Identification of a disulfidptosis-related genes signature for diagnostic and immune infiltration characteristics in endometriosis

Four GEO datasets were obtained for analysis. One analysis was showed( GSE7305 ). Volcano Plots and heatmaps were drawn to display DEGs (Fig.  2 A,B). GO and KEGG analyses were carried out with to investigate DEGs functions. The important GO-BP (biological process) terms were mostly associated with the development of urogenital system, protein zinc ion homeostasia, protein nitrosytation, organelle fission and so on. Based on GO-CC (cell component) enrichment, DEGs of the four datasets were significantly associated with cell junctions, basement membrane, collagen-containing extracellular matrix and spindle. DEGs were significantly related to GO-MF (molecular function) terms including Glycosaminoglycan binding, immune receptor activity, extracellular matrix structural constituent, and transcription coactivator activity ( GSE7305 was showed in Fig.  2 C). Using KEGG enrichment, these DEGs of four datasets were associated with Cell cycle, Cell adhesion molecules, Glutathione metabolism, IL-17 pathway, Human T-cell leukemia virus 1 infection, Human T-cell leukemia virus 1 infection, Estrogen signaling pathway, Cytokine-cytokine receptor interaction, Inflammatory bowel disease, Cholesterol metabolism, actin binding and so on ( GSE7305  was showed in Fig.  2 D). The findings indicate that disulfidptosis is closely related to Glutathione metabolism, immune response, and inflammatory factors, suggesting the significant role of disulfidptosis in EMs. Fig. 2 DEG detection. ( A ) Heatmap displays the expression levels of EMs-related DEGs in GSE7305 . ( B ) volcano plot showing significant DEGs of GSE7305 . ( C ) GO annotation displays enriched terms associated with EMs. ( D ) KEGG analysis suggests pathways enriched into EMs. GO gene ontology, BP biological process, CC cellular component, MF molecular function, KEGG Kyoto Encyclopedia of Genes and Genomes. DEG detection. ( A ) Heatmap displays the expression levels of EMs-related DEGs in GSE7305 . ( B ) volcano plot showing significant DEGs of GSE7305 . ( C ) GO annotation displays enriched terms associated with EMs. ( D ) KEGG analysis suggests pathways enriched into EMs. GO gene ontology, BP biological process, CC cellular component, MF molecular function, KEGG Kyoto Encyclopedia of Genes and Genomes. Gene Set Enrichment Analysis (GSEA) was carried out using C5.bp.v7.4.symbols.gmt reference gene set in order to explore mechanisms involved in EMs pathogenic mechanism(Fig.  3 ). Differences between individuals with EMs and normal controls were of significance, particularly in terms of B cell receptor pathway, B cell-mediated immunity, leukocyte mediated immunity, and humoral immune response. Notably, there was a significantly enhanced humoral immune response in individuals with EMs, which was shown to be closely associated with B-cell-related immunity. Fig. 3 GSEA on pathways in EMs and normal groups ( GSE7305 ). GSEA on pathways in EMs and normal groups ( GSE7305 ). For the 31 DRGs, their expression profiles in EMs versus healthy samples were analyzed. There were altogether 15 hub genes obtained through the intersection of 102,231 DEGs with 31 DRGs (Fig.  4 A), including OXSM, MYH10, RAC1, RPN1, IQGAP1, CD2AP, PDLIM1, GYS1, ACTN4, FLNB, TLN1, PSMD2, CYFIP1, WASF2, and ABI2. To evaluate immune infiltration levels among EMs patients, the CIBERSORT algorithm was employed to analyze immune cell percentages in EMs. Immune cell maps and correlation maps were drawn for 22 immune cell types (Fig.  4 D,F) in EMs samples. Therefore, EMs patients showed significantly decreased abundances of activated NK cells, T cells CD8, T cells follicular