Immune-related hub genes in intrauterine adhesions: a bioinformatics approach

Study design

We implemented four-phase analytical framework to study the pathogenesis of IUA. (1) The results of primary screening identified IUA-associated DEGs using the GSE224093 database using the following criteria: —log₂FC— ≥ 1 and Benjamini–Hochberg (BH) false discovery rate (FDR)-adjusted p < 0.05). We independently curated the immune-related genes (IRGs) from ImmPort. (2) Functional convergence analysis was conducted using Venn intersection to identify overlapping gene. This analysis was followed by characterizing these genes through Gene Ontology (GO) enrichment analysis, Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis, Sankey network visualization, and gene set enrichment analysis (GSEA) using the KEGG gene sets. (3) We used weighted gene co-expression network analysis WGCNA to identify the conserved IUA modules, the results of which were confirmed through eigengene heatmaps and DEG-IRG correlation bar plots. (4) The seven identified hub genes were experimentally verified through qPCR in human and murine models, validating four genes (IGF1, WNT5A, BIRC5, and GDF7) that showed complete transcriptional agreement (p < 0.05, Pearson’s r > 0.85). Protein-level results were obtained with immunohistochemistry (IHC) (Fig. 1). This integrated approach combined bioinformatic methods with experimental validation, providing comprehensive molecular insights into the development of IUA.

Data sources and DEG screening

GSE224093, an IUA microarray dataset, was retrieved from the GEO database. The dataset comprises endometrial samples from seven patients with severe IUA and seven normal controls, profiling 33,944 genes. All data are publicly accessible.

We used the GEO2R optimized Limma pipeline to ensure robust and unbiased differential gene expression analysis of the GSE224093 microarray dataset as well as address the inherent data challenges, including the nonunique gene identifiers and Fragments Per Kilobase of transcript per Million mapped reads (FPKM)-normalized values incompatible with count-based methods. This platform-specific approach avoided the need for arbitrary gene filtering with DESeq2, which requires unique gene symbols and integer counts, as well as problematic data transformations for Limma-voom applications. We thus preserved potentially significant biological signals. The analysis involved (1) automated probe-to-gene annotation using platform-specific GEO Platform Library (GPL) files; (2) quantile normalization with low-expression filtering, retaining features with log2 intensity >6 in ≥50% of samples; and (3) linear modeling with empirical Bayes moderation (eBayes) to assess group-wise differences and adjust for heteroscedasticity. Differential gene expression was determined using two stringent thresholds, —log2FC— ≥ 1 with Benjamini–Hochberg (BH) FDR-adjusted p < 0.05. The quality of the results was controlled using diagnostic volcano plots. The top 100 significant DEGs were visualized via hierarchical clustering in Multiexperiment Viewer (MeV).

Identifying IRGs and intersecting genes

We obtained a set of 1,794 IRGs from the ImmPort database to identify the IRGs associated with IUA. We used an online bioinformatics platform (http://www.bioinformatics.com.cn/) to intersect these IRGs with the DEGs identified from IUA samples. The results identified 22 overlapping genes as immune-related candidates involved in the pathogenesis of IUA.

Functional enrichment analysis of intersecting genes

We performed GO and KEGG pathway enrichment analyses to characterize the biological functions and pathways associated with the DEGs. The GO (http://www.geneontology.org/) and KEGG pathway (http://www.genome.jp/kegg/pathway.html) analyses were conducted using the Database for Annotation, Visualization, and Integrated Discovery (DAVID;  https://davidbioinformatics.nih.gov) (Givant-Horwitz, Davidson & Reich, 2005). Terms with a threshold of p < 0.05 were considered significantly enriched. Additionally, the Metascape database (https://metascape.org/) was used to annotate and functionally cluster the DEGs and intersecting IRGs involved in IUA.

Screening and verifying hub genes

We applied WGCNA on the GEO datasets to additionally validate the results. We first preprocessed the data through calculating the median absolute deviation and removing outliers using the WGCNA goodSamplesGenes method. We then constructed a scale-free coexpression network by computing Pearson’s correlation matrices. We established a weighted adjacency matrix (A_mn = —C_mn—^β, β = 12), which we transformed into a topological overlap matrix (TOM). We identified gene modules using TOM-based dissimilarity (1-TOM) with average linkage hierarchical clustering (minimum module size = 30, sensitivity = 3). We merged the gene modules with a dissimilarity of <0.25, identifying the significant saddlebrown module, containing 3,272 genes. The highly connected hub genes in this module were intersected with the DEGs and IRGs (DEGs–IRGs–saddlebrown), from which we selected the seven most significantly dysregulated genes (—log2FC— ≥ 1, BH-FDR p < 0.05) whose expression patterns consistently matched our preliminary DEG analysis. We thus confirmed their identification as hub genes in IUA pathogenesis.

