Identification of GCNT3 as a glycometabolism-associated biomarker in endometrial cancer Authors: Da Ke, Wenzhe Li, Jun Xu, Xian He, Ya Wang, Jie Tan, Yaling Sun Published: 2026 ABSTRACT Endometrial carcinoma (EC) incidence is increasing, with diabetes mellitus (DM) elevating EC risk. This study investigates the glycometabolism-associated gene GCNT3 in EC. We systematically integrated pancreatic tissue DM datasets (GSE25724, GSE76896, and GSE95849) with RNA sequencing data from the TCGA-UCEC cohort. Our analytical strategy incorporated a comprehensive bioinformatics workflow, primarily including differential gene expression profiling, survival outcome modeling, functional enrichment analysis, and detailed immune infiltration assessment. Three machine-learning algorithms, including LASSO regression, support vector machine-recursive feature elimination (SVM-RFE), and Random Forest, were applied for feature selection. To reduce overestimation from using the same discovery cohort alone, the EC cohort was randomly divided into a training set and a test set at a ratio of 7:3. Feature selection was performed in the training set, with 5-fold cross-validation used for LASSO and SVM-RFE, and model discrimination and decision-curve performance were subsequently evaluated in the independent test set. Finally, clinical validation was performed by immunohistochemical examination of 80 EC tissues and 40 histologically confirmed normal endometrial control tissues to validate GCNT3 expression differences and clarify its potential biological significance in EC. Molecular docking studies were then conducted to explore potential binding interactions between GCNT3 and the selected candidate drugs Afatinib and Selumetinib. GCNT3 was upregulated in EC. In unadjusted Kaplan-Meier analysis, higher GCNT3 expression was associated with improved overall survival and disease-free survival. In multivariable Cox regression analysis adjusting for age, tumor grade, and tumor stage, GCNT3 remained an independent prognostic factor in EC. High GCNT3 expression was also associated with lower tumor grade and earlier stage, together with distinct immune infiltration patterns. High GCNT3 expression was associated with lower predicted IC50 values for Afatinib and Selumetinib. Molecular docking suggested potential binding interactions between GCNT3 and these agents, supporting a possible association between GCNT3 expression and differential drug responsiveness. GCNT3 is a potential biomarker for EC prognosis and therapy, showing consistent associations with glycometabolic signatures and the immune microenvironment.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-52806-y. Introduction Endometrial cancer (EC), one of the most prevalent malignancies of the female reproductive system, has demonstrated a marked increase in both incidence and mortality rates worldwide in recent years. It has emerged as the most common gynecological malignancy in developed countries1. In Europe, EC ranks as the fourth most frequently diagnosed cancer among women2, with reported incidence rates ranging from 12.9 to 20.2 per 100,000 and mortality rates between 2.0 and 2.7 per 100,0003,4. Traditionally, EC has been classified histologically into estrogen-dependent (Type I) and non-estrogen-dependent (Type II) subtypes5. However, this classification system exhibits certain limitations in predicting therapeutic response and recurrence risk6. With rapid advancements in molecular biology, significant progress has been made in the molecular classification of EC7–9, prompting the World Health Organization to recommend transitioning from histological to molecular classification systems10. The application of molecular classification in EC provides a robust foundation for personalized treatment strategies and precise prognostic prediction, thereby substantially improving clinical outcomes and overall survival (OS) among EC patients. Diabetes mellitus (DM) is a chronic metabolic disorder characterized by persistent hyperglycemia. Epidemiological studies have demonstrated a substantial increase in both the prevalence and mortality of DM over recent decades, establishing it as one of the leading global causes of death, currently ranked as the ninth most common mortality factor worldwide11–13. Notably, patients with DM exhibit significantly higher incidence and mortality rates of malignant tumors compared to the general population14,15, with this association being particularly pronounced in EC cases16–18. Meta-analytical data reveal that DM patients face 1.65-fold (95% CI: 1.50–1.81) and 1.32-fold (95% CI: 1.13–1.55) increased risks of EC incidence and mortality, respectively, compared to non-diabetic individuals19, underscoring the significant impact of DM on both EC development and progression. The underlying mechanisms may involve multiple pathological pathways, including insulin resistance, hyperinsulinemia, and chronic low-grade inflammation20. However, the precise molecular mechanisms governing the DM-EC pathogenetic link remain incompletely understood and require further investigation. The rapid advancement of molecular medicine has created new opportunities to identify genes potentially linking metabolic dysregulation to EC biology. In cancer research, systems-level transcriptomic strategies have been widely used to identify biologically relevant gene modules and hub genes associated with tumor initiation and progression. For example, weighted gene co-expression network analysis (WGCNA) has been successfully applied to identify clinically relevant modules and hub genes in oral squamous cell carcinoma, thereby providing useful methodological context for hub-gene discovery in malignancies21. In this study, we employed an integrated bioinformatics strategy using pancreatic DM-related transcriptomic signatures as a discovery framework, followed by validation and functional characterization in EC datasets and clinical specimens. Rather than assuming that pancreatic transcriptomic alterations directly recapitulate molecular events within EC tumor cells, we used them to prioritize candidate genes potentially reflecting broader DM-associated metabolic and inflammatory perturbations. Through this approach, GCNT3 was identified as a glycometabolism-associated candidate biomarker for EC. Subsequent analyses were then conducted within EC-specific datasets to evaluate its expression pattern, prognostic relevance, immune associations, and potential therapeutic implications. Our findings suggest that GCNT3 may represent a biologically relevant intersection between metabolic dysregulation and EC progression, while the precise mechanistic relationship between DM and EC requires further dedicated investigation. Materials and methods Acquisition and preprocessing of DM-related genes Gene expression profiles associated with DM were systematically retrieved from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/) using “diabetes” as the primary search term. The search was restricted to datasets from Homo sapiens, with specific exclusion of studies focusing on methylation profiles, diabetic nephropathy, or non-pancreatic tissues to ensure tissue specificity. Importantly, pancreatic DM transcriptomic alterations were used here as a DM-associated discovery signature reflecting broader systemic metabolic and inflammatory perturbations in diabetes, rather than as direct surrogates of glycometabolic reprogramming within EC tumor cells. This design was intended to prioritize candidate genes for subsequent EC-specific validation and characterization. Following these stringent selection criteria, three pancreatic tissue-derived datasets (GSE25724, GSE76896, and GSE95849) were identified and subsequently integrated, generating a consolidated expression matrix encompassing 197 samples (67 DM cases and 130 normal controls). To improve data comparability across different datasets, we performed batch effect correction. Principal component analysis (PCA) was used as a descriptive visualization tool to compare overall sample distribution patterns before and after correction. The post-correction PCA plot suggested reduced dataset-driven separation and improved comparability among samples, although PCA alone cannot formally verify preservation of biological signal. The study workflow is schematically presented in the Graphical Abstract. Identification of differentially expressed genes (DEGs) DEGs between diabetic and normal pancreatic tissues were identified using the “Limma” R package. Significant DEGs were defined as Benjamini–Hochberg adjusted P < 0.05 and |log2FC| ≥ 0.8 (approximately a 1.74-fold change). We selected this moderately stringent cutoff to avoid discarding reproducible, biologically meaningful yet moderate expression changes across the integrated cohorts after batch-effect correction; downstream multi-algorithm feature selection and independent validation in TCGA-UCEC further mitigated false discoveries. Machine learning for biomarker identification Machine learning algorithms offer distinct advantages in processing complex biological datasets and have been widely applied in bioinformatics research22. To systematically identify robust biomarkers associated with DM, we employed three complementary machine learning approaches including Least Absolute Shrinkage and Selection Operator regression (LASSO), Support Vector Machine-Recursive Feature Elimination (SVM-RFE), and Random Forest algorithm (RF)23–25. These methods were strategically selected to leverage their unique strengths in feature selection from large-scale gene expression data. LASSO regression provided effective dimensionality reduction through L1 regularization, SVM-RFE offered high-dimensional pattern recognition capability, while RF contributed robust feature importance rankings through ensemble learning. To ensure maximal reliability of our biomarker selection, we adopted a stringent consensus approach by identifying the intersecting genes from all three machine learning methods. This integrative strategy enhanced the biological relevance and statistical confidence of the candidate DM-associated genes identified through our analysis. To provide a more rigorous internal validation of the machine-learning pipeline, the EC cohort was randomly split into a training set (70%) and a test set (30%) prior to model construction. Feature selection was