helper, T cells regulatory (Tregs), and NK cells resting; while increased abundances of B cells naïve, plasma cells, Macrophages M2, and T cells CD4 memory resting (Fig.  4 E). Among 15 differentially expressed DRGs (Fig.  4 A) in EMs, IQGAP1, ACTN4, FLNB, TLN1, and WASF2 exhibited significant overexpression. In addition, ABI2 was significantly overexpressed, while OXSM, MYH10, RAC1, RPN1, CD2AP, PDLIM1 and GYS1 were significantly down-regulated. PSMD2 and CYFIP1 were up-regulated in normal samples. Heatmaps and grouping box maps (Fig.  4 B,C) were used to visualize the expression patterns of these 15 genes. Clearly, among these key genes, OXSM showed a significant negative association with ABI2 but positive correlations with PSMD2, GYS1, and RPN1; MYH10 was negatively related to IQGAP1; RAC1 was negatively correlated with ABI2; while RPN1 demonstrated positive correlations with CYFIP1, PSMD2, GYS1, and PDLIM1. PDLIM1 showed a positively correlation with CYFIP1, whereas PSMD2 showed a positive correlation with GYS1. ACTN4 displayed a negative correlation with CYFIP1. In terms of immune cell analysis for the correlations of these 15 key genes, NK cells exhibited strong associations, indicating that unique immune cell infiltration in EMs was probably closely associated with disulfidptosis, while DRGs were likely to be vital for the regulation of immune microenvironment in EMs (Fig.  4 A,G,H). Fig. 4 DRGs levels and immune cell infiltrating degrees within EMs. ( A ) Overlap of genes between DEGs and DRGs. ( B ) Boxplot showing DRGs expression profiles. ( C ) Heatmap displaying DRGs expression. ( D ) Heatmap representing percentages of 22 infiltrating immune cells within EMs and healthy controls. ( E ) Boxplots exhibiting diverse immune infiltration degrees in EMs compared with normal samples. ( F ) Infiltrating immune cell percentages within EMs and healthy samples were compared. ( G ) Correlation analysis of 15 key genes. (H) Correlation analysis of 15 key genes and infiltrated immune cells. DRGs levels and immune cell infiltrating degrees within EMs. ( A ) Overlap of genes between DEGs and DRGs. ( B ) Boxplot showing DRGs expression profiles. ( C ) Heatmap displaying DRGs expression. ( D ) Heatmap representing percentages of 22 infiltrating immune cells within EMs and healthy controls. ( E ) Boxplots exhibiting diverse immune infiltration degrees in EMs compared with normal samples. ( F ) Infiltrating immune cell percentages within EMs and healthy samples were compared. ( G ) Correlation analysis of 15 key genes. (H) Correlation analysis of 15 key genes and infiltrated immune cells. Based on LASSO regression, random forest, SVM-REF, Boruta, and XGBoost algorithms based on 15 key genes, this study identified candidate genes and constructed the disulfidptosis-signature (Fig.  5 A–D). The standard error λ of the minimum distance in the Lasso regression model was determined to be 0.065. TLN1, CYFIP1, PDLIM1, RPN1, and IQGAP1 were selected as significant variables in this model. The use of the SVM-REFCV method for feature selection yielded two factors, IQGAP1 and PDLIM1. By adopting the Burota method for confirmation purposes, GYS1, FLNB, TLN1, CYFIP1, PSMD2, WASF2, ABI2, PDLIM1, CD2AP, IQGAP1, RPN1 and OXSM were found to be relevant genes. Variable importance analysis using XGBoost revealed that PDLIM1 had the highest importance score; while Random Forest Regressor indicated that five variables with the highest significance included PDLIM1, RPN1, PSMD2, CYFIP1, and GYS1. Finally, the intersection of these five machine learning algorithms obtained PDLIM1 as the hub gene for the construction of our model (Fig.  5 E). Fig. 5 Disulfidptosis-signature construction based on machine learning. ( A , B ) LASSO algorithm adopted for the construction of disulfidptosis-signatures. ( C ) RF algorithm applied in ranking genes according to their importance. ( D ) The Burota algorithm. Tentative is yellow, Rejected is red, Accepted is green, and Shadow is blue. ( E ) Venn diagram displaying overlap of possible genes obtained from five algorithms. Disulfidptosis-signature construction based on machine learning. ( A , B ) LASSO algorithm adopted for the construction of disulfidptosis-signatures. ( C ) RF algorithm applied in ranking genes according to their importance. ( D ) The Burota algorithm. Tentative is yellow, Rejected is red, Accepted is green, and Shadow is blue. ( E ) Venn diagram displaying overlap of possible genes obtained from five algorithms. The nomogram model was prepared to diagnose EMs. The calibration curve demonstrated minimal error between actual and predicted risks, indicating excellent prediction accuracy of the nomogram model (Fig.  6 A,B). External datasets GSE23339 , GSE6364 , and GSE51981 were utilized for further validation purposes, demonstrating the diagnostic value of hub genes with area under the curve (AUC) values being 0.889 for GSE23339 , 0.773 for GSE51981 , and 0.631 for GSE6364 (Fig.  6 C–E). The obtained findings highlight the substantial diagnostic significance of nomogram model and suggest that the gene can effectively distinguish between EMs and normal groups. Fig. 6 Diagnostic significance validated using hub genes. ( A ) A nomogram constructed for predicting EMs with hub gene. ( B ) Calibration curve plotted for evaluating the nomogram prediction value. ( C – E ) ROC curves of GSE23339 , GSE51981 , and GSE6364 . Diagnostic significance validated using hub genes. ( A ) A nomogram constructed for predicting EMs with hub gene. ( B ) Calibration curve plotted for evaluating the nomogram prediction value. ( C – E ) ROC curves of GSE23339 , GSE51981 , and GSE6364 . To validate reliability of our hub gene, samples were obtained from EMs patients ( n  = 3) and normal human endometrium ( n  = 3) to conduct qRT-PCR and IHC. As a result, PDLIM1 expression was significantly reduced in EMs at mRNA and protein levels (Fig.  7 A,B). Subsequently, we evaluated the effect of down-regulation of PDLIM1 on the proliferation of primary EMs cells by the colony formation assay. We found that cell proliferation decreased after PDLIM1 knockdown compared to the control group. (Fig.  7 C) Fig. 7 Results of RT-qPCR and IHC. ( A ) The results of RT-qPCR illustrated PDLIM1 expression in EMs tissues ( n  = 3) and healthy endometrium ( n  = 3). * p  < 0.05. ( B ) IHC analysis revealed PDLIM1 expression in EMs tissues and normal endometrium. ( C ) The Colony formation assay to determine the effect of PDLIM1 knockdown on the proliferation of EMs primary cells. Results of RT-qPCR and IHC. ( A ) The results of RT-qPCR illustrated PDLIM1 expression in EMs tissues ( n  = 3) and healthy endometrium ( n  = 3). * p  < 0.05. ( B ) IHC analysis revealed PDLIM1 expression in EMs tissues and normal endometrium. ( C ) The Colony formation assay to determine the effect of PDLIM1 knockdown on the proliferation of EMs primary cells.

To conclude, this study shows that DRGs are correlated with infiltrating immune cells in EMs. In addition, we have identified PDLIM1 as a unique gene related to disulfidoptosis. These findings hold significant potential for enhancing diagnostic assessments of EMs. Furthermore, our research reveals an unprecedented correlation between disulfidoptosis and EM development while suggesting a possible relationship to immunity. Therefore, it provies fresh perspectives on the underlying pathological mechanisms involved and proposes novel avenues for therapeutic approaches targeting EMs.