Correlation analysis between gene expression and immune cell infiltration

We conducted computational immunogenomic analysis to characterize the immune microenvironment in IUA and investigate the relationship of this microenvironment with gene expression patterns. The normalized gene expression matrix from all training cohort samples was analyzed using CIBERSORTx (version 1.0) with the LM22 signature matrix and 1,000 permutations, disabling quantile normalization to accurately deconvolute the relative proportions of 22 functionally distinct immune cell subsets. We controlled data quality by excluding samples with CIBERSORTx p-values ≥ 0.05. Correlation analysis was performed using Spearman’s rank correlation coefficient, with multiple testing correction via the BH method (FDR < 0.05 considered significant). We used deconvolution algorithms (xCell and Microenvironment Cell Populations (MCP)-counter) for comparative analysis to validate the robustness of our findings. This multialgorithm approach allowed us to comprehensively assess immune cell infiltration patterns and their potential associations with gene expression profiles in the pathogenesis of IUA.

qRT-PCR

IHC

The IUA and normal endometrial tissue samples were fixed in 10% neutral buffered formalin, embedded in paraffin, and sectioned into four µm thick slices. The samples were deparaffinized and rehydrated. Then, antigens were retrieved using citrate buffer (pH 6.0). Nonspecific binding sites were blocked with 5% bovine serum albumin solution.

The primary antibodies were incubated using a rabbit monoclonal antibody specific to the target protein. The optimal antibody concentration was determined through preliminary experiments on human and mouse endometrial tissues, including the IUA and control groups, in accordance with the manufacturer’s protocol. This predetermined concentration was then applied to all study samples.

We incubated the samples with a horseradish peroxidase (HRP)-conjugated antirabbit IgG secondary antibody, which was followed by chromogenic development with 3,3′-diaminobenzidine. The sections were counterstained, dehydrated, and mounted with neutral resin for microscopy evaluation.

All immunohistochemical slides were independently reviewed by a board-certified pathologist. The staining intensity was quantitatively analyzed using ImageJ software. The results are expressed as the percentage positive area (% Area) for statistical comparison.

Sample sources and primary reagents

We confirmed the hub gene differential expression in clinical and mice cohorts. The sample sizes across all validation platforms (qPCR: humans, 4–8/group; mice, 4–6/group. IHC: humans, 5–8 IUA vs. five controls; mice, five IUA vs. five controls) were determined considering the statistical requirements and practical constraints. We incorporated (1) prospective power calculations (G*Power 3.1), anticipating large effect sizes (Cohen’s d ≥ 1.2 from preliminary bioinformatics analyses), which confirmed 72–85% power in detecting significant differences (d ≥ 1.0, α = 0.05 two-tailed) after BH correction; (2) quality control thresholds (RNA Integrity Number (RIN) ≥7 for qPCR) that inherently limited specimen inclusion; (3) alignment with established methodological standards for nonparametric analyses (Wittwer et al., 2009); (4) ethical and logistical limitations in clinical specimen acquisition. These sample sizes enabled the robust detection of large-effect biomarkers as evidenced by the consistent confirmation of the hub genes; however, we acknowledge the low sensitivity in identifying moderate effects (d = 0.5−0.8), a limitation mitigated through our multiplatform validation approach. As such, larger cohorts should be included in future studies to comprehensively characterize subtle regulatory differences.

DEGs in IUA

The GSE224093 dataset was downloaded from the GEO website, comprising 33,944 genes. —log₂FC— ≥ 1 and p < 0.05 were used as the filters to identify the DEGs. A total of 311 DEGs were obtained: 190 down- and 121 upregulated genes, respectively, and 33,633 genes that did not significantly differ in expression between the groups (Fig. 2A). The expression levels of the top 100 significant DEGs were analyzed (Fig. 2B). We then downloaded 1794 IRGs from Immport (http://www.immport.org/home). We used the intersection with DEGs in IUA, identifying 22 intersecting genes (Fig. 2C). We identified six downregulated genes through analyzing the expressions of the intersecting genes in the GSE224093 dataset (Figs. 2D, 2E): BIRC5, LGR5, TUBB3, IGF1, GDF7, and WNT5A. We also identified 16 upregulated genes: ANGPTL1, ACKR1, EDNRB, NPR3, PTGDS, CCL14, PTGFR, APOD, CCL21, LYZ, CXCL10, PTGER3, PI15, FGF7, C3, and DES.