conducted using the training set. Specifically, LASSO regression was implemented with 5-fold cross-validation to determine the optimal penalty parameter, and SVM-RFE was also performed under a 5-fold cross-validation framework. Random Forest was used to rank candidate genes according to variable importance in the training set. After feature selection, the classification performance and clinical net benefit of the resulting models were assessed in the held-out test set using receiver operating characteristic (ROC) analysis and decision curve analysis (DCA), respectively. Nested cross-validation was not performed in the present study. Acquisition of EC-related genes and identification of candidate biomarkers RNA sequencing data were obtained from The TCGA database (https://portal.gdc.cancer.gov/), comprising 589 initial data files. Following quality control and removal of duplicated samples, the final dataset included 545 EC samples and 35 adjacent non-cancerous tissue samples for subsequent analyses. Clinical annotations encompassing age, tumor grade and staging information were systematically extracted and preserved to facilitate comprehensive correlative analyses. To further address the concern regarding the lack of independent validation, two publicly available GEO datasets containing endometrial tumor and non-tumor tissues, GSE63678 and GSE115810, were used as an external transcriptomic validation cohort. After probe annotation and expression matrix extraction, the two datasets were merged, and batch effects were corrected using the “sva” package in R. The merged validation cohort comprised 31 tumor tissues and 8 normal endometrial tissues. GCNT3 expression differences between tumor and normal groups were evaluated using the Wilcoxon rank-sum test, and the results were visualized using box plots. Construction of the prognostic model The survival prediction model was developed using multivariable Cox regression analysis. Based on the clinical variables available in the TCGA-UCEC cohort and their established prognostic relevance in EC, age, tumor grade, tumor stage, and GCNT3 expression were incorporated into the final model. Univariable Kaplan-Meier analysis was used to describe unadjusted survival associations, whereas multivariable Cox regression was used to evaluate whether GCNT3 retained prognostic significance after adjustment for major clinicopathological covariates. Based on the regression modeling strategies (RMS) framework, we constructed a comprehensive nomogram to predict 1-, 3-, and 5-year OS probabilities. This integrated scoring system assigns quantitative weights to each prognostic factor, enabling individualized risk assessment for EC patients through a visual, clinically applicable tool. Identification of GCNT3-associated DEGs and functional enrichment analyses EC samples were categorized into high and low expression groups according to median GCNT3 expression levels. Subsequent integrative bioinformatics investigations, incorporating Gene Ontology (GO) enrichment, Kyoto Encyclopedia of Genes and Genomes (KEGG, www.kegg.jp/kegg/kegg1.html) pathway analysis26–28, and Gene Set Enrichment Analysis (GSEA), were systematically conducted to delineate the functional characteristics of GCNT3-related gene expression profiles. Together, this multifaceted analytical framework facilitated characterization of the biological processes and pathways enriched among GCNT3-associated differentially expressed genes in EC. Construction of protein-protein interaction (PPI) network and gene enrichment analysis This study utilized the STRING database platform (https://string-db.org/) for protein association identification and PPI network construction. With a specified confidence score threshold of 0.400, we analyzed GCNT3-related genes and ultimately established a PPI network29. Additionally, we performed gene enrichment analysis through GO and KEGG analyses to evaluate the various biological processes in which these genes participate. These integrative bioinformatics approaches provided valuable insights into the molecular mechanisms underlying GCNT3’s role in EC. Single-cell transcriptomics analysis To elucidate the cell-type-specific expression patterns of GCNT3 in EC and its potential functional implications within the tumor microenvironment, we performed single-cell RNA sequencing analysis using the scCancerExplorer platform (https://bianlab.cn/scCancerExplorer/home)30. The GSE173682 single-cell dataset was selected to systematically investigate GCNT3 expression across diverse immune cell populations. This comprehensive analysis revealed distinct GCNT3 expression profiles and provided mechanistic insights into its role in modulating the tumor immune microenvironment. Genetic mutation and stemness index analysis To investigate the mutational profile of GCNT3 in EC, somatic mutation data were obtained from TCGA database. Mutation analysis was performed using maftools and dplyr packages to extract and characterize GCNT3-associated genetic alterations. Tumor stemness index (mRNAsi), a quantitative measure reflecting cellular regenerative and self-renewal capacity, was calculated using the machine learning-based One-Class Logistic Regression (OCLR) algorithm developed by Malta et al.31. Higher stemness indices correlate strongly with tumor dedifferentiation, a hallmark of aggressive malignancies. As tumors originate from monoclonal expansions of malignant cells, increased dedifferentiation capacity indicates poor cellular differentiation, which is a critical feature of high-grade, biologically aggressive cancers32,33. This analysis provides mechanistic insights into GCNT3’s potential role in EC tumorigenesis through both mutational landscape and stemness-associated pathways. Association between GCNT3 expression and immune checkpoints in EC Immune checkpoints are critical regulatory molecules in the immune system, primarily expressed on immune cells, which play essential roles in modulating the timing and intensity of immune responses34,35. To delineate the potential interplay between GCNT3 expression and immune checkpoint pathways, we focused our analysis on ten clinically relevant immune checkpoint genes, including CD274, CTLA4, HAVCR2, LAG3, PDCD1, PDCD1LG2, TIGIT, SIGLEC15, ITPRIPL1, and IGSF8. Expression profiles of these immune regulators were quantitatively analyzed for correlation patterns with GCNT3 expression levels. Furthermore, we employed TIMER 2.0 (http://timer.cistrome.org/) to systematically analyze the association between these immune checkpoint genes and GCNT3 expression, thereby elucidating the potential regulatory role of GCNT3 in immune checkpoint modulation within the tumor microenvironment. Analysis of GCNT3 expression and tumor-infiltrating lymphocyte levels (TILs) Immune cells play a pivotal role in tumor immune evasion and the development of therapy resistance. For instance, regulatory T cells (Tregs) suppress antitumor immune responses, thereby promoting tumor progression and angiogenesis, which facilitates immune surveillance escape by malignant cells36. To investigate the relationship between GCNT3 expression and TILs, we performed statistical evaluation using Spearman’s correlation analysis via the TIMER2.0 database (http://timer.cistrome.org/). This approach enabled systematic assessment of potential associations between GCNT3 and immune cell infiltration patterns. Association between GCNT3 and IPS Immunotherapy targeting immune checkpoints has emerged as an effective therapeutic strategy for malignant tumors. In our study, we obtained IPS data for EC from The Cancer Immunome Atlas (TCIA) database (https://tcia.at/home). Subsequently, we employed the TIDE algorithm to predict potential responses to immune checkpoint blockade (ICB) therapy37. This integrative approach enabled comprehensive evaluation of GCNT3’s potential role in modulating tumor immunogenicity and response to immunotherapy. Analysis of GCNT3 and drug sensitivity profiles To evaluate the clinical value of GCNT3 as a potential molecular biomarker, this study systematically analyzed the correlation between GCNT3 expression and drug sensitivity based on the TCGA database. The oncoPredict R package integrated with the GDSC2 database was employed, and the top five candidate drugs showing the strongest correlation with GCNT3 expression were identified38–40. Effect sizes for selected drug-sensitivity comparisons were additionally quantified using Cliff’s delta (Cliff’s d). To further interpret the associations observed in the drug sensitivity analysis at the structural level and assess potential interactions between the screened drugs and GCNT3, this study utilized molecular docking for exploratory evaluation of Afatinib and Selumetinib. Docking was performed using AutoDock version 1.2.7 with a global docking strategy41,42. Validation of glycometabolic reprogramming in EC To directly address whether EC exhibits aberrant glycometabolic activity, we evaluated tumor-versus-normal endometrial transcriptomic differences within TCGA-UCEC and quantified pathway activity using gene-set–based scoring. Specifically, glycolysis- and gluconeogenesis-related signatures and glycosylation-related pathways were assessed using correlation and enrichment analyses based on TCGA-UCEC expression data. These EC-specific analyses were used to validate the glycometabolism relevance of the candidate biomarkers identified from the DM-associated signature, and to avoid extrapolating EC metabolic phenotypes solely from pancreatic DM datasets. Clinical validation of GCNT3 expression in EC tissues For clinical validation, formalin-fixed paraffin-embedded tissue specimens were collected at Jingzhou First People’s Hospital between January 2022 and April 2025. The clinical validation cohort included 80 primary EC tissues and 40 normal endometrial control tissues. The EC cohort was intentionally expanded to 80 cases to support not only tumor-versus-normal protein expression comparison, but also exploratory stratification of EC tissues into high- and low-GCNT3-expression groups for follow-up-based prognostic assessment. The normal control cohort was primarily used for expression comparison and was constrained by the availability of eligible histologically confirmed normal endometrial tissues. Therefore, the IHC validation cohort was not designed as a strictly balanced