EMs, a benign disease, lacks effective markers and therapeutic drugs, posing significant challenges for women of childbearing age. The theory of retrograde menstruation put forward in 1921 by Sampson has been extensively studied and widely accepted as an explanation. In addition, alternative hypotheses including vascular/lymphatic metastasis and coelomic metaplasia are released for elucidating the EMs pathophysiology 31 . However, no single hypothesis comprehensively accounts for all pathological types. Furthermore, immune, genetic, endocrine, and environmental factors have been suggested to be related to the EMs pathogenesis. Disulfidptosis, which has been recently identified as the new cell death pattern, can be induced by disulfide bond accumulation. This results in the cytoskeleton collapse and later cellular demise, exhibiting a close association with disease progression 32 . PCD mechanisms, including apoptosis, necorsis or autophagy, is important for the modulation of disease progression and immune microenvironment. Pharmacological agents for inducing specific cell death pathways have demonstrated effectiveness in tumor treatment. While research on ferroptosis has reached significant maturity 33 , more attention needs to be paid to disulfidoptosis as a new cell death mode. Disulfidptosis is induced by redox imbalance related to glucose and amino acid metabolic disturbance. These two metabolisms play vital roles in biological processes including immune evasion, metastasis, or drug resistance 34 , 35 . Disulfidptosis, as the novel PCD form, is potentially correlated with tumor immune response. The precise mechanism underlying disulfidptosis and the regulation of diverse disorders, along with candidate pathways, need to be further explored. Since the emergence of disulfidptosis in 2023, studies have been carried out to explore the intricate associations between hepatocellular carcinoma, pancreatic cancer, Alzheimer’s disease, and disulfidptosis 36 – 38 . The pathogenesis of disulfidptosis and EMs has been linked for the first time in this study. Candidate hub genes and targets are explored through bioinformatics analysis. In our study, gene expression in EMs and healthy samples was explored using the GEO database. GO and KEGG analyses on differential genes from GSE7305 showed that the cells were mostly associated with the immune receptor activity, positive regulation of Glycosaminoglycan binding, extracellular matrix structural constituent, transcription coactivator activity, Cell adhesion molecules, Cell cycle, Glutathione metabolism, IL-17 pathway, Human T-cell leukemia virus 1 infection, Human T-cell leukemia virus 1 infection, Estrogen pathway, Cytokine-cytokine receptor interaction, Inflammatory bowel disease, Cholesterol metabolism, actin binding and so on. These findings indicate that the immune microenvironment in patients with EMs significantly deviates from that of healthy individuals. In addition, we conducted further analysis of the dataset using GSEA, which revealed that patients with EMs exhibited enhanced humoral immunity, particularly in relation to B-cell receptor-related pathways. Our study suggested aberrant glucose metabolism in patients with EMs, prompting us to postulate a potential association between EMs and disulfidoptosis. Based on the obtained results, we proceeded to investigate DRGs of EMs. Subsequently, 15 DRGs associated with EMs were found by selecting EMs DEGs and 31 known disulfidptosis genes. The 15 key genes are OXSM, MYH10, RAC1, RPN1, IQGAP1, CD2AP, PDLIM1, GYS1, ACTN4, FLNB, TLN1, PSMD2, CYFIP1, WASF2, and ABI2. Immune cell infiltration analysis was performed, showing significant changes in immune cell populations between normal individuals and those with EMs. Notably, NK cells exhibited the most significant changes, consistent with previous studies 39 . Moreover, they were also significant differences in B cells naive, which was also in consistence with GSEA analysis. Through