DEG functions and pathway enrichment

The genes overlapping among the IUA-associated DEGs and IRGs were functionally annotated and clustered using the Metascape database (https://metascape.org/). GO, KEGG pathway enrichment, and Sankey diagram analyses were performed via an online bioinformatics platform (http://www.bioinformatics.com.cn/). The results of GO enrichment analysis revealed that the overlapping genes in the biological process category were predominantly associated with inflammatory response, chemotaxis, and protein phosphorylation. These results suggest a strong link between dysregulated immune processes and IUA pathogenesis (Fig. 3A). The results of KEGG pathway analysis demonstrated substantial enrichment of the DEGs in the Wnt signaling pathway, indicating a role in IUA progression (Fig. 3B). The Sankey diagram illustrated the gene–pathway relationships, showing that each gene mapped to 1–10 pathways, and each pathway contained one-six associated genes (Fig. 3C). GSEA was then conducted on the —log₂FC—-ranked gene list, using KEGG gene sets to identify broader pathway-level alterations. The GSEA results highlighted considerable enrichment in cell cycle regulation, oocyte meiosis, and spliceosome activity, with consistent activation and inhibition trends across these pathways (Fig. 3D).

Gene validation

We performed WGCNA on the GEO database again to further screen the target genes. We analyzed the GEO database using a soft sign-independent threshold (Fig. 4A), an average continuity soft threshold (Fig. 4B), sample clustering (Fig. 4C), gene coexpression (Fig. 4D), a module feature vector (Fig. 4E), module phenotypic correlation (Fig. 4F), and a differentially expressed module (Fig. 4G). We obtained 3,272 differential module genes. We obtained the genes that intersected among the WGCNA differential module and DEGs, and IRGs (DEGs–IRGs–saddlebrow) (Fig. 4H). Seven significant genes were obtained, of which TUBB3, WNT5A, GDF7, IGF1, and BIRC5 were downregulated and PTGDS and CCL14 were upregulated (Figs. 4I, 4J).

Clinical relevance of key genes

We identified seven key hub genes using bioinformatics methods and explored the molecular mechanisms underlying IUA: TUBB3, WNT5A, GDF7, IGF1, BIRC5, PTGDS, and CCL14. We conducted functional enrichment and pathway analyses to determine the roles of these genes in IUA. These genes are significantly enriched in biological processes related to inflammatory responses, chemotaxis, and the regulation of cytokine production, among other immune-related processes. These genes are implicated in several key signaling pathways, such as cytokine–cytokine receptor interactions, the PI3K/Akt signaling pathway, and the TGF-β signaling pathway, all of which may play roles in the development of IUA.

We performed GSEA to validate the involvement of these genes in IUA. These hub genes were found to be enriched in biological processes, such as the cell cycle, oocyte meiosis, and RNA splicing, suggesting that these genes influence the pathological mechanisms of IUA through affecting cell proliferation, differentiation, and RNA processing. We constructed a network map of the hub genes and their related signaling pathways, illustrating their complex interactions and their crucial roles in the inflammatory response and fibrosis. Among these genes, TUBB3 is involved in the maturation of anatomical structures and is closely linked to cell cycle regulation, indicating its importance in the remodeling of IUA. The aberrant expression of TUBB3 may contribute to the overproliferation and fibrosis of uterine tissues (Puri, Barry & Engle, 2023). WNT5A was found to be enriched in the Wnt signaling pathway and inflammatory response-related GO terms, indicating roles in regulating inflammation and fibrosis (Kikuchi et al., 2011; Pashirzad et al., 2017). GDF7 was found to be enriched in the TGF-β signaling pathway, suggesting involvement in tissue fibrosis (Morikawa, Derynck & Miyazono, 2016). IGF1 primarily participates in regulating growth factor activity and the PI3K/Akt signaling pathway, implying a role in IUA formation and progression through promoting cell proliferation and inhibiting apoptosis (Werner, 2023). BIRC5 was significantly enriched in anti-apoptosis-related GO terms, indicating a role in protecting IUA tissue through inhibiting apoptosis (Sanhueza et al., 2015; Zhang et al., 2021). PTGDS was found to be mainly associated with GO terms related to prostaglandin metabolic processes and inflammatory responses, suggesting involvement in the inflammation regulation (Crookenden et al., 2023; Chandrasekaran, Weiskirchen & Weiskirchen, 2024). Finally, CCL14 plays a role in regulating chemokine activity regulation and recruiting immune cells, which was found to be enriched in GO terms related to the immune response and KEGG pathways, underscoring roles in the inflammatory response and immune regulation within the uterine cavity (Allen et al., 2009; Zhu et al., 2019).