case-control set. No formal a priori power calculation was performed. Eligible EC cases met the following inclusion criteria: (1) primary endometrial cancer confirmed by postoperative histopathology; (2) complete clinicopathological data available; and (3) adequate paraffin-embedded tissue blocks for immunohistochemical analysis. Cases were excluded if they had another concomitant malignant tumor, had received neoadjuvant therapy before surgery, represented recurrent disease, had insufficient tissue quality, or lacked key clinical information. The 40 control samples comprised a mixture of histologically confirmed normal endometrium adjacent to tumor and histologically confirmed normal endometrium obtained from benign hysterectomy specimens. Among the 80 EC and 40 control samples, 20 paired tumor-normal cases were included as a subset for paired analysis, whereas the overall unpaired comparison was performed using all 80 EC tissues and 40 control tissues. Available clinicopathological variables for the EC cohort included age, body mass index (BMI), histological type, tumor stage, tumor grade, ER status, and PR status (Supplementary Table S1). Immunohistochemical staining was performed on formalin-fixed, paraffin-embedded tissue sections. After deparaffinization in xylene and rehydration through graded ethanol, heat-mediated antigen retrieval was carried out in Tris-EDTA buffer (pH 9.0) using a pressure cooker for 20 min. Endogenous peroxidase activity was blocked with 3% hydrogen peroxide, followed by blocking of nonspecific binding. The sections were then incubated with a rabbit polyclonal anti-GCNT3 primary antibody (Solarbio, Cat. K115443P) at a dilution of 1:100 for 1 h at room temperature, followed by incubation with a ready-to-use polymer HRP-conjugated goat anti-rabbit secondary antibody (Proteintech, Cat. RGAR011). Immunoreactivity was visualized using DAB chromogen and counterstained with hematoxylin. For negative controls, the primary antibody was omitted and replaced with antibody diluent/PBS under otherwise identical conditions. GCNT3 expression was evaluated by both board-certified pathologists and quantitative digital analysis using ImageJ software (version 1.54, IHC Profiler), based on an established scoring system43,44. Staining intensity was graded and the percentage of positively stained cells was scored. Final expression scores were derived from the product of intensity and percentage scores45. Slides were independently reviewed in a blinded manner by two pathologists, and discrepant cases were resolved by consensus. For stratified analysis, GCNT3 protein expression was dichotomized into high- and low-expression groups using the median total score as the cutoff. Statistical analysis Statistical analyses were conducted using R software (version 4.4.1). Kaplan-Meier survival curves were generated and compared between groups using Cox proportional hazards regression models implemented in the “survival” and “survminer” packages. The proportional hazards assumption was assessed graphically, using log-minus-log survival plots for categorical covariates and residual plots against ranked survival time for continuous covariates (Supplementary Fig. 1). Cases with missing clinical data were automatically excluded from the corresponding regression analyses (complete-case analysis). Group comparisons were conducted using Wilcoxon rank-sum tests for two-group analyses and Kruskal-Wallis tests for comparisons involving three or more groups46. Data visualization was achieved through box plots created with the “ggplot2” package. Benjamini-Hochberg adjusted P values were applied in differential expression analysis, whereas the parallel immune-, pathway-, and drug-sensitivity-related association analyses were primarily exploratory and are reported using nominal P values. Statistical significance thresholds were set at P < 0.05*, P < 0.01**, and P < 0.001***. The study was supported by computational resources from the Sangerbox database (http://sangerbox.com/)47. Results Results of PCA and DEGs analysis Initial evaluation of DM-related gene expression datasets (GSE25724, GSE76896, and GSE95849) revealed significant batch effects across the different datasets. Following batch effect correction, PCA showed an altered sample distribution pattern with reduced dataset-driven separation in the post-correction plot, suggesting improved overall comparability among the integrated samples (Supplementary Fig. 2). Volcano plot visualization (Fig. 1A) identified 89 statistically significant DEGs, comprising 30 upregulated and 59 downregulated genes (Supplementary Table S2). Notably, compared with the healthy control group, patients with DM exhibited significantly increased expression of ID1, MARK1, and GCNT3, while the expression of KCNAB2 and SLC26A4 was significantly reduced. These keys differentially expressed genes (DEGs) provide a critical foundation for subsequent functional studies. Fig. 1Identification of prognostic biomarkers. (A) Volcano plot displaying differentially expressed genes, (DEGs) associated with DM. (B, C) Based on the Lasso regression algorithm, 19 genes corresponding to the lowest binominal deviation were identified as optimal biomarkers for DM diagnosis. (D) The top 20 genes were selected based on Support Vector Machine Recursive Feature Elimination, (SVM-RFE) with the highest accuracy for DM classification. (E, F) The top 30 genes were selected and ranked according to the importance scores derived from the random forest algorithm applied to DM. (G) Decision curve analysis of the three machine-learning models in the independent test set. (H) ROC curves of the LASSO, SVM-RFE, and Random Forest models evaluated in the independent test set. (I) Venn diagram of overlapping candidate genes. (J) Prognostic-Related genes screening. (K) Intersection analysis between DM-associated genes and endometrial carcinoma, (EC) prognostic markers. Identification of prognostic biomarkers. (A) Volcano plot displaying differentially expressed genes, (DEGs) associated with DM. (B, C) Based on the Lasso regression algorithm, 19 genes corresponding to the lowest binominal deviation were identified as optimal biomarkers for DM diagnosis. (D) The top 20 genes were selected based on Support Vector Machine Recursive Feature Elimination, (SVM-RFE) with the highest accuracy for DM classification. (E, F) The top 30 genes were selected and ranked according to the importance scores derived from the random forest algorithm applied to DM. (G) Decision curve analysis of the three machine-learning models in the independent test set. (H) ROC curves of the LASSO, SVM-RFE, and Random Forest models evaluated in the independent test set. (I) Venn diagram of overlapping candidate genes. (J) Prognostic-Related genes screening. (K) Intersection analysis between DM-associated genes and endometrial carcinoma, (EC) prognostic markers. Results of machine learning Based on these 89 DEGs, we systematically applied three complementary machine learning algorithms to identify potential biomarker candidates. The LASSO regression analysis identified 19 significant genes (Fig. 1B, C), while the SVM-RFE algorithm selected 47 genes with the lowest root mean square error (RMSE) values (Fig. 1D). The RF algorithm further ranked the top 30 genes according to their importance scores (Supplementary Table S3, Fig. 1E, F). To further compare the classification performance of the three algorithm-derived models, we evaluated ROC curves and DCA in the independent test set after feature selection had been completed in the training set (Fig. 1G, H). The ROC curves showed that all three models retained good discriminatory ability in the test set, with AUCs of 0.924 for the RF model, 0.919 for the LASSO model, and 0.894 for the SVM-RFE model. DCA further indicated that these models provided a positive net benefit across a clinically relevant range of threshold probabilities in the test set, supporting their potential predictive value. By integrating results from these three independent machine learning algorithms through intersection analysis, we ultimately identified nine overlapping candidate genes: SLC26A4, APOD, GREM1, ID1, CH25H, SLC2A2, RGS16, PON1, and GCNT3 (Fig. 1I). These consensus genes, rigorously selected through our multi-algorithm approach, represent highly reliable potential biomarkers for further investigation. Identification of candidate biomarkers To identify prognostic biomarkers, we analyzed EC transcriptomic data from TCGA using a multi-method approach. First, univariate Cox regression was performed to assess the relationship between gene expression and patient survival outcomes. This analysis was complemented by Kaplan-Meier survival curves with log-rank tests, where patient stratification was performed using both median gene expression levels (COX/KM Median) and optimal cutoff values determined by maximal selected rank statistics (COX/KM Optimal Cutoff Value). By intersecting results from these five analytical approaches (p < 0.05), we identified 1,425 genes significantly associated with prognosis (Supplementary Table S4, Fig. 1J). Further overlap analysis between these prognostic genes and the nine previously identified DM-related genes revealed GCNT3 as the most robust candidate biomarker (Fig. 1K), demonstrating consistent prognostic value and potential clinical relevance in EC. Pan-cancer expression profiling of GCNT3 As illustrated in Fig. 2A, comprehensive pan-cancer analysis revealed marked expression heterogeneity of GCNT3 across various malignancies. Comparative evaluation revealed differential GCNT3 expression in nine tumor types compared to their corresponding normal tissues, with statistically significant downregulation observed in five cancer types. Notably, GCNT3 showed specific upregulation in four malignancies: pancreatic adenocarcinoma (PAAD), stomach adenocarcinoma (STAD), breast cancer (BRCA), and uterine corpus endometrial carcinoma (UCEC). These findings suggest that GCNT3 may play tissue-specific regulatory roles in tumor development and progression. The distinct expression patterns of GCNT3 across different cancer types highlight its potential dual functionality in tumor biology, possibly acting as either an oncogene or tumor suppressor depending on the cellular context. The consistent upregulation observed in PAAD, STAD, BRCA, and UCEC particularly underscores its potential clinical relevance in these malignancies. Fig. 2Differential expression of GCNT3. (A) Analysis of GCNT3 expression across human malignancies. (B) Comparative evaluation of GCNT3 expression levels between EC specimens and matched normal endometrial tissues. (C) GCNT3 expression across EC histological grading. (D) GCNT3 expression across different EC clinical stages. (E) Them RNA expression levels of GCNT3 in EC and normal groups, (GSE63678 and GSE115810 datasets).