaction among Transcription factors, miRNAs, and their target genes. The interaction of TF miRNA co-regulation was collected from the RegNetwork reservoir. Transcription factors are proteins that can bind to specific sequences of DNA, and miRNAs are small RNA molecules that regulate the transcription and translation process of genes by binding to the miRNA of target genes, thereby affecting the expression level of genes. NetworkAnalyst was used to visualize the TF miRNA co-regulation network.

Tumor immunity infiltration online database (TIMER, https://cistrome.shinyapps.io/timer/) is applied to analyze different types of cancer immunity infiltration. TIMER can analyze the relationship between the cancer immune cell infiltration level and hub genes. The relationship between somatic copy number alterations (SCNA) of potential prognostic hub genes and infiltrating immune cells was explored through related modules. A cutoff value of P < 0.05 was applied as the threshold.

Drug molecular identification is a key component of research, and the drug-Gene Interaction Database (DGIdb, http://www.dgidb.org) is a web-based resource that weaves the genome of druggable genomes into known drug interactions and potential druggable targets. We input the co-expressed DEGs and evaluate the DEGs in DGIdb to find potential therapeutic drugs.

The human endometrial cancer cell line (HEC - 1B and Ishikawa) and the normal endometrial epithelial cell line (HEEC) were both purchased from ATCC. These cell lines were cultured in a culture flask containing DMEM medium supplemented with 10% fetal bovine serum and 1% penicillin-streptomycin in an incubator set at 37°C with 5% CO₂. Cells from the endometrial cancer cell lines and the normal endometrial cell line, both in the logarithmic growth phase, were collected separately. Total RNA was isolated from the cells following the manufacturer’s instructions using the PureLink RNA Mini Kit (Invitrogen, Carlsbad, CA, USA), and the RNA was reverse transcribed using the PrimeScript RT Kit with gDNA Eraser (Takara, Dalian, China). Subsequently, RT-qPCR amplification was conducted using PowerUp SYBR Green Master Mix (Thermo) according to the manufacturer’s protocol, with GAPDH serving as the reference gene. The PCR protocol consisted of an initial denaturation at 95°C for 15 seconds followed by 40 cycles at 95°C for 5 seconds, and 60°C for 1 minute. The primer sequences are presented in Table 2 . Each sample was prepared in triplicate.

All four datasets were analyzed for gene expression using the R limma package, and genes with P values <0.05 and Log [FC]>1.5 were designated as DEGs. As depicted in the volcano plot of Figure 1 , specific information about the four datasets is summarized in Table 1 . In the GSE7305 dataset, 2595 DEGs were identified, including 1278 upregulated and 1317 downregulated genes. In GSE2339, 2171 DEGs were identified, with 1016 upregulated and 1155 downregulated genes. In GSE25628, 3842 DEGs were identified, comprising 1819 upregulated and 2023 downregulated genes. In GSE17025, 4790 DEGs were identified, including 2246 upregulated and 2544 downregulated genes ( Table 3 ). The DEGs for each dataset were visualized using volcano plots ( Figures 1A–D ). After standardization, we identified 141 differentially expressed genes ( Figure 1E ) by comparing the DEGs present across the four datasets, with 56 differentially expressed genes being consistent. Among these, 33 were upregulated and 23 were downregulated ( Table 4 ). The top 40 DEGs from all four datasets were visualized on a heat map ( Figures 2A–D ).

KEGG and GO analyses were conducted on the 141 DEGs to explore the biological functions of the integrated DEGs. In Biological Process (BP), DEGs were significantly enriched in processes related to growth and development, regulation of the Wnt signaling pathway, and proteoglycan metabolism ( Figure 3A ). In Cellular Component (CC), DEGs were primarily involved in the collagen-rich extracellular matrix, neuronal cell bodies, and membrane structures ( Figure 3B ). Molecular Function (MF) analysis revealed significant enrichment in glycosaminoglycan binding, cholesterol, and lipid transfer activities ( Figure 3C ). KEGG pathway analysis indicated that DEGs were mainly enriched in the JAK-STAT signaling pathway and leukocyte migration through endothelial cells ( Figure 3D ). Significantly enriched terms were displayed in bubble plots using the R package, with a criterion of P <0.05.