The results of the GO, KEGG, and GSEA analyses together identified the potential roles of TUBB3, WNT5A, GDF7, IGF1, BIRC5, PTGDS, and CCL14 in IUA. These hub genes may be integral to the pathological processes associated with IUA, influencing cell proliferation, inflammatory responses, fibrosis, and immune evasion.

qPCR results

The results of qPCR validation revealed distinct expression profiles of the hub genes in the human and murine endometrial specimens (Fig. 5). IGF1, WNT5A, BIRC5 and GDF7 expression levels were significantly downregulated in the human IUA groups compared with the controls, whereas the expression levels of TUBB3, PTGDS, and CCL14 were upregulated. The TUBB3 expression level did not agree with the bioinformatic predictions. The expression levels of IGF1, WNT5A, BIRC5, GDF7, TUBB3, and PTGDS were similarly downregulated in the murine specimens. The measured PTGDS expression level in the murine specimens did not agree with the results of computational analysis. Technical limitations prevented the validation of CCL14 expression levels in the murine models due to primer unavailability and the potential gene absence in mouse endometrium. All statistical details are presented in Table 2.

IHC results

The IHC results confirmed substantial differences in the IGF1, WNT5A, BIRC5, and GDF7 protein expression patterns between the normal and IUA groups in the human endometrial and murine tissues. The observed protein expression profiles completely agreed with the transcriptomic patterns identified through our biosignature analysis (Fig. 6). All statistical details are presented in Table 3.

Importance of immune microenvironment in formation of IUA

IUA is common gynecological disorder with complex pathomechanisms that are difficult to treat. We identified an important role of the immune microenvironment in the development of IUA. Immune imbalance impacts the endometrial fibrosis process (Niu et al., 2023), whereas Chen et al. (2022) further identified that CD4+ T-cell-mediated macrophage polarization plays a critical role in the formation of fibrosis. Our results support these findings and demonstrate the changes in the expression of specific immune-related genes in IUA, which may be associated with immune cell activation, inflammatory factor release, and tissue repair processes (Torres-De La Roche, 2019).

Role of hub genes in IUA

We identified key hub genes, such as IGF1, WNT5A, BIRC5, and GDF7, through bioinformatics analysis, which may play roles in regulating inflammation, immune activation, tissue repair, and fibrosis. IGF1 is crucial for tissue repair, fibrosis, and immune responses. IGF1 is involved in regulating tissue repair and fibrosis through the PI3K/Akt and TGF-β signaling pathways, promoting cell proliferation and matrix deposition in hepatic and cardiac fibrosis (Wang et al., 2023). IGF1 downregulation is associated with aberrant fibrosis and impaired tissue repair, highlighting its potential as a therapeutic target (Wang et al., 2018). The roles of WNT5A in immune regulation, cell migration, and tissue remodeling are being increasingly supported. WNT5A is involved in pathological processes such as lung fibrosis and skin scarring via the nonclassical Wnt/Ca²⁺ pathway (Singla et al., 2023; Trinh-Minh et al., 2024). Notably, the downregulation of WNT5A expression levels may hinder fibrosis progression through diminishing local immune responses following trauma (Xue et al., 2022), indicating potential as a diagnostic and prognostic marker of IUA. BIRC5 functions as an antiapoptotic protein that enhances cell survival through inhibiting apoptosis as well as promotes tissue repair by sustaining fibroblasts during the fibrotic process (Horowitz et al., 2012). The downregulation of BIRC5 expression levels leads to reduced fibrosis and compromised tissue repair. GDF7 is a member of the TGF-β family that is primarily involved in neural and muscular differentiation. Although GDF7 has received less attention in fibrotic diseases, GDF7 may play a crucial role in tissue fibrosis through influencing fibroblast differentiation (Kong et al., 2023). Furthermore, GDF7 downregulation may adversely affect tissue repair and fibrosis progression.

In conclusion, these genes play roles in fibrosis, cellular repair, and immune regulation. These genes may be suitable as biomarkers and therapeutic targets for IUA and thus warrant further investigation.