(F, G) The EC group and the paracancerous tissues were subjected to comparative analysis through IHC scoring. (F) indicates an overall difference., (G) indicates compared differences.(H) IHC staining of GCNT3 in EC tissues, (upper: 100 μm, lower: 50 μm). (I) IHC staining of GCNT3 in normal endometrial tissues, (upper: 100 μm; lower: 50 μm). Differential expression of GCNT3. (A) Analysis of GCNT3 expression across human malignancies. (B) Comparative evaluation of GCNT3 expression levels between EC specimens and matched normal endometrial tissues. (C) GCNT3 expression across EC histological grading. (D) GCNT3 expression across different EC clinical stages. (E) Them RNA expression levels of GCNT3 in EC and normal groups, (GSE63678 and GSE115810 datasets).(F, G) The EC group and the paracancerous tissues were subjected to comparative analysis through IHC scoring. (F) indicates an overall difference., (G) indicates compared differences.(H) IHC staining of GCNT3 in EC tissues, (upper: 100 μm, lower: 50 μm). (I) IHC staining of GCNT3 in normal endometrial tissues, (upper: 100 μm; lower: 50 μm). Expression of GCNT3 in EC tissues To comprehensively evaluate the expression patterns of GCNT3 in EC, we analyzed 545 EC samples and 35 normal controls, with data distribution visualized through box plot analysis. The results demonstrated significantly elevated GCNT3 expression in EC tissues compared to normal controls, suggesting its potential functional importance in EC pathogenesis (Fig. 2B). Stratified analysis revealed distinct expression patterns across tumor grades and stages (Fig. 2C, D). GCNT3 expression showed significant variation among different histological grades, with both G1 and G2 tumors exhibiting markedly higher expression than normal tissues. Notably, expression levels progressively decreased with advancing tumor grade. Similarly, significant stage-dependent differences were observed, where stage III/IV tumors displayed significantly lower GCNT3 expression compared to stage I/II specimens. Independent external validation further confirmed that GCNT3 expression was significantly higher in endometrial cancer tissues than in normal controls (Fig. 2E), supporting the robustness of GCNT3 upregulation in EC. Clinical validation was subsequently performed in an independent tissue cohort consisting of 80 EC samples and 40 histologically confirmed normal endometrial controls. The control group included both tumor-adjacent normal endometrium and normal endometrium from benign hysterectomy specimens. In addition to the overall unpaired comparison (80 EC vs. 40 controls), a paired analysis was conducted in a subset of 20 matched tumor-normal cases derived from the same cohort. To corroborate the transcriptomic findings at the protein level, immunohistochemical validation was performed using clinical specimens. GCNT3 protein expression was significantly higher in EC tissues than in normal endometrial controls in the overall unpaired analysis (Fig. 2F-I). Consistently, paired analysis of the 20 matched tumor-normal cases also demonstrated higher GCNT3 expression in tumor tissues than in the corresponding normal endometrium (Supplementary Table S5 Fig. 2G). These findings collectively provide convergent transcriptomic and protein-level evidence supporting the upregulation of GCNT3 in EC. Prognostic evaluation of GCNT3 in EC To further investigate the prognostic value of GCNT3 in EC, Kaplan-Meier analyses were performed after stratifying patients into high- and low-expression groups by the median GCNT3 level (Fig. 3A, B). In unadjusted Kaplan-Meier analysis, high GCNT3 expression was associated with improved OS (HR = 0.515, 95% CI: 0.335–0.792; log-rank P = 0.00251) and improved DFS (HR = 0.528, 95% CI: 0.304–0.916; log-rank P = 0.023). Because these estimates were derived from unadjusted survival analyses, they were interpreted as descriptive associations rather than evidence of independent prognostic value. Fig. 3Clinical correlation and construction of prognostic model. (A, B) Kaplan-Meier survival analysis of GCNT3 in EC. (C) Sankey diagram linking GCNT3 to clinicopathological characteristics. (D) Univariate Cox regression analysis assessing prognostic factors. (E) Multivariate Cox regression analysis incorporating significant covariates. (F) Prognostic predictive model integrating GCNT3 expression with clinical parameters in EC patients. (G) Calibration curve evaluating the predictive accuracy of survival estimates. Clinical correlation and construction of prognostic model. (A, B) Kaplan-Meier survival analysis of GCNT3 in EC. (C) Sankey diagram linking GCNT3 to clinicopathological characteristics. (D) Univariate Cox regression analysis assessing prognostic factors. (E) Multivariate Cox regression analysis incorporating significant covariates. (F) Prognostic predictive model integrating GCNT3 expression with clinical parameters in EC patients. (G) Calibration curve evaluating the predictive accuracy of survival estimates. Establishment of a GCNT3-based prognostic model for EC The expression levels of GCNT3 showed significant correlations with multiple clinical characteristics in EC patients, establishing its value as a key prognostic biomarker. The Sankey diagram (Fig. 3C) effectively delineated the complex interplay between GCNT3 expression and clinical variables, including patient age, tumor stage, histological grade, and clinical outcomes, indicating bidirectional relationships between GCNT3 expression and disease progression. Univariate Cox regression analysis of DFS revealed a significant association between GCNT3 expression levels and patient prognosis (P = 0.017), with lower GCNT3 expression correlating with markedly increased recurrence risk (Fig. 3D). Subsequent multivariate analysis, adjusting for age, tumor grade, TNM stage, and GCNT3 expression, further confirmed GCNT3 as an independent prognostic factor in EC (Fig. 3E). The constructed nomogram (Fig. 3F) integrated these clinically relevant parameters, assigning weighted scores (0–100 points) to each variable, including GCNT3 expression. The total prognostic score, derived from the sum of individual component scores, enabled prediction of patient survival probabilities when mapped to the corresponding survival curves. Notably, GCNT3 expression contributed substantially to the risk stratification system, reinforcing its prognostic significance in EC. Furthermore, model validation through 1-, 3-, and 5-year survival probability analyses confirmed the robustness of this predictive tool (Fig. 3G). In summary, this comprehensive prognostic model combining GCNT3 expression with relevant clinical parameters provides a clinically applicable approach for predicting survival outcomes in EC patients, thereby offering valuable decision-support information for treatment planning and prognostic assessment in clinical practice. Functional enrichment analysis of GCNT3-associated pathways in EC We performed comprehensive functional characterization of GCNT3-associated molecular pathways in EC by stratifying 545 EC patients into high- and low-expression groups based on median GCNT3 levels. Comparative transcriptomic analysis revealed distinct gene expression patterns between the groups, as visualized in the volcano plot (Fig. 4A). To elucidate the functional significance of these DEGs under different GCNT3 expression levels in EC, we conducted systematic GO enrichment and KEGG pathway analyses to explore their biological functions (Fig. 4B). The top six enriched biological processes (BP) included cilium movement, axoneme assembly, microtubule bundle formation, axonemal dynein complex assembly, microtubule-based movement, and antimicrobial humoral response. For cellular component (CC) analysis, the six most significantly enriched terms were motile cilium, apical plasma membrane, apical part of cell, axoneme, ciliary plasm, and motile cilium. Notably, motile cilia are present in the luminal epithelial cells of endometrium, where they participate in physiological processes such as secretion from diffusely distributed glands and sensory functions48,49. Dysfunction of motile cilia may disrupt endometrial homeostasis, thereby potentially promoting EC development. The top six enriched molecular functions (MF) were G protein-coupled receptor binding, peptidase inhibitor activity, endopeptidase inhibitor activity, peptidase regulator activity, endopeptidase regulator activity, and enzyme inhibitor activity. In KEGG pathway analysis, the six most significantly enriched pathways were mineral absorption, glycosphingolipid, biosynthesis-lacto and neolacto series, viral protein interaction with cytokine and cytokine receptor, mucin type O-glycan biosynthesis, and IL-17 signaling pathway. Aberrant glycosphingolipid expression has been recognized as a hallmark of cancer cells50. Studies have found that sulfatide expression is significantly higher in well-differentiated endometrial adenocarcinomas compared to poorly differentiated ones. Increased expression of cholesterol sulfate on cell membranes may mask sulfated glycolipids (including sulfatides) and weaken their interaction with type I collagen, consequently leading to a poorly differentiated phenotype51. The IL-17 signaling pathway emerged as another critical element in EC pathogenesis, primarily through its activation of NF-κB and ERK signaling cascades. This activation subsequently alters the expression of key chemokines and cytokines within EC cells, ultimately reshaping the tumor microenvironment and indirectly influencing EC initiation and progression52. Collectively, these enrichment analyses indicate that genes differentially expressed between the GCNT3 high- and low-expression groups are enriched in biological processes related to ciliary function, membrane-associated signaling, and immune- or microenvironment-related pathways. These