The 56 DEGs were uploaded to the STRING database (https://string-db.org/cgi/) with a confidence score threshold of >0.4, and the resulting PPI network was visualized using Cytoscape software. The PPI network comprised 56 nodes and 17 edges ( Figure 4A ), and the CytoHubba plugin identified the hub genes within the PPI network. Using the MCC algorithm for sorting, the top 10 hub genes selected were APOE, FGF9, TIMP1, BGN, C1QB, MX1, SIGLEC1, BST2, ICAM1, and MME ( Figure 4B ).

TF-gene interactions were collected using NetworkAnalyst. For the hub genes (APOE, FGF9, TIMP1, BGN, C1QB, MX1, SIGLEC1, BST2, ICAM1, MME), TF-gene interactions were analyzed. The interactions between TFs and common DEGs are shown in Figure 5A . The network contains 130 nodes and 171 edges. Among them, ICAM1 is regulated by 69 TFs, and APOE and BST2 are each regulated by 29 TFs. These TFs regulate more than one interaction in the network with a common DEG. TF-miRNA co-regulatory network analysis reveals the interactions of miRNAs and TFs with common DEGs, which may be responsible for regulating the expression of DEGs. The network created by the TF-miRNA co-regulatory analysis consisted of 235 nodes and 307 edges. Figure 5B shows the co-regulatory network of TF-miRNAs.

The top 10 hub genes were used for subsequent analyses. The GEPIA database was used to investigate differences in gene expression levels between early tumor tissues and normal tissues ( Figure 6A ). The results showed that four hub genes, APOE, BGN, BST2, and C1QB, were abnormally differentially expressed in normal tissues and endometrial cancer. Additionally, the immunohistochemical (IHC) staining extracted from the Human Protein Atlas Database (HPA) also showed differential protein expression levels of hub genes, as shown in Figure 6B . Among the 10 hub genes, only the down-regulated genes APOE and BGN were associated with EC prognosis, confirming the reliability of the hub genes we identified ( Figure 6C ).

The TIMER database was used to analyze the correlation between mRNA expression of APOE, BGN, BST2, and C1QB and infiltrating immune cells in cancer tissues. APOE was correlated with the abundance of B cells (cor=0.419, P=1.00e−13), CD4+ T cells (cor=0.458, P=1.86e−16), macrophages (cor=0.332, P=6.25e−09), and neutrophils (cor=0.246, P=0.005). Dendritic cell abundance (cor=0.258, P=7.66e−06) also showed significant correlation ( Figure 7A ). BGN showed a significant correlation with the abundance of CD4+ T cells (cor=0.173, P=3.08e−03) ( Figure 7B ). BST2 was significantly correlated with the abundance of B cells (cor=0.136, P=2.09e−02), CD8+ T cells (cor=-0.238, P=4.28e−05), CD4+ T cells (cor=0.181, P=1.98e−03), neutrophils (cor=0.17, P=0.017), and CD4+ T cells (cor=0.17, P=0.016) ( Figure 7C ). C1QB was significantly correlated with B cells (cor=0.633, P=1.10e−33), CD8+ T cells (cor=0.255, P=1.15e−05), CD4+ T cells (cor=0.502, P=5.06e−20), macrophages (cor=0.464, P=0.005), neutrophils (cor=0.459, P=1.21e−16), and dendritic cells (cor=0.624, P=7.24e−36). These results provide strong evidence that APOE, BGN, BST2, and C1QB play crucial roles in infiltrating immune cells, including B cells, CD8+ T cells, CD4+ T cells, macrophages, neutrophils, and dendritic cells.

We entered the 10 most significantly expressed DEGs into the DGIdb database to identify drug-gene interactions and potential drug targets. APOE was associated with 17 drugs for treating endometriosis, ICAM1 with 5 drugs, and MME with 8 drugs for treating endometriosis ( Table 5 ). The potential applicability of these drugs as targets awaits further investigation through future cellular and animal studies.

To further validate our bioinformatics findings, we performed in vitro experiments. To further validate our bioinformatics findings, we conducted in vitro experiments. Endometrial epithelial cells HEEC were used as a control. We confirmed by qRT-PCR that the mRNA expression levels of APOE, BGN, BST1, and C1QB in endometrial cancer cell lines (HEC-1B and Ishikawa) were consistent with our previous results. The mRNA expression levels of APOE, BST1, and C1QB in endometrial cancer cell lines were significantly higher than those in normal endometrial epithelial cells, whereas BGN expression was higher in normal endometrial cell lines than in endometrial cancer cell lines ( Figures 8A–D ).