correlation analysis of 15 hub genes with immune cells, it can be observed that these key genes have a significant relationship to NK cells. These results suggest that differences in EMs immune cells are probably to be closely associated with the regulation of disulfidoptosis. To further identify hub genes with significant predictive effects among key genes, five machine learning algorithms were used. Finally, PDLIM1 was identified as the hub gene. The prediction effect was significantly improved by incorporating three external datasets for verification. PDLIM1 is an actinin-associated LIM protein (ALP) subfamily member of PDZ-LIM protein family 40 . Also referred to as CLP36, Elfin or CLIM1, PDLIM1 makes a vital impact on cytoskeleton organization, neuronal signaling, and organ development by interaction with a variety of proteins like paladin, α-actinin, EGFR and FHL1 41 . PDLIM1 exhibits the wide distribution in cardiac, pulmonary, hepatic, and splenic tissues. Through its interactions with diverse proteins in distinct tissue types, PDLIM1 is actively involved in the regulation of diverse signaling pathways while demonstrating tissue-specific functionalities. PDLIM1 is the cytoskeleton-related protein that exerts a crucial role in governing actin cytoskeleton organization. PDLIM1 is related to EMT in different cancers. It has shown significant tissue-specific effects in cancer. In addition, it is found to suppress CRC 42 and HCC 43 metastasis, as well as enhance breast cancer 43 , CML 44 and other cancer occurrence. Currently, PDLIM1 has been poorly investigated in gynecological diseases, with only one study in ovarian cancer, which shows that for OC patients, their response to anti-PDLIM1 autoantibody exhibit a positive relation to PDLIM1 up-regulation in OC samples. This reveals that anti-PDLIM1 autoantibody can serve as the complementary measure for CA125, which can thereby improve OC detection power. However, studies in EMs are lacking at present. Although our prediction model suggests that PDLIM1 demonstrates a relatively robust diagnostic efficacy, this study also utilized immunohistochemistry for the evaluation of PDLIM1 protein expression, demonstration significant differences between endometrial malignancies and normal endometrial tissues. Moreover, we showed through cell experiments that after PDLIM1 knockdown, the proliferation ability of primary EMs cells decreased significantly. However, due to the limited sample size, it is necessary to perform further comprehensive investigations. In addition, the mechanism of PDLIM1 affecting EMs through disulfidoptosis still needs to be further explored, which is also our further research direction. Furthermore, our study emphasizes the potential correlation between the immune microenvironment of EMs and disulfidoptosis, suggesting that the investigation of PDLIM1 in the context of endoheterogeneity immunity could open up new avenues for understanding the pathogenesis and progression of EMs. At the forefront of biomedical research, ML (Machine Learning) and deep learning(DL) algorithms, with their remarkable capabilities, have the potential to help doctors predict individualized treatment options for specific patients. Traditional ML methods are prone to overfitting, especially when dealing with high-dimensional omics data. This led to an understanding that the complexity of the model should be carefully managed to prevent such overfitting. However, recent theoretical advances are challenging this view, especially in the field of DL. Specifically, DL algorithm has inherent regularization characteristics in the process of backpropagation learning. Interestingly, these features actually reduce the risk of overfitting as the network scales, and contrary to what one might expect, adding more nodes or layers can make the model more robust 45 – 47 . While paying attention to this important problem, our team will make further efforts to develop more advanced algorithms.