findings provide functional context for the transcriptomic differences associated with GCNT3 expression in EC, but do not by themselves demonstrate that GCNT3 directly regulates these processes. Fig. 4Volcano plot of gene expression and enrichment analysis. (A) Volcano plot visualization of DEGs based on GCNT3 expression levels. (B) GO and KEGG enrichment analyses identifying biological processes, cellular components, molecular functions, and signaling pathways associated with GCNT3. (C, D) GSEA-based enrichment analysis showing biological processes and pathways enriched in the GCNT3 high- versus low-expression groups in EC. Volcano plot of gene expression and enrichment analysis. (A) Volcano plot visualization of DEGs based on GCNT3 expression levels. (B) GO and KEGG enrichment analyses identifying biological processes, cellular components, molecular functions, and signaling pathways associated with GCNT3. (C, D) GSEA-based enrichment analysis showing biological processes and pathways enriched in the GCNT3 high- versus low-expression groups in EC. GSEA revealed the top ten most significantly enriched GO terms, including CENP-A-containing chromatin assembly, double-strand break repair via break-induced replication, flavonoid metabolic process, postsynaptic membrane assembly, radial spoke, U1 snRNP, aldose reductase (NADPH) activity, bile acid binding, ubiquitin protein transferase inhibitor activity, and UDP-galactose beta-N-acetylglucosamine beta-1,3-galactosyltransferase activity (Fig. 4C). These findings gain particular relevance considering existing evidence that in EC, defects in double-strand break (DSB) repair mechanisms frequently occur in microsatellite instability (MSI) tumors and may actively contribute to disease progression53. The top ten enriched KEGG pathways included COPI vesicle formation, antigen processing and presentation by Major histocompatibility complex (MHC) class II molecules, mutation-induced aberrant SOD1 in retrograde axonal transport, dephosphorylation of the kinetochore, oligomeric conformation of PrPC in anterogradely transported axons, DNA replication licensing, pre-initiation complex formation, origin unwinding and elongation, organization of the outer kinetochore, and translation initiation (Fig. 4D). Of particular translational significance, current research indicates that in EC, LC3-mediated suppression of the MHC class I antigen presentation pathway occurs through NLRC5 downregulation. Since NLRC5 represents a crucial regulator of immune evasion, strategies aimed at LC3 inhibition coupled with NLRC5 activation could potentially enhance immunotherapy efficacy in EC patients54. In summary, these comprehensive analyses demonstrate widespread gene expression differences between GCNT3 low- and high-expression EC subgroups. The diverse biological functions associated with GCNT3-related genes may underline the poor prognosis observed in EC patients with low GCNT3 expression, offering valuable insights for further investigation. PPI Network and gene enrichment analysis To comprehensively characterize the molecular interactions of GCNT3, we constructed a PPI network using the STRING database and subsequently analyzed the associated genes through GEPIA2.0. The network identified key interacting partners of GCNT3, including B3GNT6, C1GALT1, C1GALT1C1, GCNT1, GCNT4, MUC13, MUC2, ST3GAL1, ST3GAL2, and C1GALT1C1L. These protein nodes were visually distinguished by distinct colors and line styles, effectively illustrating GCNT3’s central position within the network and its potential regulatory influence on other biomolecules (Fig. 5A). The identification of these interacting proteins provides valuable insights into GCNT3’s potential role in the pathological processes of EC, thereby enhancing our understanding of the molecular pathways through which GCNT3 may contribute to EC progression. Fig. 5PPI network and single-cell transcriptomic profiling of GCNT3. (A) PPI network of GCNT3. (B, C) GO and KEGG enrichment analyses of PPI network-associated genes. (D, E) Single-cell transcriptomic profiling of GCNT3 expression patterns. PPI network and single-cell transcriptomic profiling of GCNT3. (A) PPI network of GCNT3. (B, C) GO and KEGG enrichment analyses of PPI network-associated genes. (D, E) Single-cell transcriptomic profiling of GCNT3 expression patterns. To further characterize the biological relevance of these genes, we performed GO and KEGG enrichment analyses. The top seven enriched KEGG pathways included gastric cancer, amoebiasis, metabolic pathways, mucin type O-glycan biosynthesis, glycosphingolipid biosynthesis, glycosaminoglycan biosynthesis, and glycosphingolipid biosynthesis (Fig. 5B). The BP analysis identified the top three enriched terms, including O-glycan processing, protein O-linked glycosylation, and protein glycosylation. Notably, O-linked glycosylation modification (O-GlcNAcylation) has been closely linked to EC. Clinical evidence demonstrates that elevated O-GlcNAcylation levels correlate positively with adverse prognostic factors including higher histological grade, advanced FIGO stage, and reduced survival outcomes in EC patients. At the molecular level, this phenomenon appears mediated through the tumor suppressor FBXO31, which normally regulates O-GlcNAcylation homeostasis by facilitating O-GlcNAc transferase (OGT) ubiquitination. In EC tissues, downregulation of FBXO31 leads to OGT stabilization and increased cellular O-GlcNAcylation levels, thereby driving tumor progression55. For CC analysis, the top three enriched terms were Golgi apparatus part, Golgi apparatus, and Golgi membrane. The most significantly enriched MF included catalytic activity acting on glycoproteins, transferase activity, and beta-1,3-galactosyl-O-glycosyl-glycoprotein beta-1,6-N-acetylglucosaminyltransferase activity (Fig. 5C). Collectively, these findings suggest that GCNT3 and its interacting genes may participate in multiple complex biological processes central to EC development, particularly those involving protein glycosylation and post-translational modification pathways. Single-cell transcriptomic analysis To further investigate the impact of GCNT3 on the tumor microenvironment, we performed high-resolution single-cell transcriptomic analysis. The results revealed remarkably low expression levels of GCNT3 across various immune cell populations within EC tumor tissues, suggesting that diminished GCNT3 activity in EC may compromise critical immune surveillance mechanisms, potentially facilitating tumor immune evasion and disease progression (Fig. 5D, E). These findings provide a plausible molecular explanation for the adverse clinical outcomes consistently observed in EC patients exhibiting low GCNT3 expression patterns. Analysis of genetic mutations and tumor stemness index Genetic mutations play a crucial role in the pathogenesis and clinical progression of EC. For instance, p53 gene mutations demonstrate a strong association with tumor aggressiveness, exhibiting significantly higher mutation rates in serous subtypes and potentially serving as an early carcinogenic risk factor56. Similarly, ARID1A mutations, frequently observed in endometrioid adenocarcinoma, contribute to early malignant transformation by disrupting chromatin remodeling processes57. In this study, we analyzed mutation profiles in EC using data obtained from the TCGA database and visualized the results in a waterfall plot (Fig. 6A). This investigation identified PTEN as the most frequently mutated gene across EC specimens. Regarding GCNT3 specifically, our data revealed a 3% mutation incidence in EC, with missense variants representing the principal mutational class. To further investigate the relationship between GCNT3 expression and tumor dedifferentiation, we stratified samples based on GCNT3 expression levels and assessed their association with tumor stemness index. The results revealed that the normal control group exhibited the lowest stemness index, while EC specimens with reduced GCNT3 expression showed a significant increase in stemness, suggesting a higher degree of tumor cell dedifferentiation (Fig. 6B). This phenotypic shift toward stem-like properties potentially underlies the emergence of more aggressive tumor behaviors, including enhanced invasive capacity, greater metastatic propensity, and acquired therapeutic resistance mechanisms. Fig. 6Genomic alterations and tumor stemness characteristics. (A) Comprehensive mutational landscape in the EC cohort. (B) Comparative analysis of stemness indices across GCNT3 expression groups. (C) Differential immune cell infiltration patterns between EC and normal endometrial tissues. (D) Quantitative assessment of immune cell proportions stratified by GCNT3 expression levels, revealing statistically significant variations in immune subset distribution. (E) Heatmap visualization of immune-related gene expression profiles in GCNT3 high and low expression groups. (F) The bar plot visualizes the proportion of tumor-infiltrating immune cells in different samples. Genomic alterations and tumor stemness characteristics. (A) Comprehensive mutational landscape in the EC cohort. (B) Comparative analysis of stemness indices across GCNT3 expression groups. (C) Differential immune cell infiltration patterns between EC and normal endometrial tissues. (D) Quantitative assessment of immune cell proportions stratified by GCNT3 expression levels, revealing statistically significant variations in immune subset distribution. (E) Heatmap visualization of immune-related gene expression profiles in GCNT3 high and low expression groups. (F) The bar plot visualizes the proportion of tumor-infiltrating immune cells in different samples. Analysis of immune infiltration and immune checkpoints In this study, we conducted an in-depth analysis of immune cell infiltration patterns in EC to elucidate the cellular composition and functional alterations of the immune system in this malignancy. Using the CIBERSORT algorithm, we performed a comprehensive comparison of immune cell proportions between EC samples and normal controls. The results revealed significant differences in the distribution of multiple immune cell types between the two groups, suggesting potential associations with immune responses and inflammatory processes in EC (Fig. 6C). Specifically, we observed substantially higher proportions of