Role of bioinformatics methods in exploring pathogenesis of IUA

Bioinformatics methods can be used to integrate and analyze multiomics data, providing tools for comprehensively understanding the molecular mechanisms of IUA. These approaches may aid in elucidating the molecular mechanisms underpinning IUA, facilitating the discovery of biomarkers and the development of personalized therapies (Liang et al., 2024). We identified the genes closely associated with the pathological processes of IUA using GSEA. The results of GO and KEGG pathway enrichment analyses identified the biological functions of these genes and the signaling pathways involved. We used WGCNA to identify the key hub genes in IUA through constructing gene coexpression networks, through which we determined their potential roles in pathological processes and established regulatory networks. Integrating these methodologies identified potentially suitable molecular targets for diagnosing and treating IUA. These methods are powerful tools that could be used in future mechanistic studies.

Experimental validation

Although the results of bioinformatic analysis produced a large number of candidate genes, experimental validation was a step in ensuring the reliability of our findings. Our qPCR and IHC experiments validated the changes in the expression levels of some key genes in IUA, enhancing our confidence in the determined roles of these genes in IUA. However, the inconsistency between the experimental results and raw confidence analyses suggests that more samples and in-depth mechanistic studies are needed to explain these discrepancies. Further study will increase our understanding of the effects of the differences in the samples, species, and experimental conditions and technical errors on the results. Additional single- and multi-center studies may contribute to our understanding of the changes in the immune microenvironment during the pathogenesis of IUA.

The observed discrepancies between bioinformatic predictions and experimental validation, particularly for TUBB3 and PTGDS, warrant careful consideration of species-specific biological and technical factors. TUBB3 expression was upregulated and downregulated in the human and muring IUA endometrial specimens compared with the controls, respectively. However, the PTGDS expression level was up- and down-regulated in the human and murine IUA samples compared with the controls, respectively. These divergent trends may reflect (1) the fundamental differences in endometrial physiology between species, as murine models incompletely mimic human IUA pathophysiology; (2) technical limitations in cross-species primer design, exemplified by CCL14 being undetectable in mice due to potential gene absence or primer incompatibility; and (3) microenvironmental variations, where human samples represent chronic IUA states versus acute murine injury responses. Such disparities underscore the need for species-matched validation when translating computational findings. The consistent validation of IGF1, WNT5A, BIRC5, and GDF7 between the qPCR and IHC (human/mice p < 0.05) results reinforces their roles in IUA; however, the inconsistency in the genes highlights methodological challenges. The failure to amplify murine CCL14 suggests gaps in murine genome annotation or expression sparsity, urging caution in assuming gene ortholog conservation. Researchers should (1) integrate multiomics methods (e.g., single-cell RNA-seq) to clarify species-specific expression patterns, (2) optimize species-specific primer validation using long-read sequencing, and (3) incorporate patient-derived organoids to bridge translational gaps and address these issues. These strategies could mitigate the limitations posed by interspecies variability and technical artifacts, enabling further IUA biomarker discovery.

Study limitations and future directions

This study has three principal limitations. (1) The restricted sample size of the GSE224093 dataset (n = 14) limited the statistical power for moderate-effect genes (—log2FC— < 1.5), although this limitation was mitigated using FDR correction (p < 0.05) and external validation; (2) The modest validation cohort sizes (human: n = 5–8; mice: n = 5–8) provided adequate power (72–80%) for detecting large-effect biomarkers (Cohen’s d ≥ 1.0) but limited the sensitivity to moderate-effect targets (d =0.5–0.8), as evidenced by the wider confidence intervals in the effect-size estimates. The core findings (e.g., IGF1/WNT5A) demonstrated cross-species consistency; however, the small sample sizes precluded meaningful subgroup analyses and impact the generalizability of our findings to heterogeneous clinical populations. (3) Interspecies divergences—such as undetectable murine CCL14, opposing TUBB3/PTGDS regulation, and distinct inflammatory/fibrotic timelines—highlight the fundamental differences between surgical murine models and human IUA pathophysiology.

A tiered validation strategy is proposed to address these limitations. (1) Expanded multicenter cohorts (target n≥30/group) should be included, incorporating laser-capture microdissected endometrial compartments to enhance spatial resolution. (2) Humanized mouse models should be developed through the xenotransplantation of patient-derived endometrial stromal cells to bridge species gaps. (3) Single-cell multiomics (scRNA-seq + spatial proteomics) should be used with paired human/murine samples to determine the cell-type-specific expression patterns confounding bulk analyses. The CCL14 discrepancy should be addressed through long-read sequencing to confirm murine gene absence versus technical detection failure. These approaches will help with systematically addressing the current limitations and identifying clinically actionable biomarkers and targeted therapies for IUA.