Endometriosis (EMs) is the prevalent gynecological disease, commonly occurring in 10-15% of reproductive-age females 1 , 2 . EMs gives rise to dysmenorrhea or chronic pelvic pain and infertility, significantly affecting women’s health and overall quality of life. Characterized by extensive and diverse lesions, EMs exhibits an extremely invasive and recurrent nature, usually referred to as both a “benign cancer” and a “refractory disease”. Surgical treatment is still the diagnostic gold standard. However, EMs has been currently identified as the systemic condition not restricted to pelvic diseases only 3 . Recently, EMs has been identified to be closely associated with autoimmunity and inflammation 4 , 5 . In recent years, with advancements in immunology research, significant changes have been observed in humoral and cellular immune functions among EM patients 6 – 8 . Due to its unknown pathogenic mechanism, non-specific symptoms, and absence of non-invasive indicators for detection, EMs is usually not diagnosed in a timely manner. Therefore, it is crucial to explore the mechanism and early diagnosis of EMs. Various cell death patterns, including pyroptosis, apoptosis, cuproptosis or ferroptosis during EMs are widely explored to discover markers that can be used to predict prognosis and treat EMs 9 – 11 . Regulated cell death (RCD) provides a vital idea to make the investigation. In March 2023 12 , disulfide molecules accumulate at a large amount in glucose-deficient SLC7A11-high cancer cells, which causes aberrant disulfide bonding among actin cytoskeleton proteins, and interference with the tissue, finally resulting in actin network breakdown or even the novel cell death pattern known as disulfidptosis. The significance of SLC7A11 and glucose metabolism in the pathogenesis and progression of EMs has been demonstrated by a number of studies 13 , 14 . Thus, disulfidptosis may be related to EMs, even though more studies are needed to explore the underlying mechanism. Gene Expression Omnibus (GEO) database was used to investigate the possible mechanisms of EMs by examining differentially expressed genes (DEGs) in EMs compared with healthy samples. Cross-referencing analysis was also performed in DEGs with disultioptose-related genes (DRGs), aiming to identify differential DRGs and evaluate their correlation with immune cells. Then, diverse machine learning methods were employed to identify hub genes and construct predictive models. Later, the modal map and two external datasets were used to validate model performance, providing novel insights into understanding molecular mechanisms associated with EMs pathogenic mechanism.

Four raw datasets ( GSE7305 15 , GSE23339 16 , GSE6364 17 , and GSE51981 18 ) were obtained from the GEO database. GSE51981 included 77 EMs samples and 34 normal controls (without other common uterine/pelvic pathologies). GSE6364 included 21 EMs and 16 normal control samples, whereas GSE7305 involved 10 EMs and 10 normal control samples. GSE23339 dataset contains endometrial samples from 9 individual women and 10 EMs samples. The “remove Batch Effect” function was used to eliminate batch effects with R package “limma”. In addition, “normalize Between Arrays” function was applied in data standardization. Figure  1 shows the flowchart of analysis. Fig. 1 The study flow chart. The study flow chart. Differential analysis was performed with R package “limma” 19 . DEGs were screened in EMs compared with healthy samples upon thresholds of Log2| FoldChange | (Log|FC|)  1, and P  < 0.05. Then, heatmaps and volcano plots were drawn to show differential gene expression. Gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment was conducted using R package “clusterProfiler” 20 to explore gene biological functions. Based on Liu et al. 12 , 21 , 31 DRGs were obtained. Afterwards, DEGs ( GSE7305 , P  < 0.05) were crossed with DRGs to acquire 15 differentially expressed DRGs. The 22 kinds of lymphocytes were analyzed for their levels in EC and EU samples using CIBERSORT algorithm and immune cell LM22 gene set 22 . Immune cell infiltrating levels were analyzed according to expression matrix of every sample 23 . GSEA was applied in single-sample gene set enrichment analysis (ssGSEA) 24 . Spearman correlation was performed to visualize the correlation of these 15 genes. The association of immune infiltration with DRGs was analyzed with R packages dplyr, stringr, ggpubr, ggplot2, igraph, ggraph, corrr, corrplot, tidyverse, and reshape2 25 . In this study, five machine learning approaches were used for screening candidate diagnostic genes of EMs. As