plasma cells, follicular helper T cells, regulatory T cells (Tregs), M0 macrophages, and M1 macrophages in EC tissues compared to normal controls. Conversely, the EC group exhibited markedly reduced proportions of naive B cells, CD8 + T cells, resting memory CD4 + T cells, activated NK cells, monocytes, M2 macrophages, resting dendritic cells, and resting mast cells. These findings highlight distinct immune microenvironment characteristics in EC that may contribute to disease progression and therapeutic responses. To systematically evaluate the relationship between GCNT3 expression and immune cell proportions, we stratified EC samples into high- and low-expression groups and compared immune infiltration patterns. The results demonstrated distinct immune profiles between the two groups (Fig. 6D). Several immune cell populations, including Tregs, M0 macrophages, CD8 + T cells, monocytes, and M2 macrophages, showed consistent differences across analyses. These reproducible findings support a robust association between GCNT3 expression and the immune composition of the EC microenvironment, although they do not by themselves establish a direct regulatory role. To complement these findings, we subsequently generated a heatmap to analyze gene expression patterns across various immune cell types (Fig. 6E) and visualized the distribution of different immune cell populations (Fig. 6F). These analyses confirmed significant differences in immune cell distribution patterns between GCNT3 expression subgroups, providing additional evidence for GCNT3’s crucial regulatory function in shaping the immune microenvironment of EC. Emerging evidence suggests that GCNT3 may contribute to EC immunomodulation through its regulatory effects on immune system function. Comprehensive analyses reveal a significant association between GCNT3 expression dynamics and distinct alterations in immune cell composition. Notably, tumors with reduced GCNT3 expression display enhanced infiltration of both cytotoxic T lymphocytes (CTLs) and follicular helper T cells (Tfh). Studies have revealed that compared to adjacent normal tissues, EC exhibits an increased total number of CD8 + T cells but significantly reduced cytotoxic activity, characterized by decreased expression of key effector molecules including granzyme A, granzyme B, and PD-1. Furthermore, secretions from endometrial adenocarcinoma have been shown to markedly inhibit both the cytotoxic killing capacity of CD8 + T cells in peripheral blood and the intracellular expression of perforin, granzyme B, and PD-1, suggesting that the tumor microenvironment suppresses CD8 + T cell cytotoxicity through the secretion of immunosuppressive cytokines58. Another study indicates that Tfh cell classification in EC facilitates precise checkpoint immunotherapy and provides prognostic stratification59. To further elucidate the precise molecular mechanisms underlying GCNT3’s involvement in tumor immune evasion, we conducted a comparative analysis of immune checkpoint-related gene expression between GCNT3-high and GCNT3-low expression groups. This analysis identified statistically significant differential expressions of multiple clinically relevant immune checkpoint regulators, most notably CTLA4, LAG3, SIGLEC15, PDCD1LG2, and ITPRIPL1, between the two groups (Supplementary Table S6 Fig. 7A). These findings provide additional compelling evidence for the crucial regulatory role of GCNT3 in shaping the immune microenvironment of EC. Taken together, these data suggest that GCNT3 expression is linked to immune-related features and immune evasion-associated phenotypes in EC. Fig. 7Immune checkpoint analysis and tumor microenvironment correlation. (A) Immune checkpoint gene expression profiles between GCNT3 high- and low-expression groups. (B-G) Scatter plots depicting correlations between GCNT3 expression levels and various tumor- infiltrating immune cell populations. Immune checkpoint analysis and tumor microenvironment correlation. (A) Immune checkpoint gene expression profiles between GCNT3 high- and low-expression groups. (B-G) Scatter plots depicting correlations between GCNT3 expression levels and various tumor- infiltrating immune cell populations. Intriguingly, GCNT3-deficient tumors show pronounced upregulation of immune checkpoint regulators such as LAG3 and PDCD1LG2. Existing research has found that over 50% of EC patients harbor mismatch repair (MMR) mutations, and these patients show poor response to current PD-1 checkpoint inhibitor monotherapy. However, combination therapies incorporating immune checkpoint inhibitors with other treatments may potentially enhance therapeutic efficacy60. Collectively, these findings position GCNT3 as a critical modulator of the EC immune microenvironment. Its low expression appears to promote tumor immune evasion by influencing immune cell infiltration and immune checkpoint molecule expression, ultimately affecting disease pathogenesis, clinical progression, and therapeutic responsiveness in EC. Correlation analysis between GCNT3 and immune cell infiltration Through systematic interrogation of the TISIDB database (Fig. 7B-G), we identified significant associations between GCNT3 expression patterns and distinct immune cell populations in EC. Our study specifically quantified infiltration densities of key immune cell populations, encompassing cancer-associated fibroblasts (CAFs), CD4 + Th1 T cells, naive CD8 + T cells, monocytes, and CD4 + Th2 T cells. Spearman’s correlation analysis revealed that GCNT3 expression showed a strong negative correlation with CAFs (Rho = −0.360, p = 5.75e-04), CD4 + Th1 T cell (Rho = −0.362, p = 5.20e-04), naive CD8 + T cells (Rho = −0.321, p = 2.27e-03), monocytes (Rho = −0.279, p = 8.41e-03), and CD4 + Th2 T cells (Rho = −0.224, p = 3.58e-02). These findings acquire particular relevance in light of established mechanisms whereby CAFs promote endometrial carcinogenesis through activation of the SDF-1/CXCR4 axis and IL-6/STAT-3/c-Myc signaling pathway, thereby driving EC cell proliferation and tumor growth61,62. Furthermore, existing clinical evidence indicates that elevated monocyte counts correlate with inferior disease-free survival in EC patients63, underscoring the potential clinical implications of our findings. The robust immunological patterns associated with GCNT3 expression establish a compelling rationale for its investigation as a novel immunomodulatory target in EC. The comprehensive immune profiling data presented here significantly enhanced our understanding of GCNT3’s role in shaping the immunological landscape of EC. Analysis of ICB and drug sensitivity Using the TIDE algorithm, we conducted an in-depth evaluation of predicted responsiveness to ICB therapy. As shown in Supplementary Fig. 3, the TIDE score was significantly higher in the GCNT3 low-expression group compared to the high-expression group (P < 0.0001), indicating that tumors with low GCNT3 expression exhibit stronger immune evasion potential and may demonstrate poorer responsiveness to ICB treatment. This finding suggests that GCNT3 expression levels could serve as a potential biomarker for predicting ICB therapeutic response in EC patients. Further analysis using the TIDE confusion matrix revealed distinct distributions of predicted responders and non-responders between the two groups. In the GCNT3 high-expression cohort, a greater proportion of patients were classified as potential ICB responders, whereas non-responders were more prevalent in the low-expression group. These results reinforce the association between GCNT3 expression and ICB treatment efficacy, highlighting its potential clinical utility in patient stratification for immunotherapy. To assess the potential clinical utility of GCNT3 in EC treatment, we investigated the correlation between GCNT3 expression levels and the half-maximal inhibitory concentration (IC50) of various antitumor agents. This study revealed through drug sensitivity analysis that the expression level of GCNT3 was significantly negatively correlated with the predicted IC50 values of five drugs (a lower IC50 indicates higher drug sensitivity) (P < 2 × 10⁻¹⁶, FDR < 2 × 10⁻¹⁶), suggesting that higher GCNT3 expression may be associated with increased tumor cell sensitivity to the corresponding drugs. Specifically, GCNT3 expression was significantly negatively correlated with the predicted IC50 values of the following drugs: Selumetinib_1736 (Spearman = −0.413, Cliff’s d = −0.423) (Fig. 8A), Afatinib_1032 (Spearman = −0.452, Cliff’s d = −0.462) (Fig. 8B), PF.4708671_1129 (Spearman = −0.429), GNE.317_1926 (Spearman = −0.402), and SCH772984_1564 (Spearman = −0.508) (Supplementary Fig. 3). These results support a statistically significant association between GCNT3 expression and predicted sensitivity to the aforementioned drugs, with Selumetinib and Afatinib additionally showing moderate effect sizes. Notably, among these drugs, Afatinib (NCT02491099) and Selumetinib (NCT01011933, NCT05554328) have already entered phase II clinical trials for endometrial cancer, providing preliminary validation of their therapeutic potential in this disease. This also indirectly suggests that GCNT3 may hold value for pharmacological research and clinical translation. Fig. 8Association of GCNT3 expression with predicted drug sensitivity and exploratory molecular docking analysis in EC. (A) Correlation between GCNT3 expression and the predicted IC50 of Selumetinib_1736 in endometrial cancer. (B) Correlation between GCNT3 expression and the predicted IC50 of Afatinib_1032 in endometrial cancer. (C) AlphaFold-predicted three-dimensional structure of the GCNT3 protein. (D) Chemical structures of the selected small-molecule candidate drugs used for exploratory molecular docking, including Selumetinib and Afatinib. (E) Predicted molecular docking modes of Selumetinib and Afatinib with GCNT3. Representative intermolecular interactions, including hydrogen bonds, hydrophobic interactions, salt bridges, and cation interactions, are shown. Association of GCNT3 expression with predicted drug sensitivity and exploratory molecular docking analysis in EC. (A) Correlation between GCNT3 expression and the predicted IC50 of Selumetinib_1736 in endometrial cancer. (B) Correlation between GCNT3 expression and the predicted IC50 of Afatinib_1032 in endometrial cancer. (C) AlphaFold-predicted three-dimensional structure of the GCNT3 protein. (D) Chemical structures of the selected small-molecule candidate drugs used for exploratory molecular docking, including Selumetinib and Afatinib. (E) Predicted molecular docking modes of Selumetinib and Afatinib with GCNT3. Representative intermolecular interactions, including hydrogen bonds, hydrophobic interactions, salt bridges, and cation interactions, are shown. Based on the results of the drug sensitivity analysis, this study performed molecular docking of Afatinib and Selumetinib with the GCNT3 protein to evaluate their potential binding modes. The three-dimensional structure of the GCNT3 protein was predicted using the AlphaFold model, with a predicted pLDDT score of 94.07 (Fig. 8C), indicating high model reliability64,65. The chemical structures of the two drugs are shown in Fig. 8D. Molecular docking results (Fig. 8E) revealed that the binding free energies of Afatinib and Selumetinib to GCNT3 were − 8.947 kcal/mol and − 7.315 kcal/mol, respectively, both lower than the commonly referenced threshold of −5 kcal/mol, suggesting that both drugs may have structural potential to bind to GCNT3. Further analysis of the binding modes indicated that although the docking sites were not at the known classical active sites, key intermolecular interactions such as hydrogen bonds were present at the binding interface. It should be noted that the primary aim of performing molecular docking in this study was to conduct a secondary computational assessment of the candidate drugs initially screened based on correlation analysis. The goal was to preliminarily evaluate, from a structural perspective, the likelihood of interaction between Afatinib, Selumetinib, and GCNT3, thereby providing a more focused hypothesis and target for subsequent experimental validation. The current results are computational predictions only, and their reliability requires further experimental verification. Association of GCNT3 with Glycometabolic pathways in EC To address glycometabolic pathway activity in EC more directly, we analyzed EC and adjacent normal endometrial tissues in the TCGA-UCEC cohort using gene-set–based pathway activity scoring and complemented this with correlation/enrichment analyses. These EC-internal analyses support that GCNT3 is linked to glycometabolic signatures. Tumor cells frequently exhibit dysregulated nutrient uptake and aberrant metabolic pathways, with these abnormal metabolic processes driving uncontrolled proliferation and metastatic spread, thereby constituting a fundamental characteristic of metabolic reprogramming in malignant tumors66,67. To investigate whether GCNT3 expression is associated with glycometabolic features in EC, we analyzed the relationship between GCNT3 expression levels and glucose metabolism-related pathways. The results demonstrated significant correlations between GCNT3 expression and multiple glucose metabolic pathways. Specifically, GCNT3 expression showed strong positive correlations with both glycolysis (r = 0.38, P = 1.23 × 10^−20) and gluconeogenesis pathways (Fig. 9A), suggesting that GCNT3 expression showed strong positive correlations with both glycolysis and gluconeogenesis pathways and consequently influencing tumor cell energy metabolism. Concurrently, GCNT3 expression was also significantly positively correlated with the glycosphingolipid biosynthesis-globo and isoglobo series pathway (r = 0.31, P = 2.24 × 10^−13) (Fig. 9B), indicating that GCNT3 expression was also significantly positively correlated with the glycosphingolipid biosynthesis-globo and isoglobo series pathway through modulation of glycosphingolipid synthesis. Furthermore, GCNT3 expression showed significant correlation with the mucin type O-glycan biosynthesis pathway (r = 0.39, P = 4.7 × 10^−21) (Fig. 9C), suggesting that GCNT3 expression was significantly correlated with the mucin type O-glycan biosynthesis pathways in EC cells. Additionally, GCNT3 expression was significantly correlated with the glycosphingolipid biosynthesis-lacto and neolacto series pathway (r = 0.46, P = 2.41 × 10^−29) (Fig. 9D) and the amino sugar and nucleotide sugar metabolism pathway (r = 0.33, P = 4.15 × 10^−29) (Fig. 9E). These findings support an association between GCNT3 expression and multiple glycometabolic signatures in EC, but do not establish direct regulation. Fig. 9Correlation of GCNT3 expression with glucose metabolic pathways in EC. (A) Correlation analysis of GCNT3 expression with glycolysis/gluconeogenesis pathway. (B) Correlation analysis of GCNT3 expression with glycosphingolipid biosynthesis pathway. (C) Correlation analysis of GCNT3 expression with mucin O-glycosylation pathway. (D) Correlation analysis of GCNT3 expression with lacto/neolacto series biosynthesis. (E) Correlation analysis of GCNT3 expression with amino sugar and nucleotide sugar metabolism. Correlation of GCNT3 expression with glucose metabolic pathways in EC. (A) Correlation analysis of GCNT3 expression with glycolysis/gluconeogenesis pathway. (B) Correlation analysis of GCNT3 expression with glycosphingolipid biosynthesis pathway. (C) Correlation analysis of GCNT3 expression with mucin O-glycosylation pathway. (D) Correlation analysis of GCNT3 expression with lacto/neolacto series biosynthesis. (E) Correlation analysis of GCNT3 expression with amino sugar and nucleotide sugar metabolism. These correlations indicate that GCNT3 expression is associated with multiple glycometabolic pathways in EC, including glycolysis/gluconeogenesis, glycosphingolipid biosynthesis, mucin-type O-glycan biosynthesis, and amino sugar and nucleotide sugar metabolism. Discussion EC represents one of the most common gynecologic malignancies, ranking prominently among cancers of the female reproductive system. Epidemiologic data reveal a rise in the age-standardized incidence rate (ASR) of EC from 8.67 per 100,000 population in 1990 to 9.99 per 100,000 in 201968. According to the latest statistics, an estimated 69,120 new cases of EC will be diagnosed in the United States in 2025, with an anticipated 13,860 EC-related deaths occurring in the same year69. Current treatment of EC primarily relies on a comprehensive approach combining surgical resection with postoperative radiotherapy and chemotherapy. However, the nonspecific clinical presentation of early-stage EC frequently results in delayed diagnosis, with many patients presenting at advanced disease stages. In such cases, even aggressive treatment combining surgery with chemoradiation often leads to unsatisfactory clinical outcomes, with a significant decline in five-year survival rates. Therefore, there is an urgent need to explore novel diagnostic techniques for early and precise detection, as well as to develop more effective therapeutic strategies for advanced-stage patients. Molecular classification of EC, based on genetic alterations, protein expression, and other tumor molecular characteristics, enables more accurate prediction of tumor biological behavior and treatment responsiveness, thereby facilitating personalized therapy. For instance, EC patients harboring POLEmut mutations demonstrate heightened sensitivity to immune checkpoint inhibition70,71, whereas those with the MMRd subtype are more likely to benefit from targeted immunotherapies72. Genetic testing of these molecular features not only provides clinicians with more precise prognostic assessments during diagnosis but also aids in formulating more optimal selection during therapy. This study aims to further elucidate potential therapeutic targets for EC through in-depth molecular mechanism research, offering new insights and approaches to overcome this malignancy. Through integrative analysis of pancreatic DM-associated transcriptomic signatures and the TCGA-UCEC cohort, followed by EC-specific validation analyses, we identified GCNT3 as a glycometabolism-associated gene that is upregulated in EC. In the TCGA-UCEC cohort, higher GCNT3 expression was associated with improved survival in unadjusted Kaplan-Meier analysis. Importantly, multivariable Cox regression further showed that GCNT3 retained prognostic significance after adjustment for major clinicopathological factors, including age, tumor grade, and tumor stage. Therefore, the favorable prognostic association of GCNT3 cannot be explained solely by differences in these conventional EC prognostic variables, although residual confounding cannot be completely excluded. Multi-dimensional analyses further linked low GCNT3 expression to increased tumor stemness, an immune-evasive microenvironment, and reduced predicted sensitivity to Afatinib and Selumetinib, which was further explored by molecular docking. Finally, immunohistochemistry in an independent clinical cohort confirmed elevated GCNT3 protein expression in EC compared with adjacent normal endometrium. GCNT3 demonstrates elevated expression across multiple tumor types and has been primarily characterized as an oncogenic factor in previous studies. In breast cancer and hepatocellular carcinoma, GCNT3 has been shown to drive tumor cell invasion via potent activation of the PI3K/AKT pathway73,74, while in pancreatic cancer, it facilitates tumorigenesis by modulating the β-catenin/MUC4 axis75. However, our current study reveals upregulated GCNT3 expression in EC tissues relative to adjacent normal tissues, while higher GCNT3 expression levels correlate with improved patient prognosis. This apparent discrepancy reflects the multifactorial nature of cancer prognosis. The differential expression pattern establishes GCNT3’s potential as a diagnostic biomarker for EC, without necessarily implying that higher expression predicts adverse clinical outcomes. Importantly, expression patterns do not always directly translate to prognostic significance. For example, CD27 is upregulated in tumor tissues, while it functions as a protective survival factor by enhancing antitumor immunity through T cell costimulation76. Similarly, our findings reveal that low GCNT3 expression group exhibits characteristic immunosuppressive features, including higher proportions of CD8 + T lymphocytes and follicular helper T cells, and upregulated immune checkpoint molecules like LAG3 and PDCD1LG2. These observations suggest that diminished GCNT3 expression may promote tumor immune evasion, whereas elevated GCNT3 levels could