a regression approach, least absolute shrinkage and selection operator (LASSO) contributes to improving prediction performance and selecting key variables with regularization technologies 26 . SVM has been developed as the classification or regression supervised machine learning approach, requiring a training set with labels (16). SVM-RFE is another machine learning method that trains features of diverse categories for shrinking feature set and finding the most significant features 27 . Random forest (RF) is an excellent method to provide stable prediction and predict continuous variables 27 . Machine learning technique Boruta is used for feature selection to screen out every feature set associated with dependent variables 28 . The XGBoost prediction model can be used to further improve the model prediction performance 29 . Genes intersected among LASSO regression, SVM, RF, Boruta and XGBoost prediction model analyses were considered as central diagnostic hub genes of EMs. The R version was 4.2.3, and the main package version glmnet involved was 4.1.8. Subsequently, ‘RMS’ package (version 6.2-0) used to create the biomarker-based nomogram, aiming to help evaluate EMs occurrence. Then, calibration curves were drawn to evaluate the nomogram prediction accuracy. ROC analysis was conducted using three external datasets GSE23339 , GSE6364 , GSE51981 to demonstrating the significance of our prediction model in diagnosing the disease in EMs relative to normal groups. Using TRIzol reagent, total RNAs in EMs and normal control tissues were extracted in line with specific protocols. Then, RNA 1 mg, HifairIII 1st Strand SuperMix (YEASEN,11137ES60) and Hieff qPCR SYBR Green Master Mix (YEASEN,11201ES60) were adopted for reverse transcription. This reaction was found in Bio-Rad CFX96 PCR System. The 2-11Ct method was adopted for data normalization, and β-actin served as the endogenous control. Primers who were adopted for recognizing bisulfite-modified regions (-6 to -290) in PDLIM1 promoter included 5′-GTAGAGTTGTTGAGAGTTTTGG-3′ (forward, F), 5′-AATCTATCTACTAAATAATCATAACAC-3′ (reverse, R); β-actin: 5′-CCCTGGAGAAGAGCTACGAG-3′ (F), 5′-GGAAGGAAGGCTGGAAGAGT-3′ (R). Tissue samples were obtained during laparoscopic operations and washed by PBS to remove blood and impurities, followed by rapid freezing and storage. Afterwards, to prepare tissue blocks, 4% paraformaldehyde was added to fix the sample prior to paraffin embedding. Then, the sections were incubated with primary antibody PDLIM1/CLP36 Polyclonal antibody (11674-1-AP,1:500) overnight under 4 °C. After being incubated for another one hour, using Dako REAL EnVision Detection System (K5007; Dako, Jena, Germany), sections were washed by PBS, and treated with DAB + chromogen buffer (K5007, Dako) to visualize immunostaining. After counterstaining, images were obtained under the optical microscope (Nikon, DS-U3). The ectopic lesion tissues of endometriosis patients were taken, and the blood and mucus on the tissues were washed with PBS according to the method described in previous study 30 , and then cut into small pieces ina sterile petri dish (1 mm3). 0.3% collagenase IV was added to the tissue pieces (Solarbio, Beijing, China). Digest at 37℃ for 20 min. Equal-volume complete medium DMEM/F12 (Procell, Wuhan, China) was added to collect cell suspension and terminate enzyme digestion. Cell suspensions were filtered through 100 μm cell filter (Corning, New York, NY, USA), cell suspensions were centrifuged to collect cell precipitates, and cells were cultured at a temperature of 37 °C, a CO2 concentration of 5%, and a humidity level of 95%, under conditions using DMEM/F12 medium. SiRNA was synthesized by Beijing Syngentech Co., LTD. Transfection procedures were performed using Lipofectamine 3000 (Invitrogen, USA), as per manufacturer’s instructions. The target sequence is as follows: Si-PDLIM1-1 target sequence: GTCATCACAAACCAGTACA, Si-PDLIM1-2: AGGAGAAGCAAGAGTTAAA and Si-PDLIM1-3: GTGGCATCAACCTGAAACA. In the colony formation assay, EMs stromal cells were inoculated in a 6-well plate at a density of 1 × 103 cells/well. The plates were then incubated at 37 °C and 5% CO2 for 14 days. The colonies were fixed with 4% paraformaldehyde and stained with 0.1% crystal violet. Count the number of colonies in each well, each containing more than 50 cells. The experiment was repeated three times. Data were expressed as mean standard deviation (SD). Bioinformatics analyses were performed with R software (v.4.2.3). GraphPad Prism analysis software was used in statistical analysis. P  < 0.05 represented statistical significance.