potentially enhance immune surveillance and inhibit tumor progression. Consequently, high GCNT3 expressions may foster an immunologically favorable tumor microenvironment by mitigating immunosuppressive mechanisms, thereby contributing to improved clinical outcomes. The dualistic behavior of GCNT3 can be understood through the lens of glycan structure and signaling. In normal tissues, GCNT3-mediated extension of O-glycans creates complex glycan “shields” on mucins that physically protect epithelial cells and influence cell-cell interactions. Loss of these extensions (e.g. GCNT3 downregulation) leaves simpler glycans (like the T or Tn antigens) exposed. These truncated glycans alter adhesion and can serve as ligands for lectins such as galectin-3 in the microenvironment. For example, lung adenocarcinoma cells with low GCNT3 show increased display of the unbranched Thomsen-Friedenreich antigen, which binds galectin-3 and helps recruit pro-tumoral myeloid cells to form a metastatic niche77. In contrast, when GCNT3 is expressed, it masks such antigens with further N-acetylglucosamine branching, reducing harmful cell–stromal interactions. In endometrial tissue, proper O-glycan branching may plausibly contribute to epithelial homeostasis, although this possibility was not directly examined in the present study. Furthermore, our analysis of tumor staging and histological differentiation reveals a significant correlation between GCNT3 and tumor differentiation status. Specifically, elevated GCNT3 expression is predominantly observed in early-stage EC, while diminished expression characterizes advanced-stage and poorly differentiated tumors. This expression pattern suggests that high GCNT3 levels may serve as an indicator of well differentiated and less aggressive tumor phenotypes. Supporting evidence comes from stemness index analyses, which demonstrate markedly higher stemness scores in the GCNT3-low expression cohort, indicative of enhanced cellular dedifferentiation and higher biological aggressiveness. From a clinical perspective, well differentiated neoplasms typically exhibit slower proliferation rates and increased treatment responsiveness, contributing to their more favorable prognosis. Conversely, the observed reduction in GCNT3 expression in dedifferentiated malignant cells may represent either passive downregulation during tumor progression or an active functional role of GCNT3 in maintaining cellular differentiation. Collectively, these findings demonstrate that declining GCNT3 expression mirrors the pathological transition from highly differentiated to poorly differentiated states and from low-grade to high-grade malignancy. Sustained high-level expression of GCNT3 appears to play a crucial role in maintaining cellular differentiation status and suppressing invasive characteristics, consistent with the improved clinical outcomes observed in EC patients with elevated GCNT3 expression. From a molecular and functional standpoint, prior studies in other tumor types have suggested that GCNT3 may be involved in glycosylation-related processes associated with drug response, cell adhesion, and invasive behavior. For example, studies in colorectal and ovarian cancer have reported that GCNT3 overexpression is associated with enhanced sensitivity to 5-fluorouracil and reduced proliferative or invasive phenotypes78. In addition, mechanistic studies from non-EC contexts have linked GCNT3-associated glycosylation changes with stabilization of E-cadherin and reduced β-catenin nuclear translocation79, However, these observations were not generated in EC models in the present study and therefore should be interpreted as biological context rather than as established mechanisms in EC. In our study, the EC-specific data support associations of GCNT3 expression with favorable clinicopathological features, glycometabolic signatures, and predicted drug responsiveness, but do not directly demonstrate that GCNT3 inhibits EMT or attenuates proliferative signaling pathways such as NF-κB in EC. Although GCNT3 demonstrates elevated expression in EC, its fundamental role appears to oppose malignant progression, providing a plausible biological rationale for the observed association between high GCNT3 expression and favorable clinical outcomes in this malignancy. Collectively, the biological function of GCNT3 is contingent upon the tissue microenvironment and tumor microenvironment: in contexts where mucin barrier maintenance, inflammatory suppression, and oncogenesis inhibition are operative, it exerts a protective tumor suppressor role. Future research is focusing on these mechanisms: for instance, glycosylation editing strategies are being explored to tip the balance in favor of the host. Therapeutically, inhibiting GCNT3 has shown promise in breaking down the tumor mucus shield and re-sensitizing cancers to chemotherapy and immune clearance80. Conversely, enhancing GCNT3 activity in at-risk tissues might bolster mucosal protection and reduce cancer incidence. Understanding the glyco-immunometabolic networks underlying GCNT3’s dual roles will be key to harnessing this enzyme for both cancer prevention and treatment. Despite the integrated multi-omics analyses, machine learning-based screening, external GEO validation, and immunohistochemical confirmation performed in this study, several limitations should still be acknowledged. First, although the addition of a merged independent GEO cohort and a clinical IHC cohort strengthens the reproducibility of the observed GCNT3 dysregulation, the present work remains predominantly based on retrospective bioinformatics and observational validation. Second, the mechanistic basis by which GCNT3 influences glycometabolic remodeling, immune microenvironmental alterations, and drug responsiveness in EC has not yet been directly examined through functional experiments. Therefore, in vitro studies using gain- and loss-of-function approaches, together with in vivo validation in appropriate animal models, will be required to clarify the causal role and downstream regulatory pathways of GCNT3 in EC. Third, regarding the machine learning-based screening, although an internal training/test split was used to reduce overestimation in model evaluation, nested cross-validation and external validation specifically for the classification models were not performed. Moreover, although multivariable Cox analysis was performed to adjust for major clinicopathological covariates, not all potentially relevant prognostic variables could be incorporated into the retrospective TCGA-based survival model. Therefore, residual confounding cannot be fully excluded. Furthermore, no formal a priori power calculation was performed for the IHC validation cohort. Although the enlarged EC cohort was originally intended to allow exploratory follow-up-based prognostic stratification according to GCNT3 protein expression, this analysis did not yield a significant result, likely due to limited sample size, incomplete follow-up in some cases, the predominance of early-stage surgically treated tumors, and the low number of recurrence or death events. In addition, because the initial discovery step was based on pancreatic DM transcriptomic datasets rather than EC tissues from patients with diabetes, the cross-tissue inference should be interpreted cautiously. The present design supports candidate-gene prioritization and hypothesis generation, but it does not establish a direct mechanistic bridge between DM and EC. Consequently, the reported model performance should be interpreted with appropriate caution, and future studies employing nested cross-validation, independent external validation cohorts, and more comprehensive hyperparameter tuning are warranted to further assess the robustness and generalizability of the machine-learning pipeline. Accordingly, the current findings should be interpreted as providing a robust biomarker-oriented and hypothesis-generating framework, rather than definitive mechanistic proof. Conclusion This study systematically evaluated the clinical relevance of the glycosylation-related gene GCNT3 in EC through integrative bioinformatics analyses, machine learning-based screening, external transcriptomic validation, and clinical immunohistochemical confirmation. Our findings indicate that GCNT3 is significantly upregulated in EC and is associated with favorable survival outcomes, lower tumor aggressiveness, distinct immune characteristics, and altered glycometabolic signatures. Immunologically, reduced GCNT3 expression correlates with an immunosuppressive tumor microenvironment, characterized by altered infiltration patterns of specific immune cell subsets and aberrant overexpression of immune checkpoint molecules including LAG3, PDCD1LG2, and ITPRIPL1, suggesting its involvement in immune evasion mechanisms. From a therapeutic perspective, GCNT3 expression was associated with predicted sensitivity to several candidate agents, among which Afatinib and Selumetinib were selected for exploratory molecular docking analysis. These findings support the potential value of GCNT3 in therapeutic stratification, although further experimental validation is required. Finally, our immunohistochemical validation of differential GCNT3 expression patterns in clinical EC specimens substantiates its potential value as both a diagnostic and prognostic biomarker for EC. In conclusion, this study identifies GCNT3 as a promising biomarker candidate in EC, showing associations with survival, glycometabolic signatures, immune-related features, and predicted drug responsiveness. However, the mechanistic basis of these associations requires dedicated functional validation. A comprehensive elucidation of GCNT3’s functional networks and its role in EC pathogenesis will offer novel insights into the complex molecular interplay between metabolic reprogramming and tumor development. Further elucidation of GCNT3-associated functional networks in EC may provide useful insight into how metabolic dysregulation-related signatures intersect with tumor progression, although the direct mechanistic relationship between diabetes and EC remains to be clarified. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2 Supplementary Material 3 Supplementary Material 3 Supplementary Material 4 Supplementary Material 4 Supplementary Material 5 Supplementary Material 5